Intel, AMD, VIA & Freescale CPU Microcode Repositories Discussion

Windows 11 Insider Preview 22463.1000 (22H2), nothing new.

I’m joking… 27(!) new microcodes.

cpu306F2_plat6F_ver00000048_2021-05-18_PRD_849F46EB
cpu306F4_plat80_ver0000001A_2021-05-24_PRD_A3E4C533
cpu406E3_platC0_ver000000EC_2021-04-28_PRD_C2489977
cpu406F1_platEF_ver0B000040_2021-05-19_PRD_E5BABFDB
cpu50654_platB7_ver02006C0A_2021-06-16_PRD_27BE61D6
cpu50663_plat10_ver0700001C_2021-06-12_PRD_FE624729
cpu50664_plat10_ver0F00001A_2021-06-12_PRD_BF95B0FA
cpu506C9_plat03_ver00000046_2021-05-10_PRD_EF971B58
cpu506CA_plat03_ver00000024_2021-05-10_PRD_AB75708E
cpu506F1_plat01_ver00000036_2021-05-10_PRD_60ADB6C8
cpu706A1_plat01_ver00000038_2021-05-10_PRD_34C752B7
cpu706A8_plat01_ver0000001C_2021-05-10_PRD_DBA87E35
cpu806E9_plat10_ver000000EC_2021-04-28_PRD_B69469E4
cpu806E9_platC0_ver000000EC_2021-04-28_PRD_E0D55C3D
cpu806EA_platC0_ver000000EC_2021-04-28_PRD_BF88FAFE
cpu806EB_platD0_ver000000EC_2021-04-28_PRD_77AA97DD
cpu806EC_plat94_ver000000EC_2021-04-28_PRD_AB1F6D0C
cpu906E9_plat2A_ver000000EC_2021-04-29_PRD_8C68349C
cpu906EA_plat22_ver000000EC_2021-04-28_PRD_9C25B317
cpu906EB_plat02_ver000000EC_2021-04-28_PRD_2A9C4937
cpu906EC_plat22_ver000000EC_2021-04-28_PRD_96F6D3AD
cpu906ED_plat22_ver000000EC_2021-04-28_PRD_5FBCEE1C
cpuA0652_plat20_ver000000EC_2021-04-28_PRD_10B9FB72
cpuA0653_plat22_ver000000EC_2021-04-28_PRD_F3279785
cpuA0655_plat22_ver000000EE_2021-04-28_PRD_8949FD67
cpuA0660_plat80_ver000000EA_2021-04-28_PRD_E5ECDB65
cpuA0661_plat80_ver000000EC_2021-04-29_PRD_DE37D5C8

22463.1000.rar (1.4 MB)


Oh my… Haha. That’s a ~1 year early build, probably that’s why.

Now this is interesting, some of those microcodes are quite a jump in size from previous versions, as I was interested in 406F1 I noticed that it has changed from 31KB to 35KB in a single version, most previous releases have been 1 KB larger every few versions or even keeping the same size over multiple versions. Some of the 27 new ones are the same size or 1 KB larger than previous, those are boring, but the interesing ones:
306F2 34 > 38 KB (4KB larger, 12% increase)
306F4 19 > 23 KB (4KB larger, 21% increase)
406F1 31 > 35 KB (4KB larger, 13% increase)
50654 36 > 42 KB (6KB larger, 16% increase)
50663 24 > 28 KB (4KB larger, 16% increase)
50664 24 > 27 KB (3KB larger, 12% increase)

And all of these size jumps are in the window of about ~3-4 months from previous version (not comparing years old size to current size).

So either Intel is heavily optimizing all the Spectre etc recent years vulnerability workarounds or there could be a workaround for a yet to come new vulnerability.

and/or it’s windows 11 related optimizations/customizations etc.



Windows 11 is about to be released any day now. The FINAL non-beta, non-preview, the release version of Windows 11. These new microcodes are from the NEXT development branch of Windows 11 - it is HIGHLY doubtful Intel did any work on these older generation CPUs for any sort of Windows 11 "tuning".



Windows 11 is about to be released any day now. The FINAL non-beta, non-preview, the release version of Windows 11. These new microcodes are from the NEXT development branch of Windows 11 - it is HIGHLY doubtful Intel did any work on these older generation CPUs for any sort of Windows 11 "tuning".



Maybe if you consider Skylake-W Xeon and Skylake-X core i9 (79xx X/XE) old. Skylake-X is testing fine on Windows 11 beta builds with previous cpu50654 uCode from March. But, remember Microsoft added Skylake-x/w (cpu50654) late to the compatibility list late… maybe you’re right and it’s all security or maybe there is something in there that was needed to make windows 11 work better for cpu50654 Xeons and Skylake-X. Intel need to support corporate Xeon customers more than few years…

@Marvin I noticed that Intel gave up skylake (ID 506E3) on this release, and only my CPUID 906E9 among the 806Ex and 906Ex was increased by 2KB. BTW the ID 906E9 was the only one that didn’t get support for Windows 11. I think it could be a Windows 11 supportive release.

