This is not the first time NVIDIA has used this strategy: when the next change in the technological process or architecture of GPUs occurs, the top position in the new line of gaming video cards is occupied for some time by a device on a second-tier chip. The flagship GPU first appears as part of a prosumer accelerator under the TITAN brand, and then, with certain losses of computing units and functions, it reaches gaming models. However, the differences in technical characteristics between the next TITAN and the next gaming flagship on the same GPU are unique in each generation of video cards.
In the Kepler era, which gave birth to the first Titan four years ago, the GeForce GTX 780 and GTX TITAN were the least similar to each other. In addition to the significant difference in the set of computational units, the gaming video card received half as much RAM and was eight times slower in performing double-precision calculations (FP64). Based on the Maxwell architecture, NVIDIA no longer began to release a processor that simultaneously covers graphics tasks and double-precision calculations, so the GTX 980 Ti differs from the GeForce GTX TITAN X only in the amount of RAM and has almost equivalent performance in games.
With the arrival of Pascal, for the first time in its history, NVIDIA has divided its top-tier GPUs into professional and gaming categories. The transition to 16 nm FinFET technology made it possible to release a processor with an unprecedented number of computational units and a level of performance - GP100. However, the latter is used only in Quadro and Tesla accelerators, and the likelihood of seeing it under the GeForce brand tends to zero. The "big core" alternative, GP102, lacks architectural elements and features critical to computing tasks, but at the same time is not inferior to the GP100 in the throughput of FP32 operations, which underlie the shader code of 3D applications.
GP102 has already proven itself as part of the updated TITAN X, which, no less than a gaming video card, is a computing accelerator, the priority among which this time is the use of AI models created by deep learning (inference). But looking at the TITAN X, it wasn't so easy to predict how the GTX 1080 Ti would turn out.
The fact is that NVIDIA did not leave itself much room for maneuver between the GTX 1080 and the TITAN X. Based on peak FP32 throughput (which, of course, is not an exhaustive characteristic of the GPU), the TITAN X is only 24% superior to the GeForce GTX 1080 , while the difference between the GTX 980 and GeForce GTX TITAN X in this parameter was 35%. If in the Pascal generation the developers decided to significantly reduce the computing resources of TITAN X, NVIDIA would have a rather unconvincing flagship of the gaming line.
On the other hand, in TITAN X neither the frequency potential nor the set of functional blocks of GP102 are fully revealed, so the GeForce GTX 1080 Ti could well turn out to be faster than TITAN X, because NVIDIA once did exactly the same with the GeForce GTX 780 Ti and the first TITAN. As a result, something unexpected happened: the GeForce GTX 1080 Ti did indeed cut down the GPU units compared to the TITAN X, but not the units that are usually affected. However, the peak performance of the GTX 1080 Ti is, at least on paper, higher. Moreover, a gaming video card has other, not so obvious, advantages.
GPU
The GP102 crystal, produced at 16 nm FinFET at TSMC facilities, contains 12 billion transistors in an area of 471 mm2. The parameters of the next highest-ranking chip in the Pascal line, GP104, are much more modest - 7.2 billion transistors and 314 mm2.
However, the structure of the GP102 computational units is the same as that of the GP104 chip, taking into account quantitative changes at the highest level of the architecture, so all the additional transistor budget in the GP102 went to increase the speed in graphics and other calculations based on the FP32 data format. The standalone SM (Streaming Multiprocessor) in the GP102 and GP104 contains 128 CUDA cores for FP32 operations, four FP64-compatible cores, and one core capable of executing two FP16 instructions simultaneously. In addition, SM includes 40 texture mapping units, 96 KB of shared memory and 48 KB of L1 cache.
In total, the GP102 contains 30 SM blocks, which in total give a configuration of 3840 single-precision CUDA cores and 240 texture units. The back-end of the pipeline is represented by twelve 32-bit GDDR5/5X memory controllers (384-bit bus), each of which is associated with 8 ROPs and a 256 KB L2 cache section (3072 KB for the entire GPU).
Block diagram of NVIDIA GP102 in GeForce GTX 1080 Ti
Regarding the differences between the Pascal architecture and Maxwell, we recommend that you refer to our review of the GeForce GTX 1080. In this iteration, the developers developed the advantages of the previous generation and compensated for its inherent shortcomings. Let's briefly list the most significant changes:
- improved color compression with ratios up to 8:1;
- the Simultaneous Multi-Projection function of the PolyMorph Engine geometric engine, which allows you to create up to 16 projections of scene geometry in one pass (for VR and systems with multiple displays in the NVIDIA Surround configuration);
- the ability to interrupt (preemption) during the execution of a draw call (during rendering) and the command flow (during calculations), which, together with the dynamic distribution of GPU computing resources, provides full support for asynchronous computing (Async Compute) - an additional source of performance in games running the DirectX 12 API reduced latency in VR;
- display controller compatible with DisplayPort 1.3/1.4 and HDMI 2.b interfaces. High dynamic range (HDR) support;
- SLI bus with increased bandwidth.
Let's add to this list a technology that was the secret know-how of NVIDIA's latest architectures, but following an independent investigation, the results of which were spread on the Internet, the company revealed its cards. In addition to powerful color compression, Maxwell and Pascal architectures use a form of tile-based rendering to save memory bandwidth (BW).
Classic tile-based rendering, widely used in mobile GPUs, involves processing a frame in two passes. First, the driver divides the screen space into tiles (areas with a typical size of 16 × 16 or 32 × 32 pixels) and compiles an index of the polygons that are in the projection of each tile. Then, sequentially within each tile, the entire rendering procedure is performed - from transformation and intersection of polygons to filling textures and executing shaders - and the final result of all tiles is stitched into a single image. The advantage of this method is that any intermediate operations within a tile operate on a single data array, which is entirely placed in the GPU cache, and therefore the frequency of access to RAM is reduced.
However, the need for two passes to process the scene geometry itself consumes RAM bandwidth, since the GPU must first write information about the polygons that fall into a particular tile into external memory, and then retrieve it when rendering from tile to tile back. As a consequence, the effectiveness of tile-based rendering ultimately depends on whether the bandwidth savings in pixel fill rate outweigh the cost in two-pass geometry projection. In mobile applications with simple geometry, tile-based rendering is justified, but for modern desktop games, the standard method of instant rendering is better suited, in which sequential rasterization of one polygon after another takes place in a single screen space.
The Maxwell and Pascal pipeline combines the best features of both rendering methods. NVIDIA also uses tiles, but in this case, rendering frames, as in other desktop architectures, does not require a separate stage of sorting (binning) all polygons belonging to the frame between one or another tile. Instead, the GPU first performs the entire scene geometry transformation, and the rest of the rendering steps happen tile by tile. In this case, the size and number of tiles that are processed simultaneously are determined dynamically as the fixed-length geometry queue is filled, so that all data fits into the L2 cache.
⇡#Technical characteristics, prices
Like the TITAN X, the GTX 1080 Ti does not have a fully functional version of the GPU. The manufacturer has blocked two of the 30 SMs in the GPU of both video cards. Thus, the front-end of the chip and therefore the theoretical throughput in operations per clock in the GTX 1080 Ti did not suffer at all compared to the TITAN X. However, in the GTX 1080 Ti one of the 12 RAM controllers is disabled, and with it 8 ROPs and L2 cache section. As a result, the GPU memory bus has decreased from 384 to 352 bits and the video card is equipped with only 11 instead of 12 GB of GDDR5X.
However, the GTX 1080 Ti makes up for lost RAM bandwidth by increasing the bus frequency - from 10,000 to 11,000 Mbps per pin. In addition, NVIDIA raised the GPU frequencies: base by 63 MHz and Boost Clock by 51 MHz. Thus, the only parameter that differentiates the GTX 1080 Ti from the TITAN X in graphics rendering remains L2 volume.
What about computing tasks? Here, unlike games, there may be a situation where an additional gigabyte of RAM makes a difference. Another thing is surprising: NVIDIA did not at all limit the speed of the GTX 1080 Ti with various data formats. GP102 is already not suitable for “combat” calculations based on FP64 and FP16 data (execution speed at 1/32 and 1/64 of FP32), but both TITAN X and GTX 1080 Ti perform four times more int8 operations per clock cycle compared to FP32. Before the introduction of the GTX 1080 Ti, only the TITAN X in NVIDIA's consumer lineup supported int8, the instructions used to process data with deep learning models.
