AMD OpenCL issues



Message board moderation

To post messages, you must log in.
1 · 2 · 3 · Next
AuthorMessage
[PUGLIA] kidkidkid3

Send message
Joined: 20 Mar 13
Posts: 1
Credit: 1,357,582
RAC: 18
Message 8001 - Posted: 18 Sep 2023, 5:00:08 UTC
Hi,

https://asteroidsathome.net/boinc/result.php?resultid=401639015

first wu on ati gpu aborted after 1 hour running without progress.

Thanks in advance for your help.

K.
ID: 8001 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Lamberto Vitali

Send message
Joined: 14 Jun 23
Posts: 76
Credit: 5,914
RAC: 0
Message 8003 - Posted: 18 Sep 2023, 5:34:29 UTC - in response to Message 8001.  

Last modified: 18 Sep 2023, 5:35:05 UTC
first wu on ati gpu aborted after 1 hour running without progress.
I've got 8 cards (7 of 280X and 1 of Nano) running it 24/7. I have to abort about 5 tasks every 24 hours. It's not as bad as it was but they do still stick. Until manual intervention that card is sitting idle. I was going to run 2 per card so it could keep doing something, but they're too big to fit in VRAM.

We need a limiter on time on them. I believe LHC has something called a "watchdog" which stops the task if it reaches a certain time. Even better would be if it notices no progress in 15 minutes.
ID: 8003 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Georgi Vidinski
Volunteer moderator
Project administrator
Project developer
Project tester
Avatar

Send message
Joined: 22 Nov 17
Posts: 159
Credit: 13,180,466
RAC: 21
Message 8004 - Posted: 18 Sep 2023, 5:47:04 UTC
You may post here any any troubles, issues and questions related to the AMD OpenCL applications.

Let's keep things in order.
“The good thing about science is that it's true whether or not you believe in it.” ― Neil deGrasse Tyson
ID: 8004 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Georgi Vidinski
Volunteer moderator
Project administrator
Project developer
Project tester
Avatar

Send message
Joined: 22 Nov 17
Posts: 159
Credit: 13,180,466
RAC: 21
Message 8005 - Posted: 18 Sep 2023, 6:46:14 UTC - in response to Message 8001.  

Last modified: 18 Sep 2023, 6:54:38 UTC
Is there particular reason why you keep that machine with such an old software?

I can trace the AMD's driver version 01800.8 (VM) or 18.00 back to the 2015 while there is 23.10 available already.

And of course my next question is about the Windows 10 itself which is run by your machine - build 10.00.14393.00, traced back to the 2016, without a single update for the past 7 years?

Could you try to update at least the AMD's driver?
AMD Radeon™ RX 6900 Series Drivers & Support

Meanwhile I'll check my options to produce another build using hyper older MSVC tools.

Best,
Georgi
“The good thing about science is that it's true whether or not you believe in it.” ― Neil deGrasse Tyson
ID: 8005 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
ahorek's team
Volunteer developer
Volunteer tester

Send message
Joined: 1 Jan 13
Posts: 35
Credit: 5,289,630
RAC: 60,720
Message 8009 - Posted: 18 Sep 2023, 16:55:30 UTC
Radeon HD 6900 is a 13-year-old trash.
https://www.techpowerup.com/gpu-specs/radeon-hd-6950.c405

on the other hand, AMD Radeon RX 6900 is pretty recent
https://www.techpowerup.com/gpu-specs/radeon-rx-6900-xt.c3481

similar name, but very different cards :) in theory, the old card could run these tasks (it has only 2GB of memory), but don't expect ANY performance. I would estimate 6 hours / WU @ 150W ...

support for old hardware is welcome, but this doesn't really make sense. These days, the most low-end or even integrated GPUs are much more powerful & efficient than this card.
ID: 8009 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
ahorek's team
Volunteer developer
Volunteer tester

Send message
Joined: 1 Jan 13
Posts: 35
Credit: 5,289,630
RAC: 60,720
Message 8010 - Posted: 18 Sep 2023, 17:07:16 UTC
maybe just lowering the intensity for super low-end cards could do the trick. These GPUs probably can't handle kernels in time, so they're failing on TLDR protection.

