To be more precise, the GeForce GTX 1070 and GeForce GTX 1080 is sublimely the thousandth generation. “Green” true to their traditions. The company previously produced the video card, which at one time was considered the fastest, but some time later within the same line have gone by the wayside. For example, a couple of GeForce GTX 770/780 later replaced on the pedestal of a faster version of the GeForce GTX 780 Ti/TITAN. Almost two years ago went on sale the GeForce GTX 970/980, but then saw the light of truly top-end Maxwell solutions, the GeForce GTX 980 Ti/TITAN X. Someone like the cycle is like, because I want to get everything at once, but there’s an undeniable fact: GeForce GTX 1080 at this point in time is the fastest single-chip gaming graphics card NVIDIA.
In the first part of the review I will review the main features of GPU GP104 and architecture Pascal in General. And will examine the performance of the reference version of the GeForce GTX 1080 Founders Edition.
NVIDIA GeForce GTX 1080 Founders Edition
GP104 processor, architecture and memory Pascal GDDR5X
If we describe the architecture of Pascal in one sentence, we obtain the following: cosmetic rework of Maxwell plus transition to the 16-nanometer process technology. Indeed, if we discard the use of FinFET, there are no fundamental innovations development NVIDIA does not offer. However, the CPU GP104 used in the GeForce GTX 1080, noticeably faster than last generation sublimina — GM204 (GeForce GTX 980). Just look at the peak computing power. Not the most productive Pascal chip is almost twice as fast as solutions of Maxwell. Judging by the number of GFLOPS, GP104 General has no equal among gaming graphics cards, as the processor GM200 GPU used in the GeForce GTX TITAN X is only 6144 GFLOPS.
The processor name
GP104
GM204
Process technology
16 nm
28 nm
Square crystal
314 mm2
398 mm2
The number of transistors
7.2 billion
5.2 billion
The number of GPC
4
4
The number of SM
20
16
The number of CUDA cores
2560
2048
The number of texture units
160
128
The number of ROP
64
64
Clock frequency
1607 (1733) MHz
1126 (1216) MHz
Memory bus
256-bit
256-bit
The second level cache
2048 KB
2048 KB
Power consumption
180 W
165 W
Peak computing power at the Boost frequency
8873 GFLOPS
4981 GFLOPS
The transition to the new workflow has dramatically reduced the usable area of the crystal, but it is seriously increased transistor budget GPU. It is easy to calculate: for each square millimeter GP104 have ~23 million gate FinFET, while only 13 million GM204. GeForce GTX 1080 is the first graphics card that uses the processor with three-dimensional transistors, but not planar. It is obvious that 7.2 billion are not the limit for contract manufacturer NVIDIA — TSMC.
The second positive effect when switching to 16-nanometer production technology, of course, was a huge increase in clock frequency. Only one boost in the size-481 MHz gives a decent increase in speed. But compared to the GM204 in GP104 seriously increased CUDA cores and texture units.
The third plus a 16-nanometer low — power consumption. GeForce GTX 1080 is almost twice as powerful as the GeForce GTX 980 in the palna of computation, but requires only 15 watts more.
Graphics processor GP104
The construction of the key elements in GP104 same as the GM204. There is the main element in the streaming multiprocessor SM. Several such units are combined into a large computing cluster — cluster processor graphics GPC. Have Maxwell in the same GPC contains four SM, Pascal — five. Thus, NVIDIA continues to crush the blocks, increasing as the numerical number of compute elements (CUDA cores and texture units) and parallelism. If you consider that in each SM includes 128 CUDA and 8 texture units, the difference is tangible: 2560 cores vs 2048 and 128 vs 160 TMU TMU. Superimpose on all this the difference in frequency (481 MHz) — it turns out almost double superiority GP104 in computing power over GM204. Everything is logical.