| Manufacturer | NVIDIA | ||||
| Model | GeForce GTX 980 Ti | GeForce GTX TITAN X | GeForce GTX 1080 | GeForce GTX 1080 Ti | TITAN X |
| GPU | |||||
| Name | GM200 | GM200 | GP104 | GP102 | GP102 |
| Microarchitecture | Maxwell | Maxwell | Pascal | Pascal | Pascal |
| Technical process, nm | 28 nm | 28 nm | 16 nm FinFET | 16 nm FinFET | 16 nm FinFET |
| Number of transistors, million | 8 000 | 8 000 | 7 200 | 12 000 | 12 000 |
| Clock frequency, MHz: Base Clock / Boost Clock | 1 000 / 1 076 | 1 000 / 1 089 | 1 607 / 1 733 | 1 480 / 1582 | 1 417 / 1531 |
| Number of shader ALUs | 2 816 | 3 072 | 2 560 | 3 584 | 3 584 |
| Number of texture mapping units | 176 | 192 | 160 | 224 | 224 |
| ROP number | 96 | 96 | 64 | 88 | 96 |
| RAM | |||||
| Bus width, bits | 384 | 384 | 256 | 352 | 384 |
| Chip type | GDDR5 SDRAM | GDDR5 SDRAM | GDDR5X SDRAM | GDDR5X SDRAM | GDDR5X SDRAM |
| Clock frequency, MHz (bandwidth per contact, Mbit/s) | 1 753 (7 012) | 1 753 (7 012) | 1 250 (10 000) | 1 376,25 (11 010) | 1 250 (10 000) |
| Volume, MB | 6 144 | 12 288 | 8 192 | 11 264 | 12 288 |
| I/O bus | PCI Express 3.0 x16 | PCI Express 3.0 x16 | PCI Express 3.0 x16 | PCI Express 3.0 x16 | PCI Express 3.0 x16 |
| Performance | |||||
| Peak performance FP32, GFLOPS (based on maximum specified frequency) | 6 060 | 6 691 | 8 873 | 11 340 | 10 974 |
| Performance FP32/FP64 | 1/32 | 1/32 | 1/32 | 1/32 | 1/32 |
| RAM bandwidth, GB/s | 336 | 336 | 320 | 484 | 480 |
| Image output | |||||
| Image output interfaces | DL DVI-I, DisplayPort 1.2, HDMI 1.4a | DL DVI-I, DisplayPort 1.2, HDMI 1.4a | DL DVI-D, DisplayPort 1.3/1.4, HDMI 2.0b | DisplayPort 1.3/1.4, HDMI 2.0b | DL DVI-D, DisplayPort 1.3/1.4, HDMI 2.0b |
| TDP, W | 250 | 250 | 180 | 250 | 250 |
| Suggested retail price (USA, excluding tax), $ | 649 (at time of release) | 999 (at time of release) | At the time of release: 599 / 699 (FE) New price: 499 / 549 (FE) | 699 | 1 200 |
| Recommended retail price (Russia), rub. | 39,990 (at time of release) | 74,900 (at time of release) | At the time of release: 54,990 (FE) New price: 45,790 (FE) | 52,990 (FE) | 89 990 |
The GTX 1080 Ti and TITAN X are separated by a rather shaky boundary, and this, on the one hand, is good news for enthusiasts who decided to hold off on upgrading their old video card to Pascal while waiting for a top-end accelerator. On the other hand, high performance is reflected in the price of the video card. The GTX 1080 Ti became the third single-processor GeForce after the GTX 780 Ti and GTX 1080 Founders Edition to reach $699. The price in the company’s Russian online store is RUB 52,990. In addition, this time NVIDIA does not distinguish between the reference design video card (Founders Edition) and partner samples that will appear later and will receive exactly the same recommended price tag.
At the same time, NVIDIA has made the GeForce GTX 1070 and GTX 1080 more affordable. At the time of release, the latter were priced at $379/449 and $599/699 (partner cards and Founders Edition, respectively), and are now sold for $349/399 and $499/549. Russian prices for video cards on the NVIDIA website decreased from 34,990 54,990 to 31,590 and 45,790 rubles.
At the same time, NVIDIA announced that video card manufacturers will have the opportunity to equip their versions of the GeForce GTX 1060 and GTX 1080 with faster GDDR5/5X chips: 9 instead of 8 Gbit/s for the GTX 1060 and 11 instead of 10 Gbit/s for the GTX 1080. Reference specifications The models remained unchanged, it’s just that partners purchasing memory chips from NVIDIA along with a GPU for installation on their own boards now have a choice. Judging by how rarely we see RAM chips on original design video cards that differ from those found in reference accelerators, this move by NVIDIA could really make factory overclocking of video memory more common.
Review of ten GeForce GTX 1080 Ti video cards: choosing the best model
Table of contents
- Introduction
- Nvidia GeForce GTX 1080 Ti
- ASUS ROG GeForce GTX 1080 Ti Strix Gaming OC
- Gainward GeForce GTX 1080 Ti Golden Sample
- Gigabyte Aorus GeForce GTX 1080 Ti 11G
- Inno3D iChill GeForce GTX 1080 Ti X4 Ultra
- Inno3D iChill GeForce GTX 1080 Ti Black
- KFA2 GeForce GTX 1080 Ti EXOC White
- MSI GeForce GTX 1080 Ti Gaming X 11G
- MSI GeForce GTX 1080 Ti Gaming X Trio
- MSI GeForce GTX 1080 Ti Lightning Z
- Characteristics of tested video cards
- GeForce GTX 1080 Ti performance
- Conclusion
Introduction
The entire computer world froze at a low start in anticipation of the announcement of a new line of Nvidia graphics accelerators.
But for now the company is in no hurry to please users, taking a defensive position and minimizing possible leaks about the performance of video cards. However, there are about two months left until the theoretical hour X, and we are already anticipating a noticeable jump in performance. Meanwhile, the community of enthusiasts split in two: one part of gamers began to save money for a top-end solution, the cost of which, according to partners, would not be affordable, and the second part was studying models of the previous generation based on Pascal, since their prices had noticeably decreased.
And in order not to be verbose, we will continue the series of summary materials dedicated to Nvidia graphics accelerators, and will start with the big leagues. At the moment, the flagship GeForce GTX provides capabilities that allow you to enjoy modern games even in 4K resolution at 60 FPS. Namely, if you believe the rumors, it will become the main opponent of the upcoming new product.
It's time to plunge into the past, because earlier in the laboratory they had already summed up the results of the confrontation between Nvidia GeForce GTX 1xxx and GeForce GTX 9xx and selected the best models among the GeForce GTX 1060 and GeForce GTX 1080.
announcements and advertising
2080 Super Gigabyte Gaming OC for 60 rubles.
Compeo.ru - the right comp store without any tricks
RTX 2060 becomes cheaper before the arrival of 3xxx
Ryzen 4000
series included in computers already in Citylink
The price of MSI RTX 2070 has collapsed after the announcement of RTX 3xxx
Core i9 10 series is half the price of the same 9 series
The price of memory has been halved in Regard - it’s more expensive everywhere
Nvidia GeForce GTX 1080 Ti
Full name: Nvidia GeForce GTX 1080 Ti, manufacturer code: Nvidia GeForce GTX 1080 Ti. Description on the company website - link.
In the spring of 2020, the company decided to please players and enthusiasts with a new video card, and not just a new one, but the most productive one. Nvidia GTX Titan X (Pascal) was taken as the basis, as the most expensive and powerful. But persuasive leadership required more than that—attracting the buyer with price. And at the time of the announcement, the following happens: the GeForce GTX 1070 drops in price to $349, the GeForce GTX 1080 – to $499, and the flagship GeForce GTX 1080 Ti gets a price of $649.
And so that the new video card does not step on the heels of the GeForce GTX Titan X, they removed one memory chip, reducing the volume to 11 GB, and along with it one memory controller disappeared from the GPU, and installed a new bus width format of 352 bits. True, the memory itself operates at a higher frequency: 2750 MHz instead of 2500 MHz.
The changes in the power unit did not go unnoticed: with the same power consumption of 250 W, it was possible to raise the average GPU frequency to 1680 MHz. Let us remember that for the GeForce GTX Titan X the video core frequency was 1580 MHz.
Nvidia's reference cooling system has changed its design periodically, but not this time. The engineers only worked on increasing the radiator area. The evaporation chamber installed at its base has increased in size, which theoretically improved heat transfer by 10-15%. In addition, now the DVI ports do not interfere with the removal of heated air.
Initially, the flagship was based on the GeForce GTX Titan X printed circuit board, the power system of which was tinkered with, strengthening the power circuits. Thanks to this, efficiency has increased.
Now the GPU is served by a system consisting of seven dual phases. The maximum output reaches 250 A, and the video card is powered by two 6 + 8 pin connectors.
Standard operating modes of the video card: GPU – 1480 MHz, boost – 1680 MHz, memory – 1376 MHz.
Review of Nvidia GeForce GTX 1080 Ti in our laboratory.
ASUS ROG GeForce GTX 1080 Ti Strix Gaming OC
Full name: ASUS ROG GeForce GTX 1080 Ti Strix Gaming OC, manufacturer code: ROG-Strix-GTX1080TI-O11G-Gaming. Description on the company website - link.
The front part of the ASUS ROG model is covered with an original cooling system, which visually has not undergone any changes and is almost no different from its closest relatives of the ROG Strix series. The same three 90 mm fans, a black protective casing, many edges and slots for ASUS Aura lighting.
The company's developers abandoned cooling the graphics processor and power subsystem with a single radiator. And now the plate cooling the VRM and memory chips is in contact with the main heat dissipator through a thermal pad.
The engineers' next unexpected step was the transition from direct contact to the use of a nickel plate, which many had been waiting for. As for the main radiator, it is pierced by six nickel-plated heat pipes with a diameter of 6 mm each. The nickel-plated base is smooth, the polishing can be called ideal, the reflection is mirror-like.
The printed circuit board of the video card is ASUS' own design. The power subsystem is made according to a 10 + 2 scheme (ten phases are responsible for powering the graphics processor, two for the memory subsystem), using high-quality Super Alloy Power II components.
Standard operating modes of the video card: GPU – 1594 MHz, boost – 1974 MHz, memory – 1376 MHz.
Review of ASUS ROG GeForce GTX 1080 Ti Strix Gaming OC in our laboratory.
Gainward GeForce GTX 1080 Ti Golden Sample
Full name: Gainward GeForce GTX 1080 Ti “Golden Sample”, manufacturer code: 426018336-3903. Description on the company website - link.
The design of the Gainward version is close to the outlines of expensive supercars. The front side is decorated with a scarlet metal insert, the radial fan located in the front part is associated with the car’s engine, and the overall appearance is completed by a blank plastic casing of the cooling system with the abbreviation “GS”.
This is a reference model in the usual sense of the term, but from the Nvidia reference version we did not get everything that it offers: only a printed circuit board and a similar cooling system. Each element hides its own nuances. For example, the reverse side of the Gainward GeForce GTX 1080 Ti Golden Sample is protected by a massive metal “backplate” without slots or decorative inserts. And to increase reliability and rigidity, a crosspiece is used in the GPU area.
The decorative casing hides a massive cooler fixed in the side grooves. The main radiator is a classic “sandwich” of plates and a base. To save money, engineers replaced the evaporation chamber with an aluminum “blank”, and in order for the design to somehow cope with heat removal, they used a copper insert.
The video memory chips, vGPU and vMem mosfets are covered by a separate plate, which acts as a radiator and stiffener. In the area of the power subsystem, the design received a slight convexity, which cannot be called an effective solution, but placing a “turbine” in this area allows the VRM subsystem to be one of the first to blow.
The Gainward GeForce GTX 1080 Ti Golden Sample video card is built on an original printed circuit board with the performance characteristics of Nvidia using the Palit element base (Gainward is a subsidiary). Further, there are continuous differences.
The engineers redesigned part of the PCB, eliminating two slots for microcircuits in the radial fan area, and also removed the side additional power supply and the cutout in the PCB.
