The Anticipated Linux Driver Requirements For The Radeon Instinct MI50 / MI60 (Vega 20)

[phoronix]

Phoronix: The Anticipated Linux Driver Requirements For The Radeon Instinct MI50 / MI60 (Vega 20)

With this week's announcement of the Radeon Instinct MI60 and MI50 as what we previously knew as Vega 20, here's a look at what is likely required from the Linux software side for making use of these professional GPUs that will begin shipping in early 2019...

http://www.phoronix.com/scan.php?page=news_item&px=Radeon-Instinct-MI50-MI60-Linux

[Brisse]

No mention of Windows-support. Looks like the card will be Linux-exclusive.

[gorgone]

yes for now no direct windows support from amd

[pegasus]

Smart move. Do you know anyone who does ML at scale on windows?

What I'm mostly interested in is their ability to translate cuda to c++ and then build c++ to run on their GPUs. This is absolutely critical functionality if they want to even dent the huge cuda software ecosystem out there.

[Michael]

No mention of Windows-support. Looks like the card will be Linux-exclusive.

Unfortunately I wasn't invited to this event to learn more or ask questions :/

[boxie]

Unfortunately I wasn't invited to this event to learn more or ask questions :/

what do you think your chances are of getting a review sample?

[Michael]

what do you think your chances are of getting a review sample?

For the MI50/MI60? Probably very unlikely... I haven't received any AMD professional cards since before the GCN days, 2011. Nor have I been briefed on RX 590 - I assume they did some briefing/sampling during the AMD event this week after hours.

[starshipeleven]

No mention of Windows-support. Looks like the card will be Linux-exclusive.

Card mostly used for computing and AI runs only on the OS that is most commonly used for computing and AI, more news at 20:00.

[mphuZ]

ROCM 2.0 will be released in December 11-25, along with a big Radeon Software update.

[polarathene]

I wonder how these compare vs nvidia's offerings with say tensorflow?

Does Michael have any machine learning benchmarks in place with PTS? Maybe one of the AMD staff that read here would be able to help get one of the cards loaned out to Michael if AMD is not big on donating one.