Topic on Glossary talk:Central Processing Unit (CPU)
This page shows the changes between two versions of a post by Mirh in the topic "AMD hidden shenanigans" on Glossary talk:Central Processing Unit (CPU).
You can see other versions of this post at its history page.
| Line 1: | Line 1: | ||
I'm not talking about the usual just-so-frequent incompatibilities, crashes, or other obvious problems, and not even about the [https://www.overclockers.com/laptop-hacking/ various] [https://semiaccurate.com/2012/01/09/spotlight-brazostweaker/ hacked] [https://www.overclock.net/forum/10-amd-cpus/1730316-undervolting-bristol-ridge-laptops-feat-a10-9600p.html overclocking] [https://github.com/FlyGoat/RyzenAdj tools]. But something a bit more in the shadow: | I'm not talking about the usual just-so-frequent incompatibilities, crashes, or other obvious problems, and not even about the [https://www.overclockers.com/laptop-hacking/ various] [https://semiaccurate.com/2012/01/09/spotlight-brazostweaker/ hacked] [https://www.overclock.net/forum/10-amd-cpus/1730316-undervolting-bristol-ridge-laptops-feat-a10-9600p.html overclocking] [https://github.com/FlyGoat/RyzenAdj tools]. But something a bit more in the shadow: | ||
| + | |||
| + | * Original Phenoms cpu had a very [https://www.anandtech.com/show/2477/2 concerning] bug with TLB. Gaming isn't as sensitive of a task as compiling tax returns though, so bold people could try to [http://www.xtremesystems.org/forums/showthread.php?171105-Howto-disable-the-tlb-fix-with-CrystalCPU disable] the software workaround. | ||
* Llano has [https://community.hwbot.org/topic/185623-fm1-and-the-unexplainable-gpupi-performance/ quite reduced] DIV/IDIV performance <small>(I'm really [https://bugzilla.mozilla.org/show_bug.cgi?id=1281759#c39 not] the the guy recommending to undo errata fixes, but considering all the recent guarding around speculative execution.. maybe there is a chance?)</small> | * Llano has [https://community.hwbot.org/topic/185623-fm1-and-the-unexplainable-gpupi-performance/ quite reduced] DIV/IDIV performance <small>(I'm really [https://bugzilla.mozilla.org/show_bug.cgi?id=1281759#c39 not] the the guy recommending to undo errata fixes, but considering all the recent guarding around speculative execution.. maybe there is a chance?)</small> | ||
* Bulldozer has [https://web.archive.org/web/20131007224201/http://www.xtremesystems.org/forums/showthread.php?286448-The-Book-of-Bulldozer-Revelations-Episode-2-(SuperPI-x87) crippled] x87 performance <small>(there are no public errata regarding the feature, so this might be all free real estate?)</small> | * Bulldozer has [https://web.archive.org/web/20131007224201/http://www.xtremesystems.org/forums/showthread.php?286448-The-Book-of-Bulldozer-Revelations-Episode-2-(SuperPI-x87) crippled] x87 performance <small>(there are no public errata regarding the feature, so this might be all free real estate?)</small> | ||
| + | |||
| + | * Bulldozer (and I could swear Bobcat) CPUs are liable to [[Store:Steam#Steam CEG protected games crash or freeze the system (AMD Bulldozer CPU)|system locking up]] on steam CEG games loading. Given the [https://salsa.debian.org/hmh/amd64-microcode/-/blob/releases/bpo9/debian/changelog#L304 time window], and given some [https://bugs.winehq.org/show_bug.cgi?id=32515#c49 wine] [https://bugs.winehq.org/show_bug.cgi?id=46472#c43 discussions] involving debug registers, I believe this might have been [https://www.amd.com/system/files/TechDocs/48063_15h_Mod_00h-0Fh_Rev_Guide.pdf#page=57 erratum 671] <small>(Trinity's AGESA has indeed [https://github.com/coreboot/coreboot/blob/4.13/src/vendorcode/amd/agesa/f15tn/Proc/CPU/Family/0x15/TN/F15TnMsrTables.c#L159 provisions] for it)</small> | ||