The power subsystem has been simplified, and although the total number of phases has not changed, not all elements are wired. These changes are not in favor of Gainward. Now it is built according to the “7+2+1” scheme, in which seven phases are allocated to the graphics processor, two to the video memory, and one to the PLL chip.
Standard operating modes of the video card: GPU – 1506 MHz, boost – 1911 MHz, memory – 1376 MHz.
Review of the Gainward GeForce GTX 1080 Ti Golden Sample in our laboratory, in which you can get acquainted with alternative cooling systems of the reference version.
Gigabyte Aorus GeForce GTX 1080 Ti 11G
Full name: Aorus GeForce GTX 1080 Ti 11G, manufacturer code: GV-N108TAorus-11GD. Description on the company website - link.
The length of the Gigabyte model is 293 mm, width – 142 mm, height – 55 mm. The last parameter means that it will cover three expansion slots, and enthusiasts who are going to install two such video cards in SLI mode will need a spacious case and a motherboard with PCI-Express connectors spaced at the required distance from each other.
This version of the GeForce GTX 1080 Ti has an interesting appearance, which is emphasized by three large fans with a diameter of 95 mm each and a casing equipped with a backlit side panel and a crosshair in the center.
The reverse side, in the best traditions of top products, is protected by a backplate. Among the interesting features, we note that it is part of a passive cooling system, in contact with the printed circuit board through a thermal interface.
A massive cooler is responsible for cooling the Aorus GeForce GTX 1080 Ti 11G video card. Its base is made in the form of a massive copper plate covering all eight memory chips. In this case, the technology of direct contact was abandoned, perhaps it was for the better. The power subsystem is cooled by a soldered aluminum insert with two rows of thermal pads.
The design of the massive radiator is simple - it is a two-section heat dissipator made from a huge number of aluminum plates of complex shape. All connections are carefully soldered. Another factor is the use of composite heat pipes that enhance heat transfer. A total of six tubes are used here: five with a diameter of 8 mm, one with a diameter of 6 mm.
The Aorus GeForce GTX 1080 Ti 11G video card is built on an original black printed circuit board with a glossy sheen, which in fact turns out to be a special coating to protect it from moisture and dust.
The element base corresponds to the Ultra Durable VGA concept, which includes the use of high-quality capacitors, chokes with ferrite cores, field-effect transistors and double copper layers in the board.
The power subsystem is made according to the formula “12 + 2”: twelve phases are allocated to the graphics processor, and two go to the video memory.
Standard operating modes of the video card: GPU – 1569 MHz, boost – 1949 MHz, memory – 1376 MHz.
Review of Gigabyte Aorus GeForce GTX 1080 Ti 11G in our laboratory.
Inno3D iChill GeForce GTX 1080 Ti X4 Ultra
Full name: iChill GeForce GTX 1080 Ti X4 Ultra, manufacturer code: C108T4-1SDN-Q6MNX. Description on the company website - link.
First of all, the original HerculeZ Air Boss cooling system is of interest. Three 90 mm fans, a black protective casing, a somewhat futuristic but no-frills design – this video card looks visually unusual. Moreover, its length barely exceeds 30 cm.
The main difference from competitor models and our own iChill GeForce GTX 1080 Ti X3 Ultra is the use of Active Power Cooling System (APCS) technology, which is based on a separate radiator on the power circuits and a fourth fan blowing it.
The first thing that catches your eye is the plate on the memory chips and power circuits, which looks like a single whole, but in fact it is not: it consists of two separate parts. The heat dissipator of the power circuits not only cools them, but also covers part of the memory chips, tightly contacting the heatsink on the remaining chips.
Additionally, the entire structure received a pair of heat pipes and fins, which will benefit the efficiency of this type of cooling. The main CO radiator is quite massive, two-section. It is pierced by four heat pipes: two 8 mm and two 6 mm.
The printed circuit board of the Inno3D video card is made in the reference design. The difference is the presence of a DVI-D video output. On the one hand, this ensures compatibility with a large number of monitors, on the other hand, it makes it impossible to install a full-coverage EK Water Blocks water block for the GeForce GTX 1080 Ti and turn the model into a single-slot one.
The power subsystem is made according to the “7 + 1 + 2” scheme (GPU/PLL/MEM). Seven dual phases, 6 + 8 pin power supply, maximum current 250 A - we have already seen all this in the review of Nvidia GeForce GTX 1080 Ti.
Standard operating modes of the video card: GPU – 1607 MHz, boost – 2025 MHz, memory – 1425 MHz.
Review of Inno3D iChill GeForce GTX 1080 Ti X4 Ultra in our laboratory.
Inno3D iChill GeForce GTX 1080 Ti Black
Full name: Nvidia GeForce GTX 1080 Ti, manufacturer code: Nvidia GeForce GTX 1080 Ti. Description on the company website - link.
Here is a representative of the top segment with a pre-installed liquid cooling system that does not require additional maintenance other than installation in the case. In the appearance of the Inno3D iChill Black model, the emphasis is on a strict style, represented by a black metal casing with slots along the perimeter and a red stripe crossing the equator. However, the monotony disappears in the active state, when the inscriptions “iChill” with a snowflake and “Accelero Hybrid S” are illuminated by RGB lighting.
Two rubberized tubes emerge from the top end, through which coolant flows from the water block pump installed on the Nvidia GP102 GPU. Their length is 365 mm, and this is enough to place a cooler with a 120 mm fan not only on the rear wall of the case, but also in the upper and even front sections.
The cooling system of the Inno3D video card is a complex design consisting of a liquid cooling system and a modified cooling of the power unit, as well as an external 120 mm fan and power cable.
The metal casing of the CO, together with the water block pump, is attached directly to the casing along with the lighting elements; there is also an 80 mm turntable, which is responsible for blowing the power subsystem.
Manufacturers of graphics accelerators have a short list of companies that you can contact to install a maintenance-free life support system. And the fact of using Arctic Cooling CO – Accelero Hybrid II-120 – will not be a revelation. This product has received positive recommendations due to its excellent efficiency and performance: it is not surprising that the developers of Inno3D took advantage of this solution.
Inno3D iChill GeForce GTX 1080 Ti Black is built on a black reference PCB. An attentive reader will note a minimum of differences with the original design: capacitors of a different color with a nominal capacity and an additional 2 pin connector are soldered, although it is not used.
The general formula of the power subsystem has ten phases: seven phases are allocated to the graphics processor, two to the video memory, and one to the PLL.
Standard operating modes of the video card: GPU – 1607 MHz, boost – 2025 MHz, memory – 1425 MHz.
Review of Inno3D iChill GeForce GTX 1080 Ti Black in our laboratory.
⇡#Design
Since the predecessor of the GeForce GTX 1080 Ti is TITAN X, it is not surprising that the differences in the appearance of the video cards are minimal. The GTX 1080 Ti differs from the all-black Titan in the standard silver color of the casing, and inside there is exactly the same radiator with an evaporation chamber and a radial fan. This design has proven itself well back in the days of the first TITAN and GeForce GTX 780/780 Ti. Possessing outstanding acoustic characteristics for this type of cooling system, NVIDIA reference video cards stand out among partner devices, since the latter are almost always equipped with an open-design cooler. At the same time, a closed “turbine” remains the best option for compact PCs, where it is important that the video card exhausts most of the air outside. In addition, the reference cooler ensures cooling of memory chips and many other hot spots of the board due to its massive aluminum base.
Beginning with the GeForce GTX 1080, NVIDIA SLI-compatible graphics cards allow you to remove part of the metal plate that covers the back of the circuit board to allow air to flow freely to the cooling system of the adjacent graphics card.
Despite sharing the same cooling design as the TITAN X, the GTX 1080 Ti has one obvious feature: NVIDIA has removed the DL-DVI port from the external panel of the graphics card. The entire set of outputs that remains on the GTX 1080 Ti is limited to three DisplayPort 1.3/1.4 connectors and one HDMI 2.0b. But the entire half of the plate is now used to remove hot air outside the PC case, which should have a positive effect on cooling efficiency. For those who still use a monitor without DisplayPort or HDMI connectors, NVIDIA includes a DP to Single-Link DVI adapter for the GeForce GTX 1080 Ti Founders Edition.
⇡#Pay
Like the cooling system, the equipment on the GeForce GTX 1080 Ti PCB not only has not deteriorated compared to the TITAN X, but has also become better. The increased RAM bandwidth in the GTX 1080 Ti is achieved through new Micron chips (labeled MT58K256M321JA-110), for which an effective frequency of 11 GB/s is standard. By the way, Micron’s catalog also includes GDDR5X memory with a frequency of 12 GB/s, but by choosing the fastest chips, NVIDIA would probably have to exhaust the reliability reserve of the RAM controllers in the GPU.
What remains unchanged in the voltage converter of NVIDIA video cards is the uP9511 PWM controller from uPI Semiconductor, which on the GTX 1080 Ti serves the maximum number of phases for its capabilities - seven, all of which belong to the GPU power supply circuit. The two phases of the GDDR5X chips are controlled by the uP1685 controller. However, there are changes in the power part of the video card. NVIDIA placed a number of elements on the printed circuit board that were missing in TITAN X, first of all, it doubled all field-effect transistors in the “phases” of the GPU power supply, which should have a positive effect on the efficiency of the converter.
Choosing an Nvidia GeForce GTX 1080 video card: review of 9 models (page 2)
EVGA GeForce GTX 1080 SC Gaming ACX 3.0
Description on the manufacturer's website - link.
EVGA GeForce GTX 1080 SC Gaming received an interesting casing design that changed the coloring. The white and black cooling system casing, along with metal inserts, grille coverings and emblems, sets the new product apart from other models with dark colors.
The front side is equipped with two fans with a diameter of 90 mm with an interesting curved blade shape. They practically cover the entire radiator in height, which will have a positive effect on the efficiency of the cooler. This approach is more advantageous than three smaller turntables.
The length of the EVGA GeForce GTX 1080 SC Gaming is comparable to the reference model and is 267 mm, but the width due to the backlit side insert is 112 mm. In a PC, the video card will cover two expansion slots.
