Freezes with 1950x & zenith extreme

Message boards : Number crunching : Freezes with 1950x & zenith extreme
Message board moderation

To post messages, you must log in.

1 · 2 · Next

AuthorMessage
Henri Ala-Peijari

Send message
Joined: 24 Apr 02
Posts: 22
Credit: 38,248,035
RAC: 105
Finland
Message 1976572 - Posted: 23 Jan 2019, 10:02:08 UTC

My computer freezes often when there's a seti@home weekly service outage. I presume that some component doesn't like the automatic boost (when there are just a few tasks remaining the cpu overclocks the cores with work). I've tried disabling the "core performance boost" from the bios - no effect. Any help with this? I exchanged my CPU cooler recently - so the problem is likely to be elsewhere.

My PC Specs
______________
Motherboard Asus Zenith Extreme
Processor AMD Threadripper 1950x
Memory Corsair CMD32GX4M2A2666C15 Dominator Platinum Series 32 GB (2 x 16 GB)
Video Card # 1 AMD Gigabyte Vega 64
Hard Drive # 1 Samsung 960 PRO MZ-V6P512BW Interne Solid State Drive (512GB)
Power Supply Corsair AX1200i 1200 Watt, 80 Plus Platinum Certified Digital ATX Power Supply
Case Corsair Obsidian 900D
Operating System Windows 10 64bit
CPU cooler : Corsair h150i pro (I exchanged from the enermax TR4 cooler to the corsair h150ipro when I received the adapter from Corsair)

Ps. I'm only getting about 23,000 seti@home credits per day. About 13,000 from the cpu and the rest from gpu. The psu shows power usage jumping up&down from about 350 to 600, where am I having a bottleneck in performance? Why such a low performance from the GPU?
ID: 1976572 · Report as offensive
Profile Tom M
Volunteer tester

Send message
Joined: 28 Nov 02
Posts: 5124
Credit: 276,046,078
RAC: 462
Message 1976596 - Posted: 23 Jan 2019, 14:57:22 UTC - in response to Message 1976572.  
Last modified: 23 Jan 2019, 15:07:52 UTC

My computer freezes often when there's a seti@home weekly service outage. I presume that some component doesn't like the automatic boost (when there are just a few tasks remaining the cpu overclocks the cores with work). I've tried disabling the "core performance boost" from the bios - no effect. Any help with this? I exchanged my CPU cooler recently - so the problem is likely to be elsewhere.

My PC Specs
______________
Motherboard Asus Zenith Extreme
Processor AMD Threadripper 1950x
Memory Corsair CMD32GX4M2A2666C15 Dominator Platinum Series 32 GB (2 x 16 GB)
Video Card # 1 AMD Gigabyte Vega 64
Hard Drive # 1 Samsung 960 PRO MZ-V6P512BW Interne Solid State Drive (512GB)
Power Supply Corsair AX1200i 1200 Watt, 80 Plus Platinum Certified Digital ATX Power Supply
Case Corsair Obsidian 900D
Operating System Windows 10 64bit
CPU cooler : Corsair h150i pro (I exchanged from the Enermax TR4 cooler to the corsair h150ipro when I received the adapter from Corsair)

Ps. I'm only getting about 23,000 seti@home credits per day. About 13,000 from the cpu and the rest from gpu. The psu shows power usage jumping up&down from about 350 to 600, where am I having a bottleneck in performance? Why such a low performance from the GPU?


If you are not using a command line with the Gpu I have read elsewhere that it can improve the results by 33% or so.

What kind of settings are you using in the bios? All "auto" or do you have anything set to turbo? If you see anything about "cool n' quiet" you will want to disable that.

When the system freezes, does everything stay hot or does it cool down? If everything stays hot then your keyboard/mouse/monitor are non-responsive but the system is still running.

It seems like your CPU processing may be a little slow. What GHz is your CPU baseline set at? What is it running at?

