NVidia GTX 1080
“It’s insane,” nVidia CEO Jen-Hsun Huang proudly proclaimed at the GeForce GTX 1080’s launch, holding the graphics card aloft. “The 1080 is insane. It’s almost irresponsible amounts of performance… the 1080 is the new king.” He wasn’t joking. The long, desolate years of stalled GPU technology are over, and this is a beast.
A giant leap for GPU-kind
As wondrous as it is, the outrageous performance leap of the GTX 1080 doesn’t exactly come as a surprise.
Faltering graphics processor process technology left graphics cards from both nVidia and AMD stranded on the 28nm transistor node for four years – an eternity in the lightning-fast world of modern technology. Plans to move to 20nm GPUs fell by the wayside due to technical woes. That means the 16nm Pascal GPUs beating inside the GTX 1080’s heart (and AMD’s forthcoming 14nm Polaris GPUs) represent a leap of two full process generations.
That’s mad, and it alone could create a big theoretical jump in performance. But nVidia didn’t stop there. Pascal GPUs adopted the advanced FinFET ‘3D’ transistor technology that made its first mainstream appearance in Intel’s Ivy Bridge computer processors, and the GTX 1080 is the first graphics card powered by GDDR5X memory, a supercharged new version of the GDDR5 memory that’s come standard in graphics cards for a few years now.
On top of all that, nVidia invested significantly in the new Pascal architecture itself, particularly in tweaking efficiencies to increase clock speeds while simultaneously reducing power requirements, as well as many more under-the-hood goodies that we’ll get to later, including enhanced asynchronous compute features that should help nVidia’s cards perform better in DirectX 12 titles and combat a major Radeon advantage.
Let’s kick things off with an nVidia-supplied spec sheet comparison of the GTX 1080 vs its predecessor, the GTX 980 (see page 42). Here, some of the benefits to switching to 16nm jump out immediately. While the ‘GP104’ Pascal GPU’s 314mm2 die size is considerably smaller than 398mm2 die in the older GTX 980, it still manages to squeeze in 2 billion more transistors overall, as well as 25 percent more CUDA cores – 2560 in the GTX 1080, versus 2048 in the GTX 980.
The GTX 1080 indeed has a massive 1607MHz base clock and 1733MHz boost clock speeds – and that’s just the stock speeds. We managed to crank it to over 2GHz on air without breaking a sweat or tinkering with the card’s voltage. Add it all up and the new graphics card blows its predecessor out of the water in both gaming performance and compute tasks, leaping from 4981 GFLOPS in the GTX 980 to 8,873 GFLOPS in the GTX 1080.
Diving even deeper, each Pascal Streaming Multiprocessor (SM) features 128 CUDA cores, 256KB of register file capability, a 96KB shared memory unit, 48KB of L1 cache, and eight texture units. Each SM is paired with a GP104 PolyMorph engine that handles vertex fetch, tessellation, viewport transformation, vertex attribute setup, perspective correction, and the intriguing new Simultaneous Multi-Projection technology, according to nVidia.
A group of five SM/PolyMorph engines with a dedicated raster engine forms a Graphics Processing Cluster, and there are four GPCs in the GTX 1080. The GPU also features eight 32-bit memory controllers for a 256-bit memory bus, with a total of 2048KB L2 cache and 64 ROP units among them. That segues nicely into another technological advance in nVidia’s card: the memory. Despite having a 256-bit bus the same size as its predecessor, the GTX 1080 managed to push the overall memory bandwidth all the way to 320GB/s, from 224GB/s in the GTX 980. That’s thanks to the 8GB of cutting-edge Micron GGDR5X memory inside, which runs at a blistering 10Gb/s — a full 3Gb/s faster than the GTX 980’s already speedy memory.
Implementing such speedy memory required nVidia to redesign both the GPU circuit architecture as well as the board channel between the GPU and memory dies to exacting specifications – a process that will also benefit graphics cards equipped with standard GDDR5 memory, nVidia says.
Pascal achieves even greater data transfers capabilities thanks to enhanced memory compression technology. Specifically, it builds on the delta colour compression