Finally: Overclocking EPYC Rome ES
- Thread starter ExecutableFix
- Start date
Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.
Thanks for the reply, I ended up buying the "4359124M0009" as they mentioned they are OEM taken from HP systems and not vendor locked. Will test on my H11DSI-NT. Thankfully they accept returns if it does not work.
Here's an example of an -04 (used to be considered as 'QS' and should work in retail motherboards, not sure if that's the case with Genoa and newer):
Note the CPU model #: 100-000000332-04
Here's the item: AMD EPYC 64 CORE PROCESSOR CPU 100-000000332-04 | eBay
You can just search on eBay using the CPU model and post-fix with '-04'.
Note the CPU model #: 100-000000332-04
Here's the item: AMD EPYC 64 CORE PROCESSOR CPU 100-000000332-04 | eBay
You can just search on eBay using the CPU model and post-fix with '-04'.
overlocking ?
no. 7K62 has fused max. boost multiplier = multiplier locked.
anyone enabled CBS menu on the newest v3 H11DSi bios? Also, is the PCIe Gen locked to 3 or can 4 be enabled?
Last edited:
both processors must be same base clock, core count and L3. the BIOS may refuse to work, just try.
( AMD says both processors must same OPN )
Last edited:
Testing 100-0000000114-09 ES CPU (Zen3). It's a shame there isn't a utility for Zen3 parameter adjustments. This CPU has been cut off at the knees - max boost of 3.0 (but really 2.96 or so), while I estimate max wattage to be ~200W. It runs at 2.5 base, despite the published 1.6. Right now it runs under a 7702P (passmark) at around 58-60K, and looking at the 7702P TDP of 200W, it's probably it's predecessor. I'd love to see what this could do at 3.4.
Attachments
I'm having trouble overclocking my dual-socket ZS1406E2VJUG5 setup. No matter what I try, the overclock just doesn't seem to apply. In other words, nothing happens.
I'm using the modified ZenStates version by @Zhang which should be compatible with dual-socket configurations. For the BIOS, I'm using version 2.0 from the opening post by @ExecutableFix . My motherboard is the H11DSI-NT, and I have 16 x 32GB Samsung 2400MHz memory modules installed.
I'm currently (temporarily) running Proxmox 7.1, since @Wilhelm mentioned that the overclocking tool still worked on this version but breaks on Proxmox 7.2, possibly due to kernel changes. However, even on 7.1, the overclocking attempts have no effect.
The MSR driver is present, loaded, and writable, and the smu-test-message seems to be working correctly:
Here is the output of ./zenstates.py -l:
I applied everything in a single command like so:
This outputs:
But it still doesn’t seem to do anything. Monitoring CPU clocks via /proc/cpuinfo, they remain at 1400MHz and boost to 2500MHz under load—no sign of the applied overclock.
I also tried adding initcall_blacklist=acpi_cpufreq_init to my GRUB parameters (and ran update-grub afterwards), but that didn’t make a difference either.
Am I missing something? Any tips would be greatly appreciated.
Thank you!
I'm using the modified ZenStates version by @Zhang which should be compatible with dual-socket configurations. For the BIOS, I'm using version 2.0 from the opening post by @ExecutableFix . My motherboard is the H11DSI-NT, and I have 16 x 32GB Samsung 2400MHz memory modules installed.
I'm currently (temporarily) running Proxmox 7.1, since @Wilhelm mentioned that the overclocking tool still worked on this version but breaks on Proxmox 7.2, possibly due to kernel changes. However, even on 7.1, the overclocking attempts have no effect.
The MSR driver is present, loaded, and writable, and the smu-test-message seems to be working correctly:
Code:
./zenstates.py --smu-test-message
Sending test SMU message
SMU response (socket 0): 1
SMU response (socket 1): 1
Code:
P0 - Enabled - FID = 62 - DID = E - VID = 48 - Ratio = 14.00 - vCore = 1.10000
P1 - Enabled - FID = 60 - DID = 10 - VID = 58 - Ratio = 12.00 - vCore = 1.00000
P2 - Enabled - FID = 5A - DID = 14 - VID = 68 - Ratio = 9.00 - vCore = 0.90000
P3 - Disabled
P4 - Disabled
P5 - Disabled
P6 - Disabled
P7 - Disabled
C6 State - Package - Disabled
C6 State - Core - Enabled
Code:
./zenstates.py --oc-frequency 3200 --oc-vid 50 --ppt 0 --tdc 0 --edc 38
Code:
Set OC frequency to 3200MHz
Setting VID 0x50 = 1.05 V
Set OC VID to 0x50
Set PPT to 0W
Set TDC to 0A
Set EDC to 38AI also tried adding initcall_blacklist=acpi_cpufreq_init to my GRUB parameters (and ran update-grub afterwards), but that didn’t make a difference either.
Am I missing something? Any tips would be greatly appreciated.
Thank you!
Hi all,
Following up on my overclocking issues with my dual ZS1406E2VJUG5 setup:
After further research and considering the issues some have noted with newer Proxmox versions, my current hypothesis is that the problem lies with how recent Linux kernels handle direct MSR (Model-Specific Register) calls. It seems that due to ongoing security hardening efforts, newer kernels have become more restrictive about allowing direct userspace writes to certain MSRs. Even if the /dev/cpu/*/msr interface appears loaded and writable, the kernel might be silently ignoring or blocking the specific MSR writes that ZenStates uses to apply overclocks. This would explain why the settings aren't taking effect and would also align with reports of the tool breaking between Proxmox 7.1 and 7.2, likely due to kernel updates.
Given this, I've concluded that the most promising path forward is to adapt the ZenStates script (or develop a new approach) to use SMU (System Management Unit) calls instead of direct MSR manipulation. The good news is that the smu-test-message utility appears to be functioning correctly on my system, which is an encouraging sign for this approach.
I'm now in the process of trying to convert the relevant parts of the ZenStates logic to utilize SMU calls. I understand this is a non-trivial task, as it involves a different set of commands and a deeper understanding of the SMU's interface for these EPYC CPUs.
I'll be sure to post any progress I make here. If anyone has experience or resources related to SMU interaction on EPYC platforms, any advice would be greatly appreciated!
Best regards.
Following up on my overclocking issues with my dual ZS1406E2VJUG5 setup:
After further research and considering the issues some have noted with newer Proxmox versions, my current hypothesis is that the problem lies with how recent Linux kernels handle direct MSR (Model-Specific Register) calls. It seems that due to ongoing security hardening efforts, newer kernels have become more restrictive about allowing direct userspace writes to certain MSRs. Even if the /dev/cpu/*/msr interface appears loaded and writable, the kernel might be silently ignoring or blocking the specific MSR writes that ZenStates uses to apply overclocks. This would explain why the settings aren't taking effect and would also align with reports of the tool breaking between Proxmox 7.1 and 7.2, likely due to kernel updates.
Given this, I've concluded that the most promising path forward is to adapt the ZenStates script (or develop a new approach) to use SMU (System Management Unit) calls instead of direct MSR manipulation. The good news is that the smu-test-message utility appears to be functioning correctly on my system, which is an encouraging sign for this approach.
I'm now in the process of trying to convert the relevant parts of the ZenStates logic to utilize SMU calls. I understand this is a non-trivial task, as it involves a different set of commands and a deeper understanding of the SMU's interface for these EPYC CPUs.
I'll be sure to post any progress I make here. If anyone has experience or resources related to SMU interaction on EPYC platforms, any advice would be greatly appreciated!
Best regards.
