Poor CPU performance after upgrade to 6.6.1 with Threadripper

[Symon]

Hi everyone

 

I've upgraded UnRaid from 6.5.3 to 6.6.1 yesterday and was kind of shocked when I did a CPU bench with CPU-Z. The single core bench result went from 400 down to 200 (same with multicore). I don't know if this is a propblem with CPU-Z or if something changed within the update which has such a huge impact on the CPU performance. Unfortunately, I have no other results to compare than CPU-Z. All my VMs work but it takes longer to start them up and games seem to be slower // have a stutter.

 

Overall, the update whent smooth and I can't findy any errors in the logfile which woud indicate that something is wrong with the system. I shut down all dockers and VMs and the performance still stays way below what it should be according to the bench results. Can somebody with a Threadripper check what their single core bench looks like with 6.6.1?

 

My CPU is Overclocked to 3.8 GHz and the CPU Scaling Governor is set to "On demand" (I also tried "Performance"). The CPU-Cores of the VM are Isolated and I tried it with and without a dedicated core for emulation. 

 

watch grep \"cpu MHz\" /proc/cpuinfo  > shows the expected frequency on the cores of the VM.

 

I even assigned different cores to find out if the problem is related to the different dies / connections. The BIOS is up to date and I tested on a Windows 10 VM.

 

In the update notes it was mentioned that CONFIG_CRYPTO_DEV_SP_PSP: has been turned on again for Threadripper but i don't know if this could have such a big impact on the CPU performance? Is there a way to manually disable this to test it?

 

My hardware:

MB: ASUS ROG Zenith Extreme
CPU: AMD Threadripper 1950X
RAM: 64 GB HyperX Predator

 

Grateful for any help 🙂

Symon

Edited October 4, 2018 by Symon

[bastl]

@Symon I didn't noticed that big of a performance impact. CPU-Z bench gives me between 350 and 390 for singlecore tests. I'am currently running on a stable 3.9GHz without any issues in games or any instabilities or slowdowns on 6.6.1. BUT i had nearly the same issues like you a couple weeks ago. Back than i used the Enermax 360 AIO as my cooling solution and what i didn't know was, that the first gen of that rad had some manufacturing issues where corrosion happened and gunked up the rad after a couple months. At some point the CPU started trottling down because it reached the 68°C temp limit. I first noticed it in some games starting to stutter and having some hickups. I'am now back to an Noctua NH-U14S and under heavy load my temps maxing out around 63°C Tdie without throttling the clock speeds. You might have a look into your cooling and check your temps under load. With the bad radiator my idle temps where sitting around 40-50°C and are down now to 30-35. Under load they instantly spiked up to 68°+ and the CPU started to throttle down to 2GHz and sometimes down even further. Do some stresstests and check your temps with.

 

watch sensors

Edit:

Also worth to be mentioned, one of the last Windows updates reseted my energy profil back to balanced from performance mode and also the MSI-fix where reverted back so i got random sound glitches and game stutters. Might be worth a look if something changed their for you.

Edited October 4, 2018 by bastl

[Symon]

3 hours ago, bastl said:

@Symon I didn't noticed that big of a performance impact. CPU-Z bench gives me between 350 and 390 for singlecore tests. I'am currently running on a stable 3.9GHz without any issues in games or any instabilities or slowdowns on 6.6.1. BUT i had nearly the same issues like you a couple weeks ago. Back than i used the Enermax 360 AIO as my cooling solution and what i didn't know was, that the first gen of that rad had some manufacturing issues where corrosion happened and gunked up the rad after a couple months. At some point the CPU started trottling down because it reached the 68°C temp limit. I first noticed it in some games starting to stutter and having some hickups. I'am now back to an Noctua NH-U14S and under heavy load my temps maxing out around 63°C Tdie without throttling the clock speeds. You might have a look into your cooling and check your temps under load. With the bad radiator my idle temps where sitting around 40-50°C and are down now to 30-35. Under load they instantly spiked up to 68°+ and the CPU started to throttle down to 2GHz and sometimes down even further. Do some stresstests and check your temps with.

 

watch sensors

Edit:

Also worth to be mentioned, one of the last Windows updates reseted my energy profil back to balanced from performance mode and also the MSI-fix where reverted back so i got random sound glitches and game stutters. Might be worth a look if something changed their for you.

Thanks for the hint, I looked into the temperatures and they really were too high. I now reapplied thermal paste and checked that everything was set up correctly. I even reset the Bios settings and am now running on stock with 3.4 gHz. With this I'm sitting now at 39 ° idle and 58 ° stress. I might also need to replace my Kraken X62 with a better cooler.

 

However, the results from CPU-Z are now down to 167 which is kind of crazy when I was close to 400 before 😕 but maybe it's just not showing the correct value.. Do you have any other benchmark results from you VM that I can test on my system?

 

I checked the Windows settings and had to reset the MSI batch after the windows update as well.

 

I'm impressed that you got your CPU running at such a low voltage, mine is not even booting @ 1.2V....

[TType85]

I am running at 4.1 Ghz on a Noctua cooler on my 1950x and am seeing lower CPU-z scores too.  I was close to 400, now 340-350. My temps are good, 32-33c idle and I haven't seen load temps over 63c with all cores cranking away.  I have had a few stutters, mainly when I have youtube open while playing World of Warcraft but they last a few seconds and are gone. 

 

Check your setup with Cinebench R15.  My daily driver is 8c/16t (half the cpu, cinebench doesnt read it right). Single core @4.1ghz under KVM is slightly faster than bare metal at 3.4 boosting to 4.1.  The 1661 multi core falls in line with the 3300-3400 i've seen for bare metal.

 

Edit:

Just set up a windows 8.1 VM, using the same cores as the windows 10 (w10 machine off) and I get a slightly better cinebench score. I am going to pass the gfx card through and test WoW tonight.

