Titan x vs 1070 founders edition?

[aggiechase37]

I've been looking at a deal on Craigslist from a guy who has a pair of 1070's. I have a titan x and a 760. Thinking of trading the titan x for the two 1070's. You guys this this would be a good upgrade? I'd check out the bench but I'm not seeing any 1070's on there.

[oguzbir]

2x 1070 can be better againts 1genTitanX
if you do not need the 12GB vram.
I would say think twice.

There is no official benchmark with pascals...
There is one with a user made it possible with some tweaking.
viewtopic.php?f=9&t=56108&hilit=1080+benchmark&start=40
I personally have posted scores with 2 and 3 1070's
Have a look at the entire post..

[aggiechase37]

From everything I've read, just one 1070 is faster than the maxwell titan x.

[aggiechase37]

Also, my basic thinking is:

If I can get two 1070's and those are more or less going to perform like two titan x, then I can sell my titan x, sell my 760 and pretty much have a faster system and even pocket some money...

[oguzbir]

Also, my basic thinking is:
If I can get two 1070's and those are more or less going to perform like two titan x, then I can sell my titan x, sell my 760 and pretty much have a faster system and even pocket some money...

I guess that is correct. And you'll have em running at lower power. (and cooler I guess even with OC with good ventilated case ) Win win..

Just sell titan x quickly before someone reads this

[aggiechase37]

Also. I don't think I've ever been in a situation where I used even 3 gigs of vram. So 8 should be plenty.

[smicha]

Yes, if 8GB is really fine for you 1070 seems to be a great choice, especially that it is so super power efficient. As for its speed: my experience with 1080 - actually 7x 1080 on water - is that the modded benchmark does not really reflect Pascals power/speed. Namely I could OC a single 1080 (founders edition) to 2050/5500mhz (core/memory) and score something around 155, but I could not OC 7 ot them at once and run benchmark. So what I did I ran a real scene in PT changing max tile samples and parallel samples to max - otherwise Pascals will not be stressed fully (approx. 85% instead of 99%) - and using exactly the same scene with exactly the same parameters on 3 Titan X maxwell (score 426 on official bench) - I got over 1107 score on 7 of them (OC 7of them to the above values) at power draw around 900W, which makes sense (compared to 1350-1450W for 7x980ti).

I assume that having more power lanes for 1080 - not with a single 8pin cable in the founders edition but rather 8+6pin or even 2x8pin cabled 1080 - I could OC them by another 5-10% to 2200Mhz - and got a score closer to 170. If a single Titan X Maxwell scores 140 it's really hard to believe to get 1070 speed that high if it has 1/3 less of CUDA cores than 1080. A score around 120 (with OC) is more real to me, but anyway - yes, 2x1070 are a better choice than XM if 12GB is not needed.

Just please keep in mind that Octane still runs on hybrid CUDA being not fully optimized for Pascals.

[oguzbir]

Couldnt agree more Smicha.
I've been fiddling around those parallel samples and max tile samples for sometime now.
Those two changes the render times a lot. And something feels off about all these.

As for 1070's I find them as the best solution for budget rendering, not a monster maybe.
But they are very cheap, runs cool, low power, plenty of vram to fit and render heavy scenes for high dpi print works.

[smicha]

Couldnt agree more Smicha.
I've been fiddling around those parallel samples and max tile samples for sometime now.
Those two changes the render times a lot. And something feels off about all these.

As for 1070's I find them as the best solution for budget rendering, not a monster maybe.
But they are very cheap, runs cool, low power, plenty of vram to fit and render heavy scenes for high dpi print works.

Do you have them on air - 4 gpus very close each other? If so, could you post temps and fans speed when rendering?

[Lewis]

Just please keep in mind that Octane still runs on hybrid CUDA being not fully optimized for Pascals.

That's true currently but form reply i got they don't think there is going to be much change/speedup in final SDK for Pascal GPUs so it seems that's what we have for Pascal GPUs.