OTOY Forums

Do you meant Cuda Cores?
You cant really compare Titan/780 and the GTX 600 series.
The Titan/780 is crippled Tesla GPU , crippled mean for example reduced Double-Precision performance.
But this dosen´t matter for Octane performance.
The GTX 770 is a rebranded GTX 680 GK104 Kepler, all a bit confusing.
Tesla 2070 6GB is a "Pro" card for 24/7 usage with ECC VRam.

With ECC on, a portion of the dedicated memory is used for ECC bits, so the available user memory is reduced by 12.5%. (e.g. 3 GB total memory yields 2.625 GB of user available memory.)

So Tesla is slower, less memory and 3000 $ compare to 1000 $.
If you don´t really need 24 Hours 7 days full load Tesla makes no sense.

Cheers, mib.

Many thanks for your reply Mib2Berlin.
It's confusing indeed
Yes indeed I meant CUDA cores.

So if I sum up what you said...
- You can't compare Titans to GTX cards (that won't help me!!! )
- and Teslas are useless unless dedicated to rendering over a long period of time, 24/7
I just read a customer's review on a website (www.ldlc.com) : the guy said that his TITAN (999€) worked better with Octane than his other GPU which is a TESLA K20c (3500€)

So if I wanted to buy a Titan card to use with Octane, what specs of the card should I be looking at ?

said that his TITAN (999€) worked better with Octane than his other GPU which is a TESLA K20c

Yes, the Tesla cards are downclocked because of lower temperatures.
The only thing I mention is don´t buy a factory overclocked card
Some user already use Titan, may one can help with experience.

You can't compare Titans to GTX cards

Hm, Titan has 2688 Cores the GTX 680 1536 and Titan is nearly double as fast than GTX 680.
So you can.

Cheers, mib.

mib2berlin wrote:Hm, Titan has 2688 Cores the GTX 680 1536 and Titan is nearly double as fast than GTX 680.
So you can.
Cheers, mib.

Hi again,
So what would be the difference between a TITAN card and a GTX690 then ?
Same price and nearly same performances, aren't they ?

....
=> EDIT : Just had a look and they are actually quite different :
- the number of GPUs is 1 for TITAN whereas it is 2 for the GTX690
(I think I read somewhere that 1 is better, because Octane manages the TITAN's 6Gb as such, whereas it manages the GTX690's 4Gb as 2x2Gb - is that correct ?)
- TITAN uses less power consuption (250W) than the GTX690 (300W)
- both have approximately the same number of CUDA cores (slightly more for the GTX690), so rendertime should be nearly the same...

So between a GTX690 and a TITAN, I think the TITAN wins it

whereas it manages the GTX690's 4Gb as 2x2Gb - is that correct ?)

Yes and the GTX 690 are two downclocked GTX 680.
I read in forum one Titan is faster than a GTX 690 but cant find the tread anymore.

Good luck, mib.

hi everybody, I'm finding this all very interesting. I'm in a similar position, looking for the best solution and trying to wrap my head around GPU tech info to help make the decision. One of the things that confuses me most is that Octane say that the renderer is almost directly scalable in terms of CUDA cores, aka 4 gtx 580s with 512 cores each would still be less powerful that one Titan which has 2688 cores, and on top of that that core clock on the Titan is higher which would lead me to believe that 1 Titan is more equivalent to 5 GTX580s in terms of pure Octane render times. When looking at the benchmarks the Titan outperforms the 580 by only 60% which seems to make it like an expensive option.

If someone can clarify all of that it'd be much appreciated!

I'm hoping the Titan renders at least twice at fast on PMC as my current 580, making a 2 titan system a very good option for fast drafts and stills as well as single frame tests through different stages of an animation / tweaking, and then could send final renders over to the Otoy cloud system when that's online, could be a pretty cool workflow.

I'm hoping the Titan renders at least twice at fast on PMC as my current 580

It does not, you can compare only Cuda cores with the same architecture.
384 Fermi cores render ~200% of 192 Fermi cores.
But 2688 Titan cores are not 5 times faster than 512 from GTX 580, not even double as fast.
Take another look to the Octane benchmark.

Cheers, mib.

I'd figured it would have been too good to be true! thanks for clarifying... different architecture between the cards so comparing the amount of CUDA cores isn't the thing to do. The benchmark is very useful, thanks

mib2berlin wrote:It does not, you can compare only Cuda cores with the same architecture.
384 Fermi cores render ~200% of 192 Fermi cores.
But 2688 Titan cores are not 5 times faster than 512 from GTX 580, not even double as fast.

Thanks for clarifying this point Mib, indeed that's where I was confused.
As Rico says, it's not as easy as comparing the number of CUDA cores.
So it seems like reading benchmarks is the only way to get around this.

So, going back to my initial question, I could tell my friend/computer-reseller that a nice renderfarm would have something like 4 TITAN cards + a 4 PCI mother board to support them.
Could be a good system, couldn't it?

Gonna be pricey though, even at cost-price !!!

4 titans certainly sounds insane! if 1 titan is 60% faster than a 580 and you 4 of them I would imagine you'd get some pretty quick frames, but still for an animation you'd have a lot of frames to render. Wouldn't the running costs of 4 titans (and cooling them / the room) on top of initial outlay cost make it a really inefficient solution? On the other hand I have no idea what else would be more cost effective other than perhaps the Otoy cloud rendering when that comes out. I guess though if you can get it all at cost price then it's probably worth doing! (does your mate fancy making a system for me too )