Anyhow, The new MC gave me a noticeable better results on CPU-Z benchmark in Windows 11, but there was a huge discrepancy although there was no delay testing the same MC twice.

Old MC EA:

Screenshot 2021-09-23 011826.png

Screenshot 2021-09-23 011857.png



New MC EC:

Screenshot 2021-09-23 013037.png

Screenshot 2021-09-23 013123.png

Finally tested on real HW:
cpu506C9_plat03_ver00000046_2021-05-10_PRD_EF971B58.bin

CPU is now reporting:
IA32_ARCH_CAPS=0xc79
on previous version it was:
IA32_ARCH_CAPS=0x79

So it is now reporting 2 undocumented feature bits not mentioned in:
https://software.intel.com/content/www/u…ml#inpage-nav-2

Just like in the past when spectre/mds/etc workarounds were being implemented with newer microcode revisions it changed from 0x1 to 0x19 to 0x79.

@Marvin in the microcode EC My CPUID 906e9 got this new feature that was no supported. From HWINFO:

MC EA: IA32_ARCH_CAPABILITIES MSR Not Present

MC EC: IA32_ARCH_CAPABILITIES MSR Present

It’s a long story but Intel has basically provided multiple ways of telling if a CPU was invulnerable to Spectre etc bugs via reporting different MSR bits (basically CPU features), they are now implementing it via IA32_ARCH_CAPABILITIES on at least in your example for the first time.

Nothing new in Windows 11 build 22468.1000.

Intel cpuA0671_plat02_ver00000050_2021-08-29_PRD_04B1CE99

cpuA0671_plat02_ver00000050_2021-08-29_PRD_04B1CE99.rar (100 KB)

Intel cpu90675_plat03_ver00000011_2021-09-06_PRD_504BBAC9
Intel cpu90672_plat03_ver00000011_2021-09-06_PRD_504BBACC

uCodes.rar (188 KB)

Nothing new in Windows 11 build 22471.1000.


hi JEN1
what CPU settings are you testing it on ??

stock OC 3.9ghz //75A / 2400mhz RAM
??

Intel cpu90675_plat03_ver00000012_2021-09-26_PRD_2585AF9E
Intel cpu90672_plat03_ver00000012_2021-09-26_PRD_2585AFA1

uCodes.rar (190 KB)

Nothing new in Windows 11 build 22478.1000.

cpu806C1_plat80_ver0000009A_2021-08-06_PRD_E62EB895

cpu806C1_plat80_ver0000009A_2021-08-06_PRD_E62EB895.rar (108 KB)

[quote="NeXt3R, post:695, topic:32301"]
Zitat von jen1 im Beitrag #687
@Marvin I noticed that Intel gave up skylake (ID 506E3) on this release, and only my CPUID 906E9 among the 806Ex and 906Ex was increased by 2KB. BTW the ID 906E9 was the only one that didn't get support for Windows 11. I think it could be a Windows 11 supportive release.

Anyhow, The new MC gave me a noticeable better results on CPU-Z benchmark in Windows 11, but there was a huge discrepancy although there was no delay testing the same MC twice.

Old MC EA:



New MC EC:

[/quote]
hi JEN1

what CPU settings are you testing it on ??

stock OC 3.9ghz //75A / 2400mhz RAM
??

what was it about

I was afraid that OC would not work on this microcode
7820hk 4.5GHZ + EC mc works fine/or better

I did not notice any problem on "gt75 7re"
thanks

Tested the new EC ucode on the latest Insider Win11 build, on my 8700K/906EA (kept at stock for reliable comparison results).
Performed a number of tests, CB15 and CB23, CPUZ, AIDA64. Repeated to increase reliability.
Results are a further decrease in performance, both single core and multi, from the previous EA ucode, which already was the latest in a line of 2-3 ucodes that were really bad.
Don’t get me wrong, you won’t feel the performance loss in stuff like games or just normal OS usage. But it’s there and it’s pretty substantial.
For example, my 8700k with pre-Spectre ucode would score like 1470 CB / R15, 1452 with B4 (the last “compromise” ucode before the nasty ones started to drop), 1405 on CA-DE, 1387 on EC. The single core dropped from 205-ish to 188 over the time. Somewhat depressing, considering there was not one attack with these vectors. Also the latest ucodes since CA or so decreased the default cache multi from 44 to 43.
I will probably eventually mod the last BIOS (EA ucode in it now) to B4 and then delete microcode.dat when offered to stay on the older ucode. I think that’s the procedure? I’m quite done testing the new ones. Seems obvious that they’re simply getting worse and worse.

EDIT:
After renaming the mcupdate thing and going back to the EA ucode, CB R15 is back to its 1405/191 “normal”, but CPUZ is showing single core scores as low as 460, so EA is malfunctioning in Insider Dev I guess, at least in some apps/benchmarks.
What a shitshow. Imagine dropping from 530 to 460 in single core performance for the same CPU over 3 years.