1080 = new 680, Polaris 10 = new 7870?

[renderstate]

If CUDA programs can be interrupted at any time, then does it automatically follow that non CUDA programs can also be similarly interrupted? I understood that in the past there were things that CUDA could do with the GPU that were not possible with DX.



I'm asking because NV specified CUDA programs, and as I read it, that's a specific limitation that they wouldn't have inserted unless it was a genuine limitation. If it were universally available, they wouldn't have phrased it in that way.



Edit: I should have checked the white paper first. NV's phrasing doesn't specify CUDA programs, so what I wrote above is moot.

I mentioned CUDA because most of the code running on GP100 is likely going to be written in CUDA. Whether the same feature is available with computer shaders.. I don't know 🙂

 

[NTMBK]

If CUDA programs can be interrupted at any time, then does it automatically follow that non CUDA programs can also be similarly interrupted? I understood that in the past there were things that CUDA could do with the GPU that were not possible with DX.

I'm asking because NV specified CUDA programs, and as I read it, that's a specific limitation that they wouldn't have inserted unless it was a genuine limitation. If it were universally available, they wouldn't have phrased it in that way.

Edit: I should have checked the white paper first. NV's phrasing doesn't specify CUDA programs, so what I wrote above is moot.

Doesn't mention whether we will see this feature in the 1080, only GP100. Remember, the GTX 680 was missing quite a few compute oriented features that the original Titan had (e.g. Hyper-Q).

 

[badb0y]

Is it too much to hope a GTX 280/4870 situation plays out again? nVidia getting real bold charging $699 for 1080.

 

[Timmah!]

It is nowhere premature to say P10x2 will be cheaper than the 1080 founders edition, and CRUSH it in performance in games where CF scaling works. I think that's a given.

Just like how massively OCing a 980Ti won't make it faster than a 390x when CF scaling works....

CF Scaling is just good, and SLI scaling sucks. So P10s multi card configurations will be performance winners.

Unless you want to bet on Nvidia's 1070SLI/1080SLI being great performers at great prices....
Lol..

Not that i care about multi GPU setups one bit, but nothing is given. You just assume that because 900 series cards (and possibly 600,700 etc) scale worse than Crossfire and SLI in their case isgenerally horrible, this has to continue with 1000 series. Well, it may or it may not. We shall see. But its certainly not given at this point. I recall throughout the years, as the GPU generations passed by, SLI and Crossfire were trading blows and at different points one was better than another and vice-versa. Even though it was Crossfire on top lately, the tides may turn again.

My 2 cents.

 

[C@mM!]

Is it too much to hope a GTX 280/4870 situation plays out again? nVidia getting real bold charging $699 for 1080.

Nvidia's getting squeezed on wafer costs out of TSMC due to ARM demand. Hence the shift to the high end market.

AMD on the other hand has favourable buys due to its minimum wafer buy out of GloFo. This was an issue in the past because of TSMC's fab lead, but the licensing of Samsungs 14nmFF has AMD on a superior process with cheaper wafers with a higher amount available.