Meet T-rex: Nvidia’s Ti­tan RTX is the new graph­ics card mega-mon­ster

But gamers should stick with the Geforce RTX 2080 Ti in­stead.

PCWorld (USA) - - News | Meet Qualcomm’s Snapdragon 855 - BY BRAD CHA­COS

When the Geforce RTX 2080 Ti ( go.pcworld. com/20ti) launched with a sky-high $1,000 the­o­ret­i­cal price tag and $1,200-plus ac­tual price tag, some peo­ple spec­u­lated that it sim­ply took the place of the Ti­tan at the top of Nvidia’s graph­ics card lineup. Nope. Re­cently, Nvidia re­vealed the Ti­tan RTX, a $2,499 be­he­moth that the com­pany also calls “T-rex.”

It’s cer­tainly mon­strous enough. Like its Ti­tan V pre­de­ces­sor ( tinv), the Ti­tan RTX re­turns to the Ti­tan’s roots as a pro­sumer card, with a fo­cus on AI, data science, and con­tent cre­ation tasks. Nvidia’s flag­ship TU102 GPU packs 72 ded­i­cated RT cores for real-time ray trac­ing and 576 ten­sor cores so beloved by ma­chine learn­ing tasks. That’s iden­ti­cal to the Geforce RTX 2080 Ti’s load­out.

While Nvidia’s an­nounce­ment didn’t spec­ify the Ti­tan RTX’S CUDA core count, an Nvidia spokesper­son con­firmed that it holds more than the Geforce card. The Ti­tan RTX matches the 4,608 CUDA cores in­side the Quadro RTX 6000 and RTX 8000 ( go. pro­fes­sional GPUS,

rather than the RTX 2080 Ti’s 4,352 cores. That gives the new­est Ti­tan fewer to­tal CUDA cores than its pre­de­ces­sor—the Ti­tan V crammed in 5,120—but the Tur­ing GPU’S CUDA cores are much more ef­fec­tive ( go. than the ones in­side older Pas­cal GPUS. The ex­tra cores also help the card per­form deep-learn­ing tasks more ef­fi­ciently: The RTX 2080 Ti per­forms FP16 tasks at 110 ter­aflops, while Nvidia said the Ti­tan RTX churns through the same tasks at 130 TFLOPS.

Data tasks can re­quire much more mem­ory than gam­ing, so Nvidia loaded the T-rex with VRAM. The Ti­tan RTX comes with 24GB of cut­ting-edge GDDR6 mem­ory, for a to­tal mem­ory band­width of 672GB per sec­ond. That matches the Quadro RTX 6000’s con­fig­u­ra­tion, and more than dou­bles up the Geforce RTX 2080 Ti, which has “only” 11GB of VRAM.

It’s an in­ter­est­ing de­ci­sion by Nvidia. The Ti­tan V in­cluded only 12GB of on­board mem­ory, pre­sum­ably to dif­fer­en­ti­ate it from the pricier Quadro op­tions. While the Ti­tan RTX’S $2,499 price may stag­ger gamers, the Quadro RTX 6000 costs $6,300.

Such a pow­er­ful arse­nal of cut­ting-edge hard­ware should make the Ti­tan RTX stomp through ma­chine-learn­ing tasks. Nvidia’s sup­port­ing it on the soft­ware side with RAPIDS open-source soft­ware li­braries ( go.pcworld. com/rpds) that rely on CUDA. Nvidia also says that the ten­sor cores and ray-trac­ing hard­ware un­lock new pos­si­bil­i­ties in creative ap­pli­ca­tions, and the card’s fear­some fire­power and mas­sive mem­ory pool gives it po­tent chops in tra­di­tional ren­der­ing tasks, too. The Ti­tan RTX can even per­form real-time 8K video edit­ing, Nvidia says.

The new Ti­tan sports the up­graded du­alax­ial cooler de­sign in­tro­duced with the con­sumer Geforce RTX cards, but its alu­minum shroud comes clad in gold, not sil­ver.

Look for the Ti­tan RTX to launch in the United States and Europe later in De­cem­ber for $2,499. Thanks to those ex­tra CUDA cores, T-rex should (slightly) out­punch the Geforce RTX 2080 Ti flag­ship in gam­ing, but at twice the price, so gamers who want to drive a 4K, 144Hz G-sync HDR mon­i­tor like the sub­lime Acer Preda­tor X27 ( go.pcworld. com/ax27) should leave the Ti­tan RTX for the data sci­en­tists and stick to the Geforce RTX 2080 Ti in­stead.

Newspapers in English

Newspapers from USA

© PressReader. All rights reserved.