Are you running 100% cpu's are have you dropped a couple of cores offline to speed up the rest? 90% of the available cores seems to be a good starting point.

HTH,
Tom
A proud member of the OFA (Old Farts Association).
ID: 1976596 · Report as offensive
rob smith Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer moderator
Volunteer tester

Send message
Joined: 7 Mar 03
Posts: 22160
Credit: 416,307,556
RAC: 380
United Kingdom
Message 1976619 - Posted: 23 Jan 2019, 17:21:03 UTC

Adding a command line it may not fix the problem, but if you use the right values it should improve the overall performance.
What is needed are is a bit more information about what you mean by "freezes" - does the computer stop responding to anything but the power switch? Does the screen blank? or what??
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?
ID: 1976619 · Report as offensive
Henri Ala-Peijari

Send message
Joined: 24 Apr 02
Posts: 22
Credit: 38,248,035
RAC: 105
Finland
Message 1976626 - Posted: 23 Jan 2019, 18:04:59 UTC - in response to Message 1976619.  

What is needed are is a bit more information about what you mean by "freezes" - does the computer stop responding to anything but the power switch? Does the screen blank? or what??


Pretty much so (stops responding to nothing except the power button) - The screen shows whatever was on the screen when the system stopped. If I remember correctly the led's on the side of the motherboard still change color. There's definitely no activity on the hard drive light on the motherboard and pressing the caps lock won't light up the keyboards indicator led. Moving the mouse won't make the pointer move on the screen.

I'm kicking myself now for posting this and not being home for a few days afterwards.
ID: 1976626 · Report as offensive
Henri Ala-Peijari

Send message
Joined: 24 Apr 02
Posts: 22
Credit: 38,248,035
RAC: 105
Finland
Message 1976628 - Posted: 23 Jan 2019, 18:24:08 UTC - in response to Message 1976596.  
Last modified: 23 Jan 2019, 18:26:04 UTC


If you are not using a command line with the Gpu I have read elsewhere that it can improve the results by 33% or so.

I didn't realize that. Can you point to a tutorial on how to do that! I've used the standard software from seti/boinc.


What kind of settings are you using in the bios? All "auto" or do you have anything set to turbo? If you see anything about "cool n' quiet" you will want to disable that.

My settings are factory settings except the setting for "core performance boost" - which I turned off. I'm not sure if this is the setting to disable the automatic overclocking when the other cores are idle. That's what I was aiming at.


When the system freezes, does everything stay hot or does it cool down? If everything stays hot then your keyboard/mouse/monitor are non-responsive but the system is still running.

The hard drive indicator led on the motherboard won't blink. The case doesn't have a hdd led. Makes me think it's pretty much frozen. The keyboard won't light the indicator led for num lock or caps lock.


It seems like your CPU processing may be a little slow. What GHz is your CPU baseline set at? What is it running at?

Um, I'm not at home to check this but it should be stock values.


Are you running 100% cpu's are have you dropped a couple of cores offline to speed up the rest? 90% of the available cores seems to be a good starting point.

I'm running 100% CPUs. I would think that my cooling solution (corsair h150i pro) would be able to deal with 100% load.


HTH,
Tom


--Henri[/quote]
ID: 1976628 · Report as offensive
rob smith Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer moderator
Volunteer tester

Send message
Joined: 7 Mar 03
Posts: 22160
Credit: 416,307,556
RAC: 380
United Kingdom
Message 1976629 - Posted: 23 Jan 2019, 18:54:41 UTC

Don't kick yourself too hard - computers have little stress detectors built in which fire off faults when we can't do a thing about them....
You may think you are running CPU tasks only, but you have returned quite a few from your (built in?) GPU....
There is something very strange going on - for one, I would expect this sort of thing to happen when the system is under load, and so running at its hottest, but it is when it is running under light load.
A couple of things to try (in reverse order), and go back to where you were after giving each one a couple of hours to really bite
- turn off network communications, this will simulate, more or less, what happens during an outage.
- set "no New tasks until your cache has drained down a bit - keep an eye on what is happening.
- force use of the GPU OFF (this can be done in quite a few ways, probably the easiest is to right click on the BOINC task icon in the tasks bar on your computer - there are other ways to do this including from the BOINC GUI on your computer (advanced view, computing options).

