AMD Posts Open-Source Linux Driver Support For "NAVI 14" GPU

[phoronix]

Phoronix: AMD Posts Open-Source Linux Driver Support For "NAVI 14" GPU

While the Radeon RX 5700 "Navi 10" series is launching on 7 July and there was the recently presented open-source Linux patches (so far for the AMDGPU kernel driver and RadeonSI OpenGL; Vulkan support still pending), today was the surprise move of posting the kernel patches for an unannounced "Navi 14" graphics processor...

http://www.phoronix.com/scan.php?page=news_item&px=AMD-Open-Source-NAVI-14

[chuckula]

I'm assuming "Navi 14" is a follow-on small Navi but I'll give AMD credit for posting source code longer in advance so that it might actually be working in mainline when the cards launch.

[boxie]

I'm assuming "Navi 14" is a follow-on small Navi but I'll give AMD credit for posting source code longer in advance so that it might actually be working in mainline when the cards launch.

<speculation>
1. I wonder if it is a workstation/server card with it's "virtual display" - seems like something remote desktop environments would like.
2. I wonder if it is the card that Google will be using for Stadia? again, virtual display would seem like a good thing here
</speculation>

[cb88]

<speculation>
1. I wonder if it is a workstation/server card with it's "virtual display" - seems like something remote desktop environments would like.
2. I wonder if it is the card that Google will be using for Stadia? again, virtual display would seem like a good thing here
</speculation>

Stadia is already built on customized variant's of Vega56 so no it doesn't require anything that Navi has. The datacenter variants of Vega 10 can support SR-IOV which would obviate the need for any virtual display.

Virtual display sounds alot like virtual super resolution on windows... which just downscales from a higher resolution to the physical display meaning there is more data there than can be displayed unlike all the various antialiasing schemes which basically just smooth existing images.

I suppose virutal display could also be an alterantive to SR-IOV at the consumer level but why implement something like that instead of just allowing SR-IOV... to beging with.

[boxie]

Stadia is already built on customized variant's of Vega56 so no it doesn't require anything that Navi has. The datacenter variants of Vega 10 can support SR-IOV which would obviate the need for any virtual display.

Virtual display sounds alot like virtual super resolution on windows... which just downscales from a higher resolution to the physical display meaning there is more data there than can be displayed unlike all the various antialiasing schemes which basically just smooth existing images.

I suppose virutal display could also be an alterantive to SR-IOV at the consumer level but why implement something like that instead of just allowing SR-IOV... to beging with.

Just re-read the announcement doc, they do not specifically call out a uArch - they do say "SR-IOV, 56 CU's and HBM2" Custom GPU.

Currently a vega56 DC GPU fits that - but being that AMD does a lot of custom stuff, I wouldn't be surprised if it is a custom Navi (given launch timelines) for power saving.

and yes, I had forgotten about VSR on the windows driver, it could be that feature making its way into the product line.

[cb88]

Just re-read the announcement doc, they do not specifically call out a uArch - they do say "SR-IOV, 56 CU's and HBM2" Custom GPU.

Currently a vega56 DC GPU fits that - but being that AMD does a lot of custom stuff, I wouldn't be surprised if it is a custom Navi (given launch timelines) for power saving.

and yes, I had forgotten about VSR on the windows driver, it could be that feature making its way into the product line.

Maybe, I was thinking custom more as in how can we cram as many of these into a box as possible and get low latency video encoding in there.... so I'd think a custom Vega or Navi die + custom video hardware + network uplink straight off the card.

[boxie]

Maybe, I was thinking custom more as in how can we cram as many of these into a box as possible and get low latency video encoding in there.... so I'd think a custom Vega or Navi die + custom video hardware + network uplink straight off the card.

that certainly all makes sense. Output video directly into video encoder -> pcie transit directly out of the hosts NIC would certainly be an interesting custom solution.

[DebianLinuxero]

What's the difference between AMDGPU and AMDGPU PRO?

What hides the closed source implementation?

[starshipeleven]

What's the difference between AMDGPU and AMDGPU PRO?

What hides the closed source implementation?

The "PRO" contains:
a more updated kernel driver, but that part is upstreamed regularly into AMDGPU
the proprietary userspace blob that provides OpenGL/CL/Vulkan/whatever and replaces Mesa (which is what AMDGPU uses)

[asdfgh]

<sarcasm>Will this driver be as stable as vega's one?</sarcasm>