Despite the fact that NVIDIA has kept tight-lipped about its upcoming Turing-based GeForce graphics cards, it really does seem like the worst-kept secret by this point. More information is starting to trickle out, and unfortunately for NVIDIA, its own partners are doing that leaking.
Today, we learn of the specs of the so-called GeForce RTX 2080 Ti – notice the “RTX” there in place of “GTX”. If it proves true, and it seems more and more likely that it is, then the GTX name is gone – at least in the high-end. At the professional graphics conference SIGGRAPH, which took place in Vancouver last week, NVIDIA claimed that RTX was the most monumental shift in graphics since the introduction of CUDA over ten years ago. That’s easier to believe when the gaming cards also see RTX reflected in their name – above the gaming-inspired GTX.
Thanks to PNY, we have some specs to chew on until the final announcement. On a product page that has since been removed, a GTX 2080 Ti is revealed to have 4352 CUDA cores based on Turing, which by itself is a gain of 512 cores beyond last-gen’s top-end Pascal-based TITAN Xp. For another comparison, the 1080 Ti bundled in 3584 CUDA cores, giving the RTX 2080 Ti a ~20% gain in cores alone.
Of course, Turing isn’t just about raw gains in single-precision performance; it’s also set to light ray tracing workloads on fire with “Real-Time Ray Tracing” support. While the RTX moniker itself leaves no secret about a feature on these cards, this is “confirmation” that the gaming cards will have it as a focus, as well. That means that hopefully, NVIDIA will have the support of game developers to talk about how they’ll be supporting the technology.
Also worth noting is the fact that the RTX 2080 Ti has 11GB of memory, just like the 1080 Ti, although that memory has been upgraded from GDDR5X up to GDDR6. This marks another generation where NVIDIA skipped over HBM on the gaming side. If another spec is to be believed, that new memory makes a considerable difference: 1080 Ti clocked in at 484GB/s on the memory, whereas PNY claims that the 2080 Ti pushes things to 616GB/s (TITAN Xp was 547GB/s).
There’s one more juicy takeaway to… take away. That’s that SLI does in fact seem to be dead – to be replaced with NVLink. On the desktop. NVLink has been an enterprise feature up to this point, to be paired up with $500 bridges, so let’s hope NVIDIA has a much cost-friendlier gaming variant coming. Speaking of pricing – the leaked page also shows a $1,000 price tag for the Ti, and $800 for the non-Ti. That of course is higher than the generation we’re leaving, so if it holds true, it’ll be interesting to learn about NVIDIA’s justification for it.
As we’ve covered before, NVIDIA will be unveiling its next GeForce adventure in Cologne, Germany, at the massive Gamescom event. After a lot of waiting, we’re finally on the cusp of some brand-new gaming GPUs from the green team. Gamescom is going to be very interesting, to say the least! And that’s ignoring the onslaught of game previews we’ll be seeing as well!