Let's see what happens.
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?
ID: 1976629 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13161
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1976633 - Posted: 23 Jan 2019, 19:31:42 UTC

You can't run 100% cpu tasks and still support gpu tasks. You are running stock apps so the stderr.txt WU output does not show cpu clock frequency the WU was crunched at. Your times are lousy for both cpu and gpu tasks. Definitely overcommitted on the cpu.

Read the readme for the gpu app for suggested command line tuning or read through the dedicated forum thread. https://setiathome.berkeley.edu/forum_thread.php?id=80859

Something like -spike_fft_thresh 4096 -tune 1 64 1 4 -oclfft_tune_gr 256 -oclfft_tune_lr 16 -oclfft_tune_wg 256 -oclfft_tune_ls 512 -oclfft_tune_bn 64 -oclfft_tune_cw 64

The gpu app requires one dedicated cpu thread to support the gpu task. You will need to make a app_config.xml file to change the default cpu and gpu usage values so that each gpu taks gets a cpu thread to support it.

<app_config>
<app_version>
<app_name>setiathome_v8</app_name>
<plan_class>opencl_ati5_SoG_nocal</plan_class>
<avg_ncpus>1</avg_ncpus>
<ngpus>1</ngpus>
</app_version>
</app_config>

I would use the website account pages and the Computing Preferences menu to change the number of cpu cores used to somewhere between 60-75%
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1976633 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13161
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1976634 - Posted: 23 Jan 2019, 19:39:11 UTC

Typically on Ryzen/Threadripper a "frozen" system is indicative of not enough cpu voltage for the loading and clock frequency. The automatic overclocking mechanisms of Ryzen/TR will boost a few cores higher under lighter loading and your reported symptom suggests that a few cores are boosting higher but are not being supplied with enough voltage. Without resorting to a manual voltage in the BIOS, you could either turn off Core Performance Boost or turn off Auto for the core multiplier and lock it to a lower multiplier inline with the base clock of the 1950X which is 3.4Ghz. That will cause all cores to run at 3.4Ghz and no cores will boost whether under heavy or light load. You could also change the LLC for the CPU and SoC in the BIOS to supply a stiffer voltage regulation under changing loads.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1976634 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13161
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1976642 - Posted: 23 Jan 2019, 20:07:06 UTC - in response to Message 1976636.  

I would disable virtualization, disable SMT and use 93% of CPU freeing up 1 core for the GPU.
Your WU times are all over the place from 3 to 50 minutes for GPU wus and 1.5 to ~8 hrs for CPU wus.
A command line entry won't fix that.

Yes, this brute force suggestion would work for a newbie that doesn't want to invest the time to learn how to tune Ryzen/TR. Both Ryzen/TR CAN be made to run at boost clocks with stability. My four AMD hosts are proof of that case. My 2920X runs 16 threads supporting four gpus and is still able to keep all cores around 4.15Ghz.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1976642 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13161
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1976660 - Posted: 23 Jan 2019, 21:51:17 UTC

Went back to the OP and finally noticed he only has two sticks of memory. So already handicapped with dual channel memory configuration instead of stock four channel memory. The memory speed is not that great either and also double sided so another handicap. Lots of BIOS tuning options available for optimization.

He might want to start reading the dedicated OCN ZE thread. https://www.overclock.net/forum/11-amd-motherboards/1636566-asus-rog-zenith-extreme-x399-threadripper-overclocking-support.html

But he really is leaving performance on the table with an untuned Vega64 card running the SoG app. First things first, get the cpu core usage down, and then start tackling the gpu.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1976660 · Report as offensive
Profile RueiKe Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 14 Feb 16
Posts: 492
Credit: 378,512,430
RAC: 785
Taiwan
Message 1976881 - Posted: 25 Jan 2019, 7:17:29 UTC - in response to Message 1976572.  