 

 

Edited October 4, 2018 by TType85

[bastl]

2 hours ago, Symon said:

I'm impressed that you got your CPU running at such a low voltage, mine is not even booting @ 1.2V....

I guess i won the chip lottery, the first time ever. I get my system stable on 4GHz @1.25V only 😂 If you check a couple youtube guides how to OC that chip most of them pushing 1.35-1.4V into it to get it stable.

 

Your temperatures aren't looking that bad. The 1950x starts throttling at 68°C TDIE. You definitly have a bit headroom to play around with. I'am using the same 8c/16t as my daily driver like @TType85 and seeing like the same results as him considering my lower clocks. I've isolated the cores 8-15 and 24-31 from unraid and only using them for this VM. Nothing else has access to them. 

 

1 hour ago, TType85 said:

I have had a few stutters, mainly when I have youtube open while playing World of Warcraft but they last a few seconds and are gone. 

Thats the exact behaviour i got without the MSI fix. Checkout the Unraid wiki, their might be a bit to tweak for you.

https://wiki.unraid.net/index.php/UnRAID_6/VM_Guest_Support#Enable_MSI_for_Interrupts_to_Fix_HDMI_Audio_Support

 

The MSI Util v2 for Windows you can find here:

https://forums.guru3d.com/threads/windows-line-based-vs-message-signaled-based-interrupts.378044/

[TType85]

I'll take a look at the MSI fix again, I am running sound on a usb adapter so I never get oddities there. 

 

I just did a CPUz bench on the w8.1 that is a full 75pts higher than the win 10 machine.  I re-ran Cinebench and got 1707/164 in back to back runs.

[bastl]

You said you quickly set up a fresh 8.1 VM, right? Thats kinda unfair against a 10 VM that might have a lot more software installed and runs a bunch of crap in the backround 🙄And as we all should already noticed, Win10 tends to reset configurations with almost every update. You can disable stuff to tweak the performance and a day later it's set back to defaults in the backround without noticing you. I can't compare my CPU-Z scores to the older Unraid or Windows versions, cause i never tested it before, but as i said, performance mode where changed back to balanced at some point and the MSI fix was gone for me. If i compare my todays CB scores against some tests i did end of august on 6.5.3 the score went down a couple points. Could be a Windows issue, could be to some changes in UnRaid 6.6.1 also be possible that the BIOS update i did a couple weeks ago is the reason. Who knows.

 

Btw. did you guys noticed with the CB single core bench and opened taskmanager, the CPU load jumps across the cores and not stay at 1 core and max this one core out as it should do?

[TType85]

5 minutes ago, bastl said:

You said you quickly set up a fresh 8.1 VM, right? Thats kinda unfair against a 10 VM that might have a lot more software installed and runs a bunch of crap in the backround 🙄And as we all should already noticed, Win10 tends to reset configurations with almost every update. You can disable stuff to tweak the performance and a day later it's set back to defaults in the backround without noticing you. I can't compare my CPU-Z scores to the older Unraid or Windows versions, cause i never tested it before, but as i said, performance mode where changed back to balanced at some point and the MSI fix was gone for me. If i compare my todays CB scores against some tests i did end of august on 6.5.3 the score went down a couple points. Could be a Windows issue, could be to some changes in UnRaid 6.6.1 also be possible that the BIOS update i did a couple weeks ago is the reason. Who knows.

 

Btw. did you guys noticed with the CB single core bench and opened taskmanager, the CPU load jumps across the cores and not stay at 1 core and max this one core out as it should do?

The windows 10 image I used to test is only a few weeks old, fully updated. I actually have very little installed. The scores from when it was a fully fresh install are around the same as they are now. CPUz is the odd one.  I was able to get to 400 single core before but mostly it has been between 350 and 375 with no consistency.  There were times it dipped as low as 320 but I attributed that to background processes.

 

I did notice that about CB.  CPUz stays on one core.  We will see how WoW is, It will put 1 core at 95-99% all the time..

[bastl]

@Symon Which cores did you assigned for the VM? Mixing up cores for a VM from both dies can have a performance impact more or less depends of the application. Can you post a screenshot of the VM settings page?

[Symon]

I only assigned them from die 1 as shown in the picture  for Die 1 as my Graphic card should be connected to that one. When I run Cinebench the temperature of that die goes up to 72 degrees while the other one stays at 43 ° (according to output of "watch sensors"). It may be possible that better results from cores on both dies are because the temperature will spread better over the cpu and it wont start throttling as fast rather than the connection now ..

After some testing with the core assignment if found out that my die 2 only gets up to 58° with 10 cores assigned whild die 1 goes up to 68 ° 🤯

 

Im running the cpu currently on 3.875 GHz @  1.25V  > I might even be able to push it a bit higher but I guess a better cooler (altought the Kraken X62 I'm currently using should be good 😕) would help to get better results .. 

 

I tested the results on my Windows 10 VM with several CPU Pinning and below are the results. Overalll, the results all seem to be to low compared to the ones you guys get but that's maybe also dependent on the VM that the test is running on .. 

Also, a dedicated emulater pin seemed to have a negative impact on the single core performance

 

Die 1 (isolated):

 

Die 1 / 2

 

 

 

Die 2:

[bastl]

Wait, you see temperatures for each die separated? Their should be only one output. Tdie is the actuall temp where Tctl is the same temperature with an offset of +27°C

[Symon]

8 hours ago, bastl said:

Wait, you see temperatures for each die separated? Their should be only one output. Tdie is the actuall temp where Tctl is the same temperature with an offset of +27°C

At least that is what i think it is  

The results make sense when I run tests on VMs that are on a specific die, the temperature on that die goes up..

Maybe MB specific?