If you are using an Intel 13th Generation (Raptor Lake) CPU and you have not disabled either the E-Cores or P-Cores, you may run into a PSOD when powering on a VM with the following exception:
GP Exception 13 in world
Since ESXi does not support Intel's hybrid consumer CPU architecture, Option 1 below is the current recommended approach that provides the best customer experience as ESXi will see one consistent set of cores, ideally the Performance Cores (P-Cores), as they include Hyper-threading (HT). If you can not or prefer not to disable the E-Cores or P-Cores, then Option 2 can be used with two additional workarounds.
Option 1 (Recommended) -
- Disable Efficiency Cores (E-Cores) within system BIOS
Option 2 -
- Add ESXi kernel boot option to disable the CPU uniformity check, please see this video HERE for the detailed instructions
- To workaround the PSOD mentioned in this blog post, you will need add the additional ESXi kernel setting after installing ESXi which ignores the MSR Faults and prevents the PSOD from happening by running the following:
esxcli system settings kernel set -s ignoreMsrFaults -v TRUE
Hi William, do you have insights if the ESXi scheduler would eventually support it in the future? I've noticed the latest 4th Generation Intel Xeon Scalable Processors have something called High Priority Cores and Low Priority Cores. It's not the same as little-big concept, but still there's some difference in the Cores and it might be nice to have the option eg on the VM config (prefer which type of core).
No, there are no plans. The upcoming Xeon will support either all P-Cores or E-Cores with same socket, so no big.LITTLE which has different CPU features which makes them non-uniform such as HT as example. For now, outside of Arm which uses big.LITTLE, this type of architecture doesn’t seem like it’ll make it into Datacenter which would influence the type of CPU we’d support since these hybrid CPUs are found only in consumer systems like Intel NUC which are not officially supported
Sad to hear as newer Intel consumers CPUs are becoming less suitable for home labs. And from the AMD world there are lot of relatively cheap NUC-sized Ryzen based systems, but their downside is unsupported NIC.
Hi William, seems (I cannot remember exactly if it was GP Exception 13) the mentioned PSOD happens on 12th gen CPUs as well with ESXi8.0u1(was OK on ESXi8.0b, maybe 8.0u1 adds this new checking?) during VM startup (with CPU uniformity check disabled)
I have a system with i7-12650H, hits into this problem, the option 2 workaround resolved the issue.
Hi William, it seems that when adding a host to a cluster, the clusteragent generates the GP#E13 PSOD. I will have a closer look later, but for now it makes my 3 node cluster a two node team :).
Niels
Hi William, is there a permanent solution to option 2? When I restart esxi and turn on a vm it shows the PSOD. I would then need to turn off esxi and turn it back on, press shift O and type ‘esxcli system settings kernel set -s ignoreMsrFaults -v TRUE‘ which fixes the problem temporarily. But this goes away again after a restart
Not sure I follow ... using ESXCLI does permanently set the value ... unless your system somehow isn't persisting the setting. Are you booting off of USB or HD/SSD? Alternatively, if you're using USB, you can manually add ignoreMsrFaults=True to the kernel boot line, which accomplishes the same behavior but again, using ESXCLI command will persist the setting, which is the whole point
Hi William, thanks for responding. Just troubleshooted further into this and it looks like the PSOD only happens when I start vcenter on the host. All other vms work as normal. The ESXCLI command above does solve the problem but if I was to reboot the host the settings would not persist. I am booting off SSD.
And just to confirm if I am doing this correctly, you are supposed to input the esxcli command (esxcli system settings kernel set -s ignoreMsrFaults -v TRUE) when the host first starts by pressing Shift + O or do you need to input this into some file?
Thanks for all you do!