The number of ROP units has remained the same — 64 units for the entire processor. And this is perhaps the weakest point GP104. Have GM200 such elements 96, which may affect the performance in Ultra HD resolution in some games. More low in Full HD and WQHD have a GeForce GTX 1080 sure everything will be fine.
Finally, the same principle works the cache. Each streaming multi-processor has 256 KB for registers, 96 KB shared memory and 48 KB of first level cache. The amount of second level cache, shared between all cluster graphs of CPU, is 2 MB.
Block diagram GP104
In addition to extensive transformations performed by switching on the 16-nanometer process technology, Pascal chips got improved engine Polymorph Engine 4.0, which includes hardware unit Simultaneous Multi-Projection (SMP) responsible for the generation of multiple projections of the same flow geometry. This is especially true in connection with the popularization of monitors and TVs with curved screens, and virtual reality helmets. The SMP-unit handles up to 16 projections of the same flow geometry. Or projection 32, divided into two center for devices VR.
Simultaneous Multi-Projection (SMP) in GP104
The next important element in the Pascal architecture — support for asynchronous computation. As practice shows, their performance impact in games with the software interface DirectX 12. The classical (static) scenario of the GPU (like Maxwell) suggests that two types of load — handling calculations and graphics (e.g., physics) is performed by different computing units for the same amount of time. But not the GPU always performs calculations of a different sort at the same time. As a result, some blocks simply left idle. Dynamic algorithm for asynchronous computation solves this problem by directing the unused elements of the processor for processing computing.
The second scenario involves the performance of tasks, time-critical performance. It can be an algorithm asynchronous time warp in virtual reality, for example. If he fails to to scan (scan out), then the frame is discarded. Therefore, the chip must be able to quickly switch between tasks. Pascal is NVIDIA’s first architecture that uses a “pixel interruption of the operations” (Pixel Level Preemption). The algorithm monitors the status of the rendering, if necessary, stops (aborts) work on the coloring of the triangle, gives a command to perform another action, and then returns to the coloring of the triangle, and at the same level at which it was interrupted.
Asynchronous computations in NVIDIA Pascal
The GeForce GTX 1080 has received eight gigabytes of memory standard GDDR5X. The main characteristic of the new “brain” is a real very high frequency of 2,500 MHz. Coupled with the 256-bit bus memory bandwidth is 320 GB/s. Not a record, of course, but it’s almost 100 GB/s more than the GeForce GTX 980. But less than that of the GeForce GTX 980 Ti.
Junior flagship — GeForce GTX 1070 — received ordinary GDDR5 (1750 MHz), but the same 8 GB. First gaming graphics card with HBM2, probably, will appear only next year.
One more thing. It is obvious that in 2016, the presence of 4 GB is the required minimum for a gaming graphics card. And therefore do not exclude that in the next year adapters of GeForce GTX 970/980 will slowly begin to “suffocate” from lack of memory. Unfortunately, the system requirements of modern games have grown disproportionately to the improvement of the quality of the graphics.
Video memory production Micron GDDR5X
The problem of consumption of resources and RAM, in particular, one of the most pressing today. Therefore, data compression algorithms fall as it is impossible by the way. Pascal immediately implemented in two new methods of image compression with the ratio of 4:1 and 8:1. The first case is effective when the difference in the color value of pixel unit is very small. The second algorithm combines compression 4:1 blocks of size 2×2 pixel with a twofold compression of the Delta between the blocks. On average, save up to 20% of the memory, if we compare the new architecture with Maxwell.
The effectiveness of texture compression
I was wary of the fact that game developers lately, frankly, “scored” support AMD CrossFire and NVIDIA SLI. And still “green” do not abandon this technology. However, in one array really only connect two GeForce GTX 1080. From support 3-Way SLI and Quad SLI in NVIDIA not completely abandon, but it is now necessary to obtain special software from key company representatives.