The radiator is a two-section heat dissipator made of many nickel-plated aluminum plates of complex shape. The fins are pierced by five heat pipes with a diameter of 6 mm each. They are made in an interesting configuration: two are U-shaped, three are I-shaped. An aluminum insert is soldered into the base.
announcements and advertising
2080 Super Gigabyte Gaming OC for 60 rubles.
Compeo.ru - the right comp store without any tricks
RTX 2060 becomes cheaper before the arrival of 3xxx
Ryzen 4000
series included in computers already in Citylink
The price of MSI RTX 2070 has collapsed after the announcement of RTX 3xxx
Core i9 10 series is half the price of the same 9 series
The price of memory has been halved in Regard - it’s more expensive everywhere
Instead of individual radiators, a massive plate is used, somewhat reminiscent of a “backplate”. It is devoid of fins, which reduces the useful dispersion area, but it covers all elements and contacts them through thermal pads.
The EVGA GeForce GTX 1080 SC Gaming video card is built on the GeForce GTX 1080 reference PCB. Many manufacturers release models on the reference PCB, thereby simplifying the task of releasing new products. Among the advantages, we note the unification, the ability to install “full covers” and third-party cooling systems.
The power subsystem is built according to a “5+1” scheme, where five phases are allocated to the graphics processor and one to the video memory. There is space on the printed circuit board for one more phase, but the developers did not solder it.
EVGA GeForce GTX 1080 SC Gaming differs from the reference version not only in the cooling system, but also in increased clock frequencies. The developers have raised the performance bar for their model by several points. The core frequency is now 1709 MHz, in Boost mode – 1848 MHz, in games it dynamically rises to 2000 MHz at a voltage of 1.062 V. The memory operates at the reference value.
Ultimately, even the Power Limit level increased to 120% did not allow us to overcome the frequency bar of 1789 MHz and 1401 MHz for the core and memory, respectively. In games, the GPU frequency dynamically increases to 2114 MHz, and the average value is 2063 MHz.
You can learn more about the video card in the material: “Review and testing of the EVGA GeForce GTX 1080 SC Gaming ACX 3.0 video card.”
Gigabyte GeForce GTX 1080 G1 Gaming
Description on the manufacturer's website - link.
The appearance of the Gigabyte GeForce GTX 1080 G1 Gaming has undergone changes compared to the ninth series. For a top-class video card, the distinguishing feature is a trio of fans with a diameter of 80 mm; the dimensions of the impeller are slightly more modest and amount to 77 mm. The fans are highlighted by a new shroud design that features distinctive bright orange stripes on a black background.
The length of the Gigabyte video card is 280 mm, width – 117 mm, in the case the new product will cover two expansion slots.
The Windforce 3X cooling system was introduced to us quite a long time ago and has survived not only several generations of video cards, but also its own modifications.
The design includes a two-section radiator, composed of many aluminum plates, pierced by three heat pipes with a diameter of 8 mm each. All connections are carefully and neatly soldered, which increases the efficiency of the cooler.
The base is smooth, but not mirror-like. There is also an aluminum pad for cooling video memory chips. Very thin gray thermal pads are used as a thermal interface. The situation is similar with the VRM area, where engineers soldered the base to the radiator fins. In addition to two holes with fasteners, one has been added in the center, which increases the clamping force and improves the quality of cooling of power transistors. In a good sense, the chokes also got a “hit”: they were given a separate frame and thermal pad.
The Gigabyte GeForce GTX 1080 G1 Gaming model is built on an original black PCB, the surface of which is slightly glossy. PCB length is 267 mm.
The high-quality element base is made according to the Ultra Durable VGA concept, which in practice means the use of high-quality capacitors, chokes with ferrite cores and field-effect transistors.
The power subsystem is made according to the “8+2” formula, where eight phases are allocated to the graphics processor and two to video memory. Such an enhanced VRM implementation allows you to count on high overclocking and proper load redistribution, as well as lower transistor temperatures.
The Gigabyte GeForce GTX 1080 G1 Gaming version operates at increased clock speeds, which provides higher performance compared to other versions of the GeForce GTX 1080. The core frequency is 1696 MHz instead of the reference 1607 MHz, in Boost mode it reaches 1835 MHz, the video memory frequency is 1251 MHz. The average GPU voltage is 1.043-1.05 V.
From the first attempts, we managed to overclock the new product, increasing the core and memory frequencies by 80 MHz and 163 MHz, respectively.
Based on the results of a laboratory test, the Gigabyte GeForce GTX 1080 G1 Gaming model was able to demonstrate its outstanding character. Don’t pass by and study at your leisure the article “Review and testing of the Gigabyte GeForce GTX 1080 G1 Gaming video card.”
⇡#Test stand, testing methodology
| Test bench configuration | |
| CPU | Intel Core i7-5960X @ 4 GHz (100 MHz × 40), constant frequency |
| Motherboard | ASUS RAMPAGE V EXTREME |
| RAM | Corsair Vengeance LPX, 2133 MHz, 4 × 4 GB |
| ROM | Intel SSD 520 240 GB + Crucial M550 512 GB |
| power unit | Corsair AX1200i, 1200 W |
| CPU cooling system | Thermalright Archon |
| Frame | CoolerMaster Test Bench V1.0 |
| Monitor | NEC EA244UHD |
| operating system | Windows 10 Pro x64 |
| Software for AMD GPUs | |
| All video cards | Radeon Software Crimson ReLive Edition 17.3.1 (Tesselation: Use application settings) |
| NVIDIA GPU software | |
| All video cards | GeForce Game Ready Driver 378.78 WHQL |
| Benchmarks: synthetic | |||
| Test | API | Permission | Full screen anti-aliasing |
| 3DMark Fire Strike | DirectX 11 (feature level 11_0) | 1920 × 1080 | Off |
| 3DMark Fire Strike Extreme | 2560 × 1440 | ||
| 3DMark Fire Strike Ultra | 3840 × 2160 | ||
| 3DMark Time Spy | DirectX 12 (feature level 11_0) | 2560 × 1440 | |
| Benchmarks: games | ||||
| Game (in order of release date) | API | Settings | Full screen anti-aliasing | |
| 1920 × 1080 / 2560 × 1440 | 3840 × 2160 | |||
| Crysis 3 + FRAPS | DirectX 11 | Max. quality. Start of the Swamp mission | MSAA 4x | Off |
| Metro: Last Light Redux, built-in benchmark | Max. quality | SSAA 4x | ||
| GTA V, built-in benchmark | Max. quality | MSAA 4x + FXAA + Reflection MSAA 4x | ||
| DiRT Rally, built-in benchmark | Max. quality | MSAA 4x | ||
| Rise of the Tomb Raider, built-in benchmark | DirectX 12 | Max. quality, VXAO off | SSAA 4x | |
| Tom Clancy's The Division, built-in benchmark | Max. quality, HFTS off | SMAA 1x Ultra + TAA: Supersampling | TAA: Stabilization | |
| Ashes of the Singularity, built-in benchmark | Max. quality | MSAA 4x + TAA 4x | Off | |
| DOOM | Vulkan | Max. quality. Foundry Mission | TSSAA 8TX | |
| Total War: WARHAMMER, built-in benchmark | DirectX 12 | Max. quality | MSAA 4x | |
| Deus Ex: Mankind Divided, built-in benchmark | Max. quality | MSAA 4x | ||
| Battlefield 1 + OCAT | Max. quality. Start of mission Through Mud and Blood | TAA | ||
| Benchmarks: video decoding, computing | |
| Program | Settings |
| DXVA Checker, Decode Benchmark, H.264 | Files 1920 × 1080p (High Profile, L4.1), 3840 × 2160p (High Profile, L5.1). Microsoft H264 Video Decoder |
| DXVA Checker, Decode Benchmark, H.265 | Files 1920 × 1080p (Main Profile, L4.0), 3840 × 2160p (Main Profile, L5.0). Microsoft H265 Video Decoder |
| LuxMark 3.1 x64 | Hotel Lobby Scene (Complex Benchmark) |
| Sony Vegas Pro 13 | Sony benchmark for Vegas Pro 11, duration - 65 s, rendering in XDCAM EX, 1920 × 1080p 24 Hz |
| SiSoftware Sandra 2020 SP1, GPGPU Scientific Analysis | Open CL, FP32/FP64 |
| CompuBench CL Desktop Edition X64, Ocean Surface Simulation | — |
| CompuBench CL Desktop Edition X64, Particle Simulation— 64K | — |
Test participants
The following video cards took part in performance testing:
- AMD Radeon R9 Fury X (1050/1000 MHz, 4 GB);
- NVIDIA TITAN X (1417/10000 MHz, 12 GB);
- NVIDIA GeForce GTX 1080 Ti (1480/11010 MHz, 11 GB);
- NVIDIA GeForce GTX 1080 (1607/10008 MHz, 8 GB);
- NVIDIA GeForce GTX TITAN X (1000/7010 MHz, 12 GB);
GeForce GTX 1080 Ti
At the beginning of spring 2020, the Californian company officially presented its new “combat machine” for fans of virtual battles - GeForce GTX 1080 Ti . This video adapter is aimed at particularly demanding gamers who do not have any special budget restrictions. The GTX 1080 Ti model is perhaps the final, but at the same time the loudest “major chord” in the line of video cards based on the Pascal architecture. What can the most powerful single-chip video card of 2020 offer? Let's get a look.
At first glance at the new product, you can pay attention to the almost complete external similarity with the GeForce GTX 1080. However, in terms of internal content, the GeForce GTX 1080 Ti is rather a relative of the previous absolute flagship TITAN X (Pascal). Yes, yes, exactly the “past”, since NVIDIA, neither more nor less, calls its new brainchild as the “fastest” gaming video card ever released by the corporation. According to the “greens”, the performance superiority over the above-mentioned GTX 1080 is more than 30 percent. The validity of this information, as well as the magnitude of the superiority over TITAN X Pascal, will be discussed below based on the results of the tests performed.
Like the latest Titan, the GeForce GTX 1080 Ti is based on a GPU codenamed GP102 and built on the Pascal graphics architecture. The same number of CUDA cores (3584) makes these models similar, but the amount of video memory in the new product is slightly reduced - from 12 to 11 GB. However, the GDDR5X memory of the GTX 1080 Ti video chip operates at a higher effective frequency of 11 GHz, instead of 10. As a result, the new model has increased throughput to 484 GB/s, which is even more than the last Titan. The new product has a slightly “narrowed” memory bus - 352 bits, instead of 384. But the processor clock frequency in boost mode can reach 1582 MHz, while for TITAN X Pascal this value is 1531 MHz.
