Is Microsoft intentionally Kneecapping AMD? (Die Shrink Telegrams)
Added 2024-09-05 22:49:45 +0000 UTCFrom bidding Intel against AMD for the Next Gen XBOX, to preventing Strix from getting Copilot+ certification before Lunar Lake launches - what is going on with Microsoft and AMD?
It's a broad and open-ended question, but we think it is a good one for this week's Die Shrink! You have only ~3 hours to write-in below! Sorry for the short notice on this one!
https://youtu.be/izqEZmjTfuM?si=Fek5uXc-Ix_NV2OH
https://youtu.be/yDzVWqncMFA?si=vayrc2spa-QJyMmg
Comments
It really looks that way. Think about how CCX parking was a mess for years, but when Alder Lake launched they worked overtime to fix P-core parking. Hell, back on the Radeon 7000 remember how AMD magically got their pixel shaders revoked so Microsoft could use Nvidia’s implementation?
Ketafuki
2024-09-06 14:59:57 +0000 UTCI watched this hardware unboxed https://www.youtube.com/watch?v=izqEZmjTfuM where Steve showed unexplainable and significant performance regressions in Windows 11. Clearly gamers believed Windows 11 had bad juju and these go back to the release of the AMD's 5900 series.
Kiln God
2024-09-06 01:42:01 +0000 UTCI'm pretty sure it's to avoid the latency of communicating between the CCDs. In the current state of things 6 or 8 Zen 4/5 cores is enough to get the job done and the latency of communicating between the dies is a bigger detriment than the gain from adding more cores. Game mode parks one of the CCDs
Deadeyes
2024-09-06 01:28:30 +0000 UTCPat keeps playing double or nothing like the loser he is.
Paul Hope
2024-09-06 01:25:28 +0000 UTCHey Tom like you I'm remaining on Windows 10 until Microsoft literally rips it away from me. Any words on Windows 12 release date? Saw Microsoft was aiming for 2024. I rather upgrade to that than Windows 11.
Kinihun25
2024-09-06 01:21:51 +0000 UTCRemember the on stage handshake with Lisa Su asking "you promise?"
Shadhahvar
2024-09-06 01:16:42 +0000 UTCAt least with relation to Zen5 issues, another take on the question: Detail is still light on what *actually* has changed in the codepath to speed up Zen4/5. Some reports indicate Virtualization-Based Security (VBS) and memory integrity features are involved. Has Zen 4/5 under-invested in speeding up process/vm isolation feature speed? How will this impact the server chips vs Intel when/if Intel get moving again? On the client disabling/reducing security may be a reasonable patch. Maybe this is not the case on server.
David
2024-09-06 01:08:49 +0000 UTCOh and USBC and Thunderbolt (even on an Asus Pro art Certified thunderbolt motherboard) still do not work nearly as well as on intel)
lewis ruskin
2024-09-06 01:01:10 +0000 UTCI get to wondering what the real reason is that the multi-die AMD CPUs (7900x3D; 7950x3D; 9900X; 9950X) require Game Mode to function properly. It definitely got to making less sense when neither chiplet had 3-D V-Cache. At first it sounded like it was to adjust between different clocked chiplets with different purposes, but now that doesn't make sense as the chiplets sound like they're the same speed. It would be very nice to not have to do such strange steps just to get a CPU running smoothly.
Trevor Renfro
2024-09-06 00:56:28 +0000 UTCMicrosoft’s milking the brand before it’s in the toilet.
Paul Hope
2024-09-06 00:53:38 +0000 UTCIt still amazes me how much legacy work windows relies on. One of the best things intel made were its Nic's. So often (linux or windows regardless) their nice just work with software. Often a compatibility list is intel nic's with a couple broadcom or marvel. Does AMD need to get into that business, or buy another maker just to get the basic drivers for wifi and ethernet?
lewis ruskin
2024-09-06 00:47:12 +0000 UTCIt’s not the first time that Windows has just happened to nerf performance on Zen. It’s possible that Microsoft doesn't even bother to test development versions of Windows on AMD processors to check for performance regressions. https://m.hexus.net/tech/news/software/148544-microsoft-windows-11-ryzen-l3-cache-latency-fix-now-available/ https://www.amd.com/en/resources/support-articles/faqs/PA-400.html
Chris Rijk
2024-09-06 00:40:45 +0000 UTCOne curious thing about the recently made patch that improves Zen performance is that AMD have refused to go into any detail about what actually changed. I can think of only two possible reasons for this - either AMD made a big mistake by not getting this resolved sooner and don’t want to admit it or they’ve been sworn to secrecy by Microsoft.
