Xeon E5-2680 v3 : CPU Mark only 0.32 of average reported results for the model :(

[InBalance]

Hi, I’m a licensed PT8 user, and would like help please with troubleshooting. After I initially built a custom system, I used a competing benchmark software, which rated my system in the 99th percentile of all reported results. So that was great news. But in retrospect, I wish I had known about your product at the time!

A few weeks later, that same competing product ranked my system in the 37th percentile. I realized only then that this other benchmarking software wouldn’t be helpful in troubleshooting. So I uninstalled it, and after further research decided on PT8.

I’m obviously missing a true baseline PT8 rating for this system (i.e. immediately after physical build and Windows installation). However, the other software and PT8 give similar actual-vs-baseline-expected results. My PT8 CPU Mark is (5809) vs reported PT8 Average (18437) = 0.32

1. http://www.cpubenchmark.net/cpu.php?....50GHz&id=2390

I have read through the following posts & resources on your site:

• http://www.passmark.com/forum/showth...a-I7-2600K-CPU
• http://www.passmark.com/support/perf...erformance.htm
• http://www.passmark.com/forum/showthread.php?1057-Causes-and-Solutions-for-a-slow-PC

I have checked the following settings:

• Intel Virtualization and Hyper-threading are both OFF in the BIOS.
• In Control Panel / Power Options / Advanced / Processor, settings are 100% with active cooling
• In msconfig / Boot, number of processors is unchecked.
• >check the number of processes in the Edit / Preferences window matches the number of CPU's and cores that you have in the machine
  • I did this, and verified 24 logical processors.

I have made a few hardware changes since the initial build and Windows install:

• Intel TPM module added for secure boot
• 2 additional DIMMs added (all single-rank slots now populated : 4 per CPU)
• SATA BD-ROM drive added

Any advice or suggestions appreciated,

Thanks,
Brad

[Richard (PassMark)]

Do you have a baseline number we can take a look at?

[David (PassMark)]

Why do you have Virtualization & Hyperthreading off?
I assume you are benchmarking the physical machine and not a VM?

[InBalance]

Hi Richard and David,

Thank you both for your replies.

>Do you have a baseline number we can take a look at?

I saved my test as a baseline, and uploaded it to the site.

• I don't see how to access that page from my profile, nor from the PT8 application?

I see the results under My Baselines in the application. I have selected my baseline, along with one of the reference choices. I don't see how to bring those results up in another window.

• How can I copy the results from My Baselines for pasting here.
• Is the only option to manually copy results out by hand from the My Baseline tabs? The only options at the bottom of the tab are Help and Close.

>Why do you have Virtualization & Hyperthreading off?
>I assume you are benchmarking the physical machine and not a VM?

The research I've done seems to show that Hyperthreading is typically a net negative on performance. I have read that Virtualization is also typically a net negative. I will re-enable (in order to use TPM) once I get this issue resolved.

At this point, I'm trying to simplify as much as possible to determine what could be causing the dramatically poor CPU results?

[Richard (PassMark)]

I don't see how to access that page from my profile, nor from the PT8 application?

When you upload the baseline from within the software, it will give you an URL to your baseline. If you did not note down the number, then it can be hard to track down. You may be able to search for baselines (from Manage Baselines) and locate your baseline which then you can look at the ID.

If you saved the baseline (.ptx) you can send it to us and we can take a look as well.

How can I copy the results from My Baselines for pasting here.

Under the File Menu, there are several options to Export baselines (text, image, etc).

The research I've done seems to show that Hyperthreading is typically a net negative on performance. I have read that Virtualization is also typically a net negative. I will re-enable (in order to use TPM) once I get this issue resolved.

Most the baselines submitted are with hyperthreading on. At the very least you should enable Hyperthreading and run the test again to eliminate that as a factor on performance.

[InBalance]

Hi Richard,


I was able to resolve my performance issue, completely by chance! :\


As I was bench-testing my system, I went back-and-forth between having multiple monitors plugged into my Quadro video card, and having only one monitor plugged in. During these changes, I noticed that Windows 10 would “update” the video driver and then require a reboot.


Over several update-and-reboot cycles, Task manager showed the CPU speed change from ~3 GHz to 1.X GHz and back again. The Performance Test ratings followed suit with TaskMan CPU speeds. In all cases, TaskMan shows the same 2.50 GHz nominal CPU speed above the graph in Performance / CPU, while the effective CPU speed is shown below the graph.


The only changing variable in this experience appeared to be Windows’ video driver updating. I don’t understand this at all, but it seems to be the only explanation I can find.


Thankfully, my system is consistently running now at high CPU speed per TaskMan, with 3 monitors connected. I am continuing to benchmark with HyperThreading off (but thanks for the suggestion). Current CPU Mark is 24955.

It's scary to think that such a subtle and non-obvious Windows configuration, for the most part out of the user's control, could have such dramatic performance implications! What are your thoughts on this video-driver factor being the crucial variable?


Thanks,
Brad

[David (PassMark)]

Adding or removing monitors should not result in a video card device driver update. So that is not normal.

It is also not normal that it effects the CPUs clock speed.
Maybe there is some other subtle factor, like the CPU getting too hot and temperature throttling.

[InBalance]

Hi David,

Thanks for your comments and suggestions.

I happened to be using my laptop last week (Intel i7-3630QM @ 2.40 GHz), and noticed in Task Manager that the actual speed is 1.15 GHz, just as I was experiencing on my desktop machine mentioned above.

This seems like a unlikely coincidence. I Googled the phrase "task manager cpu speed 1.15" and found quite a few postings that describe the same situation. I think it's likely that, as you said, there's some kind of subtle common factor here.

In my case, I don't feel it's probably temperature throttling. I have seen the same 1.15 GHz CPU speed on both machines, at 70F ambient room temperature, just after startup.

I have a couple of MSDN premium support instances via my subscription. Do you feel it would be worth using one of them on this issue?

Thanks,
Brad

[David (PassMark)]

It is normal that the CPU throttles itself down when under low load. All new CPUs do this to save power and reduce cooling requirements. When when you run the CPU test in PerformanceTest, you should see it lift back up to the nominal level for the CPU that you have however.

If you can't get it to move from 1.15GHz, even when under high sustained CPU load, there is something wrong (like thermal throttling).

[InBalance]

If you can't get it to move from 1.15GHz, even when under high sustained CPU load, there is something wrong (like thermal throttling).

I followed up on this with Microsoft Technical Support. The level 1 rep checked my system remotely and did extensive research. He also checked with his technical lead.

The two of them found numerous similar reported cases, where Windows users reported the same 1.18 GHz CPU speed, well below rated speed. They also found that disabling processor virtualization in the BIOS was able to resolve the issue.

We found via Task Manager that Hyper-V was disabled on my machine, but virtualization enabled. The rep requested that I disable. After doing so, on next system boot, my CPU speed increased to above rated speed (without overclocking). Problem solved!

As we closed the case, I confirmed with the tech agent and his manager, that Microsoft currently has no public documents explaining that virtualization is a potential cause of abnormally slow CPU speed. I requested that the documentation team consider publishing a public article reflecting the tech support findings. The manager agreed. He told me there is enough evidence in the support database to support submitting this to a research team for corroboration and article publication.

Hopefully we will see something published in the near future, and others can benefit from this knowledge.

[David (PassMark)]

Thanks for the follow up. We have added the issue to the list we maintain of things that slow PCs down.