Announcement

Collapse
No announcement yet.

Linux 5.11 Drops AMD Zen Voltage/Current Reporting Over Lack Of Documentation

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • #31
    Originally posted by Sonadow View Post
    People are getting too damn entitled here.
    I think you're trying to say "self-entitled", but that of course doesn't make sense as these people paid money for AMD products and are thus AMD customers. Paying customers get to voice complaints about their purchases.

    Originally posted by Sonadow View Post
    AMD should just drop all their open source support this instant, abandon every single open source driver they have ever worked on and cease adding support to any new products in the existing OSS drivers to show the users who's boss. Let the so-called community pick up the pieces themselves if they think they are so great, like what is going on with Nouveau.
    You're getting pretty heated reading these forum posts, I'd suggest taking a break.

    Comment


    • #32
      Originally posted by duby229 View Post
      I'm absolutely positive Linux users contribute a much larger percentage of AMD's profit than gets recognized by most people. Especially now that we have really good graphics drivers to use.
      I don't have the numbers, I take it you don't either and that no one from AMD is going to share a good breakdown of such to the public. The 2% metric afaik is only regarding desktop OS, not other slices like server market where I think AMD has been cutting into Intel market share with EPYC and we know that Linux is a pretty dominant OS in this area. Workstations, like for VFX work can often use Linux too at big studios, you'd get some high value desktop OS scenarios there but these aren't machines that'd be overclocked or running on battery power.

      Whatever the metrics of value of linux users to AMD, you have to narrow the scope down to those where this reporting would be valued highly like you do. For me it matters a bit if I want to buy a laptop and make the most out of battery life and performance when needed, but I'm expecting schedulers in the kernel to leverage that information without my involvement.

      The 5xxx series is meant to have an undervolting feature that works via firmware update AFAIK, but it can undervolt cores and boost voltage on a core that needs more perf, I forget the marketing name for it. I'm waiting on the next APU generation with Zen 3, so not sure how much more useful temp/power reporting is going to be for that. I understand it may be more valuable to earlier generations of their Zen series, especially for enthusiasts, but while there may be real world benefits to some degree for most users on linux indirectly, many probably view it as "nice to have" vs critical purchasing decision.

      Originally posted by duby229 View Post
      The second problem that I have is that Alex has expressed multiple times now that he doesn't really care about linux support.
      He responded on a linux focused forum community, he didn't have to. To me that shows he must care to some degree, considering the response could generate backlash like it did which can wrongly as representing AMD as a whole. From what Bridgman said, his role is not responsible for handling CPU features/support, his position might hold some weight and be able to relay the interest through the correct channels, but berating the guy isn't going to encourage nice gestures like that.

      Getting angry at customer service might work, but I rarely see it have a positive outcome against those whom are closer to the actual process of making things happen. I get why you're unhappy about it, why you feel it's quite important to have prioritized, but if you already view Alex as not caring...why would saying negative things directed at him encourage him to help enable raising the priority of the feature? It works against your benefit to take that approach and can cause the opposite effect.

      Better approaches are to express in a positive manner how useful the feature would be, that you look forward to seeing it and finding like minded customers that share your desire for it (like those silly petitions). Then for those at AMD involved, it's received by users more like an exciting gift that is praised vs disgruntled customers complaining "About f**king time FFS! AMD should have had this ready on release!". Which reaction sounds more motivating to invest time working on a feature, which reaction seems more positive towards a company image by making such a business decision, etc.

      Without knowing more, I assume it's on the TODO list and that there is currently more important tasks being worked on, whatever those are. The problem was identified as a lack of available talent to allocate that much time for a feature not considered top priority. No need to blame individuals, but fair enough to question why AMD isn't able to grow the linux team further to address that bottleneck. As Bridgman pointed out, there's also a slower process to go through as is expected at larger businesses, especially with all the IP/Legal barriers that have to go through approval before releasing more info or contributions to public. Seems like the higher priority was as he mentioned, having the CPU division adopt the improved processes that the GPU division now has in place.

      Comment


      • #33
        Originally posted by duby229 View Post

        Thank you for the level headed response.

        So I have basically two issues,
        If only.