Chris Rijk
2024-09-06 00:40:22 +0000 UTCPerhaps the closest analogy is how some game developers let Nvidia put in whatever they want, except in this case it would be Intel developers and Windows. Intel has contributed various things to Windows over the years and perhaps they contributed some of the security mitigations for security flaws that showed up about 5 years ago. That might have led to a situation where those mitigations were being run unnecessarily on more recent AMD processors but neither Microsoft or Intel particularly cared.
Chris Rijk
2024-09-06 00:40:15 +0000 UTCI suspect what happened was that AMD did not have an army of coders to help with Windows the way that Intel has in the past few years. For instance, Intel threw oodles of engineers at the scheduling issue between P and E cores. I wonder how much more difficult it was for AMD to get the scheduler fix for the X3D CCD problem for the 7900X3D and 7950X3D. Microsoft obviously isn't blameless. Causing AMD processors to lose what looks like a half generation of performance seems like a dumb move that I believe is stupidity or just lack of testing. If anything Microsoft should want Windows to work well on everything. Don't they want Windows to be everywhere?
Woody Chang
2024-09-06 00:08:16 +0000 UTCIt's not called Wintel for a reason, is it not true Intel developers are deeply invloved with Windows kernels? We've has performance issues with AMD ever since AMD started being extremely competitive 5900 series to compete with 12900K, 13900K and 14900K. While I do not think Microsoft intentionally handicaped AMD. I am not blind to the 6 to 12 billion in revenue Intel ould be out it Intel lost their crown. What is the likelihood a kernel developer got free vacations or kickback to weight the scale in Intels favor? Moreso, latest release to magically cure all AMD processors back to 5900 series which should have beat Intel. More so the fix is suddenly delivered with the release of Intels next generation of processors. I say "coincidence not" someone tilted the scales for Intel.
Kiln God
2024-09-05 23:58:57 +0000 UTCPeople noticed Linux was faster, and M$ loves Linux more than M$ loves AMD. /s
Nicholas Buckner
2024-09-05 23:55:10 +0000 UTCI wrote in a while back asking what Linux could do to make you switch over... On the other hand, what could MICROSOFT do (to windows or in general) to get you to make the switch? I think MS has made some very questionable decisions, like screwing AMD over with Surface, and it's not long until their poor decisions push people over the edge.
Trogdor
2024-09-05 23:46:18 +0000 UTCIt seems to me is that Microsoft wants more control over the platforms they run on which is why they seem to be pushing Qualcomm over AMD. I think they gradually saw Intel as an increased risk to their business and now want more control rather than a partnership regardless of technical prowess or product performance. As far as the Windows update and 24H2 optimizations that benefit Zen architecture cores, *my assumption* (I have no knowledge of the actual situation) is that the performance deficiencies on Windows are aftermath of the particular way that Microsoft implemented mitigations for the multiple Spectre-class vulnerabilities from around 2019 and later. I don't have knowledge about this particular case, but I have personally programmed x86 assembly code and have studied processor architecture and design at university. My understanding is that they collaborated with Intel engineers during that time they were recoding those security boundaries for dealing with Meltdown and Spectre that involve context switches between Ring 0 (most privileged execution mode) and Ring 3 (unprivileged execution) under some situations—such as when performing hardware I/O to a graphics card or storage devices which happens much more frequently in games. This would make sense with observed data with Administrator and elevated process access assumed to have kernel-level access anyway so the mitigation code likely skipped the cache/speculation invalidation instructions that were added for normal user-level and/or sandboxed processes. That cache invalidation or other mitigation method they used (one possibility could be a memory barrier such as a fence instruction) could easily cause an expensive pipeline flush or execution hazards (technical term for a situation that could cause incorrect/invalid results) in the processor which would hurt branch prediction as a side-effect. I doubt Microsoft did more than very basic testing on AMD platforms back then, so I don't think it was intentional. I think some discussions such as Microsoft "disabling branch prediction" aren't an accurate model of what likely happened in light of these possibilities. EDIT to add: I do think there is an illogical negative perception about AMD at Microsoft management (other than maybe the Xbox/gaming division) based on how often they make trade offs that hurt AMD for little or no realized gain. Hell, they refactored major portions of the Windows thread scheduling library for Intel's thread director for Alder Lake and later, and can't seem to spend any of Microsoft's own engineering time on improving AMD's thread scheduling which forced AMD to hack together core parking as a work-around to somewhat mitigate Windows's poor thread scheduling and management algorithm. AMD processors don't need core parking on Linux to perform well in games. Microsoft seems almost incapable of doing technical changes that benefit AMD unless there is public knowledge about a defect and they get pushback. There are probably a number of engineers at Microsoft that would like to improve more tooling, system libraries, and compiler optimizations for AMD hardware but there seems to be a pervasive pattern of never get allocated any time to work on such projects unless it's for Intel or their shiny new Qualcomm investment.
