I have pretty much learned all these through the "hard way" a.k.a failures. Not pleasant experience, but for the sake of anyone who is still interested in this processor, I hope this thread can give out some info to avoid going through the same pain I had gone through.
Some background: I was trying to use this CPU in a 4-way configuration in a Supermicro 8049U-E1CR4T, without success. Here are some highlights:
1. Xeon 8273CL has been locked down to S2S (2-way) only
This is a dead end, well, at least for me. If you are trying to put 4x 8273CL on a X11QPH+, during POST it will stuck at BIOS code A3. This is a pretty mysterious code, and there are only few Supermicro Q&A mentioned this (e.g. FAQ Entry | Online Support | Support - Super Micro Computer, Inc.) Since I am using all same model CPUs rather than like mentioned in the Q&A, I believe this code has a more general meaning like "CPU layout configuration / init" stuff. Note that if you are using an older version of BIOS like 3.0a, it will stuck without display output. Newer BIOS like 3.8a/4.0 will report code A3. My initial suspect was something wrong with UPI connectivity, and confirmed that Intel has locked these 8273CL down to "S2S" (2-way) physically from the HWiNFO64 report, even though Platinum SKUs are usually "S8S".
Anyway, this is a show stopper for me - I wanted to use them in a 4-way configuration. End result was... returning the CPUs....
2. Xeon 8273CL requires special microcode revision (0x05000014)
I spent some time to play around with 2-way, since I believe others could be interested in this more common configuration. There are already a lot of posts mentioned that, with Supermicro motherboards, only 3.0a version of BIOS can work, and any older ones (e.g. I tried 3.0c) would result in POST stuck at code 79. I can confirm this is the same for X11QPH+.
The interesting thing is, I extracted the microcode from BIOS 3.0a (e.g. cpu50657_platBF_ver05000014_2018-12-17_PRD_E1FFC46B) and injected it to BIOS 4.0, and the modded BIOS worked fine. So obviously the code 79 issue was caused by microcode rather than other content in the BIOS. I think this is relatively a good news, because even we have to use an older revision of microcode, we can use any version of BIOS we want in Suprtmicro systems, which can guarantee at least other modules, such as ME / UEFI drivers etc. are the up to date.
If you look closely at those microcode files, you can see that the microcode files would "shrink" in size from time to time, which looks like Intel was "purging" something from the microcode files. I suspect that Intel had put everything, including support of OEM CPUs, into the initial microcode, but a while later (e.g. BIOS v3.0b/c) Intel purged these OEM specific support from their newer version of microcode files. Therefore later versions of BIOS in Supermicro systems can no longer spin up those OEM CPUs properly, causing code 79 stuck.This could be applicable to other CPUs like 8272CL / 8275CL etc. as well.
This has an interesting implication: 8273CL could theoretically work with ANY motherboard! As long as the correct microcode present in the BIOS, I think 8273CL at least can work nicely with any Supermicro systems in 1-way/2-way.
UPDATE: thanks @RolloZ170 confirming that 0x0500001C can also work, see replies below.
3. 8237CL is optimized for Machine Learning and has a higher AVX2/AVX512 clock than 8276L under full load
We often compare 8273CL with 8276L as they are really close in terms of specs. I once was wondering why the lower CPU clock compare to 8276L, besides it could be from a lower bin. It turns out that under 28-core 100% workload, 8273CL has a slightly higher AVX2/AVX512 clock than 8276L, which is obviously tuned for Machine Learning workload. See attached HWiNFO64 report. (comparing to 8276L data: Xeon Platinum 8276L - Intel - WikiChip)
4. Microcode revision 0x05000014 or 0x05003003 !?!?
I noticed that in the HWiNFO64 report, the actual detected microcode was 0x5003003, which is a bit surprising, because I believe the microcode I injected into the BIOS must be 0x05000014. I guess it could be either of the following:
Some background: I was trying to use this CPU in a 4-way configuration in a Supermicro 8049U-E1CR4T, without success. Here are some highlights:
1. Xeon 8273CL has been locked down to S2S (2-way) only
This is a dead end, well, at least for me. If you are trying to put 4x 8273CL on a X11QPH+, during POST it will stuck at BIOS code A3. This is a pretty mysterious code, and there are only few Supermicro Q&A mentioned this (e.g. FAQ Entry | Online Support | Support - Super Micro Computer, Inc.) Since I am using all same model CPUs rather than like mentioned in the Q&A, I believe this code has a more general meaning like "CPU layout configuration / init" stuff. Note that if you are using an older version of BIOS like 3.0a, it will stuck without display output. Newer BIOS like 3.8a/4.0 will report code A3. My initial suspect was something wrong with UPI connectivity, and confirmed that Intel has locked these 8273CL down to "S2S" (2-way) physically from the HWiNFO64 report, even though Platinum SKUs are usually "S8S".
