OTOY Forums

moritzw wrote:it's been 8 months, still no news on that topic?

Hi there, Jules stated that OTOY made some progress with help from nvidia optimising =) let's wait a bit more.

Overall Pascal cards are pretty fast already =)

Yes - Marcus tracked this to its source in the CUDA stack, and we have reported and shared these findings with NVIDIA at their request. NVIDIA has been extremely supportive and helpful during this process, and we all hope to use the results of this work to try to avoid a similar issue when Volta comes .

Octane on Pascal (in my testing) is faster than Maxwell on the same TDP. After this addressed (we are working to get this done ASAP), Octane will be even more optimized on Pascal, and we will finally add Pascal officially to OctaneBench.

Thanks for the update!


Now if NVIDIA would release Pascal drivers for OSX - or - MAC OS, as they prefer to refer to this. Anyone who uses web drivers for Maxwell cards in Mac Pro, take a couple mins and go to NVIDIA site, chat them up about getting the web drivers up to Pascal.

./rrb

Thanks Glimpse, Goldorak for the update.

It's been 6 months and like many pros here the #1 motivation to buy a next-gen card from Nvidia is for GPU rendering.

All of Pascal's amazing improvements are mute if the software pipeline can't capitalize. We understand that Nvidia's Cuda 8.x may be at fault so we aren't fault-finding w/Otoy...just expressing our eagerness.

I purchased a 1080 last summer but I'll be stay on the sidelines waiting to buy two new 1080 TI. Once this is worked out I'll proceed with that purchase. That will likely drive me to buy a third Octane license as well...

I hope to do my wee little part in rewarding all parties engaged in the resolution here.

Last update from Marcus was that we may get Pascal support wrapped up in 3.06 and, if that is the case, release a new OB based on 3.0.6 shortly thereafter.

We may refresh OB again after 3.1 is out, due to all the changes we want to capture with OSL. We expect (and hope) relative speed on modern cards won't change much, and OB will likely remain normalized with a 980 as ~100 OB.

Hi there.

I made a test yesterday and to my disappointment my new 1080 (which i choose, after one of my 980 Tis said goodbye) was slower than my old 780s. And much slower, than my remaining 980 Ti. For example i had around 25 M/sec on 1080 and in the same simple scene, just refresh on 980 Ti i had around 45 M/sec.

So, is there any idea what can cause this? Is there a possiblility, that this gap will be closed with 3.0.6 ?

Best Chris

Chriz wrote:Hi there.

I made a test yesterday and to my disappointment my new 1080 (which i choose, after one of my 980 Tis said goodbye) was slower than my old 780s. And much slower, than my remaining 980 Ti. For example i had around 25 M/sec on 1080 and in the same simple scene, just refresh on 980 Ti i had around 45 M/sec.

So, is there any idea what can cause this? Is there a possiblility, that this gap will be closed with 3.0.6 ?

Best Chris

viewtopic.php?f=9&t=58820

Thanks, i already knew that threat.

And first: Someone mentioned there he has no problem with mixing maxwell and pascal cards
second: of cause i deselected the cards in Octanesettings and tested them seperatly.

So i am not sure, what i could learn from that threat? Did i miss something?

@ Elvissuperstar007

best chriz

Chriz wrote:Thanks, i already knew that threat.

And first: Someone mentioned there he has no problem with mixing maxwell and pascal cards
second: of cause i deselected the cards in Octanesettings and tested them seperatly.

So i am not sure, what i could learn from that threat? Did i miss something?

@ Elvissuperstar007

best chriz

What are your temps like, maybe your card is being throttled.

Temps are low.

Both, the 980 TI and the 1080 are connected via PCI 16x to 1x USB3 raisers. But this is not be the problem, because both card working well this way and 1080 is just computing much slower than expected. The only effect by this raiser stuff is a little slower data transfer to the gpus. The 980 Ti works as great as if its directly plugged into the board.

The both 780s are directly plugged into the mainboard.

Any suggestions?