My computer freezes often when there's a seti@home weekly service outage. I presume that some component doesn't like the automatic boost (when there are just a few tasks remaining the cpu overclocks the cores with work). I've tried disabling the "core performance boost" from the bios - no effect. Any help with this? I exchanged my CPU cooler recently - so the problem is likely to be elsewhere.

I have experienced this on all of my ZE builds. I found that by default, the Asus ZE uses a very aggressive LLC, resulting in Vcore being much higher at high loads, even with Freq set to a static value. I have discussed my observations in this thread.

My solution is to use the new -L1 LLC level for 1950X and -L2 level for the 2990WX. You will need to increase your Vcore setting significantly in order to get the same Vcore available at the chip. I had used a meter to verify actual Vcore in my case, but you could probably just estimate the offset from the curves I have posted.
GitHub: Ricks-Lab
Instagram: ricks_labs
ID: 1976881 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13161
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1976887 - Posted: 25 Jan 2019, 7:38:48 UTC - in response to Message 1976881.  
Last modified: 25 Jan 2019, 7:50:46 UTC

Thanks for the post Rick. I had completely forgotten that thread. I never did pursue investigating the LLC issues on the ZE as I ended up with the Asrock as you know. I still do regularly visit the OCN ZE thread but there is very little traffic there since the TR+ chips debuted. Maybe one post a week is all that forum see nowadays. And the forum content is mainly about gaming like the regular Ryzen threads.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1976887 · Report as offensive
Henri Ala-Peijari

Send message
Joined: 24 Apr 02
Posts: 22
Credit: 38,248,035
RAC: 105
Finland
Message 1977655 - Posted: 29 Jan 2019, 13:18:09 UTC - in response to Message 1976629.  

Thanks for all the replies. I just started trying out the suggestions. I lowered cpu/core usage to 75%, which made the GPU work much faster. Looking at the power supply graph (below) it would seem that there's still more punch in the system but something is keeping it from reaching full potential. My power usage has gone up from where it was with 100% CPU utilization.

I noticed that when I ran the ethereum miner, I would only get about 28MH/s. I read somewhere that this GPU should get around 40MH/s. Maybe this is a hardware issue? Or would I need watercooling on the GPU to have full crunching power? I'll try the suggested motherboard BIOS values later.


ID: 1977655 · Report as offensive
rob smith Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer moderator
Volunteer tester

Send message
Joined: 7 Mar 03
Posts: 22160
Credit: 416,307,556
RAC: 380
United Kingdom
Message 1977669 - Posted: 29 Jan 2019, 14:43:05 UTC

For now you can ignore the power drawn by your system, you have more than enough spare capacity on that 1200W PSU for your system as you are only drawing about 500W.
you need to be looking further at getting the GPU utilisation up.
The first thing to do is run the Lunatics installer which will get you a faster GPU application which you can download from http://mikesworld.eu/
Then as you have an AMD Vega GPU I would suggest you look for some advise from Mike or Zaltser, who appear to know more about tuning AMD GPUs for maximum performance.
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?
ID: 1977669 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13161
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1977764 - Posted: 30 Jan 2019, 5:06:51 UTC - in response to Message 1977669.  

Zalster wouldn't be the man. He has always been a Team Green man. Mike OTOH, is very conversant in AMD tuning being a Team Red aficionado I believe.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1977764 · Report as offensive
Profile Zalster Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 27 May 99
Posts: 5517
Credit: 528,817,460
RAC: 242
United States
Message 1977778 - Posted: 30 Jan 2019, 7:17:02 UTC - in response to Message 1977764.  

Zalster wouldn't be the man. He has always been a Team Green man. Mike OTOH, is very conversant in AMD tuning being a Team Red aficionado I believe.