Full HD
WQHD @60Hz
WQHD @120Hz
Ultra HD
Ordinary bridge
Compatible
Compatible
—
—
HB-bridge
Compatible
Compatible
Compatible
Compatible
Each graphics card is equipped with dual “scallops”. You can connect and old floppy bridges, but their bandwidth, according to NVIDIA, there will be enough for resolutions of Full HD and WQHD. In other cases it is better to use a new hard connector.
SLI-an array of NVIDIA GeForce GTX 1080 Founders Edition
NVIDIA will release three variations for three types of distances between PEG ports on the motherboard. SLI controller now operates at an increased to 650 MHz frequency. That is why it is recommended to use the new bridges.

Special SLI bridges for NVIDIA GeForce GTX 1080 Founders Edition

Special SLI bridges for NVIDIA GeForce GTX 1080 Founders Edition

Special SLI bridges for NVIDIA GeForce GTX 1080 Founders Edition
The advent of DirectX 12 contributed to the emergence of new SLI modes: MDA (Multi Display Adapter) and Linked Display Adapter (LDA). The second type has two Podgorny: Implicit, which is used by programmers of NVIDIA, and Explicit, which is used by the developer of the computer game. The main and long-awaited feature of the LDA-mode was the possibility of combining the array of video memory in a single Bank. However, keep in mind that work a similar algorithm is accompanied by a certain delay, since the same adapter will have to seek the memory of another via the PCI Express bus. MDA-type involves a classic principle of operation, wherein each GPU works only with its memory chips. LDA-mode is intended to merge cards with identical chips, and MDA — with different, including from different manufacturers at all. The second case requires a precise optimization of game code. Studying modern pipeline, you can immediately assume that anybody will not do it.
SLI-an array of NVIDIA GeForce GTX 1080 Founders Edition
Of the most important innovations endorsed a new type of vertical sync Fast Sync. Describe in detail the principle of operation will not function. The main thing — the result. The algorithm allows to significantly reduce delays in the output image on the screen. To do this, use the optional third frame buffer, called Last Rendered Buffer (LRB).
Delay vertical sync
Finally, the display controller GP104 received the support of several important standards. For example, BT.2020 with support 10/12 bit and real HDR. I/O panel the GeForce GTX 1080 are equipped with HDMI 2.0 and 1.4 b DisplayPort.
“Green” has presented several technologies and functions. So, paid attention to the sound in VR. Vivid video attached. NVIDIA introduced the platform ANSEL, carefully and creatively able to work with screenshots.
On paper, Pascal and her first-born looks very interesting:
- 16-nanometer process technology has significantly increased the performance of the GPU. Including due to the high clock frequency.
- Engine-Simultaneous Multi-Projection improved and accelerated image processing in virtual reality devices.
- Cache resources and memory are utilized more economically.
- SLI technology now adds VRAM.
- Improved asynchronous computation.
- Introduced support BT.2020 and HDR.
Not so little for a new generation.
NVIDIA GeForce GTX 1080 Founders Edition
To our test lab arrived the reference version of the GeForce GTX 1080. It is now called Founders Edition. Most unexpected is the reference modification is now more expensive custom bikes. In Russia, the Founders Edition was estimated at 55 000 rubles. To put it mildly, not cheap. Full list of specifications given below.