Anyway, this is a show stopper for me - I wanted to use them in a 4-way configuration. End result was... returning the CPUs....
2. Xeon 8273CL requires special microcode revision (0x05000014)
I spent some time to play around with 2-way, since I believe others could be interested in this more common configuration. There are already a lot of posts mentioned that, with Supermicro motherboards, only 3.0a version of BIOS can work, and any older ones (e.g. I tried 3.0c) would result in POST stuck at code 79. I can confirm this is the same for X11QPH+.
The interesting thing is, I extracted the microcode from BIOS 3.0a (e.g. cpu50657_platBF_ver05000014_2018-12-17_PRD_E1FFC46B) and injected it to BIOS 4.0, and the modded BIOS worked fine. So obviously the code 79 issue was caused by microcode rather than other content in the BIOS. I think this is relatively a good news, because even we have to use an older revision of microcode, we can use any version of BIOS we want in Suprtmicro systems, which can guarantee at least other modules, such as ME / UEFI drivers etc. are the up to date.
If you look closely at those microcode files, you can see that the microcode files would "shrink" in size from time to time, which looks like Intel was "purging" something from the microcode files. I suspect that Intel had put everything, including support of OEM CPUs, into the initial microcode, but a while later (e.g. BIOS v3.0b/c) Intel purged these OEM specific support from their newer version of microcode files. Therefore later versions of BIOS in Supermicro systems can no longer spin up those OEM CPUs properly, causing code 79 stuck.
Code:
...
cpu50657_platBF_ver05000013_2018-12-03_PRD_3B5B26B8 => 48KB
cpu50657_platBF_ver05000014_2018-12-17_PRD_E1FFC46B => 48KB
cpu50657_platBF_ver05000021_2019-02-27_PRD_14AE68DF => 46KB
...
cpu50657_platBF_ver05003006_2020-12-31_PRD_B36FD6DE => 52KB
cpu50657_platBF_ver05003101_2021-01-15_PRD_40827521 => 30KB
...
UPDATE: thanks @RolloZ170 confirming that 0x0500001C can also work, see replies below.
3. 8237CL is optimized for Machine Learning and has a higher AVX2/AVX512 clock than 8276L under full load
We often compare 8273CL with 8276L as they are really close in terms of specs. I once was wondering why the lower CPU clock compare to 8276L, besides it could be from a lower bin. It turns out that under 28-core 100% workload, 8273CL has a slightly higher AVX2/AVX512 clock than 8276L, which is obviously tuned for Machine Learning workload. See attached HWiNFO64 report. (comparing to 8276L data: Xeon Platinum 8276L - Intel - WikiChip)
4. Microcode revision 0x05000014 or 0x05003003 !?!?
I noticed that in the HWiNFO64 report, the actual detected microcode was 0x5003003, which is a bit surprising, because I believe the microcode I injected into the BIOS must be 0x05000014. I guess it could be either of the following:
- Windows has loaded 0x5003003 microcode after startup. Operating systems can dynamically load microcode after POST, so I think to fix critical issues, Windows has patched the microcode because verison 0x05000014 was simply too old. This is also a bit confusing because I can't explain why 0x5003003 would cause code 79 stuck during POST. Could it be once the 8273CL CPU has gone past POST, then newer microcode can work? If so, then this CPU can potentially be used in production, e.g. we use older microcode 0x05000014 to start the machine, then once OS is loaded, we further patch the CPU with latest microcode like 0x05003604...via tools like this => VMware CPU Microcode Update Driver
- I possibly did put 0x5003003 int he BIOS..... !? I don't know, maybe you can try that out yourself if you are feeling adventurous
Attachments
-
28.9 KB Views: 30
Last edited: