Processors
Intel® Processors, Tools, and Utilities
14404 Discussions

Intel Performance Maximizer comments

Dobratz__Glenn
1,235 Views

I just used the IPM to overclock my 10980XE processor.  The software worked as advertised, but the overall process and result was very unsatisfying.  On the off chance that Intel is listening:

Testing issues:

1. Is it really necessary to spend 20+ seconds on every test iteration revalidating the tests?

    One would think that once is enough.

2. When performing the AVX2 and AVX3 tests, why go all the way back to where the non-AVX tests started minus the assumed default AVX Offsets.  Wouldn't it be sufficient/better (certainly faster) to start where the non-AVX tests ended (instead of where they began), minus the assumed default AVX Offsets.  (If the first test frequency fails, then just keep moving backwards.)  It would seem that this approach would let you find the needed AVX offset in 2-3 iterations instead of the 12-15 iterations that it takes now.

Results issues:

1.  On my processor the UEFI software announced that the best all-cores frequency was 4.7 GHz.  However, the Windows component announced the best all-cores frequency was 4.6 GHz (this is what is actually being used by the IPM drivers).  This should be investigated.  Especially since the 4.6 value results in CPU temperatures 10C below the thermal limits.

2.  Only the best all-cores frequency is reported publicly, however, many other values are changed (VCore, AVX Offsets, etc.).  I would like to see the Windows component provide a printable log that shows every value that was changed (so I can have them set by the BIOS).

3.  After booting back to Windows, the TjMax value had been changed from the Intel-delivered 86C to 110C.  Why?  Is 110C really safe?

4.  I use HWInfo64 to monitor the system during use.  Before using IPM, HWInfo64 reported the CPU Package Power properly.  After IPM, this value is always reported as 1W.

5.  The AVX Offset test results were low.  The tests resulted in AVX-2/512 Offsets of 5/8.  Subsequent manual testing by me, showed that offsets of 2/5 were sufficient.

 

Please consider addressing some of these issues.  Thanks.

Labels (1)
0 Kudos
8 Replies
David_G_Intel
Moderator
1,217 Views

Hello @Dobratz__Glenn


Thank you for posting on the Intel️® communities. In order to investigate this, please share the Intel® System Support Utility (Intel® SSU) results with us:


Regards, 

David G 

Intel Customer Support Technician 


0 Kudos
Dobratz__Glenn
1,211 Views

Here is the requested scan.

 

0 Kudos
David_G_Intel
Moderator
1,204 Views

Thanks for the update, let me investigate this and I will post the updates on the thread.


Regards, 

David G 

Intel Customer Support Technician 


0 Kudos
David_G_Intel
Moderator
1,178 Views

After further investigation, here is more information that will help with this request:


Testing issues:

  1. This varies from silicon to silicon.
  2. The different iterations of the tests are programmed to guarantee the most stable results from the overclocking, while it may not seem efficient the goal is to obtain best performance paired with the highest stability thus the duration of the overall process should not be a concern.


Results issues:

  1. The behavior can vary from silicon to silicon, on the i9-10980XE we guarantee 4.6Ghz and anything above that is silicon lottery, The UEFI component might have its own overclocking set that can push to 4.7Ghz.
  2. Noted, this request will be pushed to the IPM dev team for consideration.
  3. No, it is not safe and may in fact reduce the products useful life, but the silicon was determined to withstand it during the overclocking tests, exactly why overclocking voids the product's warranty.
  4. We can't validate results from 3rd party tools and 1W is a misreading.
  5. This can vary from silicon to silicon, ideally use another benchmarking tool to validate before vs. after results.


Finally, for points 4. and 5. we recommend using the Intel® Extreme Tuning Utility (Intel® XTU)

Let us know if you have any questions.


