Nvidia’s Quadro RTX GPUS are built on a rev­o­lu­tion­ary Tur­ing GPU with ded­i­cated hard­ware for real-time ray trac­ing. Here’s what they her­ald for PC gamers and next-gen Ge­force graph­ics cards.

PCWorld (USA) - - News - BY BRAD CHACOS

It looks like the fear­some GPU in­side Nvidia’s Ti­tan V ( go.pc­world.com/ttnv) will never make it to gamers af­ter all. Poor Volta. Nvidia CEO Jensen Huang re­vealed at Sig­graph 2018 in mid-au­gust ( go. pc­world.com/rytr) the Quadro RTX 8000, 6000, and 5000 graph­ics pro­ces­sors, pow­ered by an all-new and long-ru­mored “Tur­ing” graph­ics architecture. Friends, Tur­ing looks fe­ro­cious—and it’s brim­ming with

por­tent about Nvidia’s nextgen Ge­force graph­ics cards.

The Quadro RTX GPUS are over­flow­ing with CUDA cores, Ai-boost­ing ten­sor cores, and an all-new tech­nol­ogy dubbed “RT cores” de­signed to make Nvidia’s new graph­ics architecture up to 25x more ef­fi­cient at cut­ting-edge real-time ray-trac­ing op­er­a­tions ( go.pc­world.com/dxry). “Tur­ing is Nvidia’s most im­por­tant in­no­va­tion in com­puter graph­ics in more than a decade,” Huang said. “The ar­rival of real-time ray trac­ing is the Holy Grail of our in­dus­try.”

We don’t nor­mally cover data cen­ter and work­sta­tion graph­ics here at Pc­world, but with Nvidia teas­ing a Ge­force event with “spec­tac­u­lar sur­prises” ( go.pc­world.com/ nwgc) at Gamescom in a mere week’s time, let’s ex­am­ine what the an­nounce­ment of these Quadro RTX GPUS po­ten­tially mean for PC gamers and the con­sumer Ge­force lineup, in­clud­ing the ru­mored GTX 2080. Or should I say RTX 2080, as a Ge­force tweet heav­ily hints?


For months, ru­mors have been swirling about po­ten­tial next-gen Ge­force GTX 1180 or 2080 graph­ics cards, but each leak fea­tured a dif­fer­ent name for the would-be GPUS. Volta? Am­pere? Tur­ing? What would Nvidia’s new architecture be called?

The Quadro RTX an­nounce­ment made Tur­ing GPUS of­fi­cial, and their guts will likely make up the in­side of new Ge­force cards, too. There’s still the pos­si­bil­ity that Nvidia could strip Tur­ing of its Ai-boost­ing ten­sor cores and call the re­sult­ing con­sumer GPU architecture “Am­pere,” though that doesn’t seem likely.


“This is a sig­nif­i­cant mo­ment in the his­tory of com­puter graph­ics,” Jon Peddie, the well­re­spected CEO of an­a­lyst firm JPR and a noted graph­ics ex­pert, said in Nvidia’s press re­lease. “Nvidia is de­liv­er­ing real-time ray trac­ing five years be­fore we had thought pos­si­ble.”

The Quadro RTX GPUS are the first GPUS to in­clude hard­ware ded­i­cated to real-time ray trac­ing, a gor­geous visual ef­fect that mod­els how light­ing works in the phys­i­cal uni­verse by send­ing light rays out from il­lu­mi­nated sources in the 3D scene. Light might be re­flected by other ob­jects, or look dif­fer­ent af­ter pass­ing through wa­ter, or be blocked by another ob­ject com­pletely and cre­ate a shadow. The ob­jects the rays bounce off even af­fect the fi­nal color you see, just like in real life. (Tra­di­tional “ras­ter­i­za­tion” ren­der­ing con­verts a game’s 3D mod­els into pix­els on your 2D screen, then ap­plies the color in­for­ma­tion af­ter.)

Ray-traced scenes can look much more re­al­is­tic than ras­ter­ized vi­su­als, but they’re very com­pu­ta­tion­ally ex­pen­sive. That’s why ray trac­ing is tra­di­tion­ally found in movie CGI, which doesn’t need to be ren­dered on-the-fly, but not games. The dark era ap­pears to be end­ing though. Microsoft laid the ground­work soft­ware-wise with the Directx Ray­trac­ing API ( go.pc­world.com/msry) in Win­dows ear­lier this year, and it’s likely that Tur­ing’s RT cores will be com­ing to at least some Ge­force GPUS.