* Possibly every architecture before Ryzen [https://www.overclock.net/forum/10-amd-cpus/1499562-undervolting-editing-p-states-piledriver-using-amd-mrstweaker-mini-guide.html could hit] the TDP ceiling before even reaching the maximum frequencies <small><strike>(unclear whether this couldn't just be specific to some motherboards and/or workload though)</strike></small> | * Possibly every architecture before Ryzen [https://www.overclock.net/forum/10-amd-cpus/1499562-undervolting-editing-p-states-piledriver-using-amd-mrstweaker-mini-guide.html could hit] the TDP ceiling before even reaching the maximum frequencies <small><strike>(unclear whether this couldn't just be specific to some motherboards and/or workload though)</strike></small> | ||
| − | :* ok this is probably just [https://www.tomshardware.com/reviews/fx-8150-zambezi-bulldozer-990fx,3043-8.html Turbo Core] ([https://www.anandtech.com/show/3674/amds-sixcore-phenom-ii-x6-1090t-1055t-reviewed/2 whatever] [https://www.xtremehardware.com/eng-articles/llano-architecture-201107015403/?start=12 its] [https://www.hotchips.org/wp-content/uploads/hc_archives/hc24/HC24-3-ManyCore/HC24.28.325-Trinity-Nussbaum-AMD-v2.pdf features] [https://www.guru3d.com/articles-pages/amd-fx-8320e-processor-review,3.html unclear] [https://www.hotchips.org/wp-content/uploads/hc_archives/hc25/HC25.30-SoC2-epub/HC25.26.312-Richland_Hotchips25_PraveenDongara_FINAL_presented1.pdf#page=5 progression]) being dynamically adjusted according to the current load. It can be fooled then, and some OEMs may actually be enforcing ''All Core Turbo'' from the firmware ([https://forum.ixbt.com/topic.cgi?id=9:68421 ''Turbo CPB''], [https://fcenter.ru/online/hardarticles/motherboards/37219-Gigabyte_G1_Sniper_A88X_plata_Socket_FM2_igrovoj_serii_G1_Killer#05 ''Turbo Performance Boost Ratio''] and [https://www.techpowerup.com/120454/asus-develops-own-technology-to-rival-amd-turbo-core-and-intel-turbo-boost ''Turbo Unlocker''] being some likely names?). Or at least I can't explain why else the same CPU getting [https://overclockers.ru/lab/show/77524_3/obzor-i-testirovanie-processora-amd-athlon-x4-845-v-ispolnenii-socket-fm2-znakomstvo-i-njuansy-razgona downclocked] for clear [http://oc.jagatreview.com/2016/03/hands-on-review-overclocking-athlon-x4-845-carrizo-murah-tanpa-igp/4/ TDP reasons] would be reported elsewhere [https:// | + | :* ok this is probably just [https://www.tomshardware.com/reviews/fx-8150-zambezi-bulldozer-990fx,3043-8.html Turbo Core] ([https://www.anandtech.com/show/3674/amds-sixcore-phenom-ii-x6-1090t-1055t-reviewed/2 whatever] [https://www.xtremehardware.com/eng-articles/llano-architecture-201107015403/?start=12 its] [https://www.hotchips.org/wp-content/uploads/hc_archives/hc24/HC24-3-ManyCore/HC24.28.325-Trinity-Nussbaum-AMD-v2.pdf features] [https://www.guru3d.com/articles-pages/amd-fx-8320e-processor-review,3.html unclear] [https://www.hotchips.org/wp-content/uploads/hc_archives/hc25/HC25.30-SoC2-epub/HC25.26.312-Richland_Hotchips25_PraveenDongara_FINAL_presented1.pdf#page=5 progression]) being dynamically adjusted according to the current load. It can be fooled then, and some OEMs may actually be enforcing ''All Core Turbo'' from the firmware ([https://forum.ixbt.com/topic.cgi?id=9:68421 ''Turbo CPB''], [https://fcenter.ru/online/hardarticles/motherboards/37219-Gigabyte_G1_Sniper_A88X_plata_Socket_FM2_igrovoj_serii_G1_Killer#05 ''Turbo Performance Boost Ratio''] and [https://www.techpowerup.com/120454/asus-develops-own-technology-to-rival-amd-turbo-core-and-intel-turbo-boost ''Turbo Unlocker''] being some likely names?). Or at least I can't explain why else the same CPU getting [https://overclockers.ru/lab/show/77524_3/obzor-i-testirovanie-processora-amd-athlon-x4-845-v-ispolnenii-socket-fm2-znakomstvo-i-njuansy-razgona downclocked] for clear [http://oc.jagatreview.com/2016/03/hands-on-review-overclocking-athlon-x4-845-carrizo-murah-tanpa-igp/4/ TDP reasons] would be reported elsewhere [https://old.reddit.com/r/Amd/comments/bcsppb/athlon_x4_845_sustained_allcore_turbo_and_is_it/ happily running] dozens of watts above the official limit. |
