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
For me in order to have a more compact high-end graphics card, I may have to wait for the next generation. Hopefully by then HBM would have worked its way down to the x80 line.
Especially people that can afford dual titans, the price isnt an issue. They have no problem with upgrading their titans every year to continue to be on the bleeding edge of performance.
The only titan that could fit into that catagory was the titan black, being so upstaged by the titan x. But I dont know any miserable titan or titan X owners. Know of a few happy ones though.
I am not sure about the name even though I can guess where it came from. But then again it's just a name, let's see this thing in action!
AMD is doomed this year. I seriously doubt that VEGA will even touch 1080, let alone this monster.
But it will still sell like hotcakes......
The GTX 1080 has 2,560 cuda cores and a 256 bit bus of GDDR5X, which gives it a 320gb/s memory bandwidth
(the equation for memory bandwidth on any type of DDR type RAM is as follows:
(bus width / 8 ) * memory frequency = bandwidth)
Therefore on the GTX 1080 we calculate it as so: (256 bit / 8 = 32) * 10,000mhz = Bandwidth)
32 * 10,000mhz = 320gb/s (exactly what 1080 has)
So the GTX 1080 TI with 384 bit bus of GDDR5X would have: (384 / 8 = 48) * 10,000 = Bandwidth
48 * 10,000 = 480gb/s!!!
So not only does the GTX 1080 TI have ~50% more cores (~3,840) Combine that with a whopping 50% extra memory bandwidth (at MINIMUM! As it's been confirmed that GDDR5X CAN hit ~12,000mhz; so if they crank the VRAM on the 1080 TI to 12,000mhz the bandwidth raises to (384 / 8 = 48) * 12,000mhz = 576gb/s!! Which is a massive EIGHTY PERCENT FASTER BANDWIDTH compared to GTX 1080's 320gb/s!
So how exactly is it that you think a card with between 40% and 50% more cores, and between 50% and 80% more memory bandwidth, will only be 30% faster?!?!?!?
Bare minimum it will be around 50% faster, which is what these leaks actually said was the rough performance of the GP102 chip. You're looking at possibly 60% faster if it does indeed use 12,000mhz G5X VRAM. And with the 1080 being ~25-30% faster than a 980 TI, that means the 1080 TI will likely (just like the 1080 and 1070) be as fast as SLI of its predecessor (i.e. GTX 1070 was as fast as 970 SLI in most cases. GTX 1080 was as fast as 980 SLI in most cases. And GTX 1080 TI will likely be as fast as GTX 980 TI SLI!!)
I noticed the HUGE similarities between Pascal's release and the GTX 600 series Keplar release.
1) GTX 680 and 670 were the first 28nm GPU's. GTX 1080 and 1070 are first 16nm GPUs
2) Due to low yields on new process node, you will see fairly large supply shortages; just like on the 600 series where it took MONTHS for supply to stabilize enough to meet even a portion of market demand
3) Low power consumption compared to past cards. Compare 980 TI and TITAN X's 250w TDP to GTX 1080's 180w TDP, it's a large increase in efficiency, along with a drastic increase in clock speed to go with it! Same with 600 series: GTX 680 had a significantly lower TDP and power consumption than the UBER hot 480 and still pretty hot 580.
Then if you assume that Nvidia will follow their tried and true 600 and 700 series release schedule this all makes sense:
GTX 600 Series =
Start 28nm GPU manufacturing
GK104 GTX 680 = 1,536 cuda cores. 2GB GDDR5 VRAM, 1006mhz 1058mhz boost, 256 bit, 6000mhz, 192gb/s bandwidth
A few months later =
GTX TITAN GK110 comes out: 2,688 cuda cores. 6GB GDDR5 VRAM, 837 876mhz boost, 384 bit, 6000mhz, 288.4gb/s bandwidth
The following year =
GTX 700 series launches!
GK210 GTX 780 = 2304 cuda cores, 3GB GDDR5 VRAM, 863-900mhz boost, 384 bit, 6000mhz, 288.4gb/s bandwidth
GK210 GTX 780 TI = 2880 cuda cores, 3GB GDDR5 VRAM, 875-928mhz boost, 384 bit, 7000mhz, 336gb/s bandwdith
So my prediction was as follows:
Q2 2016 = GP104 launches with GTX 1080 = 2,560 cuda cores, 8GB GDDR5X VRAM, 1607-1733mhz boost, 256 bit, 10,000mhz, 320gb/s bandwidth
Q4 2016 = GP102 launches with GTX 1080 TI or TITAN model = 3,584 cuda cores, 16GB GDDR5X VRAM, 384 bit, 10,000mhz, 480gb/s bandwidth
Q2 2017 = GP100 "GTX 1180" launches = 3,840 cuda cores, 12GB HBM2 VRAM, 3072 bit, ??? mhz, ~683gb/s bandwidth
GP100 GTX 1180 TI = 4,480 cuda cores, ,12GB HBM2 VRAM, 4096 bit, ??? mhz, ~1tb/s bandwidth (1024gb/s)
And honestly it appears to be coming true for the most part. My specs on the upcoming 1080 TI/TITAN and the 1180/1180 TI etc.. are based on the exact same "ratio math" that the 600 and 700 series followed. For example, the GTX 780 has 2304 cores compared to 1536 on the 680; thats exactly 50% more cores; and matches up with the 1080 having 2,560 cores which is exactly 50% less than the 3,840 cores that the pascal TITAN P or 1080 TI etc.. will have. And the GTX 780 TI has 2880 cores compared to the 2304 cores on the 780; which is a 25% increase in cores; which matches EXACTLy with the 4,480 cores i predicted for the "1180 TI".
Of course there's still a fair bit of speculation, but with the GTX 1080 TI/ TITAN P or whatever they're calling it, being released SO soon after 1080; i'm almost POSITIVE that this is the type of plan Nvidia is going for. The most important thing about doing it this way, is that adding in a "pascal refresh 1100 series" will "hold us over" until Volta in 2018 which likely ended up with delays since it was moved up from ~2016-2017 original planned release for volta. I'm betting on Volta being similar to Pascal too! focusing far more on energy efficiency and architectural mastery/improvement.
980 -> 2048 cores
1080 -> 2560 cores
1180 -> you predict 3840 cores?
anyway, i ended up reselling it to a relative at cost because the performance gain over my 980 Ti that boosts to 1600MHz wouldn't be worth the time it takes to rebuild my computer. the MSRP of the 1080 Ti will probably be $700 or so. maybe $730. or less if AMD actually releases a proper competitor.People who buy titans usually don't care about prices. i know this may comes as a shock to you, but there are people out there who can drop thousands, even tens of thousands of dollars on dumb things constantly and not even see a dent in the checking accounts.