        Comment


        • #34
          This is really bad - https://github.com/ocerman/zenpower/issues/39

          bridgman I understand the current situation and your role but may be possible to provide some raw information under NDA to some of these guys quickly?
          Because in opposite we will have such temp and voltage lack (yes it's not only about voltage!) for next ~6 months. Proper documentation and all legal procedures can be done somehow after.

          Comment


          • #35
            Originally posted by sandy8925 View Post
            Cut the bullshit.
            Please, learn some manners.

            Originally posted by sandy8925 View Post
            This is extreme important information, especially given AMDs weird boost behaviour. Using a CPU like the 3900X to it's full extent, requires the ability to see voltage, current, temperature etc. from userspace. This is very much critical functionality.
            I got my 5950X a couple of days ago and I didn't even realize that I don't have that sensor data available. What I did notice is that my system bootet up and worked flawlessly on the very first try, the massive amount of processing power I got and that my productivity increased as a direct result of that.

            So in total I'm simply grateful for an awesome product, the support I as a Linux user get from AMD and the work Alex and his team does for me. Totally worth the money.

            Comment


            • #36
              AMD: We don't have enough resources to deal with non-critical issues ATM.
              "not entitled" and "reasonable" Linux users: You treat us like shit!

              Comment


              • #37
                Originally posted by niner View Post
                So in total I'm simply grateful for an awesome product, the support I as a Linux user get from AMD and the work Alex and his team does for me. Totally worth the money.
                I got my 5950X also a couple of days ago and manually tuned it with an all-core frequency/voltage setting in the BIOS because the AMD supplied curve optimizer was glitchy at best or outright crashing my system (even with very small iterations and doing the tuning steps by guide)
                Now I have 4.6GHz all-core frequency at 1.25V and the CPU doesn't even reach 60°C under full load (custom watercooling)

                In short: You can still manually tune your CPU (and a lot of applications benefit from an all core setting vs dynamic one), but the disadvantage is that no power saving features (nor any safeties) are present with such a manual tune.

                In general, if you want a dynamic auto-tuning, its hard to throw some raw numbers into some sort of configuration file, as the silicon lottery spreads the f/v curve so far in both directions, that you either have an unstable system if you are too agressive or you throw away a lot of energy for nothing if you are too conservative.

                I liked 1usmus approach of testing every core with an auto-benchmark while retaining all the clocking-features build into the CPU. This way you set you CPU close to the optimal f/v curve for every single core, evade a whole bunch of outcries of unstable CPU from the community and still run the CPU with all power saving features enabled. This would also benefit the newer CPU scheduler approaches and could ride in synergy with them.

                Anyhow, this also requires a stable AGESA and the current one right now isn't. A significant amount of people report crashes under stock settings.

                Comment


                • #38
                  When I understood that right, AMD has just do deliver documentation. Writing the drivers is done by awesome other people.
                  So I do not really understand AMDs problem. No one told them to put resources into it.

                  And I cant believe, that they do not already have documentation. I don't think the AMD official (windows) tools where done by reverse engineering.

                  Comment


                  • #39
                    Originally posted by obri View Post
                    When I understood that right, AMD has just do deliver documentation. Writing the drivers is done by awesome other people.
                    So I do not really understand AMDs problem. No one told them to put resources into it.
                    I think they were asked exactly that, to do work to publish documentation that doesn't yet exist.

                    Originally posted by obri View Post
                    And I cant believe, that they do not already have documentation. I don't think the AMD official (windows) tools where done by reverse engineering.
                    That Windows software might have used some internal documentation that might be scattered around and is not publishable as such. They would need to put it together and ensure that it's something they can release. I would think it's not minor amount of work to do.

                    Anyway, I don't know what it's really like and I don't have any inside knowledge but I doubt it's just matter of publishing.

                    Comment


                    • #40
                      Originally posted by duby229 View Post
                      The second problem that I have is that Alex has expressed multiple times now that he doesn't really care about linux support. And you might say who cares if he doesn't care, except he has a really high level position where caring about linux support is critical. You might also say that he never did express that he doesn't care, but come on, read between lines, what he definitely did say was that proper reporting of critical metrics isn't critical. If that doesn't mean he doesn't he doesn't care, then what does it mean?
                      Wow - what fake news alternate reality do you live in? agd4f has been working on OpenSource AMD drivers for over 10 years now (including before he worked at AMD).

                      I'm surprised AMD bother posting here. I wouldn't.

                      Comment

                      Working...
                      X