Let's take a closer look at the technical characteristics of the mentioned graphics adapters in comparison with each other.
| GeForce GTX 1080 Ti | TITAN X Pascal | GeForce GTX 1080 | |
| Chip code name | GP102 | GP102 | GP104 |
| Architecture | Pascal 1.0 | Pascal 1.0 | Pascal 1.0 |
| Technical process | 16 nm | 16 nm | 16 nm |
| Number of transistors | 12 billion | 12 billion | 7.2 billion |
| Clock frequency, MHz: base / Boost | 1480/1582 | 1417/1531 | 1607/1734 |
| Texture units (TMU) | 224 | 224 | 160 |
| Universal processors | 3584 | 3584 | 2560 |
| Rasterization Units (ROP) | 88 | 96 | 64 |
| Video memory type | GDDR5X | GDDR5X | GDDR5X |
| Video memory capacity | 11 GB | 12 GB | 8 GB |
| Bandwidth GB/s | 484 | 480 | 320 |
| Streaming multiprocessors | 28 | 28 | 20 |
| Bus width, bits | 352 | 384 | 256 |
| Effective memory frequency | 11 GHz | 10 GHz | 10 GHz |
| Interface | PCI-Express 3.0 | PCI-Express 3.0 | PCI-Express 3.0 |
| Additional connector nutrition | 6+8 | 6+8 | 6+8 |
| Power consumption | 250 W | 250 W | 180 W |
| Maximum temperature | 91 °C | 94 °C | 94 °C |
According to the manufacturer's recommendation, the power supply for the system should be at least 600 W. The reference model uses a seven-phase power subsystem with VRM voltage regulators, as well as dual FET switches to reduce load. At the same time, MOSFET transistors also receive lower operating temperatures. All these components made it possible to reduce the noise of the GeForce GTX 1080 Ti to 2.5 dB, and lower the temperature under load by 5°C compared to the GeForce GTX 1080.
The Founders Edition measures 268 (length) x 102 (height) x 37 (width) mm. The video card in question has one HDMI 2.0b connector and three DisplayPort 1.4. Technologies supported by GeForce GTX 1080 Ti are: Multi-Projection, NVIDIA GameStream, NVIDIA SLI Ready, VR Ready, NVIDIA Ansel, NVIDIA G-SYNC, Microsoft DirectX 12 API (12_1), Vulkan API, OpenGL 4.5.
Testing GeForce GTX 1080 Ti Founders Edition in games* (average FPS)
| Games/Settings | Ultra 1920x1080 | Ultra 2560x1440 | Ultra 3840x2160 |
| Red Dead Redemption 2 | 60 fps | 45 fps | 31 fps |
| The Witcher 3: Wild Hunt | 124 fps | 86 fps | 48 fps |
| The Division | 154 fps | 112 fps | 60 fps |
| DOOM 2016 | 157 fps | 121 fps | 74 fps |
| FarCry Primal | 112 fps | 90 fps | 51 fps |
| Fallout 4 | 159 fps | 117 fps | 62 fps |
*The following bench configuration was used for testing: Intel Core i7-6700K processor, mat. ASUS Maximus VIII Extreme board, G.Skill 32 GB RAM, Plextor PX-256M6Pro SSD, Corsair AX1500i power supply, Windows 10 (64 Bit) operating system. NVIDIA GeForce driver version: 378.92.
Thus, NVIDIA's loud statement about the release of the world's fastest gaming video card built on a single GPU is entirely justified. When tested in 3DMark Fire Strike, the GeForce GTX 1080 Ti scored 27,805 points. In gaming and synthetic tests, the superiority of the new product over the GeForce GTX 1080 actually reaches 30 (and sometimes 35) percent.
Regarding TITAN X Pascal, despite the reduction in bus width, as well as the reduction in video memory, the GeForce GTX 1080 Ti shows an advantage of up to 5%. This happens due to increased bandwidth and increased clock speed. If you pay attention to the difference in performance with the previous GeForce GTX 980 Ti, then the new model of the GeForce line is about 60% more powerful.
Comparing NVIDIA's new product with the most productive video card from AMD at the beginning of spring 2020 - Radeon R9 Fury X, we see the absolute superiority of the GTX 1080 Ti model. Even the new Radeon RX Vega, which is expected to be released in the near future, is unlikely to be able to fully compete in performance with NVIDIA's latest flagship.
And finally, about the price. At the time of announcement (March 2020), the GeForce GTX 1080 Ti costs $699. And this amount is $500 less than the cost of TITAN X Pascal. In Russian monetary units, the recommended price for the new product is 52,990 rubles. Official deliveries of the new video card began on March 10, and in addition to the reference sample, we will see a variety of partner models.
⇡#Performance: 3DMark
Synthetic tests ignore the loss of one gigabyte of memory in the GeForce GTX 1080 Ti, but are sensitive to the GPU clock speed. As a result, in various tests, either the GTX 1080 Ti and TITAN X are equivalent in terms of points scored, or the GTX 1080 Ti has an insignificant advantage of 1-3%.
No matter which of the other single-processor gaming video cards we take to compare with the GTX 1080 Ti, the latter demonstrates a completely different level of performance. The new flagship is on average 32% ahead of the GTX 1080 in 3DMark, and in the case of the GeForce GTX 980 Ti and Radeon R9 Fury X, the GTX 1080 Ti's advantage reaches 70% and 75%.
| 3DMark (Graphics Score) | |||||||
| Permission | NVIDIA GeForce GTX 1080 Ti (1480/11010 MHz, 11 GB) | NVIDIA GeForce GTX TITAN X (1000/7012 MHz, 12 GB) | NVIDIA GeForce GTX 1080 (1607/10008 MHz, 8 GB) | NVIDIA TITAN X (1417/10000 MHz, 12 GB) | AMD Radeon R9 Fury X (1050/1000 MHz, 4 GB) | 0 | |
| Fire Strike | 1920 × 1080 | 28 003 | 17 115 | 21 694 | 27 987 | 16 279 | 0 |
| Fire Strike Extreme | 2560 × 1440 | 13 453 | 7 928 | 10 264 | 13 520 | 7 745 | 0 |
| Fire Strike Ultra | 3840 × 2160 | 6 594 | 4 042 | 5 001 | 6 557 | 3 942 | 0 |
| Time Spy | 2560 × 1440 | 9 480 | 5 106 | 7 111 | 9 149 | 5 084 | 0 |
| Max. | −39% | −23% | +0% | −40% | −100% | ||
| Average | −41% | −24% | −1% | −43% | −100% | ||
| Min. | −46% | −25% | −3% | −46% | −100% | ||
Performance: Gaming (1920 × 1080, 2560 × 1440, 3840× 2160)
Tests in real games paint a more complex picture of comparison between the GeForce GTX 1080 Ti and TITAN X. In all resolutions, the new product has an average advantage of 2%. At the same time, there is sometimes no difference of even 1 FPS between the two cards, and rare examples of TITAN X victory come down to one frame per second over the result of the GTX 1080 Ti, which clearly fits into random fluctuations in performance measurements. However, there is one game that clearly favors the GTX 1080 Ti - Tom Clancy's The Division. In it, the advantage of the GTX 1080 Ti is from 8 to 10%.
Compared to the GTX 1080, the new flagship delivers 25-33% higher frame rates, but the biggest difference is seen at high resolutions, from 1440p to 2160p. The processing power of top gaming video cards is so great that in 1080p mode, CPU speed becomes an important factor. Additionally, DOOM already hits the game's frame rate limit of 200 FPS on both GPUs.
The advantage of the GTX 1080 Ti over the GTX TITAN X based on Maxwell architecture in smaller games than in synthetic games, but is still extremely large - from 54 to 67% depending on the resolution. Radeon R9 Fury X, on the contrary, turned out to be an even easier opponent, because a small amount of RAM (4 GB) plays against the AMD accelerator. As a result, while in 1080p mode the GTX 1080 Ti provides 70% higher frame rates, then at 1440p and 2160p the difference increases to 82 and 96%.
We will pay special attention to the results of the GTX 1080 Ti in 4K. Only NVIDIA video cards based on GP102 are currently unconditionally suitable for gaming at this resolution with maximum graphics quality settings (albeit without full-screen anti-aliasing using resource-intensive methods). In most of the games in our test frame, the average frame rate exceeds 50–60 FPS, and only in Deus Ex: Mankind Divided does it drop significantly below this level.