Nvidia al­ready said its next-gen graph­ics card will in­clude ded­i­cated “RTX tech­nol­ogy” for en­hanced ray-trac­ing sup­port, and just days be­fore the Quadro an­nounce­ment, Video­cardz ( go.pc­world.com/vcdz) un­earthed new Nvidia fil­ings at the U.S. Patent and Trade­mark Of­fice for “Tur­ing,” “Quadro RTX,” and “Ge­force RTX.” The same day, Youtuber Adoredtv ( go. pc­world.com/adtv) pub­lished a video claim­ing that Nvidia’s next high-end Ge­force GPUS would be called the RTX 2080 and RTX 2070, ditch­ing the long-stand­ing “GTX” moniker. (Lesser-pow­ered graph­ics are still branded with GTX, Adoredtv claims.)

The tim­ing of the two leaks, the week­end ahead of this Sig­graph unveiling, seems sig­nif­i­cant. With Tur­ing and Quadro RTX prov­ing ac­cu­rate, and real-time ray trac­ing be­ing pushed so hard by the graph­ics in­dus­try, it’s very pos­si­ble that Nvidia’s more po­tent Ge­force cards could in­deed adopt RT cores and the RTX brand­ing, as the Ge­force ac­count it­self hinted at later in the even­ing. More on that later.

The Quadro RTX GPUS are the first GPUS to in­clude hard­ware ded­i­cated to re­al­time ray trac­ing, a gor­geous visual ef­fect that mod­els how light­ing works in the phys­i­cal uni­verse by send­ing light rays out from il­lu­mi­nated sources in the 3D scene.


Nvidia’s stream­ing mul­ti­pro­ces­sors are the build­ing blocks of its ar­chi­tec­tures, the de­sign of which dic­tates how its GPUS are cre­ated. The Tur­ing GPUS over­haul Nvidia’s cur­rent model with some new tricks, in­clud­ing fea­tures added to the Volta GPUS for AI and data cen­ters.

To quote the com­pany’s press re­lease:

“Tur­ing-based GPUS fea­ture a new stream­ing mul­ti­pro­ces­sor (SM) architecture that adds an in­te­ger ex­e­cu­tion unit ex­e­cut­ing in par­al­lel with the float­ing point data­path, and a new uni­fied cache architecture with dou­ble the band­width of the pre­vi­ous gen­er­a­tion. Com­bined with new graph­ics tech­nolo­gies such as vari­able rate shad­ing, the Tur­ing SM achieves un­prece­dented lev­els of per­for­mance per core.”

Nvidia also says Tur­ing works in “new graph­ics tech­nolo­gies such as vari­able rate shad­ing.” Deep-level tech­ni­cal de­tails weren’t re­leased, but there’s no ques­tion that the Ge­force GTX 2080’s Tur­ing GPU will be a step for­ward from the Pas­cal architecture in to­day’s GTX 10-series graph­ics cards.


The Quadro RTX cards fea­ture cut­ting-edge GDDR6 VRAM, an up­grade that mem­ory sup­pli­ers have been hint­ing at for months. De­pend­ing on which sup­plier you ask, GDDR6 should be about 40-per­cent faster than the GDDR5X VRAM in­side GTX 10-series

graph­ics cards, and of­fer twice the band­width of non-x GDDR5 mem­ory. There’s no rea­son to think that the more pow­er­ful cards in Nvidia’s next-gen Ge­force lineup, in­clud­ing the ru­mored RTX 2080 and RTX 2070, would skip GDDR6 mem­ory.


Fi­nally, the Quadro RTX GPUS add na­tive sup­port for Vir­tu­allink ( go.pc­world.com/vrtl), the new USB-C al­ter­nate mode that a con­sor­tium of PC giants an­nounced in July. Vir­tu­allink stan­dard­izes vir­tual-re­al­ity con­nec­tions around a sin­gle USB-C ca­ble that bakes in four high-speed HBR3 Dis­play­port lanes, a USB 3.1 data chan­nel for sen­sors and cam­era in­for­ma­tion, and up to 27 watts of power.

“Sim­u­lat­ing re­al­ity re­quires in­cred­i­ble visual fi­delity and pro­cess­ing power,” Jason Paul, Nvidia’s gen­eral man­ager of gam­ing and VR, said in a state­ment when the stan­dard was an­nounced. “With a sin­gle, high-band­width ca­ble, Vir­tu­allink un­locks the full po­ten­tial of the PC to power amaz­ing VR ex­pe­ri­ences.”

With Vir­tu­allink al­ready mak­ing an ap­pear­ance in the Quadro RTX 8000, 6000, and 5000, I fully ex­pect to see it sup­ported in the Ge­force GTX 2080, Ge­force RTX 2080, or what­ever Nvidia’s next high-end con­sumer graph­ics card will be called.

Nvidia CEO Jensen Huang holds up a Quadro RTX GPU built on the Tur­ing architecture.

Nvidia’s Tur­ing GPU.


Newspapers in English

Newspapers from USA

© PressReader. All rights reserved.