I have a question regarding this latest and most powerful NVidia card...will it work in Octane? The thing is, those 2 GPUs onboard are basically plugged into permanent SLI, which i do not think can be switched off, like when you have 2 separate cards...and obviously you cannot run Octane in SLI mode (at least according to manual, never tried it myself).
I am going to earn some extra money, so i am thinking about selling my 2 gtx 460 (or for the beginning one of them, i already have a buyer), as they are unfortunately quite slow compared to 470 o have at work. Not to mention running this dual card thing on my mobo is a disaster, the cards have to be plugged to adjacent PCIe slots, so i have to deal with high temps and very unpleasant noise when running both in Octane... so i was thinking about getting 570, then 580 and now 590 seems really good, 2x 512 CUDA cores on single PCB, so no temp/noise problems....however i am not sure, if its possible.
This was debated, but IIRC without any real conclusions, because nobody has this card. People speculate that it would not work and / or you will get only 1.5 GB of RAM.
matej wrote:This was debated, but IIRC without any real conclusions, because nobody has this card. People speculate that it would not work and / or you will get only 1.5 GB of RAM.
You might try it and then report the facts
Haha, thanks. The 1,5 GB of RAM is given, i am pretty sure its not going to be 3GB card.. but i can live with that. The issue is the SLI, but i made a little inquiry, wrote a question into the Anandtech forums and apparently it can be done (switched off), you can then one GPU core assign to doing PhysX stuff (for gaming). But obviously i cant confirm it until i try it myself
Anyway, i was hoping for answer for this from Radiance/ someone else from the Octane team. Even if they do not own the card, perhaps they know or they can contact Nvidia about this...not to mention, they never quite explained, why SLI has to be switched off, what will happen if it wont (at least i am not aware they did). Maybe the explanation is just technical jibberish, so we are not going to understand anyway (well i wont for sure ), but then again maybe not.
OK guys, i just bought and installed the GTX590. You can indeed switch the Multi-GPU mode in the Nvidia drivers off and Octane sees both GPU cores as independent cards, so you can use them both for rendering, or one as video adapter only and second GPU core for rendering. So basically it works exactly in same manner as my previous GTX460 SLI. The upside is definitely higher performance and far less noisy...really the upper GTX460 in that SLI had 85-90 degrees and the fan was screaming like hell, as it had run at 90/100 percent to cool the thing. Now the highest temp i saw with both cores rendering was 86 degrees C, but the fan was at 63 percent max, which is still relatively quiet.
Granted dual 580s will be even faster, but i learned a hard way, i need bigger case (i have Fractal Design Define R2, brilliantly looking case IMHO) and different mainboard layout for such a solution...so i am satisfied, really. Now if only i could make such godlike renders as some people over here
BTW where i can get that benchmark scene, the one with the boxy walls on sides, which looks like surface of the Death Star? I will benchmark it and post the results.
So, with Octane 2.46b, pathtracing on at the default resolution (whatever it is) i get:
one GPU: cca 2,65 Megasamples
both GPUs: cca 4,85 Megasamples
both numbers are fluctuating, these are actually the middle values
the worrying thing is, obviously there is not 2x performance improvement with 2 GPUs rendering, it should be 5,3... however on 1920x1200 its actually 5,31/5,39 Megasamples, so more than on lesser resolution? WTF?
the other worrying thing, i read in some older thread, that GTX480 is getting cca 4 Megasamples on this benchmark, but it was some older Octane version...can anyone confirm the numbers for 480/580 on 2,46b?
Well given GTX590 is 2x full 512shader core on 610MHz frequency (compared to 480cores of GTX480 on 700MHz), i expected the speed of about 2 480s in SLI, i have not seen any benchmark of such solution, but if one GTX480 is getting 4 Megasamples, 4,8 Ms for GTX590 definitely does not seem right...
Little update... i mildly overclocked the card from 607core/1214shaders/1707memory to 670/1340/1800...
the speed in the Trench benchmark on 1920*1200 bumped to 5,81/5,84 Megasamples... i tested it for 20 minutes for stability and it sampled 2740 samples in that period... i suppose now i am pretty much on par with 2x480s performance wise...
And yeah, i look forward to the final 1.0 version built on CUDA 4.0, then we shoud see some more performance increase...however with the new kernel it might be actually slower as well...we will see..
Timmaigh wrote:Little update... i mildly overclocked the card from 607core/1214shaders/1707memory to 670/1340/1800...
the speed in the Trench benchmark on 1920*1200 bumped to 5,81/5,84 Megasamples... i tested it for 20 minutes for stability and it sampled 2740 samples in that period... i suppose now i am pretty much on par with 2x480s performance wise...
And yeah, i look forward to the final 1.0 version built on CUDA 4.0, then we shoud see some more performance increase...however with the new kernel it might be actually slower as well...we will see..
So what speeds are you getting now, im also considering getting one, would you advise me to do so are rather get two 580s?
Regards
Nehale
Windows 10 x64 | 2x GTX680 | i7 990X | 24GB
Cinema 4D R17 Studio + 3Ds Max 2017 OctaneRender™ for Cinema 4D 3.01 + OctaneRender® for 3ds max® v3.14
and now 590 seems really good, 2x 512 CUDA cores on single PCB, so no temp/noise problems....however i am not sure, if its possible.