| 1920 × 1080 | ||||||
| Full screen anti-aliasing | NVIDIA GeForce GTX 1080 Ti (1480/11010 MHz, 11 GB) | NVIDIA GeForce GTX TITAN X (1000/7012 MHz, 12 GB) | NVIDIA GeForce GTX 1080 (1607/10008 MHz, 8 GB) | NVIDIA TITAN X (1417/10000 MHz, 12 GB) | AMD Radeon R9 Fury X (1050/1000 MHz, 4 GB) | |
| Ashes of the Singularity | MSAA 4x + TAA 4x | 58 | 45 | 45 | 55 | 32 |
| Battlefield 1 | TAA | 144 | 90 | 118 | 140 | 92 |
| Crysis 3 | MSAA 4x | 112 | 66 | 79 | 110 | 61 |
| Deus Ex: Mankind Divided | MSAA 4x | 53 | 30 | 38 | 52 | 34 |
| DiRT Rally | MSAA 4x | 128 | 84 | 101 | 124 | 66 |
| DOOM | TSSAA 8TX | 200 | 151 | 200 | 200 | 156 |
| GTA V | MSAA 4x + FXAA + Reflection MSAA 4x | 91 | 67 | 84 | 91 | 54 |
| Metro: Last Light Redux | SSAA 4x | 122 | 74 | 92 | 121 | 70 |
| Rise of the Tomb Raider | SSAA 4x | 85 | 47 | 63 | 84 | 43 |
| Tom Clancy's The Division | SMAA 1x Ultra + TAA: Supersampling | 127 | 67 | 94 | 116 | 61 |
| Total War: WARHAMMER | MSAA 4x | 77 | 59 | 71 | 78 | 51 |
| Max. | −23% | +0% | +1% | −22% | ||
| Average | −35% | −19% | −2% | −41% | ||
| Min. | −47% | −29% | −9% | −52% | ||
| 2560 × 1440 | ||||||
| Full screen anti-aliasing | NVIDIA GeForce GTX 1080 Ti (1480/11010 MHz, 11 GB) | NVIDIA GeForce GTX TITAN X (1000/7012 MHz, 12 GB) | NVIDIA GeForce GTX 1080 (1607/10008 MHz, 8 GB) | NVIDIA TITAN X (1417/10000 MHz, 12 GB) | AMD Radeon R9 Fury X (1050/1000 MHz, 4 GB) | |
| Ashes of the Singularity | MSAA 4x + TAA 4x | 47 | 34 | 34 | 47 | 26 |
| Battlefield 1 | TAA | 111 | 65 | 89 | 108 | 60 |
| Crysis 3 | MSAA 4x | 72 | 41 | 53 | 70 | 40 |
| Deus Ex: Mankind Divided | MSAA 4x | 34 | 19 | 25 | 34 | 20 |
| DiRT Rally | MSAA 4x | 96 | 61 | 73 | 93 | 49 |
| DOOM | TSSAA 8TX | 177 | 100 | 136 | 175 | 107 |
| GTA V | MSAA 4x + FXAA + Reflection MSAA 4x | 78 | 48 | 63 | 78 | 40 |
| Metro: Last Light Redux | SSAA 4x | 72 | 44 | 52 | 69 | 42 |
| Rise of the Tomb Raider | SSAA 4x | 51 | 29 | 38 | 52 | 27 |
| Tom Clancy's The Division | SMAA 1x Ultra + TAA: Supersampling | 91 | 49 | 66 | 83 | 47 |
| Total War: WARHAMMER | MSAA 4x | 64 | 40 | 49 | 64 | 35 |
| Max. | −28% | −19% | +2% | −40% | ||
| Average | −40% | −25% | −2% | −45% | ||
| Min. | −46% | −28% | −9% | −49% | ||
| 3840 × 2160 | ||||||
| Full screen anti-aliasing | NVIDIA GeForce GTX 1080 Ti (1480/11010 MHz, 11 GB) | NVIDIA GeForce GTX TITAN X (1000/7012 MHz, 12 GB) | NVIDIA GeForce GTX 1080 (1607/10008 MHz, 8 GB) | NVIDIA TITAN X (1417/10000 MHz, 12 GB) | AMD Radeon R9 Fury X (1050/1000 MHz, 4 GB) | |
| Ashes of the Singularity | Off | 57 | 44 | 44 | 58 | 35 |
| Battlefield 1 | 64 | 40 | 54 | 64 | 17 | |
| Crysis 3 | 50 | 31 | 36 | 47 | 28 | |
| Deus Ex: Mankind Divided | 38 | 21 | 28 | 36 | 13 | |
| DiRT Rally | 66 | 41 | 50 | 64 | 33 | |
| DOOM | 97 | 54 | 75 | 97 | 57 | |
| GTA V | 70 | 41 | 52 | 68 | 34 | |
| Metro: Last Light Redux | 64 | 39 | 47 | 61 | 38 | |
| Rise of the Tomb Raider | 61 | 35 | 44 | 61 | 31 | |
| Tom Clancy's The Division | TAA: Stabilization | 54 | 29 | 39 | 50 | 32 |
| Total War: WARHAMMER | Off | 52 | 31 | 39 | 52 | 32 |
| Max. | −23% | −16% | +3% | −38% | ||
| Average | −40% | −25% | −2% | −49% | ||
| Min. | −46% | −28% | −7% | −73% | ||
Performance: Video Decoding
Chips from the Pascal family are equipped with the same multimedia unit, the performance of which depends solely on the GPU clock speed. As a result, there are no practically significant differences in the speed of decoding H.264 and HEVC video formats between top “green” video cards. In this discipline, Pascal has moved far ahead compared to previous video cards, especially the GeForce GTX TITAN X, which, unlike the GTX 960 and 950, cannot fully decode HEVC in hardware and shifts some of the work to the CPU. AMD, too, although it has improved the performance of its codec in H.264, cannot compete on equal terms with Pascal.
Note: since decoders within the same GPU line usually do not differ, the diagrams show one device from each family (or more if this rule is violated).
Review and testing of the Nvidia GeForce GTX 1080 video card
Table of contents
- Introduction
- New opportunities
- Specifications
- Nvidia GeForce GTX 1080 Review
- Appearance and dimensions
- Printed circuit board
- Cooling system
- Star Wars Battlefront
- Battlefield 4
- The Witcher 3: Wild Hunt
- Sniper Elite III
- Fallout 4
- Hitman (2016)
- Ashes of the Singularity
- Grand Theft Auto V
- Tom Clancy's The Division
- Rise of the Tomb Raider
- 3DMark
Introduction
The main innovation in the GeForce GTX 1080 is the manufacturing process - 16 nm. Thanks to it, a medium-sized graphics processor (314 mm2) fits 7.2 billion transistors. For comparison, the rather large GM200 has only 8 billion. Thus, the GP104 is an analogue of the GM204, but in terms of capabilities it surpasses the GM200.
As part of the 16 nm process technology, Nvidia has managed to significantly increase the operating frequencies of the GPU, and in GPU Boost mode an incredible frequency of 1.7-1.8 GHz is achieved. Overclocking will also please enthusiasts. On production models you can expect “air” 2.0 GHz. In this case, you do not have to use a high thermal package; for this frequency, 210-220 W is enough.
The GPU is paired with Micron GDDR5X high-frequency memory, rated at 10 GHz. It has a small frequency reserve - 100-200 MHz, which gives a resulting ~400 MHz. Thanks to this, the total throughput of the new product has reached 1.7x the speed of the GeForce GTX 980. And the improved texture compression algorithm allows developers to claim superiority even over the GTX Titan X.
Doesn't everything look very interesting?
announcements and advertising
2080 Super Gigabyte Gaming OC for 60 rubles.
Compeo.ru - the right comp store without any tricks
RTX 2060 becomes cheaper before the arrival of 3xxx
Ryzen 4000
series included in computers already in Citylink
The price of MSI RTX 2070 has collapsed after the announcement of RTX 3xxx
Core i9 10 series is half the price of the same 9 series
The price of memory has been halved in Regard - it’s more expensive everywhere
New opportunities
Nvidia truly believes in the future of 3D reality, and therefore many innovations are aimed at supporting solutions coming out soon. The main improvements inside the GPU are aimed at optimizing work with 3D glasses and we can talk about this endlessly. Let's focus on the main thing.
First, it is a single-pass stereo image for VR. Thanks to the technology, the work of the GPU “in geometry” is reduced, which significantly increases performance without compromising quality. Secondly, the driver and video card take into account the geometric features of the lenses in 3D glasses and use special calculation algorithms.
New Sli mode with double bridges. Initially, many observers misinterpreted Nvidia's announcement of the new connection.
GeForce GTX 1080 can still be combined in a 3- or 4-Way configuration, but the emphasis is gradually shifting towards 2-Way, due to the sufficient performance of the GP104. Even one video card is superior to the GTX Titan X, and a pair is enough for 100 frames per second and 4K resolution in any game. The old Sli bridges can still be used, but greater resolutions are available with the new ones.
Since we're talking about the graphics processor, let's take a closer look at it. In fact, we have a middle-class solution, and until more expensive and powerful models are presented, the Nvidia GeForce GTX 1080 will be considered the fastest video card.
In GP104, each SM is paired with the Polimorph engine. But to meet modern demands, a new Multi-Projection block was installed in it.
The pair of SM and Polimorph engines form the TPC block.
Each SM contains four GPCs with 20 Pascal multiprocessors and eight memory controllers. Five GPCs form one SM. Each SM contains 128 Cuda cores and a 256 KB cache, 96 KB of shared memory and 48 KB of L1 cache, as well as eight texture units.
Any SM can work with a queue of up to 32 commands. Eight 32-bit G5X memory controllers remain on the outer edge. Boost-adjusted raw speeds are nearly double in GFOP compared to the GTX 980.
DX12 support was expressed in a new mode of operation of the task scheduler. The Pascal video core has dynamic allocation. To prevent the GPU from being idle when processing graphics data is finished, the scheduler assigns a computational task to the cores. Thus, the time of inactivity of the GP is reduced, and significantly.
Specifications
| Name | Radeon R9 Fury X | GeForce GTX 980 | GeForce GTX 980 Ti | GeForce GTX 1080 |
| Codename | Fiji | GM204 | GM200 | GP104 |
| Version | GCN 1.2 | Maxwell 2.x | Maxwell 2.x | Pascal |
| Technical process, nm | 28 | 28 | 28 | 16 |
| Size of core/cores, mm2 | 596 | 398 | 601 | 314 |
| Number of transistors, million | 8900 | 5200 | 8000 | 7200 |
| Core frequency, MHz | – | 1126 | 1000 | 1607 |
| Core frequency (Turbo), MHz | 1050 | 1216 | 1075 | 1734 |
| Number of shaders (PS), pcs. | 4096 | 2048 | 2816 | 2560 |
| Number of texture units (TMU), pcs. | 256 | 128 | 176 | 160 |
| Number of rasterization units (ROP), pcs. | 64 | 64 | 96 | 64 |
| Maximum fill rate, Gpix/s | 67.2 | 72 | 96.2 | 102.8 |
| Maximum texture sampling speed, Gtex/s | 269 | 144.1 | 176 | 257.1 |
| Memory type | H.B.M. | GDDR5 | GDDR5 | GDDR5X |
| Effective memory frequency, MHz | 500 | 1750 | 1750 | 2500 |
| Memory capacity, GB | 4 | 4 | 6 | 8 |
| Memory bus, bit | 4096 | 256 | 384 | 256 |
| Memory bandwidth, GB/s | 512 | 224.3 | 336.5 | 320.3 |
| Power, Pin connectors | 8 + 8 | 6 + 6 | 6 + 8 | 8 |
| Power consumption (2D/3D), Watt | -/275 | -/165 | -/250 | -/180 |
| CrossFire/Sli | V | V | V | V |
| Announcement price, $ | 649 | 500 | 650 | 599-699 |
| Replacement model | Radeon R9 295X | GeForce GTX 780 Ti | GeForce GTX Titan Black | GeForce GTX 980 |
⇡#Performance: Computing
The results of computing tests, unlike games, are far from being as uniform. Too much depends on the specifics of applications in which different accelerator characteristics take priority: the number of shader ALUs, clock speed, memory size and bandwidth. Of all the tasks, the TITAN X was faster than the GTX 1080 Ti only in ray tracing in LuxMark and mathematical functions in SiSoft Sandra. In other tests, there is either no difference between TITAN X and GTX 1080 Ti, or the latter has an advantage. There are also types of workloads that are still best handled by the AMD architecture. Video rendering in Sony Vegas 13 Pro and water surface simulation in CompuBench CL give preference to the Radeon R9 Fury X.
