Hi!!

We have 4 Teslas in raid, will H12 work at 100% with this type of machine?

Thanks.

Sick machine. I think we need a bigger picture.

Sorry for that.

alejandro
Hi!!

We have 4 Teslas in raid, will H12 work at 100% with this type of machine?

Thanks.

Sorry for me intervention, but this is NOT a RAID, as a RAID is only a Redundant Array (of) Independent Disks (or Inexpensive, but I'd prefer first definition)

But OTOH, I envy you for that gear……. :cry: :cry: :cry:

To your question: I would not see any reason why it should not profit from your setup. As it makes no difference if you have few, many or a whole lot of processing ressources. If it is parallelizable (you know what I mean!) that's it! Problem is only how much data interaction is required and what the available bandwidth is. If you want more details ask me tomorrow.

It's 05:00 Zulu

Yes, Houdini 12 will run on that machine. No, 4 tesla cards will not make it any faster than 1 tesla card. We can make use of your GPU in H12 for a significant performance increase but we do not support multi-GPU processing at this time.

Thank you!!!

chrism
Yes, Houdini 12 will run on that machine. No, 4 tesla cards will not make it any faster than 1 tesla card. We can make use of your GPU in H12 for a significant performance increase but we do not support multi-GPU processing at this time.

Hi chris!

I never did CUDA or other GPU parallel stuff yet, but a lot of other CPU based parallelization stuff (MPI on SPARC&PowerPC unices). Is it such a different approach, or are there any other reasons H12 will not benefit from using multi Tesla setups ?

THX !

I guess one main reason coulc be they do not share memory, and data exchange via PCIExpress is slow compared to videoram

When Houdini 12 will go on shelf I would like to know excatly,
what hardware it will able to support and what not.
The Only 1 GPU unit support is very important fact before you go and spend money on your irons.
&
It will be nice to hear about the team who worked on the new Houdini Engine in the last ~5 years (~5 years right ?).

sanostol
I guess one main reason coulc be they do not share memory, and data exchange via PCIExpress is slow compared to videoram

If it was the only reason, cards with two GPUs on same card should be good solution.I guess there are more reasons …

sanostol
I guess one main reason could be they do not share memory, and data exchange via PCIExpress is slow compared to videoram

That is what the “buzzword” NUMA is referring to. You could avoid that by partitioning your data to stay local to a compute cluster, just like you would do in rendering when building your octree. This is also the reason why doubling your horses does not imply double the horsepower, from a certain level on, the administration overhead eats up your ressources…

But your reason seems reasonable