Regards, 

David G 

Intel Customer Support Technician 


0 Kudos
Dobratz__Glenn
1,170 Views

Testing issues:

  1. "This varies from silicon to silicon."
    But why waste this time on ANY silicon?
  2. The different iterations of the tests are programmed to guarantee the most stable results from the overclocking, while it may not seem efficient the goal is to obtain best performance paired with the highest stability thus the duration of the overall process should not be a concern.
    I understand that the duration of the testing does not effect the final result, but if there is a way to produce the same results while wasting less of the user's time then you should do it.

Results issues:

  1. The behavior can vary from silicon to silicon, on the i9-10980XE we guarantee 4.6Ghz and anything above that is silicon lottery, The UEFI component might have its own overclocking set that can push to 4.7Ghz.
    You seemed to have missed the entire point here.  There is an inconsistent reporting of results between the UEFI portion of the product and the Windows portion.  For the same test run, they report different results.  Inconsistent reporting reduces the user's confidence in the product.  What is the real number?  Should I be using 4.6 GHz or 4.7 GHz?
  2. Noted, this request will be pushed to the IPM dev team for consideration.
    Thanks.
  3. No, it is not safe and may in fact reduce the products useful life, but the silicon was determined to withstand it during the overclocking tests, exactly why overclocking voids the product's warranty.
    Then this raises the question: Why is this product doing something that is not safe?  This seems to be something that the development team needs to investigate/address.  The point here is that this unsafe change was made and remains in effect even if the user subsequently decides to not use the overclocking settings.
  4. We can't validate results from 3rd party tools and 1W is a misreading.
    Again, the point seems to have been missed.  Why does running this produce cause third-party products to begin misbehaving?
  5. This can vary from silicon to silicon, ideally use another benchmarking tool to validate before vs. after results.
    I obtained these results using XTU.  The system has been running under load for 10 days now and the lower AVX offsets are working fine.  Of course the results will vary from silicon to silicon, but isn't it the whole point of the produce to find the best sppeds / voltages / offsets for the silicon that it is running on?  It did not do this for this silicon.
0 Kudos
David_G_Intel
Moderator
1,116 Views

To answer your questions:

1. "But why waste this time on ANY silicon?"

- This is necessary for security. 


2. "I understand that the duration of the testing does not effect the final result, but if there is a way to produce the same results while wasting less of the user's time then you should do it."

- Current test times and loops were optimized for system stability. We offer the Intel Speed Optimizer tool if you need something fast but non-optimized.


3. " There is an inconsistent reporting of results between the UEFI portion of the product and the Windows portion. For the same test run, they report different results. Inconsistent reporting reduces the user's confidence in the product. What is the real number? Should I be using 4.6 GHz or 4.7 GHz?"

- Intel PM provides more robust overclock so this is expected and done to provide the customer with more secure OC.


4. "Why is this product doing something that is not safe? This seems to be something that the development team needs to investigate/address. The point here is that this unsafe change was made and remains in effect even if the user subsequently decides to not use the overclocking settings."

- We don’t alter temperature, we leave the temperature settings alone for the system.


5. "Why does running this produce cause third-party products to begin misbehaving?"

- We cannot control how 3rd party software reads and reports processor information from our programmed registers.


6. "The system has been running under load for 10 days now and the lower AVX offsets are working fine. Of course the results will vary from silicon to silicon, but isn't it the whole point of the produce to find the best sppeds / voltages / offsets for the silicon that it is running on? It did not do this for this silicon."

- Intel PM optimizes for stability over time which may result in overclocks that are lower than what can be achieved with manual tuning


Regards, 

David G 

Intel Customer Support Technician 


0 Kudos
David_G_Intel
Moderator
1,084 Views

Were you able to check the previous post?  

Let us know if you still need assistance.  

  

Best regards,  

David G.  

Intel Customer Support Technician  


0 Kudos
David_G_Intel
Moderator
1,058 Views

We have not heard back from you, so we will close this thread. If you need any additional information, please submit a new question as this thread will no longer be monitored. 


Best regards,  

David G. 

Intel Customer Support Technician  


0 Kudos
Reply