NVIDIA GeForce GTX 1080 Founders Edition
NVIDIA GeForce GTX TITAN X
NVIDIA GeForce GTX 980 Ti
NVIDIA GeForce GTX 980
The name of the chip
GP104
GM200
GM200
GM204
Process technology
16 nm
28 nm
28 nm
28 nm
The number of transistors
7.2 billion
8 billion
8 billion
5.2 billion
The number of GPC
4
6
6
4
The number of SM
20
24
22
16
The number of CUDA cores
2560
3072
2816
2048
The number of texture units
160
192
176
128
The number of ROP
64
96
96
64
The second level cache
2048 KB
3072 KB
3072 KB
2048 KB
Core frequency
1607 (1733) MHz
1000 (1075) MHz
1000 (1075) MHz
1126 (1216) MHz
Memory
GDDR5X
GDDR5
GDDR5
GDDR5
Amount of memory
8 GB
12 GB
6 GB
4 GB
Memory interface
256-bit
384 bit
384 bit
256-bit
Memory frequency
2500 (10 000) MHz
1750 (7000) MHz
1750 (7000) MHz
1750 (7000) MHz
Memory bandwidth
320 GB/s
336,5 GB/s
336,5 GB/s
224 GB/s
TDP
180 W
250 W
250 W
165 W
Food
8 pin
6+8 pin
6+8 pin
6+6 pin
Interfaces
1x DVI
3x DisplayPort
1x HDMI
1x DVI
3x DisplayPort
1x HDMI
1x DVI
3x DisplayPort
1x HDMI
1x DVI
3x DisplayPort
1x HDMI
The recommended price at the time of the announcement
$700 (55 000 RUB.)
$1000
$650
$550
The actual price
N/A
$1000
$600
$500
The GeForce GTX 980 Ti and GeForce GTX TITAN X and more CUDA cores and texture units. However, due to the different design of the cluster processor graphics their efficiency is slightly lower than the GeForce GTX 1080. And don’t forget that the GPU itself is “hustler” at a very high clock frequency.
Reference, he is a Founders Edition, very beautiful. Personal opinion. It is clear that for the graphics card, this fact does not matter, the main thing — efficiency and reliability. And yet. The card itself turned out in standard lengths — 267 mm.
NVIDIA GeForce GTX 1080 Founders Edition
The adapter takes up two expansion slots. Closer to the outputs there is a pair of connectors for connecting the SLI bridge. The graphics card you must connect one 8-pin female wire from the power supply.
NVIDIA GeForce GTX 1080 Founders Edition, top view
The flip side Founders Edition got beklet. That’s fair, we in the editorial was not able to understand what material it is made. Kind of plastic. Or metal. However, not much difference. Part backplate removed. So, according to engineers NVIDIA, the graphics card will fare better paired with another GeForce GTX 1080. Something similar was used in the reference version of GeForce GTX 980.

The downside of the NVIDIA GeForce GTX 1080 Founders Edition

The downside of the NVIDIA GeForce GTX 1080 Founders Edition

The downside of the NVIDIA GeForce GTX 1080 Founders Edition
On I/O panel is one DVI, three DisplayPort and one HDMI. All the rest of the space is occupied by a grille through which the heated air is output.
Video outputs NVIDIA GeForce GTX 1080 Founders Edition
PCB from GeForce GTX 1080 simple. Graphics processor tied with eight memory chips GDDR5X. Today only Toshiba produces. For meals GPU answer five phases.
PCB NVIDIA GeForce GTX 1080 Founders Edition
The cooling system of the Founders Edition is the same as that of other references. Namely, the radiator is used with a vaporization chamber and a tangential blower.
Cooling system NVIDIA GeForce GTX 1080 Founders Edition
Under load turbine spins up to ~2150 rpm, which is 54% of maximum possible speed. Cooling system while keeping the temperature at 81 degrees Celsius. The maximum rate for GP104 — 92 degrees Celsius. The noise was noticeable. Distance meter measuring device is fixed to 41.5 dB. It’s louder than the reference GeForce GTX 980, but quieter than the reference versions of GeForce GTX 980 Ti and GeForce GTX TITAN X. I Note that the fan rotates all the time. Apparently, the driver for the press is not activated semi-passive mode of operation of the cooler when the impeller starts to spin only when reaching a certain GPU temperature.
Chokes “podsvechivayut”.
Monitoring of the performance of the NVIDIA GeForce GTX 1080 Founders Edition
Besplat well inhibits the heat. So the card will not heat the other components. The hot item from the insights of steel memory chips GDDR5X. But nothing critical. The reference GeForce GTX 1080 can be called relatively cool.