⇡#Clock speeds, power consumption, temperature, overclocking
In terms of the frequency mode and operating profile of the cooling system, the GeForce GTX 1080 Ti differs from the TITAN X. In the Crysis 3 test, the GPU frequency of the new product ranges from 1734 to 1772 MHz, and the limit value is even higher than that of the GTX 1080: 1873 MHz versus 1860 MHz. In TITAN X, the zone of typical frequencies under load, judging by our measurements, is located below: 1557–1671 MHz.
At the same time, naturally, the GTX 1080 Ti needs more efficient heat removal from the GPU. In addition to both cards increasing their default temperature target by 3°C compared to the GTX 1080 (85 vs. 82°C), the GTX 1080 Ti spins its fan to a higher RPM than the TITAN X (2628 vs. 2472 RPM) . It's not for nothing that NVIDIA removed the DVI port from the GTX 1080 Ti to make it easier for hot air to escape. The maximum supply voltage for GP102 also exceeds that for GP104: 1.062 and 1.05 V, respectively.
In terms of power, the GTX 1080 Ti and TITAN X are the same from a practical point of view, although the system bundled with the latter consumes 7 W less under a gaming load, which is explained by lower GPU and RAM frequencies. Both the GTX 1080 and the previous generation flagships (GeForce GTX TITAN X and Radeon R9 Fury X), and especially the GeForce GTX 1080, are more economical video cards.
To overclock the GTX 1080 Ti, we increased the maximum GPU supply voltage to 1.093 V, ran the CO fan at maximum speed (4815 rpm) and, as a result, achieved stable operation of the chip at a base frequency of 1600 MHz. The TITAN X sample we tested earlier was overclocked to 1617 MHz, but this difference can be safely attributed to the quality of the individual GPU crystals. Memory chips with a higher nominal frequency reached the 12 GHz mark for the first time in our experience. At the same time, the average GPU frequency that the video card supported in games was 1974 MHz, and the maximum was 2025 MHz. By the way, the peak GPU frequency in our experiments with overclocking the GTX 1080 was not much higher - 2126 MHz.
It is interesting to compare the overclocked GeForce GTX 1080 Ti and TITAN X in terms of temperature and power consumption. Although insignificant from a practical point of view, the GTX 1080 Ti has lower these parameters, despite the fact that the high RAM frequency on the GTX 1080 Ti covers the difference in the frequencies of the overclocked GPU. The credit for this goes to the enhanced power supply and cooling system of the new card.
⇡#Performance: overclocking
The effect of overclocking the GeForce GTX 1080 Ti is not as great as in the case of TITAN X, because the graphics processor here initially operates at higher frequencies. However, overclocking brought an additional 10% of points in 3DMark, and in games the frame rate increased by 7-12% depending on the resolution (the higher the resolution, the greater the difference).
| 3DMark (Graphics Score) | |||
| Permission | NVIDIA GeForce GTX 1080 Ti (1480/11010 MHz, 11 GB) | NVIDIA GeForce GTX 1080 Ti (1600/12310 MHz, 11 GB) | |
| Fire Strike | 1920 × 1080 | 28 003 | 30 571 |
| Fire Strike Extreme | 2560 × 1440 | 13 453 | 14 882 |
| Fire Strike Ultra | 3840 × 2160 | 6 594 | 7 334 |
| Time Spy | 2560 × 1440 | 9 480 | 10 515 |
| Max. | +11% | ||
| Average | +10% | ||
| Min. | +9% | ||
| 1920 × 1080 | |||
| Full screen anti-aliasing | NVIDIA GeForce GTX 1080 Ti (1480/11010 MHz, 11 GB) | NVIDIA GeForce GTX 1080 Ti (1600/12310 MHz, 11 GB) | |
| Ashes of the Singularity | MSAA 4x + TAA 4x | 58 | 63 |
| Battlefield 1 | TAA | 144 | 154 |
| Crysis 3 | MSAA 4x | 112 | 120 |
| Deus Ex: Mankind Divided | MSAA 4x | 53 | 59 |
| DiRT Rally | MSAA 4x | 128 | 139 |
| DOOM | TSSAA 8TX | 200 | 200 |
| GTA V | MSAA 4x + FXAA + Reflection MSAA 4x | 91 | 93 |
| Metro: Last Light Redux | SSAA 4x | 122 | 137 |
| Rise of the Tomb Raider | SSAA 4x | 85 | 92 |
| Tom Clancy's The Division | SMAA 1x Ultra + TAA: Supersampling | 127 | 140 |
| Total War: WARHAMMER | MSAA 4x | 77 | 77 |
| Max. | +13% | ||
| Average | +7% | ||
| Min. | +0% | ||
| 2560 × 1440 | |||
| Full screen anti-aliasing | NVIDIA GeForce GTX 1080 Ti (1480/11010 MHz, 11 GB) | NVIDIA GeForce GTX 1080 Ti (1600/12310 MHz, 11 GB) | |
| Ashes of the Singularity | MSAA 4x + TAA 4x | 47 | 51 |
| Battlefield 1 | TAA | 111 | 112 |
| Crysis 3 | MSAA 4x | 72 | 80 |
| Deus Ex: Mankind Divided | MSAA 4x | 34 | 38 |
| DiRT Rally | MSAA 4x | 96 | 103 |
| DOOM | TSSAA 8TX | 177 | 197 |
| GTA V | MSAA 4x + FXAA + Reflection MSAA 4x | 78 | 84 |
| Metro: Last Light Redux | SSAA 4x | 72 | 80 |
| Rise of the Tomb Raider | SSAA 4x | 51 | 57 |
| Tom Clancy's The Division | SMAA 1x Ultra + TAA: Supersampling | 91 | 101 |
| Total War: WARHAMMER | MSAA 4x | 64 | 71 |
| Max. | +12% | ||
| Average | +9% | ||
| Min. | +1% | ||
| 3840 × 2160 | |||
| Full screen anti-aliasing | NVIDIA GeForce GTX 1080 Ti (1480/11010 MHz, 11 GB) | NVIDIA GeForce GTX 1080 Ti (1600/12310 MHz, 11 GB) | |
| Ashes of the Singularity | Off | 57 | 63 |
| Battlefield 1 | 64 | 75 | |
| Crysis 3 | 50 | 55 | |
| Deus Ex: Mankind Divided | 38 | 42 | |
| DiRT Rally | 66 | 73 | |
| DOOM | 97 | 110 | |
| GTA V | 70 | 75 | |
| Metro: Last Light Redux | 64 | 72 | |
| Rise of the Tomb Raider | 61 | 69 | |
| Tom Clancy's The Division | TAA: Stabilization | 54 | 61 |
| Total War: WARHAMMER | Off | 52 | 57 |
| Max. | +17% | ||
| Average | +12% | ||
| Min. | +6% | ||
⇡#Conclusions
Contrary to what NVIDIA did with top GPUs in previous generations of accelerators, this time it seems that everything has been done to ensure that the GeForce GTX 1080 Ti is not only not inferior to the TITAN X in performance, but also receives additional advantages. The GTX 1080 Ti GPU suffers certain losses in the set of functional units compared to the GPU in the TITAN X, which is also not fully unlocked. But the absence of one RAM controller, eight ROPs and one of the 12 GB of memory is almost invisible in games and is fully compensated by a slight increase in GPU clock speeds and new GDDR5X chips operating at an effective frequency of 11 GHz. Oddly enough, even the computational capabilities of GP102 in the GTX 1080 Ti were not affected: the video card can aggregate four operations on int8 data in one clock cycle of the 32-bit CUDA core - just like TITAN X, for which, as an accelerator for pre-trained deep learning networks, this is one of the priority functions.
Thus, a small compromise in the specifications of the GeForce GTX 1080 Ti looks more like an attempt to reduce the cost of the video card than to differentiate it from TITAN X. Moreover, NVIDIA has improved the “physical” parameters of the board - cooling efficiency and power system efficiency, ultimately depriving TITAN X of any remaining practical values. The latter, at a price $500 higher, can only give the buyer the moral satisfaction of owning a slightly more powerful GPU. TITAN X also has specific scenarios for computational tasks, in which it is absolutely necessary to have 12 GB of RAM.
It looks like NVIDIA is willing to sacrifice TITAN X in order to give a decisive battle to AMD's Vega architecture in the near future, which will revive competition in the upper segment of gaming video cards. It is possible that an updated version of TITAN X based on GP102 will soon appear, which will receive a fully functional version of the GPU, and maybe the “big Pascal” - GP100 - will still find its way into prosumer devices.
Whatever happens in the future, for now NVIDIA still enjoys the lack of competition among top video accelerators and the GeForce GTX 1080 Ti can only be compared with its predecessors in the leadership position - the GTX 980 Ti (GTX TITAN X) and GeForce GTX 1080. Judging by our tests, the Pascal family received a better flagship than Maxwell. The GTX 1080 Ti delivers 25-33% higher frame rates in games compared to a lower-class card (GTX 1080), something the GTX 980 Ti couldn't boast. As a result, those gamers who did not find the GTX 1080 a convincing offer at the time may decide to upgrade, even if the once mighty GTX TITAN X is to be replaced, which the GTX 1080 Ti is 54-67% ahead of.