* Steamroller CPUs are [https://www.tomshardware.com/reviews/amd-a10-7890k-gaming-performance-benchmark,4491-2.html preemptively throttled] when their iGPU is in use <small>(which is not as bad as it sounds [https://www.overclock.net/threads/amd-kaveri-refresh-a10-7870k-apu-arrives.1555615/page-5 usually], even in games it should be hard to be CPU-limited without a dedicated graphics card after all, but it is not impossible)</small>. If a [https://www.overclock.net/forum/10-amd-cpus/1588994-disabling-igpu-usage-induced-cpu-throttling-geapm-steamroller-apus.html modded BIOS] isn't available, editing p-states and disabling [https://community.hwbot.org/topic/83268-the-official-asrock-fm2-a10-oc-competition-thread/page/2/#comment-258042 disabling] APM may still do something, but [https://www.overclock.net/threads/amd-kaveri-a10-7850k-overclock-guide.1460028/page-63 not very] much: | * Steamroller CPUs are [https://www.tomshardware.com/reviews/amd-a10-7890k-gaming-performance-benchmark,4491-2.html preemptively throttled] when their iGPU is in use <small>(which is not as bad as it sounds [https://www.overclock.net/threads/amd-kaveri-refresh-a10-7870k-apu-arrives.1555615/page-5 usually], even in games it should be hard to be CPU-limited without a dedicated graphics card after all, but it is not impossible)</small>. If a [https://www.overclock.net/forum/10-amd-cpus/1588994-disabling-igpu-usage-induced-cpu-throttling-geapm-steamroller-apus.html modded BIOS] isn't available, editing p-states and disabling [https://community.hwbot.org/topic/83268-the-official-asrock-fm2-a10-oc-competition-thread/page/2/#comment-258042 disabling] APM may still do something, but [https://www.overclock.net/threads/amd-kaveri-a10-7850k-overclock-guide.1460028/page-63 not very] much: | ||
| Line 13: | Line 17: | ||
:* this is unlike [https://bjorn3d.com/2012/10/amd-virgo-platform-2nd-generation-apu/ Trinity] being able to sustain even turbo on both simultaneously (though with its [https://bjorn3d.com/2013/06/amd-richland-june-5-2013-1201am-edt-nda/ idiot balls]), Llano just barely knowing how to turbo [https://www.anandtech.com/show/4444/amd-llano-notebook-review-a-series-fusion-apu-a8-3500m/4 its CPU], Jaguar [https://www.tomshardware.com/reviews/kabini-a4-5000-review,3518-3.html somehow] regressing with just [https://web.archive.org/web/20140508165702/http://techreport.com/review/26377/a-first-look-at-amd-mullins-mobile-apu/ one CPU model] out of stone age (Puma came back with the very interesting concept of [https://hwbot.org/news/10740_amds_beema_and_mullins_architecture_stapm_lower_power_nothing_for_enthusiasts_but_ivr_in_2015/ frequency sensitivity] though) and Bobcat being a ([https://setiathome.berkeley.edu/forum_thread.php?id=73431 very]) [https://www.nordichardware.se/nyheter/amd-updates-brazos-with-turbo-core-and-3d.html mixed bag]. | :* this is unlike [https://bjorn3d.com/2012/10/amd-virgo-platform-2nd-generation-apu/ Trinity] being able to sustain even turbo on both simultaneously (though with its [https://bjorn3d.com/2013/06/amd-richland-june-5-2013-1201am-edt-nda/ idiot balls]), Llano just barely knowing how to turbo [https://www.anandtech.com/show/4444/amd-llano-notebook-review-a-series-fusion-apu-a8-3500m/4 its CPU], Jaguar [https://www.tomshardware.com/reviews/kabini-a4-5000-review,3518-3.html somehow] regressing with just [https://web.archive.org/web/20140508165702/http://techreport.com/review/26377/a-first-look-at-amd-mullins-mobile-apu/ one CPU model] out of stone age (Puma came back with the very interesting concept of [https://hwbot.org/news/10740_amds_beema_and_mullins_architecture_stapm_lower_power_nothing_for_enthusiasts_but_ivr_in_2015/ frequency sensitivity] though) and Bobcat being a ([https://setiathome.berkeley.edu/forum_thread.php?id=73431 very]) [https://www.nordichardware.se/nyheter/amd-updates-brazos-with-turbo-core-and-3d.html mixed bag]. | ||
