NVidia GTX 1080

Tech Advisor - - REVIEWS -

“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

Newspapers in English

Newspapers from UK

© PressReader. All rights reserved.