The GTX 1080 Ti is the first GeForce to be unconditionally capable of 4K gaming, a goal that graphics cards since the GeForce GTX 780 Ti and Radeon R9 290X have unsuccessfully strived for. Before the GTX 1080 Ti, only the TITAN X or a pair of GTX 1070/1080 in SLI mode could cope with this task, but the low price of the flagship compared to the TITAN X makes it a more profitable solution, devoid of all the costs of multi-processor configurations.
New king! Testing GeForce GTX 1080 Ti
NVIDIA continues to conquer new heights of graphics performance. AMD is still not ready to release Radeon Vega for the segment of older solutions that could compete with last year’s “green” flagships, and now new NVIDIA video adapters are rushing to empty the wallets of enthusiasts. The latest high-profile premieres were the GeForce GTX 1080 Ti and Titan Xp video accelerators. The second is an exclusive product with a small circulation and an extremely high price. But the GeForce GTX 1080 Ti will become the new flagship of mass production models, which significantly raises the performance bar. We'll talk about it in this article.
GeForce GTX 1080 Ti is based on the GP102 graphics processor, built on the Pascal architecture. This chip was previously the basis for the ultra-powerful Titan X (Pascal) solution, and the even faster Titan Xp was recently released. The GP102 processor has six graphics clusters of five SM multiprocessors instead of the four clusters of the GP104 (GeForce GTX 1080). The internal structure of the clusters is absolutely identical; it is discussed in detail in a separate review of the Pascal architecture.
In its full configuration, the GP102 has 3840 CUDA stream processors and 240 texture units. Memory bandwidth is increased to 384 bits thanks to 12 memory controllers. The crystal is produced using a 16-nm process technology and has 12 billion transistors. Perhaps due to such a complex structure, the first Titan X based on GP102 received an incomplete configuration with 3584 cores and 224 texture units. The new serial flagship GeForce GTX 1080 Ti has a similar configuration. The newcomer also has 88 ROP blocks with 2.8 MB of L2 cache (GP104 has only 2 MB). The flexible structure of the new memory system made it possible to achieve an unusual bus configuration with a total capacity of 352 bits, which is why the memory capacity of the video card is 11 GB. Thus, a certain advantage of the Titan X, which has 12 GB, is retained. However, for the average player, even 11 GB is enough for the most modern games. At the same time, the memory of the GeForce GTX 1080 Ti operates at a record frequency of 11 GHz.
After the release of the GeForce GTX 1080 Ti, NVIDIA introduced the full version of the GP102 in the form of the Titan Xp. GeForce GTX 1080 Ti is identical to Titan Xp in terms of operating frequency configuration. The GPU base speed of 1480 MHz is ensured even under the most extreme operating conditions. GPU Boost technology increases core frequency while power and temperature limits are not exceeded. The average Boost value is 1582 MHz, but the maximum is traditionally higher.
You can compare all current solutions based on GP102 with their predecessors using the table below.
| Video adapter | Titan Xp | Titan X | GeForce GTX 1080 Ti | GeForce GTX 1080 | GeForce GTX 980 Ti |
| Core | GP102 | GP102 | GP102 | GP104 | GM200 |
| Number of transistors, million pieces | 12000 | 12000 | 12000 | 7200 | 8000 |
| Technical process, nm | 16 | 16 | 16 | 16 | 28 |
| Core area, sq. mm | 471 | 471 | 471 | 314 | 601 |
| Number of stream processors | 3840 | 3584 | 3584 | 2560 | 2816 |
| Number of texture blocks | 240 | 224 | 224 | 160 | 176 |
| Number of rendering units | 96 | 96 | 88 | 64 | 96 |
| Core frequency, MHz | 1480–1582 | 1417–1531 | 1480–1582 | 1607–1733 | 1000–1075 |
| Memory bus, bit | 384 | 384 | 352 | 256 | 386 |
| Memory type | GDDR5X | GDDR5X | GDDR5X | GDDR5X | GDDR5 |
| Memory frequency, MHz | 11008 | 10008 | 11008 | 10010 | 7010 |
| Memory capacity, MB | 12288 | 12288 | 11264 | 8192 | 6144 |
| Supported DirectX Version | 12.1 | 12.1 | 12.1 | 12.1 | 12 |
| Interface | PCI-E 3.0 | PCI-E 3.0 | PCI-E 3.0 | PCI-E 3.0 | PCI-E 3.0 |
| Power, W | 250 | 250 | 250 | 180 | 250 |
The GeForce GTX 1080 Ti combines all the advantages of the Pascal architecture and supports technologies that are also relevant for the GeForce GTX 1080. Among them, high bandwidth efficiency thanks to advanced data compression methods in the frame buffer. Tile caching also provides a certain advantage, when the image is divided into fragments (tiles), which allows you to optimize the processing of geometry and pixels for each tile. This method allows you to use L2 cache resources more efficiently, reducing the load on external memory.
And all this is combined with a 352-bit bus and a record frequency of GDDR5X chips. With a physical throughput of 484 GB/s, the theoretical peak can reach 1.2 TB/s. However, the effectiveness of additional improvements greatly depends on each specific application.
NVIDIA solutions are highly optimized for DirectX 11, backed by years of support from engineers and programmers. Thanks to the availability of convenient tools, this API still remains the main one for many game developers. DirectX 12 gives you more control over GPU resources, allows you to better use the multi-threading capabilities of processors and more efficiently perform combined operations, combining graphics and other calculations, including thanks to Async Compute. NVIDIA Pascal generation solutions fully support all the capabilities of the new API at the hardware level, including promising asynchronous computing. But the implementation of DirectX 12 requires more costs on the part of developers, and the benefits from this are still small. Many games only benefit from DirectX 12 performance benefits several months after release through a series of updates. Examples of this are Rise of The Tomb Raider, Deus Ex: Mankind Divided and the seasonal Hitman. And these are good examples, because sometimes DirectX 12 only brings a drop in fps and nothing more.
NVIDIA is trying to improve the situation by providing software support for its products and introducing new developer tools for debugging code. Microsoft recently introduced the PIX utility for debugging under DirectX 12, and NVIDIA offers nSight Visual Studio Edition 5.3 and NVIDIA Aftermat. These programs will help developers work with DirectX 12 when finding and fixing errors.
The situation with software support is also constantly improving. Along with the launch of the GeForce GTX 1080 Ti, a new driver was released that seriously speeds up a number of games running DirectX 12. And improvements to the drivers made it possible to achieve noticeable acceleration in the already mentioned Rise of The Tomb Raider, Hitman and other games.
The company is also expanding the use of GameWorks effects by introducing support for the new DirectX 12. This initiative is based on a toolkit of ready-made graphics technologies and effects that are easy to integrate and can improve the visual performance of the game. NVIDIA GameWorks support is present in many modern games; certain technologies are supported in the latest versions of the Unreal Engine, Unity and Amazon Lumberyard engines.
The HBAO+ background shading technology has been widely used, which gives a higher quality and more detailed picture compared to SSAO and other algorithms. The technology for simulating volumetric lighting, Volumetric Lighting, has shown itself to be excellent. This is exactly the kind of lighting you saw in Fallout 4, and this game also already has integrated support for HBAO+ and physical effects with Flex-based shards.
Another GameWorks example is HairWorks technology, which allows you to simulate complex hair and fur. The technology was actively used in The Witcher 3: Wild Hunt. HairWorks now runs DirectX 12, and we may soon see new games with similar effects.
Among the new effects, Turf Effects should be highlighted for simulating complex grass with separate blade geometry.
The technology is used in Tom Clancy's Ghost Recon: Wildlands. The game itself shows an excellent picture and supports many other NVIDIA technologies. Learn more about the technological capabilities of Ghost Recon: Wildlands with a demonstration of GameWorks effects in a special video:
Simulating different dynamic environments is a complex task that different GameWorks tools can solve. Water that changes dynamically in real time can be implemented based on WaveWorks, Flex technology allows you to simulate the behavior of crumbly solids, Flow simulates flows and gases, Blast can be used for destructible objects, NvCloth allows you to implement realistic fabrics.
Most techniques work under DirectX 12, including the promising Flex and Flow. As an example of Flow implementation, here is a short technical video demonstrating the new technology.
Optimization for the new API includes support for asynchronous computing, which allows the GPU to more efficiently calculate such effects and get real acceleration on new video cards. And according to NVIDIA experts, Async Compute support provides a twofold acceleration in rendering Flex effects on the same GeForce GTX 1080.
The new GameWorks supports DirectCompute and runs on any graphics device. So you can play with new effects on AMD video cards. However, at the moment almost all effects work on Radeon.
Neural networks in artificial intelligence systems have become a new area of application for GPUs. In particular, NVIDIA is active in the field of artificial intelligence devices for self-driving cars. But there are certain practical benefits for ordinary PC users. Based on self-learning neural networks, face recognition technologies and advanced image scaling methods are implemented.
The GeForce Experience software package is gaining new capabilities, which allows you to automatically optimize game settings for your video card, update drivers and use additional functions.
Part of these features is video capture using ShadowPlay. The encoding is handled by the built-in video unit in the GPU, there is practically no drop in performance in the game, the output video quality is very high, 4K resolution is supported. And the new ShadowPlay Highlights extension makes it easier to capture the brightest moments. The program automatically tracks the player's activity and records the most intense gaming episodes.
Ansel gives players incredible opportunities. The application allows you to take screenshots from unusual angles thanks to the free camera, apply color filters, create panoramic pictures and increase the resolution of the final image.
Players liked Ansel's features. It is noted that users have already taken more than 200 thousand screenshots in The Witcher 3 using Ansel.
In addition to the above game, Ansel is supported in War Thunder, Wathc Dogs 2, Dishonored 2, Mass Effect: Andromeda and many other games. The SDK is publicly available, so every developer can integrate Ansel into their project.
Virtual reality VR is one of the rapidly developing areas. VR systems have found application not only in the entertainment industry, but also in specialized tasks. In an article on Pascal technologies, we wrote that special image projection techniques can save GPU resources and speed up VR processing. NVIDIA VRWorks technologies provide VR rendering acceleration from 20% to 60% or more.
The combination of maximum performance and technological support makes the GeForce GTX 1080 Ti the best choice for those who want to touch the world of virtual reality.
After reviewing the general architectural features and technological innovations, it's time to take a look at the reference design graphics card.
