At its CES press conference today, NVIDIA teased a new flagship GPU: the RTX 3090 Ti. It says more details will arrive soon, but handed out a few specs to tide its fans over until then.
As a refresher, NVIDIA currently has the RTX 3090 at the top of its stack, with the RTX 3080 Ti close behind and the RTX 3080 as the mainstream flagship. All three are based on the same GA102 chip, with the number of active cores, clock speeds and memory configurations being the key differentiators. The RTX 3090 Ti will usurp the 3090 as the ultra high-end GPU outside of its creator line.
Like the 3090, the 3090 Ti will have 24GB of GDDR6X memory, except it’ll be running at 21Gbit/s, as opposed to the 19.5Gbit/s of the 3090’s memory. NVIDIA also says the GPU is capable of calculating 40 shader teraflops, 78 RT teraflops and 320 tensor (AI) teraflops That compares to the 3090’s 35.6 shader teraflops, 69.5 RT teraflops and 285 tensor teraflops.
Those figures represent a 12.5-ish percent increase over the 3090 across the board, and will likely make the 3090 Ti the most powerful gaming card ever released. We won’t know exactly how NVIDIA has achieved this until it shares full specs, but we can make some educated guesses.
The 3090 only has 82 out of the GA102’s 84 streaming multiprocessors activated, so it’s likely the 3090 Ti has all 84 running. Then, in addition to the memory clocks being a little faster, it’s likely NVIDIA has elevated the main GPU clock speeds slightly. Some napkin math suggests the boost clock will need to be at around 1,850MHz on the 3090 Ti (compared to 1,695MHz on the 3090) to reach the 40-teraflop figure NVIDIA has provided.
NVIDIA says “more details will be coming later this month,” so we’d expect a small event focused on the 3090 Ti to be announced soon, where we’ll find out how exorbitantly priced the 3090 Ti will be – remember that the 3090 itself has an RRP of $1,500 – and when people will theoretically be able to buy one.
Also announced at CES is the RTX 3050, which is set to become the cheapest 30-series desktop GPU to date. NVIDIA hasn’t given us full specifications on this one either, but it has 8GB of GDDR6 memory, and will be able to calculate 9 shader teraflops, 18 RT teraflops and 73 tensor teraflops. NVIDIA says it’ll be able to power “the latest ray-traced games at over 60 frames per second,” which is probably true if factoring in the company’s DLSS upscaling tech.
As is usually the case with lower-end cards, there won’t be a “founders” edition sold directly by NVIDIA, but the RTX 3050 will go on sale through the company’s various hardware partners on January 27th, with an RRP of $250.
Follow all of the latest news from CES 2022 right here!Continue Reading