Today’s the day. Reviews of Nvidia’s newest, greatest and most powerful graphics card, the GeForce GTX 1080, are about to hit the ‘net. But there’s also more to this card than its raw performance data. Here’s what’s going on under the hood of Nvidia’s big, bold attempt at cornering 4K and VR gaming in 2016.
Here’s All The Data You Want To See
This is the fastest graphics chip yet in terms of pure, unbridled Hertz. 1.61GHz is the GTX 1080’s base clock and 1.73GHz is its turbo boost clock, while Nvidia is also talking about chip overclocks of past 2GHz, with a demonstration running at 2144MHz air-cooled at 67 degrees Celsius. The GTX 1080’s Pascal GP104 GPU has 7.2 billion transistors, 20 streaming multiprocessors with 128 cores each and 2560 CUDA cores.
Nvidia is heavily talking up the potential of the GTX 1080’s brand new GDDR5X 10Gbps memory, of which it has 8GB on board. The G5X memory, made exclusively by Micron, is the fastest non-HBM memory used on a graphics card, and Nvidia’s own delta colour compression allows up to 8:1 memory compression in some instances, a doubling of the same potential from last generation’s Maxwell cards like the GTX 980.
That memory compression, as well as the faster overall memory bandwidth, means games require massively less resources than the previous card — we’re talking 1.4x memory bandwidth improvement versus last gen, and 1.2x memory compression improvement, for an overall improvement of 1.7x. Pascal is 11 per cent more efficient in Star Wars: Battlefront, for example, and 21 per cent more efficient in Battlefield 4.
What Makes The Founder’s Edition A Founder’s Edition?
The Founder’s Edition is the board designed and built by Nvidia. Founder’s Edition is essentially a new name for the ‘reference’ design, but at the same time it’s not just the most basic offering possible. It uses premium materials, a metal backplate and vapor-chamber cooling with the tried-and-tested blower fan design. Partners will be using a custom design, and will have hover around the £450 / $599 RRP suggested by Nvidia for their cards while the Founder’s Edition is a more expensive £619.
This is the first time that you — the regular gamer that just wants awesome gaming performance — will be able to buy the Founder’s Edition (nee reference edition) version of Nvidia’s top GeForce GTX graphics card through the extent of its lifespan. That’s an important thing worth noting for anyone that’s intending on initially buying one and seeing how it’ll perform, and then potentially upgrading in the future to a two-card SLI setup.
Two cards, by the way, is the defacto standard for Nvidia’s SLI setup in the future. While three- and four-card super-SLI setups will still be supported, it’ll be through a roundabout way that requires extra driver support and won’t be set up straight straight out of the box in Nvidia’s GeForce Experience. Of course, if you’re the kind of serious enthusiast that’s setting up a quad-SLI rig, you’ll probably be more than happy to go the extra mile to tweak your system for every game you play anyway.
Software Tweaks: Simultaneous Multi-Projection
Simultaneous Multi-Projection is Nvidia’s term for a new display technology that massively improves the angle of projection for multi-monitor setups, including multiple flat monitors or individual or multiple curved monitors. Single-pass stereo distorts the image projection of the shapes being rendered — and only the shapes being rendered, to preserve resources — and massively reduces the level of distortion seen within a VR headset.
It means huge improvements especially for VR. Where a non-SMP example render takes 4.2 megapixels for a stereoscopic frame on the Oculus Rift, a SMP render will take 2.8 megapixels, which means a 1.5x horsepower boost and 5 times the pixel throughput. Geometry throughput, too, is doubled as the information behind the stereoscopic frame’s two images is rendered once rather than once per eye.
Even outside of VR, a technology like this can mean performance boosts within games — especially on lower-powered hardware. A developer in one of Nvidia’s tech demos used SMP to render the centre of the flat, single-display frame within their game and the outer area of the display at a slightly lower (apparently “almost imperceptible”) resolution, giving a performance benefit of around 40 per cent over standard.
Software Tweaks: 4K Streaming Standards
Pascal is also the first GPU family that supports both Microsoft and Netflix’s recommended standards for 4K streaming content, including 10-bit 4K HDR video support. It’s the first that supports HDR proper, too, allowing the 2x boost in visible colours and support for wider ranges of brightness and larger content ratios with a compatible display — currently only large-size LED-LCD and OLED displays from LG, Samsung, Sony and Panasonic.
The new chip support DisplayPort 1.4 and HDMI 2.0b, with 4K at 60fps 10-bit and 12-bit HEVC (h.265) content decoding and 4K at 60fps 10-bit HEVC encoding for video recording or streaming, and BT.2020 colour gamut support means Nvidia’s new GeForce GTX 1080 and GTX 1070 cards are the cards to get if you’re trying to build a 4K-capable, HDR-friendly home theatre PC. Nvidia’s Shield media streamers, too, support HDR streaming from a GTX 1080-equipped PC.
The GTX 1080 is the first card from Nvidia with GPU Boost 3.0 support, which lets the card change its frequency per voltage point — unlike GPU Boost 2.0, which only worked to its maximum potential at the maximum GPU voltage, the new tech lets the driver or user set adjustable frequency at multiple points across the range. It’s now possible for a user to run an overclock scanner program and get a bespoke maximum overclock for their individual GPU.
Software Tweaks: Nvidia Fast Sync
High FPS games, running at many hundreds of frames per second, have previously had two compromised choices: V-Sync on, which controls frame output and then produces no image tearing but high latency (sometimes 100 milliseconds), or V-Sync off with low latency but screen tearing.
Nvidia’s new decoupled render and display model, new to Pascal chips like the GeForce GTX 1080 and GTX 1070, but also older cards — expected to be “widely supported” — lets the card blast through as many frames as it can, and then picking the last rendered frame to put on the screen with no tearing.
That means no or barely increased latency levels, with no back-pressure from V-Sync, with massively high frame rates for older and lower-resolution competitive games like Counter-Strike: Global Offensive. This is basically the esports gamer’s nirvana — low latency, and no tearing.
It’s different to older, traditional OpenGL triple buffering in that it won’t have any potential for the frame buffer to fill up and back-pressure because there’s the potential for frames to be discarded — with triple buffering, every frame rendered will eventually show and this leads to overall latency.
The default Nvidia driver setting is for the application to specify its level of V-Sync support, but the new range of options — including adaptive V-Sync which enables V-Sync when the render rate is lower than the refresh rate and disables it above the refresh rate, and half frame rate adaptive V-Sync and Fast Sync — will give esports fanatics a chance to choose their own preference.
There’s no effective impact on frame pacing, either — there won’t be stutter or jitter — because pacing itself is done on the display side of the equation. The discarded frames, which aren’t displayed despite being rendered in a Fast Sync situation, simply disappear. When the frame render is extremely fast, the impact on smoothness — terrible without V-Sync, perfect with it — is minimal.
We’ll have a full review of the Nvidia GeForce GTX 1080, and a comparison to the previous card of choice within the 9-Series GTX line-up, the GTX 980Ti, later this week. Stay tuned.