but I would rather mark these cards as not supported. Some protections like a watchdog/memory limits could help with hanging tasks, so I second that.
ID: 8010 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Ian&Steve C.
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 23 Apr 21
Posts: 70
Credit: 55,092,955
RAC: 540,735
Message 8011 - Posted: 18 Sep 2023, 17:08:37 UTC - in response to Message 8009.  
Radeon HD 6900 is a 13-year-old trash.
https://www.techpowerup.com/gpu-specs/radeon-hd-6950.c405

on the other hand, AMD Radeon RX 6900 is pretty recent
https://www.techpowerup.com/gpu-specs/radeon-rx-6900-xt.c3481

similar name, but very different cards :) in theory, the old card could run these tasks (it has only 2GB of memory), but don't expect ANY performance. I would estimate 6 hours / WU @ 150W ...

support for old hardware is welcome, but this doesn't really make sense. These days, the most low-end or even integrated GPUs are much more powerful & efficient than this card.


agreed. I'm not sure how much VRAM is being used by the AMD app here, but very possible 2GB isnt enough. and yes it will be VERY slow.

ID: 8011 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Georgi Vidinski
Volunteer moderator
Project administrator
Project developer
Project tester
Avatar

Send message
Joined: 22 Nov 17
Posts: 159
Credit: 13,180,466
RAC: 21
Message 8012 - Posted: 18 Sep 2023, 17:09:29 UTC - in response to Message 8009.  
Radeon HD 6900 is a 13-year-old trash.

This is some very strong language.
It seams you cannot even imagine that not everybody can afford RX 6900. So, try to be more polite next time.

Everyone is welcome to contribute to this project with no matter if his equipment is aged.
As to what hardware is or will be supported is up to the Asteroids@home's team and it's developers.
“The good thing about science is that it's true whether or not you believe in it.” ― Neil deGrasse Tyson
ID: 8012 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Lamberto Vitali

Send message
Joined: 14 Jun 23
Posts: 76
Credit: 5,914
RAC: 0
Message 8014 - Posted: 18 Sep 2023, 17:34:33 UTC - in response to Message 8011.  
I'm not sure how much VRAM is being used by the AMD app here, but very possible 2GB isn't enough. and yes it will be VERY slow.
So is my phone but I run Asteroids on it because it's there. If old stuff works, I use it. Why throw it away? Run it into the ground.
ID: 8014 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
ahorek's team
Volunteer developer
Volunteer tester

Send message
Joined: 1 Jan 13
Posts: 35
Credit: 5,289,630
RAC: 60,720
Message 8015 - Posted: 18 Sep 2023, 17:35:56 UTC
ok, sorry for the language. The card was high-end in 2010, but it's a really long time...

The performance will be very bad and power consumption high. If it's ok for the owner to pay that price (in the electricity bill) and for developers to spend extra effort supporting it, I can't argue with that. Of course, everyone is welcome to contribute and each WU counts, I just personally don't think it's a reasonable choice.
ID: 8015 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
ahorek's team
Volunteer developer
Volunteer tester

Send message
Joined: 1 Jan 13
Posts: 35
Credit: 5,289,630
RAC: 60,720
Message 8016 - Posted: 18 Sep 2023, 17:46:46 UTC
I understand some people can't afford a new GPU, but people don't realize running such old hardware is actually wasting money, even the cheapest 2-hand GPU will pay itself in no time, and most likely offering better performance too = more contribution. It's just a friendly suggestion.
ID: 8016 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Ian&Steve C.
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 23 Apr 21
Posts: 70
Credit: 55,092,955
RAC: 540,735
Message 8017 - Posted: 18 Sep 2023, 17:46:54 UTC - in response to Message 8014.  
I'm not sure how much VRAM is being used by the AMD app here, but very possible 2GB isn't enough. and yes it will be VERY slow.
So is my phone but I run Asteroids on it because it's there. If old stuff works, I use it. Why throw it away? Run it into the ground.


the RAM use by an Arm app has nothing to do with the VRAM use by a GPU app. apples to oranges.

ID: 8017 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Lamberto Vitali

Send message
Joined: 14 Jun 23
Posts: 76
Credit: 5,914
RAC: 0
Message 8018 - Posted: 18 Sep 2023, 17:49:06 UTC - in response to Message 8017.  
the RAM use by an Arm app has nothing to do with the VRAM use by a GPU app. apples to oranges.
I wasn't talking about RAM, I was talking about using something when it isn't very fast.
ID: 8018 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Lamberto Vitali

Send message
Joined: 14 Jun 23
Posts: 76
Credit: 5,914
RAC: 0
Message 8019 - Posted: 18 Sep 2023, 17:51:11 UTC - in response to Message 8015.  

Last modified: 18 Sep 2023, 17:51:42 UTC
ok, sorry for the language.
I never knew "trash" was a naughty word. Americans say it all the time to mean "rubbish" or "destroy".

