Nvidia has unveiled its brand new GTX 1080 graphics card at a new price point of $599 MSRP. We already knew the specifications and timeline of the launch weeks before but Nvidia still managed to impress us in terms of performance numbers. While the benchmarks come directly from Nvidia (third party reviews aren’t out yet) they do serve to highlight the amazing potential of Pascal architecture and the FinFET process.
Nvidia launches the Pascal based GTX 1080 - doubles the performance of the GTX 980
Nvidia has achieved something truly remarkable with the GTX 1080. According to the official numbers, the card is roughly equivalent to a GTX 980 in SLI. That is a performance increase of 100%! In specific VR use-cases, the performance per watt increase can actually be as high as 200%. So at any given time you are looking at a performance of 2x to 3x depending on the application in question. The Geforce GTX 1080 is the first graphics card to utilize the brand new GDDR5X standard from Micron.
Rocking the GP104-400-A1 GPU flavor the card is expected to feature 2560 CUDA Cores and if the chart revealed by nvidia is to-scale then we are looking at around a 20% increase over the GTX 980 Ti (at the clock speeds shown in the demo). Nvidia’s Pascal architecture is manufactured on the 16nm FinFET process (which inherently brings a 2x Perf/Watt increase). GP100 focused on increasing gimped FP64 performance (that Maxwell suffered from) so the GP104 focuses primarily on single precision performance (good news for games).
Official Gaming Performance Benchmarks of the GTX 1080
Nvidia posted some early figures of their next-generation graphics card on the Geforce website and they tested three different applications: a VR demo, a DX12 and an overall intensive AAA title. The Witcher 3: Wild Hunt is a beautiful game but it is well known as a card killer. The GTX 1080 manages to get almost twice the amount of performance as its predecessor clocking somewhere around 1.6x. Keep in mind that while we do not know the exact resolution these were run at, intensive games like the Witcher are GPU-bound, which makes the benchmark legitimate (low resolutions plus light load equals a CPU bound scenario in which GPU performance cannot be measured accurately).
The DirectX 12 title here is the latest Tomb Raider. The GTX 1080 over here manages to achieve 1.7x the performance of its predecessor, which is pretty impressive. The last generation of Nvidia GPUs had some trouble under DX12 with some specific games, and it remains to be seen whether Pascal architecture managed to correct that flaw. Finally, we have the VR "Barbarian" demo in which the GTX 1080 is able to achieve not 2x but almost 3x the performance of its older counterpart! That is incredibly impressive and is the result of Nvidia employing new technologies that help lighten the VR workload (detailed below).
Improved Power Engine
Power supplied to the GPU is absolutely essential in any architecture. Many AIB models which provide a higher overclocking headroom actually improve the capabilities of the card by providing a more stable power supply. The GTX 1080 builds upon this ideology and aims to increase not only the power efficiency of the GPU under heavy workloads but also provide a more consistent and stable power supply at the same time. The GTX 1080 is actually 80%+ power efficient under a full workload, while as the 980 did not quite reach that mark.
The new card is also able to limit the ripples in the power supply to just 120mV which is almost twice as power efficient as the card it succeeds - which has ripples of 209 mV peak to peak. This new power efficiency not only allows Nvidia to provide an incredibly powerful GPU with just 180W TDP and a single 8-pin connector, but also provide more overclocking headroom. This is something that was readily apparent in today's demo, which brings us to the overclocking aspect of the 1080:
Incredible Overclocking Potential of the GTX 1080
The official clocks of the GTX 1080 are listed as 1733 Mhz, but the demo that was shown on the Nvidia special event was actually running at a cool 2.1 Ghz. And that too on an air cooler! This means that the demo was actually running almost 23% higher than the official boost clock (which is already 40% higher than Maxwell by the way). Needless to say we are looking at some pretty impressive overclocking headroom over here. The shift to FinFETs is partly to be thanked for the high clock rate achievable and the fact that the chip can remain so cool while doing so. It goes without saying that the graphics card used in the demo probably had a hand-picked GPU so whether that exact amount of OC can be recreated on air with off the shelf GTX 1080s remains to be seen.
Support for HB SLI Bridge Technology
Nvidia also sneakily launched a brand new iteration of SLI bridges, dubbed the "HB SLI Bridge", which I assume stands for High Bandwidth SLI Bridge. The official Geforce site has the following to say about the new tech:
NVIDIA’s new SLI HB bridge doubles the available transfer bandwidth compared to the NVIDIA Maxwell™ architecture. Delivering silky-smooth gameplay, it’s the best way to experience surround gaming—and it’s only compatible with the GeForce GTX 1080.
So these new HB SLI bridges will be able to offer twice as much transfer bandwidth as opposed to the older variants. AMD shifted long ago to a bridge-free approach and it would have been nice to see Nvidia following suite. SLI Bridges usually do not require a high amount of bandwidth since they are only used for syncing and timing so the true impact of this change remains to be seen.
Simultaneous Multi Projection Technology
Much of the performance increase you aw in the VR benchmark above is attributable to this new tech in Pascal based GPUs. The GTX 1080 introduces "Simultaneous Multi Projection" which allows the GPU to create different viewpoints of the same scene with significantly decreased performance hit. Virtual Reality is the next frontier that GPUs must conquer and outputting two different viewpoints simultaneously without incurring a huge performance penalty is one of the major challenges. With this new tech, Nvidia seems to have made significant progress towards the same:
The first frame shows the VR demo running without the "single pass stereo" filter. The GPU is outputting frame rates consistently around the 60 fps mark. This is however low enough to cause motion sickness in VR setups. Once the Single Pass Stereo filter is turned on however, the frames instantly jump to the adequete 90 fps mark and remain there consistently. Nvidia achieved this using some serious technical wizardry and is something which will allow it to get increased performance in VR titles.
Official Geforce GTX 1080 and Geforce GTX 1070 Specifications
| WCCFTech | Nvidia Geforce GTX 1080 | Nvidia Geforce GTX 1070 |
|---|---|---|
| Architecture | Pascal | Pascal |
| Transistors | 7.2 Billion | 7.2 Billion |
| CUDA Cores | 2560 | 1920 |
| Core Clock | 1607 Mhz | TBA |
| Boost Clock | 1733 Mhz | 1683 Mhz |
| Memory Type | G5X (GDDR5X) | GDDR5 |
| Memory Speed | 10 Gbps | 8 Gbps |
| Memory Configuration | 8GB | 8GB |
| Bus Width | 256-bit | 256-bit |
| Memory Bandwidth | 320 GB/s | 256 GB/s |
| Multi Projection | Yes | Yes |
| HB SLI Bridge Support | Yes | Yes |
| Nvidia GPU Boost | 3.0 | 3.0 |
| DirectX 12 Feature Level | 12_1 | 12_1 |
| OpenGL | 4.5 | 4.5 |
| Vulkan API | Yes | Yes |
| Maximum Digital Resolution | 7680x4320@60Hz | 7680x4320@60Hz |
| Display Connectors | DP 1.42, HDMI 2.0b, DL-DVI | DP 1.42, HDMI 2.0b, DL-DVI |
| HDCP | 2.2 | 2.2 |
| Power Draw | 180W | 150W |
| Power Connector | Single 8-Pin | Single 8-Pin |
| Maximum Operating Temp | 94 C | 94 C |
| Partner Price (MSRP) | $599 | $379 |
| FE Price (MSRP) | $699 | $449 |
