NVLINK Support (Tesla P100)

[mrfactory]

Hello, I'm cosidering purchasing a HPC server with 8 Tesla P100, but maybe it's a good thing to buy this server with NVLINK Tesla instead of PCI Tesla, for possible new things to come.

OCTOPUTER from MicroWay
https://www.microway.com/product/octopu ... er-nvlink/

The question,

Is Octane Render compatible with NVLINK?
Is there any advantage for Octane Render, with NVLINK?
Is there any plan to take advange of this new technology?

Thanks

[Goldorak]

Hello, I'm cosidering purchasing a HPC server with 8 Tesla P100, but maybe it's a good thing to buy this server with NVLINK Tesla instead of PCI Tesla, for possible new things to come.

OCTOPUTER from MicroWay
https://www.microway.com/product/octopu ... er-nvlink/

The question,

Is Octane Render compatible with NVLINK?
Is there any advantage for Octane Render, with NVLINK?
Is there any plan to take advange of this new technology?

Thanks

Not yet, but we are working on this right now (along with Volta support with NVLink 2) sometime around the 3.10 stable release. ThisIt would allow you to combine P100 and V100 cards to get 64/128 GB of VRAM, so current single VRAM geometry size limits largely go away - and with CPU textures also out of core you could load and render huge datasets at full speed.

[glimpse]

Not yet, but we are working on this right now (along with Volta support with NVLink 2) sometime around the 3.10 stable release. ThisIt would allow you to combine P100 and V100 cards to get 64/128 GB of VRAM, so current single VRAM geometry size limits largely go away - and with CPU textures also out of core you could load and render huge datasets at full speed.

Great news! (apart from the fact this NVLINK is so expensive, 8-9k per GPU, but hopefully it will come down into some quadro/tesla variant of TitanV for more reasonable price..)

[mrfactory]

Thank for the info, this is good news, I'm considering purchasing 10 HPC render servers like the octoputer in next 6 moth, for a big project. It seems that the 3.1 is going to be launched before next 6 month, so i will start with the engineering and budgeting.

I've already tested HPC from ASUS esc8000g3 with 8 GTX 1080, and quite happy with Octane Render.