The performance will be very bad and power consumption high. If it's ok for the owner to pay that price (in the electricity bill) and for developers to spend extra effort supporting it, I can't argue with that. Of course, everyone is welcome to contribute and each WU counts, I just personally don't think it's a reasonable choice.
Not everyone pays the same for electricity. If you run entirely on solar for example, you have spare power in summer. Selling it to the grid gets you peanuts. And for anyone, in winter the card is also a heater, so it costs you nothing.
ID: 8019 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
ahorek's team
Volunteer developer
Volunteer tester

Send message
Joined: 1 Jan 13
Posts: 35
Credit: 5,289,630
RAC: 60,720
Message 8020 - Posted: 18 Sep 2023, 19:14:05 UTC - in response to Message 8019.  

Last modified: 18 Sep 2023, 19:28:05 UTC
I really didn't want to offend anyone :) my original point is Georgi Vidinski was referring to the wrong GPU model with a similar name, that's why he couldn't find recent drivers.

I don't like throwing away old hardware that still works and I appreciate your efforts to give cards like this a second life. At least it's fun to compare what progress has been made after these years in the HW world.
Btw my 386 still works and it should have all the capabilities to run Asteroids tasks, but I don't think a potential contribution to the project will be meaningful. Yes, it was the best CPU once, but these days it's not the right tool for the job (just joking). There's a point when it's time to move on and drop the historical burden. I don't mean bleeding edge, but that card is a 13-year-old and unsupported by AMD for 8 years. Is it really worth it?
ID: 8020 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Lamberto Vitali

Send message
Joined: 14 Jun 23
Posts: 76
Credit: 5,914
RAC: 0
Message 8021 - Posted: 18 Sep 2023, 19:29:05 UTC - in response to Message 8020.  

Last modified: 18 Sep 2023, 19:30:45 UTC
I don't mean bleeding edge, but that card is a 13-year-old and unsupported by AMD for 8 years. Is it really worth it?
I've got 12 year old stuff. Four 12 core Xeon CPUs, twelve R9 280X GPUs (with 3GB of memory, only 1.5 times the "trash"), they do a lot of work. Even my main card for playing 3D games is 4GB.

I've just worked out I produce half a pint of nuclear waste per year with the electricity I use :-)

Next round's on me.
ID: 8021 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
ahorek's team
Volunteer developer
Volunteer tester

Send message
Joined: 1 Jan 13
Posts: 35
Credit: 5,289,630
RAC: 60,720
Message 8022 - Posted: 18 Sep 2023, 19:42:38 UTC - in response to Message 8021.  
6900 is a terrascale (VLIW4) architecture
your GPU R9 280X is a more modern GCN which was a big leap in performance back then, especially for computing
ID: 8022 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Lamberto Vitali

Send message
Joined: 14 Jun 23
Posts: 76
Credit: 5,914
RAC: 0
Message 8023 - Posted: 18 Sep 2023, 19:52:10 UTC - in response to Message 8022.  

Last modified: 18 Sep 2023, 19:52:49 UTC
6900 is a terrascale (VLIW4) architecture
your GPU R9 280X is a more modern GCN which was a big leap in performance back then, especially for computing
What about my dual core E5200? :-)

We're going to get spanked in a minute for going off topic....
ID: 8023 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
ahorek's team
Volunteer developer
Volunteer tester

Send message
Joined: 1 Jan 13
Posts: 35
Credit: 5,289,630
RAC: 60,720
Message 8025 - Posted: 18 Sep 2023, 20:19:03 UTC - in response to Message 8023.  
Let's be constructive. I have an old mobile GCN gpu that doesn't work. I can help with testing if devs are interested, but I have no intention to use it in a long term. In many cases (except games), it's more effective to just use a cpu instead.
ID: 8025 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Lamberto Vitali

Send message
Joined: 14 Jun 23
Posts: 76
Credit: 5,914
RAC: 0
Message 8026 - Posted: 18 Sep 2023, 20:21:37 UTC - in response to Message 8025.  

Last modified: 18 Sep 2023, 20:22:05 UTC
Let's be constructive. I have an old mobile GCN gpu that doesn't work. I can help with testing if devs are interested, but I have no intention to use it in a long term. In many cases (except games), it's more effective to just use a cpu instead.
Why not aswell? Or stick it on Einstein? It's cruelty to silicon to leave it switched off.
ID: 8026 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
1 · 2 · 3 · Next