AMD Publishes Micro Engine Scheduler "MES" Firmware Documentation

[phoronix]

Phoronix: AMD Publishes Micro Engine Scheduler "MES" Firmware Documentation

As expected, AMD today published the Micro Engine Scheduler "MES" firmware documentation for RDNA3 graphics processors as part of better engaging with the open-source community and aiming to address some gaps in their open-source GPU compute stack...

https://www.phoronix.com/news/AMD-MES-Firmware-Docs

[Jbk0]

What's up with that PDF's metadata though lol? "Burst Clock Data Recovery for 1.25G/2.5G PON Applications in UltraScale Devices Application Note"

[ms178]

The bad news is, that AMD wants to get rid of MES for RDNA2: https://lists.freedesktop.org/archiv...ay/107975.html

I seriously hoped it would go the other way instead, opening up MES on RDNA2 and making it available on Linux as well.

[cend]

With slim chance amdgpu could end up becoming a thing in Linux-libre...

[cb88]

The bad news is, that AMD wants to get rid of MES for RDNA2: https://lists.freedesktop.org/archiv...ay/107975.html

I seriously hoped it would go the other way instead, opening up MES on RDNA2 and making it available on Linux as well.

Did you even READ that message? It was just dropping MES 10.1... because no GPU ever used it, and the card in question uses MES 11 version. So the feature is fine in newer cards, AMD frequently does prototype implementaitions of upcoming features in cards that never see the light of day because they have bugs in the implementation they needed to work out still.

[jindam]

nice , hope they will live upto other expectations

[timofonic]

This is overview documentation, not full programming documentation. It's just marketing with a bit of superficial tech info.

I'll wait to the day they release COMPLETE documentation, AMD never does that.

[ms178]

Did you even READ that message? It was just dropping MES 10.1... because no GPU ever used it, and the card in question uses MES 11 version. So the feature is fine in newer cards, AMD frequently does prototype implementaitions of upcoming features in cards that never see the light of day because they have bugs in the implementation they needed to work out still.

I've had my conversations with AMD developers in this forum about this feature some time ago. And yes, I did READ that message but it seems there is a misunderstanding on your side that makes your attacking tone even more inappropriate. MES is already used on Windows on RDNA2 cards, so RDNA2 cards are technically perfectly capable to use that feature. The Linux devs just don't want to invest further work to switch over to use it with RDNA2 on Linux, too. That's a pity in my eyes due to the opening up of the MES firmware and all the third-party work to improve it that won't trickle down to RDNA2 users on Linux if that plan goes through. AMD's RDNA2 users should object to that code deletion to leave the door open for the community to finish the work that AMD seemingly doesn't have an interest in any longer.

[agd5f]

I've had my conversations with AMD developers in this forum about this feature some time ago. And yes, I did READ that message but it seems there is a misunderstanding on your side that makes your attacking tone even more inappropriate. MES is already used on Windows on RDNA2 cards, so RDNA2 cards are technically perfectly capable to use that feature. The Linux devs just don't want to invest further work to switch over to use it with RDNA2 on Linux, too. That's a pity in my eyes due to the opening up of the MES firmware and all the third-party work to improve it that won't trickle down to RDNA2 users on Linux if that plan goes through. AMD's RDNA2 users should object to that code deletion to leave the door open for the community to finish the work that AMD seemingly doesn't have an interest in any longer.

MES was not productized on RDNA2 on any OS. It was a validation vehicle for both Windows and Linux for the MES for RDNA3 which was productised. It also doesn't really bring any additional functionality over the existing hardware scheduler that it would ultimately replace in RNDA3.

[ms178]

MES was not productized on RDNA2 on any OS. It was a validation vehicle for both Windows and Linux for the MES for RDNA3 which was productised. It also doesn't really bring any additional functionality over the existing hardware scheduler that it would ultimately replace in RNDA3.

Even if it was only used as a validation vehicle on RDNA2, I cannot see a downside of opening it up and letting the community work on it if there was interest in it? I refer to the talk of TinyCorp writing a Fuzzer, fixing some bugs for the MES for RDNA3 and potentially optimize it further for AI workloads which I assume could also trickle down or be extended in scope to RDNA2 if there were no fundamental differences on the firmware/hardware level, right? As I see it, opening up the MES firmware and letting the community work on MES for RDNA2 could be an opportunity to pull in some fixes for RDNA2, too.