Heating of the NVIDIA GeForce GTX 1080 Founders Edition

Heating of the NVIDIA GeForce GTX 1080 Founders Edition
Announced for the GeForce GTX 1080 maximum consumption corresponds to reality. The flagship Pascal is located exactly between the GeForce GTX 980 and GeForce GTX 980 Ti.
Energy consumption under load NVIDIA GeForce GTX 1080 Founders Edition
With overclocking there are a couple of nuances. First, at the time of writing, the process of frequency was available only through utilities EVGA PrecisionX 16. For example, MSI Afterburner did not react to the change in the position of the sliders. Secondly, the utility GPU-Z, the latest version of which already recognize the GeForce GTX 1080 does not display frequency after overclock. Presumably, the updated algorithm GPU Boost 3.0. The third version of the technology more closely adapts to the overclocking potential of the chip. It is bearing fruit.

3.0 GPU Boost compared to GPU Boost 2.0

3.0 GPU Boost compared to GPU Boost 2.0
I, for example, to the nominal GeForce GTX 1080 could fling another 271 MHz chip and 256 MHz for memory. Under load the GPU, if you believe the EVGA PrecisionX 16 gauge, hovered around 2050-2076 MHz. There is a stable 2 GHz! I will continue to explore the overclocking potential of the GeForce GTX 1080, but we can already say that the novelty with this feature all right.
Overclocking the NVIDIA GeForce GTX 1080 Founders Edition
In Heaven benchmark 4.0 manual overclock increased number of FPS: 11% in Full HD resolution and 9% respectively in WQHD.
1080p
1440p
NVIDIA GeForce GTX 1080 @1607 (1733) / 2500 (10 000) MHz
109 FPS
64,4 FPS
NVIDIA GeForce GTX 1080 @1878 (2004) / 2756 (11 024) MHz
120,9 FPS
Of 70.4 FPS
Gaming performance
Test stand:
- CPU: Intel Core i7-4790K @4.5 GHz
- CPU cooler: Corsair H75
- Motherboard: MSI Z97 MPOWER
- Drive: Patriot SSD 480 GB Blast
- RAM: DDR3-2133 2x 8 GB
- Power supply: Corsair HX850i, 850W
- Operating system: Windows 10 x64
At the presentation, CEO Jen-sen Huang stated that the GeForce GTX 1080 faster SLI array GeForce GTX 980 and GeForce GTX TITAN performance X. And it did not disappoint! In the Heaven benchmark the new flagship ahead of the NVIDIA GeForce GTX TITAN X to 25.7% in 1080p, compared to 67.3% in 1440p and 13.3% in 2160p. Interesting to have the variation. In the resolution of 2K video card feels like a fish in water.
The test results of NVIDIA GeForce GTX 1080 Founders Edition in Heaven
In an old Metro: Last Light no 30 FPS in Ultra HD! However, the rest of the graphics look even worse. So, the GeForce GTX 1080 is ahead of the GeForce GTX 980 Ti 44.6% in Full HD, 50.4% in WQHD and 48.7% in Ultra HD. A serious advantage! It is a pity that the game itself, in addition to the benchmark, does not represent any value.
The test results of NVIDIA GeForce GTX 1080 Founders Edition Metro: Last Light
In Battlefield 4, sky-high FPS in all resolutions. Watch almost 60 fps in Ultra HD, whatever say, but nice. In this mode, the faster GeForce GTX 1080 and GeForce GTX 980 Ti by 35%.
The test results of NVIDIA GeForce GTX 1080 Founders Edition in Battlefield 4
Tomb Raider is another game in 2013, which at maximum settings “bends” in Ultra HD, even the GeForce GTX 1080.
The test results of NVIDIA GeForce GTX 1080 Founders Edition in tomb Raider
Dragon Age Inquisition — the only list a game in which the GeForce GTX 1080 does not have managed to become the first. In this application gained the upper hand “Titan”.
