Xeon E3-1285 v6 ignoring MSR_UNCORE_RATIO_LIMIT

[MartinMoe]

Hello,

my environment is a Xeon E3-1285 v6 CPU under a non-mainstream OS. HWP is disabled, Core P-State Targets are set to 44, Uncore Ratio Limits are set min 42 and max 42.

I observe a performance reduction (10%) after several minutes of runtime in parts of my application that communicate with other threads on other cores via memory (i.e. via L3 Cache/LLC-Ring).

None of the PERF_LIMIT_REASONS MSRs gives a hint, no thermal or power limits are reached.

I discovered that I can revert this state by writing to MSR_UNCORE_RATIO_LIMIT (0x620) and first set min to 41 and to 42 again (two writemsr instructions). Then, after some minutes, performance drops again.

I tried to measure the uncore frequency with MSRs 0x394/0x395 but hat no success. I got something counting at 10 to 90 MHz but that does not seem to be the UCLK.

Question: How can I prevent the CPU from throttling the Uncore?

(Writing to MSR_UNCORE_RATIO_LIMIT every now and then is not applicable for the application because of the introduced Jitter)

Further information:
CPU DisplayFamily_DisplayModel: 06_9E
CPU Brand String: Intel(R) Xeon(R) CPU E3-1285 v6 @ 4.10GHz
BIOS correctly sets up Power Limit 1 to be 79 Watts
Consumed power never reaches above 30 Watts
Package and core temperatures never above 60°C
Hyperthreading is disabled (only using the 4 native cores)
Core P-State Targets: 44
MSR_UNCORE_RATIO_LIMIT: min 42, max 42 (behavior is the same with 41 and 41)

Kind Regards,

Martin

[IntelSupport]

Hello MartinMoe,

Thank you for posting your question on this Intel® Community.

We understand that your system is experiencing a drop in performance.

To better assist you, please provide us with the following information about your environment:

• Server/system model.
• What is the non-mainstream OS the system is running?
• Could you please provide us additional details about the application you are running?
• Have you optimized the cooling system of the system?

Wanner G.

Intel Customer Support Technician

[MartinMoe]

Hello,

My company's guidelines do not allow me to disclose too much detail about hardware and software. As the OS has no power management of any kind it cannot be the cause.

To clarify things I provide the following chart:

Obviously something inside the CPU package (I think it is UCLK) slows down, because it can be "reset" by "refreshing" the MSR_UNCORE_RATIO_LIMIT (0x620).

If you can tell me which MSRs or fields in the MCHBAR memory area I could monitor, I will rerun and report results.

Martin M.

[IntelSupport]

Hello MartinMoe,

Thank you for providing additional details about your environment.

Please allow me to review the information you have provided to better assist you. I will update this thread soon.

Wanner G.

Intel Customer Support Technician

[IntelSupport]

Hello MartinMoe,

I would like to update this thread, and share the following link that contains additional details about Throttling.

https://software.intel.com/content/www/us/en/develop/documentation/vtune-help/top/analyze-performance/platform-analysis-group/throttling-analysis.html

If you are developers, and have additional questions, our recommendation is that you submit them to the following Community:

Analyzers

https://community.intel.com/t5/Analyzers/bd-p/analyzers

Wanner G.

Intel Customer Support Technician

[IntelSupport]

Hello MartinMoe,

I hope you found the information provided helpful.

If you need any further assistance, do not hesitate to update this thread.

Wanner G.

Intel Customer Support Technician

[MartinMoe]

Hello,

The Link provided did not help me. I am not using an Analyzer Tool. I have to write it (integrate it into the application). I am software developer and have access to MSRs of the given CPU. As said in the opening post, MSR_RING_PERF_LIMIT_REASONS does not indicate any reason, nor does MSR_CORE_PERF_LIMIT_REASONS (as the cores are not the thing clocking down that's OK). Analyzer Tools like Intel XTU and Intel VTune read those registers like I do.

I re-posted the topic there:

https://community.intel.com/t5/Software-Tuning-Performance/Xeon-E3-1285-v6-CPU-throttles-uncore-for-no-reason/m-p/1262306/highlight/true#M7837

It would be nice to have one of the CPU developers could shed some light on the matter. The problem is specific to a Xeon E3-1285 v6 CPU. A Xeon E3-1275 v5 does not show the behavior.

[IntelSupport]

Hello MartinMoe,

Thank you for your response.

Based on the description of the issue, I would also recommend that you submit your questions to the Intel® Developer Zone.

Intel® Developer Zone - Contact Page

https://software.intel.com/en-us/support

You can submit your ticket through the Online Service Center at https://www.intel.com/supporttickets for support.

See How to Create a Support Request at the Online Service Center.

https://software.intel.com/en-us/articles/how-to-create-a-support-request-at-online-service-center

Regards,

Wanner G.

Intel Customer Support Technician