GeForce GTX 1080

[mib2berlin]

Hi, the new cards need Cuda 8 and there is only a release candidate for it.
I also read from a Cycles developer Cuda 8 RC1 slows down older cards.
May a testbuild is possible with RC1 but not for a stable version of Otane.

Cheers, mib

[nuno1980]

Arion Cuda test 1080 GTX = 3006 980Ti = 2288

I've EVGA GTX 780 Ti Classified@stock speeds-OC version (1020MHz core normal and 1150MHz boost) and I got 2516.04 (1min59sec) using Arion Benchmark 2.5.0 x64 running Win 10 x64 with 368.22WHQL (HQ). My GPU is faster than GTX 980 Ti!?!!

[oguzbir]

I've just read this.
https://devblogs.nvidia.com/parallelfor ... -revealed/

Where it says
Conclusion
CUDA 8 is the most feature-packed and powerful release of CUDA yet. CUDA 8 will be available in August 2016 and there will be a release candidate available around June. To get started with CUDA, download the latest CUDA Toolkit.

I'm not sure if developers will start working on the 8 RC build...

[DinoMuhic]

The RC is already out.

Let's hope they do already start with the RC. It would be a nightmare for all 1080 users to wait 2-3 months before they can use them.

[Notiusweb]

1080 reasons disappointed in the 1080:
1) 8 GB Vram
2) SLI limitations
3) meh CUDA count
4) repeat 1-3, 360 times

[glimpse]

1080 reasons disappointed in the 1080:
1) 8 GB Vram
2) SLI limitations
3) meh CUDA count
4) repeat 1-3, 360 times

let's counter argument for the sake of keeping conversation:

1. 8GB of vram is 2GB more than 980Ti or any other Titans You could buy in recent years, that mean,s You can power up screens (up to 4x 4K) & still use this card for render with the rest of 6GIgers on Your system

2. SLI limitations doesn't matter for us at all, & then nvidia do allow 4way SLI if You wish (I do not care about Your money & experience, considerign scaling is far from perfect on four or even three cards)

3. cuda count matters as clock speeds compensate performance. Looking from SP number standpoint, this card should be faster than 980TI for same, ..& still offer mroe vram (see note 1)

extra, bonus part, it will iuse less of power, require less connectors (only single 8pin)..so in the end..I see little to no point trashing this card for the sake of enjoyment =)

the problem with Pascal was that hype turned over itself, backfired & left some users unhappy, 'cos nvidia seems to overpromise & under-deliver. However it's first step & in Pascal & nvidia simply trying to buy time before releaseing trully new cards.

even if 1080 was simply shrinked Maxwell rather than trully Pascal based GPU it's still an achivement to meet performance of Higher end chip from previous architecture with entry/mid product here (AMD does the same by the way with their new release).

The msot interesting news comes from AMD though (& considerign the fact OctaneRender will run on AMDs hardware), that's what interests me most =) Team read jsut anounced card for ~200$ giving 5TFLOPs of performance (that's in the range of 980 or 1070) for half the price!

so the future of VR & the same GPU rendering is goign to be interesting..especially few months ahead =) as nvidia will be forced to lower the price if AMD is going to deliver what they are trying to do with their new release =)

[nuno1980]

@glimpse: Have you saw my result surprised of ArionBench running my GPU (it's faster than GTX 980 Ti)?

[brasco]

1. 8GB of vram is 2GB more than 980Ti or any other Titans You could buy in recent years

* apart from the Titan X

[glimpse]

1. 8GB of vram is 2GB more than 980Ti or any other Titans You could buy in recent years

* apart from the Titan X

ou yes =) sorry, that card is still on it's own league =) I'm not even comparing it.
Personally I would still be looking to X instead of 1080s =)..

- that's a point of those high end models - You keep them for longer.. saving You money =)

[brasco]

- that's a point of those high end models - You keep them for longer.. saving You money =)

Possibly! I suppose it depends on what the user's uses are and whether they need all of that VRAM (I do) - if you're on a first generation Maxwell (GM104s) or Keplers, then the 1080 is going to be a solid upgrade (as long as CUDA performance isn't limited in the drivers).

cheers
brasc