Tuesday, July 5th 2016
NVIDIA to Unveil GeForce GTX TITAN P at Gamescom
NVIDIA is preparing to launch its flagship graphics card based on the "Pascal" architecture, the so-called GeForce GTX TITAN P, at the 2016 Gamescom, held in Cologne, Germany, between 17-21 August. The card is expected to be based on the GP100 silicon, and could likely come in two variants - 16 GB and 12 GB. The two differ by memory bus width besides memory size. The 16 GB variant could feature four HBM2 stacks over a 4096-bit memory bus; while the 12 GB variant could feature three HBM2 stacks, and a 3072-bit bus. This approach by NVIDIA is identical to the way it carved out Tesla P100-based PCIe accelerators, based on this ASIC. The cards' TDP could be rated between 300-375W, drawing power from two 8-pin PCIe power connectors.
The GP100 and GTX TITAN P isn't the only high-end graphics card lineup targeted at gamers and PC enthusiasts, NVIDIA is also working the GP102 silicon, positioned between the GP104 and the GP100. This chip could lack FP64 CUDA cores found on the GP100 silicon, and feature up to 3,840 CUDA cores of the same kind found on the GP104. The GP102 is also expected to feature simpler 384-bit GDDR5X memory. NVIDIA could base the GTX 1080 Ti on this chip.
Sources:
VR World, Many Thanks to okidna for the tip.
The GP100 and GTX TITAN P isn't the only high-end graphics card lineup targeted at gamers and PC enthusiasts, NVIDIA is also working the GP102 silicon, positioned between the GP104 and the GP100. This chip could lack FP64 CUDA cores found on the GP100 silicon, and feature up to 3,840 CUDA cores of the same kind found on the GP104. The GP102 is also expected to feature simpler 384-bit GDDR5X memory. NVIDIA could base the GTX 1080 Ti on this chip.
58 Comments on NVIDIA to Unveil GeForce GTX TITAN P at Gamescom
My Prediction is
980 -> 2048 cores
980 TI -> 2816 cores
1080 -> 2560 cores
1080 TI -> 3584 cores
1180 -> 3,840 cores
1180 TI -> 4,480 cores
Because that follows the exact same core "ratios" that you get with the 680 and the 700 series
GTX 680 = 1536 cores
GTX 780 = 2304 cores (which is exactly 50% more cores than the 680)
GTX 780 TI = 2,880 cores (which is exactly 25% more cores than the 780)
So if you apply that same math to Pascal, you get my numbers
1080 = 2560 cores
1180 = 3,840 cores (50% more than 1080)
1180 TI = 4,480 cores (25% more than 1180)
You have to understand that the fact that the 780, 980, and 1080, have kept relatively similar core counts (2304, 2048, 2560) is that they were ALL new architectures. But pascal is the same architecture for both the 1080 and the new "1180" etc.. in 2017. So it will end up in a pattern similar to how we went from 2048 on 980 to 3072 on the TITAN X (which was also EXACTLY 50% more cores, just like my predictions)
It's entirely possible that things will be a "bit" different, but i think that overall these numbers are fairly accurate. There's just SO many similarities between Keplar 600/700 series and Pascal for it to be a coincidence.
Another interesting pattern that fits my theory is that EVERY architectural generation since Keplar has had basically the SAME ratio math between the "104 die" x80 chip and the next generation full die implementation chip.
GTX 680 = 1536
GTX 780 = 2304 (50% increase)
GTX 980 = 2048
GTX TITAN X = 3072 (50% increase)
S0,
GTX 1080 = 2560
50% increase would = 3,840 for GTX 1180
There's been an exponential increase in cuda core counts on each new generation of cards; the gap between each x80 and x70 etc.. card gets larger and larger. Each new die shrink allows more and more cores and transistors to be put on a similarly sized GPU die, did you know that the full GP100 in the TESLA P100 physics computing GPU that Nvidia showed not long ago doesn't actually have 3,840 cuda cores like they said? It actually has a MASSIVE "5,760 CUDA CORES!!!!" that's nearly SIX THOUSAND!! But 1,920 of them are double precision cores; however that shows you how many cores are able to be fit on a full ~600mm2 size GP100 die. So my estimates of ~4,480 cores aren't that strange reallyIt's hard to say. It seems that if they are releasing a 980 TI AND a TITAN that we will get: 1080 TI = 12GB 384 bit GDDR5X (480gb/s bandwidth if running at 10,000mhz, or 576gb/s bandwidth if running at 12,000mhz), and TITAN P = 16GB 3072/4096 HBM2 (768gb/s if 3072 bit, or 1tb/s if 4096 bit)
Or if we JUST get the TITAN, or JUST the 1080 TI it will likely just be the 384 bit G5X; and the 2017 GTX 1180 and 1180 TI full GP100 ~600mm2 die chips will be the ones with HBM2 most likely, in that case. (with 1180 = 3072 bit 768gb/s, and 1180 TI = 4096 bit 1tb/s etc..)Yeah exactly. Go watch the linustechtips review of the founders edition 1080, they said they could run EVERY SINGLE game they tested at 60fps+ 4K resolution at high settings or higher with 1x or 2x AA. So will you be able to run The Witcher 3 with Max settings including Hairworks etc.. at 4K 60fps on a single 1080? Nope, probably not; but you can sure get ~45fps average, which is what my 2,150mhz overclocked MSI GAMING X GTX 1080 gets even in the most demanding areas typically.
As for the GTX 1080 TI, all we have to do to get a rough idea of performance is look at how Maxwell compares to the GTX 1070 and GTX 1080.
GTX 1070 is roughly equal to 2 way SLI GTX 970's in most cases (might fall short in a COUPLE cases, but once overclocked it's definitely there)
GTX 1080 is pretty much exactly equal to 2 way SLI GTX 1080's in most cases (again, might fall short in a few specific games, but ties or even slightly beats two 980s in the vast majority of games)
And the GTX 1080 TI leaked specs indicate it likely having roughly 3,584 - 3,840 cuda cores or so (3,584 cores = 40% increase. 3,840 cores = 50% increase), and at least a 384 bit G5X VRAM config, meaning a 50% increase in bandwidth (480gb/s over 320gb/s) So a card with 40-50% more cores and 50%+ more bandwidth will be at least around 50% faster or so. And with the GTX 1080 already being 25-30% faster than a GTX 980 TI, we get the exact same pattern as the rest of the Pascal cards, which is:
"GTX 1080 TI is roughly equal to 2 way SLI GTX 980 TI's!!!"
Call me crazy, but i recall two GTX 980 TI's being just fine for 4K gaming at 60fps. In fact considering that the 1080 TI will have 12GB or 16GB of VRAM and that VRAM will have AT LEAST have 50% more bandwidth (if it actually has HBM2 it will have FOUR TIMES more bandwidth...lol) it would do even BETTER at 4k than SLI 980 TI's in my opinion. I just CAN NOT see a 1080 TI with those kind of specs (12GB of 480gb/s G5X + 3,584-3,840 cuda cores) not being able to push 60fps at 4K in even the most demanding titles at damn near or full on max settings!