The test results of NVIDIA GeForce GTX 1080 Founders Edition in Dragon Age Inquisition
Far Cry 4 all video cards returned on their places. GeForce GTX 1080 is leading by a large margin. Given our settings, the flagship Pascal gives more than 30 fps. Playable.
The test results of NVIDIA GeForce GTX 1080 Founders Edition in Far Cry 4
GTA V — very protsessorostroeniya game. Do not be surprised if even at Full HD with 8x anti-aliasing FPS “rested” in the CPU. However, this issue I will devote a separate article. In any case, 1440p and 2160p GeForce GTX 1080 bypasses its nearest competitor by 24% and 23%, respectively.
The test results of NVIDIA GeForce GTX 1080 Founders Edition in GTA V
In the third “the Witcher”, the new product showed excellent FPS at the maximum graphics quality in Ultra HD resolution!
The test results of NVIDIA GeForce GTX 1080 Founders Edition “the Witcher 3: wild Hunt
Calculations
GeForce GTX 1080 gaming graphics card. GP104 performance in double precision calculations, about the performance when working with FP32 is the same as GM104, — 1/32. Card does not stand out against the top-end Maxwell-brothers. Proof — below.
The test results of NVIDIA GeForce GTX 1080 Founders Edition in SiSoftware Sandra
And yet novelty, as in games, in professional applications is much faster and the GeForce GTX TITAN X and GeForce GTX 980 Ti. And even the main competitor in the face of Radeon R9 X in most patterns beaten to it.
The test results of NVIDIA GeForce GTX 1080 Founders Edition in SPECviewperf 12
A similar situation in a comprehensive benchmark CompuBench. Judging by the results, the GeForce GTX 1080 is a great solution for hashing!
The test results of NVIDIA GeForce GTX 1080 Founders Edition in CompuBench
Ray tracing has always produced better on NVIDIA graphics cards. Specifically, in the pattern Room the new leader “giraffes” ahead of the GeForce GTX TITAN X is only 7%.
The test results of NVIDIA GeForce GTX 1080 Founders Edition in LuxMark
In conclusion
It has long been clear that video card manufacturers had stayed at the 28-nanometer process technology. Let me remind you, the first video card was released in late 2011. In the end, the GeForce GTX 1080 having excellent characteristics and a very high frequency, “brings” the former flagship single-chip 30-40% depending on the specific application. If you do not abuse the smoothing, with one such videokarty quite possible to play modern games in Ultra HD. The new product came out cold and well dispersed. What else is necessary for happiness?
At least $ 700. Although Europe and Russia have traditionally inflated price: 789 Euros and 55 000, respectively. For the first time a reference sample is sold at a recommended price more expensive than custom. Soon will appear the so-called MSRP version that NVIDIA was estimated at $ 600. More precisely, from $ 600. That’s interesting! At this price, most likely, will release the cheapest models with ugly plastic cooler, shumnyi and without factory overclock. Something tells me that cool and overclocked at the factory custom bikes with larger radiators and more fans will cost a potential buyer a lot of the same $ 700, if not more.
The test lab is only beginning to explore new Pascal’I. Therefore, the first part I will abstain from categorical conclusions and statements. There are facts — they are listed. In the second part we consider the performance of the GeForce GTX 1080 in a large number of modern games, and also more pay attention to overclocking of the graphics card. Interesting to know, how to behave in the new architecture with DirectX 12. It is necessary to answer one important question: which processor is “pumped” GeForce GTX 1080? Finally, we have to go custom versions of the flagship “Pascal”.
NVIDIA GeForce GTX 1080 Founders Edition
Pros:
Cons:
- High performance. With a quick odnosi NVIDIA;
- Effective cooling;
- Good overclocking potential;
- Support BT.2020 and HDR.
- Overpriced for the Founders Edition.