+1
ID: 1977778 · Report as offensive
Profile Wiggo
Avatar

Send message
Joined: 24 Jan 00
Posts: 34744
Credit: 261,360,520
RAC: 489
Australia
Message 1977783 - Posted: 30 Jan 2019, 10:06:29 UTC

Mike is the man for all. ;-)

Cheers.
ID: 1977783 · Report as offensive
Henri Ala-Peijari

Send message
Joined: 24 Apr 02
Posts: 22
Credit: 38,248,035
RAC: 105
Finland
Message 1979083 - Posted: 7 Feb 2019, 11:57:02 UTC - in response to Message 1976881.  
Last modified: 7 Feb 2019, 12:35:24 UTC

My computer freezes often when there's a seti@home weekly service outage. I presume that some component doesn't like the automatic boost (when there are just a few tasks remaining the cpu overclocks the cores with work). I've tried disabling the "core performance boost" from the bios - no effect. Any help with this? I exchanged my CPU cooler recently - so the problem is likely to be elsewhere.

I have experienced this on all of my ZE builds. I found that by default, the Asus ZE uses a very aggressive LLC, resulting in Vcore being much higher at high loads, even with Freq set to a static value. I have discussed my observations in this thread.

My solution is to use the new -L1 LLC level for 1950X and -L2 level for the 2990WX. You will need to increase your Vcore setting significantly in order to get the same Vcore available at the chip. I had used a meter to verify actual Vcore in my case, but you could probably just estimate the offset from the curves I have posted.


I set the LLC value (the Digi+ settings) to -L1. and I set the Vcore to 1.3. CPU-Z now shows it as 1.286. Is there a drawback to using such a high Vcore setting? Or should I just set it to auto?

Disabled the core performance boost.I also did the other Digi+ settings which were on the graph on the other message thread:

VRM switch freq. to 750kHz and
Phase control to extreme and
Duty cycle to extreme

Did I understand the directions correctly? Should my computer be stable now? I guess I'll find out at the next outage or the one after that at the latest.

Before making the changes (besides the core performance boost, which was off) to my system, my computer froze again on the 5th February (2019) at 20:33. UTC. I don't know exactly what phase the outage was at that time. I also noticed that the led lights on the motherboard wouldn't change color after the freeze even though I said they would in the earlier message. Setting "no new tasks" didn't reproduce the freeze (tried just once so far).
ID: 1979083 · Report as offensive
Henri Ala-Peijari

Send message
Joined: 24 Apr 02
Posts: 22
Credit: 38,248,035
RAC: 105
Finland
Message 1981154 - Posted: 19 Feb 2019, 20:20:21 UTC - in response to Message 1979083.  

My computer didn't crash during the outage! Thanks to the support at Asus (see message below). I just wonder why it's called "Global C-states" rather than cool'n'quiet. Anyway, I hope this problem is solved.

Hello Henri,

Am not certain how int BOINC's software works, but it does sound like that your CPU goes to idle and then does not respond fast enough when GPU requests next task.
Does this issue happen if you just leave system to sit for a while without this software running on the background?

BIOS Options to mandate CPU idle state are "AMD Cool'n'Quiet" / Global C-states. (AMD equivalent to Intel SpeedStep). Note that disabling these will leave your CPU to default clock rates and may increase electricity consumption due to CPU not entering idle mode / underclockign itself.



Best Regards,
Marko H
ASUS Nordic
Technical Support Department
ID: 1981154 · Report as offensive
Profile bloodrain
Volunteer tester
Avatar

Send message
Joined: 8 Dec 08
Posts: 231
Credit: 28,112,547
RAC: 1
Antarctica
Message 1981623 - Posted: 22 Feb 2019, 4:23:12 UTC

its due to some boost setting on asus mobo. no is on a asrock set up of mine
ID: 1981623 · Report as offensive
1 · 2 · Next

Message boards : Number crunching : Freezes with 1950x & zenith extreme


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.