nijiko13
2024-09-05 23:40:32 +0000 UTCHanlon's razor states that we should never attribute to malice that which is adequately explained by stupidity. I've had Windows updates repeatedly disable my graphics card, and when that wasn't enough, an update bricked the updater. At some point it's *too* stupid to be malicious.
Benjamin Oechsli (ex-lee)
2024-09-05 23:26:41 +0000 UTCThe upcoming Windows 11 Update 24H2 is slated to provide massive performance gains to Zen-based CPUs, as covered by Hardware Unboxed. It's welcome news as it's effectively free performance for AMD users, but it also implies that Windows had been effectively bottlenecking the Zen architecture for some time. What prompted Microsoft to release this update now? Has AMD allocated more of their own engineers to work on Windows optimizations? Can we expect to see more performance gains from Windows updates like these in the future?
kjm015
2024-09-05 23:23:03 +0000 UTCDid AMD let slip "Our great hardware can't fix your crappy software."?
Nicholas Buckner
2024-09-05 23:20:47 +0000 UTCGiven how M$ being crazy about "Games for Windows Live" caused Steam to buddy up with Linux, it wouldn't surprise me if M$ was hostile to AMD if they didn't instantly kneel.
Nicholas Buckner
2024-09-05 23:18:08 +0000 UTCOld habits die hard. I don't know why Microsoft hates AMD, but clearly they do. I can say that Microsoft has long had expectations of achieving Apple's M architecture: low battery drain with next gen performance (e.g. Windows RT). Intel failed many times with any promise of achieving this (e.g. Atom). Someone at Microsoft may have sworn off the x86 architecture and/or they hold a grudge against AMD. But the handwriting on the wall and the patience of execs wearing thin may be an intervention to acknowledge that AMD may have been right in their decisions. After long neglecting AMD, Microsoft may be at a disadvantage. A.k.a. their incompetency is showing.
Ivalenz
2024-09-05 23:14:13 +0000 UTCDo you think that they're actively/maliciously doing this or is this a case of despite Intel being in the bag they're still in the majority of devices in use?
CompressedAIBlocks
2024-09-05 23:06:54 +0000 UTCI don’t think they are doing it on purpose, I do think it’s just old habits of the “wintel” years that they don’t even notice because they’ve been doing it for so long they don’t notice. Since hurting AMD is only hurting themselves and has no monetary benefit from my perspective, I’ll choose the good faith option till definitive evidence is out. I will say though I do think it’s possible there’s a windows software engineer high enough that just hates AMD.
Sad XTX 999
2024-09-05 23:06:28 +0000 UTCIt's not the first time MS kneecap AMD, and probably not the last. But I honestly don't understand the vendetta when they are actively having a relationship with the Xbox SoC.
Dark Side of the Force
2024-09-05 23:01:28 +0000 UTC1) If MS is/was kneecapping AMD, what was the motive in your opinion? Was it just an outdated policy based on intel formerly having the vast majority of the market or was there something more? 2) How much more performance for AMD do you think is hidden behind sub-optimal windows code? Or in other words, are there other areas where Linux on AMD outperforms WIndows on AMD?
Lo_Res_Gamer
2024-09-05 22:57:15 +0000 UTCIt’s because of AMD’s bad GPU drivers.
Jen-Hsun Huang
2024-09-05 22:55:02 +0000 UTCI would add zen 2 for surface being cancelled which ended up in the steam deck, ironically promoting Linux over windows
coltmarshmallow
2024-09-05 22:54:08 +0000 UTCMaybe Microsoft is still mad ATI didn’t bid cheap enough to win the OG Xbox, and ended up paying us lots of money because we wouldn’t reduce our GPU price later, And transferred that hate onto AMD when they bought ATI?
Jen-Hsun Huang
2024-09-05 22:51:54 +0000 UTC