| − | * speaking of APM, that's also [https://www.overclock.net/forum/10-amd-cpus/1328938-what-core-c6-state-exactly.html#post24500096 '''''necessary''''' condition] for turbo boost states (that are in turn a [https:// | + | * speaking of APM, that's also [https://www.overclock.net/forum/10-amd-cpus/1328938-what-core-c6-state-exactly.html#post24500096 '''''necessary''''' condition] for turbo boost states (that are in turn a [https://old.reddit.com/r/Amd/comments/536khk/oced_the_athlon_x4_845_to_40ghz/ godsend] for non-K-unlocked models). Unless ''sure'' it's holding the processor back (or overclocking some [https://www.guru3d.com/articles-pages/amd-fx-8150-processor-review,8.html plutonium core]) it's very much [https://superuser.com/questions/1226526/is-it-preferable-to-disable-apm-or-to-enable-load-line-calibration-for-overcloc better off] enabled. |
| + | |||
| + | * 2014 core designs were somewhat of a backstage [https://forums.anandtech.com/threads/new-zen-microarchitecture-details.2465645/page-35 trainwreck], on the other hand their power management started to rely more and more on the SMU. Nothing of interest in the HEDT market, but on laptops it introduced a buttload of [https://www.overclock.net/threads/carrizo-fx-8800p-a10-8700p-info-and-optimizations.1573937/page-2#post-24437135 hidden] [https://www.overclock.net/threads/jagatreview-hands-on-amd-fx-8800p-carrizo.1560230/page-22#post-24306130 interesting] variables. | ||
| + | |||
| + | * all PCIe 3.0 motherboards should [https://www.spinics.net/lists/linux-pci/msg65658.html support] [https://github.com/JuliaComputing/nvidia-driver-pcie-rebar/issues/2 somehow] Resizable-BAR (needed for Smart Access Memory), but [https://old.reddit.com/r/Amd/comments/k6cbk7/zen_zen_and_zen_2_cannot_support_sam_due_to_the/gek8sop/ before] Zen 3 CPUs a [https://twitter.com/SebAaltonen/status/1327191532177420288 critical instruction] for such graphical applications was emulated slow enough to make it useless | ||
| + | |||
| + | * I couldn't find [https://www.overclock.net/forum/78-pc-gaming/1487105-want-higher-fps-what-timer-do-you-use-2.html#post26423128 anybody] on the net [https://www.anandtech.com/show/12678/a-timely-discovery-examining-amd-2nd-gen-ryzen-results/4 seriously] reporting that forcing HPET helped them, save for touching BCLK [https://web.archive.org/web/20150919163144/http://forum.notebookreview.com/threads/must-read-setfsb-o-c-game-problems-fixed.569893/ live] or under [https://web.archive.org/web/20160314074421/https://www.ocaholic.ch/modules/smartsection/item.php?itemid=1071&lang=english W8](+?). Still for some reason [https://forums.guru3d.com/threads/another-look-at-hpet-high-precision-event-timer.368604/page-42 something] in the AMD drivers always does/did that. | ||
| − | * | + | * fTPM <small>([https://bugzilla.kernel.org/show_bug.cgi?id=216989#c94 necessary] but not sufficient condition?)</small> ''was'' [https://hothardware.com/news/amd-bios-updates-resolve-ryzen-stuttering-issue confirmed] to be able to [https://www.phoronix.com/news/Torvalds-fTPM-RNG-Woes cause] hard stutters |
| − | And I could swear I knew more, but time was hard on my memory. | + | And I could swear I knew more, but time was hard on my memory. I wonder if all of this couldn't be enough to resurrect the old [[Special:Undelete/AMD_(CPU)|AMD (CPU)]] page we had. |
<small>p.s. Ryzen is so [https://www.bitsandchips.it/english-news/xfr-and-its-anchestors stacked] [https://www.gamersnexus.net/guides/3491-explaining-precision-boost-overdrive-benchmarks-auto-oc complex] and [https://www.techpowerup.com/review/1usmus-power-plan-for-amd-ryzen-new-developments/ wide] that you'd need a [https://www.anandtech.com/show/14873/reaching-for-turbo-aligning-perception-with-amds-frequency-metrics- manual] just to understand the stock situation.</small> | <small>p.s. Ryzen is so [https://www.bitsandchips.it/english-news/xfr-and-its-anchestors stacked] [https://www.gamersnexus.net/guides/3491-explaining-precision-boost-overdrive-benchmarks-auto-oc complex] and [https://www.techpowerup.com/review/1usmus-power-plan-for-amd-ryzen-new-developments/ wide] that you'd need a [https://www.anandtech.com/show/14873/reaching-for-turbo-aligning-perception-with-amds-frequency-metrics- manual] just to understand the stock situation.</small> | ||