log in

Advanced search

Message boards : Number crunching : computation error RYZEN

Author Message
bud20
Send message
Joined: 27 Jan 20
Posts: 3
Credit: 405
RAC: 1
Message 4318 - Posted: 24 Feb 2020, 15:58:20 UTC
Last modified: 24 Feb 2020, 15:58:20 UTC

i am getting lots of work tasks come through but every single one of them gets a computation error straight away not sure exactly why likely due to the new GEN CPU i have. I've also noticed that asteroids@home work tasks dont work with the TURING GPU's maybe similar issue here.

My system
Ryzen 5 3600
24GB DDR4 Ram
256GB NVME HDD
windows 10 Pro

unrelated but gtx 1660Ti and GTX 960 GPU's

Tom Miller
Send message
Joined: 8 Jan 13
Posts: 2
Credit: 253,745
RAC: 3,556
Message 4321 - Posted: 25 Feb 2020, 0:17:45 UTC - in response to Message 4318.
Last modified: 25 Feb 2020, 0:18:20 UTC

i am getting lots of work tasks come through but every single one of them gets a computation error straight away not sure exactly why likely due to the new GEN CPU i have. I've also noticed that asteroids@home work tasks dont work with the TURING GPU's maybe similar issue here.

My system
Ryzen 5 3600
24GB DDR4 Ram
256GB NVME HDD
windows 10 Pro

unrelated but gtx 1660Ti and GTX 960 GPU's



I would consider backgrading your gpu drivers for Nvidia to about 390 or so. The current version of the Nvidia drivers throws errors left and right across multiple projects.

Another thing I would do is limit the number of cpu threads that boinc uses to about 90% of your available threads. Try to free up 2-4 cpu threads.
This gives the cpu more time to take care of business and reduces errors caused by an over-committed cpu.

Tom M.

bud20
Send message
Joined: 27 Jan 20
Posts: 3
Credit: 405
RAC: 1
Message 4322 - Posted: 25 Feb 2020, 8:38:53 UTC - in response to Message 4321.
Last modified: 25 Feb 2020, 8:38:53 UTC

I don't really see how downgrading my GPU drivers will fix issues with a CPU based task? asteroids@home has issues running on turing architecture because its new and the programmers over there have not updated their project to work on them yet its a known issue.

As for the mind modelling tasks i thought it may be the same issue that the project has not been updated to work correctly with the newer ryzen 3000 CPU's. I have tried several different computational settings, even tried running only 1 mind modelling task on its own but the same happens.

I was hoping maybe someone in the forums has a ryzen 3000 CPU and has it working might be able to point to a fix or if they are having the same issue as myself then it is likely the project itself as stated above.

ChelseaOilman
Send message
Joined: 9 Jan 19
Posts: 4
Credit: 641,187
RAC: 1,698
Message 4325 - Posted: 25 Feb 2020, 15:55:39 UTC
Last modified: 25 Feb 2020, 15:55:39 UTC

I have two Ryzen 3600 systems and one 3600X system. All were running the tasks just fine until this morning when the server is almost out of work. Possibly you have memory issues.

ChelseaOilman
Send message
Joined: 9 Jan 19
Posts: 4
Credit: 641,187
RAC: 1,698
Message 4326 - Posted: 25 Feb 2020, 16:15:41 UTC
Last modified: 25 Feb 2020, 16:15:41 UTC

@bud20 I just noticed you posted that you have 24GB DDR4 Ram. I'm guessing that means you have 3 x 8GB dimms. I would stick with 2 dimms in the correct slots and maybe run MemTest86 for a few passes to make sure your not getting errors.

bud20
Send message
Joined: 27 Jan 20
Posts: 3
Credit: 405
RAC: 1
Message 4330 - Posted: 26 Feb 2020, 16:32:23 UTC - in response to Message 4326.
Last modified: 26 Feb 2020, 16:32:23 UTC

My Ram is extensively tested I did 12 hour memtest when i purchased my first 2 sticks another 12 hours when I bought my 3rd stick i also run several RAM intensive software's such as unreal engine, blender, AutoCAD etc.. i also use benchmarking software often such as 3DMark with no memory issues at all. I also run many projects through BOINC without issue.

Amicable Numbers (can have very high ram usage)
Asteroids@home
CAS@home
Citizen science grid
Collatz
Cosmology@home
Denis@home
Einstein@home
GPU Grid
LHC@home
milkyway@home
Mindmodelling
NanoHUB@home
NFS@home
Numberfields@home
Primegrid
RNA World
Rosetta@home
Seti@home
universe@home
World community grid
yoyo@home

The only CPU based project i have issues with is this one all other projects work fine. i've tried downclocking my ram from 3200mhz to 3000 and 2800 changed nothing, I tried with 2 sticks and one stick, i've tried changing computation settings with the USE CPU % and time %, i have made sure the "suspend when in use" is un-ticked. I have suspended all other projects and only allowed mindmodelling to do any work but it seems no matter what i try as soon as they begin it instantly fails and gives the computation error. actually really bugging me lol my last resort is to try a windows refresh but i wont try that yet.

ChelseaOilman
Send message
Joined: 9 Jan 19
Posts: 4
Credit: 641,187
RAC: 1,698
Message 4331 - Posted: 26 Feb 2020, 17:37:19 UTC
Last modified: 26 Feb 2020, 17:37:19 UTC

I know it can be very frustrating when you run into these kinds of issues. I've been building my own computers and running distributed computing projects on them all for almost 30 years. I currently have a very large collection of computers at my house right now. More than 20. I have a wide range of builds using CPUs from the Intel 4770K to AMD Threadripper 2990WX. GPUs from AMD 280X to Nvidia RTX 2080. None of my systems had issues with computation errors on this last batch of Mindmodeling tasks. As I mentioned earlier I have three Ryzen 3600 systems in the bunch. It's not uncommon for me to do a fresh install of Windows when I run into issues I can't find another solution to. I mentioned memory as a possible problem because with these latest Ryzen systems finding compatible memory and getting it set correctly can be a real issue. You didn't mention you were overclocking your CPU so I'll assume your not. Is it possible your CPU was getting to hot. I kind of doubt that would be the issue because you say the tasks fail instantly. Maybe update the BIOS if there's a newer BIOS released for your board? Given how cheap SSD's have gotten recently, maybe get one and switch it with your current drive. Then install a fresh copy of Windows on it and see if the problem goes away. If it doesn't you can always reinstall the old drive and use the new one for extra storage. Hopefully you find the issue at some point. Good luck.

Josh Z
Project administrator
Project developer
Send message
Joined: 15 Oct 17
Posts: 18
Credit: 2,283,552
RAC: 55,100
Message 4333 - Posted: 28 Feb 2020, 14:34:41 UTC
Last modified: 28 Feb 2020, 14:34:41 UTC

I'm sorry you are having issues with our project! We often find issues with the models our scientists submit, which can cause these issues. However the error rates we saw with this latest job were quite low. Looking at the errors returned, it seems like they are not being caused the model itself, but of course, that might be incorrect. I try to look into these issues, but it can be tough to recreate and track down. Like ChelseaOilman said, I personally haven't seen any errors on our machines.

Message boards : Number crunching : computation error RYZEN


Main page · Your account · Message boards


Copyright © 2020 MindModeling.org