NVIDIA Counts Down to GeForce GTX 1080 Ti Launch

[kruk]

Eh, well when it comes to the GTX 980ti it was beaten by the Titan X (Maxwell) so I am not sure if I would say I am expecting it to beat the Titan XP.

You are probably talking about the reference card, but most (if no tall) AIB 980Ti's beat the Titan X with a quite a big margin.

 

[XiGMAKiD]

Vega10 and Vega11 are two separate chips which can both produce 2+ cards.

You're right

With two separate chips means they can cover wider price range, and If this really is another 780Ti (a full chip top dog) maybe Vega10 is that good. But I have my doubt it's gonna use full GP102, not enough leak to guesstimate

 

[GhostRyder]

You are probably talking about the reference card, but most (if no tall) AIB 980Ti's beat the Titan X with a quite a big margin.

I guess they did looking back at it I forgot that Maxwell Titan X was held back alot by voltage and Maxwell responded decently well to Voltage. However, this generation as we have all see with the cards currently out Pascal seems not to respond well to voltage beyond a certain point (Comparing reference GTX 1080 to the Aftermarket ones) resulting in similar clocks across the board. So this generation I am going to assume the Titan XP and GTX 1080ti will be roughly the same on the overclocking front (Though I do believe the GTX 1080ti will boost higher, we will just have to see if its enough).

 

[efikkan]

It remains to be seen if the company endows the SKU with more CUDA cores than the TITAN X Pascal, or less. A lot will depend on what NVIDIA's product managers learned about AMD's upcoming Radeon Vega.

What the product managers have recently learned about Vega doesn't matter, the binning for 1080 Ti was decided many months ago, actually before the release of Titan X (Pascal). The only parameters they've been able to adjust since is clock speeds, voltage, fan profiles, etc. The final details have been decided a while ago, since the final products have starting to come of the production line for a while now. So anything Nvidia has learned in the past few weeks doesn't matter.

Nvidia will not launch a GeForce of a higher binning than the current Titan X (Pascal) until they have a higher Titan ready, so GTX 1080 Ti will be a slightly more "cut" version, most likely featuring higher clock speeds.

 

[jabbadap]

What the product managers have recently learned about Vega doesn't matter, the binning for 1080 Ti was decided many months ago, actually before the release of Titan X (Pascal). The only parameters they've been able to adjust since is clock speeds, voltage, fan profiles, etc. The final details have been decided a while ago, since the final products have starting to come of the production line for a while now. So anything Nvidia has learned in the past few weeks doesn't matter.

Nvidia will not launch a GeForce of a higher binning than the current Titan X (Pascal) until they have a higher Titan ready, so GTX 1080 Ti will be a slightly more "cut" version, most likely featuring higher clock speeds.

Well one possibility is that gtx1080ti is indeed a full gp102 and next titan is not gp102 but gp100. Just thinking of that aida64 pascal family list, there's that thing was bugging me a bit. Tesla and Quadros are usually with GL suffixes like GP100GL-A or GP100GL-B. But there's non GL GP100:s device ids too:
1725 Graphics Device (GP100-B)
172E Graphics Device (GP100-B)
172F Graphics Device (GP100-B)

So what if GP100GL-B is that Quadro P100 and GP100-B is actually Titan P100(Tesla p100s are GP100-A)?

 

[efikkan]

Well one possibility is that gtx1080ti is indeed a full gp102 and next titan is not gp102 but gp100. Just thinking of that aida64 pascal family list, there's that thing was bugging me a bit. Tesla and Quadros are usually with GL suffixes like GP100GL-A or GP100GL-B. But there's non GL GP100:s device ids too:
1725 Graphics Device (GP100-B)
172E Graphics Device (GP100-B)
172F Graphics Device (GP100-B)

So what if GP100GL-B is that Quadro P100 and GP100-B is actually Titan P100(Tesla p100s are GP100-A)?

The GP100/GP102 products released so far are:

Quadro GP100            15F0 Graphics Device (GP100GL-A)
Tesla P100-PCIE-12GB    15F7 Graphics Device (GP100GL-A)
Tesla P100-PCIE-16GB    15F8 Graphics Device (GP100GL-A)
Tesla P100-SXM2-16GB    15F9 Graphics Device (GP100GL-A)

Titan X (Pascal)        1B00 Graphics Device (GP102-A)
Quadro P6000            1B30 Graphics Device (GP102GL-A)
Tesla P40               1B38 Graphics Device (GP102GL-A)

If you dig through the drivers you'll see that there's three codes for GTX 1070, GTX 1080, etc. We can't guess the binning based on these codes.

GTX 1080 Ti was planned for end of 2016 but was postponed due to the supply of GP102 chips, Nvidia can't even make enough to meet the demand of the expensive Titan X (Pascal) and Tesla models, so we can't expect too much from the 1080 Ti.

BTW, there is also the following:

NVIDIA_DEV.1D81 = "NVIDIA Graphics Device [GV100]"

So there's already driver support for something coming next year…