Message boards : Number crunching : Gpu App?
Message board moderation

To post messages, you must log in.

Previous · 1 . . . 5 · 6 · 7 · 8 · 9 · Next

AuthorMessage
[VENETO] boboviz

Send message
Joined: 21 Feb 15
Posts: 52
Credit: 318,272
RAC: 0
Message 2845 - Posted: 30 Apr 2018, 14:56:18 UTC

Please, remove the opencl app from applications page.
ID: 2845 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
GolfSierra

Send message
Joined: 30 Jun 22
Posts: 1
Credit: 9,333
RAC: 0
Message 5691 - Posted: 1 Jul 2022, 17:53:54 UTC

April 30, 2018 was the last time that the GPU question was discussed. Now it's 4 years later. Are there any plans to develop an AMD/NVidia GPU capable app?

It's confusing seeing "allow AMD GPU" selectable with a check box on the Universe@home preferences page when there is no GPU capable application available?!
ID: 5691 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Dave Studdert
Avatar

Send message
Joined: 9 Jun 22
Posts: 1
Credit: 25,675,333
RAC: 68
Message 5923 - Posted: 31 Oct 2022, 10:10:23 UTC - in response to Message 5691.  

April 30, 2018 was the last time that the GPU question was discussed. Now it's 4 years later. Are there any plans to develop an AMD/NVidia GPU capable app?

It's confusing seeing "allow AMD GPU" selectable with a check box on the Universe@home preferences page when there is no GPU capable application available?!


I just saw the AMD GPU checkbox too and was wondering the same thing. Perhaps the option could be removed from the preferences if there is no GPU app available, to avoid confusion.
ID: 5923 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
[VENETO] boboviz

Send message
Joined: 21 Feb 15
Posts: 52
Credit: 318,272
RAC: 0
Message 5943 - Posted: 11 Nov 2022, 14:45:09 UTC - in response to Message 2567.  


Some news from OpenCL's world:
- ROCm is now at 1.7 version (with support to Tensorflow and Caffe).
- SyCL is now at 1.2.1 version, with a lot of new features.
- Codeplay released ComputeCPP 0.5.1 version to write SyCl/OpenCl code with Visual Studio
- KronosGroup released recently the unified and updated specifications of SPIR-V
- CodeXL from AMD is now at 2.5 version
- ARM released OpenCL Compute Library 17.12 version
- Intel released it's OpenCL SDK 2017 R2


- Rocm is now at 5.3.2 version
- Sycl 2020 release 5
- Spir-V is at 1.6 version
- Arm computer library 22.08 version
- Intel OpenCL SDK 2020 R3
ID: 5943 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
AllenIN

Send message
Joined: 10 Apr 20
Posts: 2
Credit: 914,400
RAC: 0
Message 6198 - Posted: 20 Jun 2023, 4:01:47 UTC

Are we using Radeon GPUs here yet? Been awhile since I've been here and trying to catch up.
ID: 6198 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Grant (SSSF)

Send message
Joined: 23 Apr 22
Posts: 154
Credit: 69,772,000
RAC: 1,613
Message 6199 - Posted: 20 Jun 2023, 5:56:00 UTC - in response to Message 6198.  

Are we using Radeon GPUs here yet? Been awhile since I've been here and trying to catch up.
There is no GPU application.
CPU only.
Grant
Darwin NT
ID: 6199 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Mr P Hucker
Avatar

Send message
Joined: 30 Oct 16
Posts: 180
Credit: 18,395,933
RAC: 301
Message 6200 - Posted: 20 Jun 2023, 18:16:06 UTC - in response to Message 6198.  
Last modified: 20 Jun 2023, 18:20:53 UTC

Are we using Radeon GPUs here yet? Been awhile since I've been here and trying to catch up.
Only place to use those for astrophysics is Einstein I think, since Milkyway has finished seperation as of today. Nvidias can also run Asteroids, but no faster than a CPU so a waste of time.
ID: 6200 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
AllenIN

Send message
Joined: 10 Apr 20
Posts: 2
Credit: 914,400
RAC: 0
Message 6202 - Posted: 20 Jun 2023, 18:52:21 UTC - in response to Message 6200.  

Looks like Einstein is the place to be then. Thanks!
ID: 6202 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Bill F
Avatar

Send message
Joined: 23 Jun 16
Posts: 60
Credit: 7,001,632
RAC: 96
Message 6207 - Posted: 21 Jun 2023, 13:21:13 UTC

Asteroids shows two windows apps that show "(cuda55)" in the Name / version. Is this an indication of a NVIDIA Application ?

Bill F
ID: 6207 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Keith Myers
Avatar

Send message
Joined: 10 May 20
Posts: 308
Credit: 4,733,484,700
RAC: 36,187
Message 6208 - Posted: 21 Jun 2023, 17:00:59 UTC - in response to Message 6207.  
Last modified: 21 Jun 2023, 17:02:14 UTC

Yes. CUDA is a Nvidia compute language. Only runs on Nvidia cards. Much more advanced and capable compared to OpenCL.

But there are no gpu apps here.

A proud member of the OFA (Old Farts Association)
ID: 6208 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Mr P Hucker
Avatar

Send message
Joined: 30 Oct 16
Posts: 180
Credit: 18,395,933
RAC: 301
Message 6209 - Posted: 21 Jun 2023, 21:34:14 UTC - in response to Message 6207.  

Asteroids shows two windows apps that show "(cuda55)" in the Name / version. Is this an indication of a NVIDIA Application ?Bill F
Yes, but those who have run it find it no faster than a CPU. Why use a more expensive and power hungry GPU to achieve the same, when that GPU could be going 50 times faster in another project like Einstein?
ID: 6209 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Keith Myers
Avatar

Send message
Joined: 10 May 20
Posts: 308
Credit: 4,733,484,700
RAC: 36,187
Message 6210 - Posted: 22 Jun 2023, 18:56:33 UTC
Last modified: 22 Jun 2023, 18:58:00 UTC

Unless you are running an absolutely ancient card from the Kepler or Maxwell families, you will never be issued a CUDA55 application. Most people will be issued the CUDA102 or CUDA118 applications which are much faster and require much newer drivers.

Those CUDA applications are much faster than running a task on the cpu. Productivity will be much greater also even if running cpu tasks on all cores.

A proud member of the OFA (Old Farts Association)
ID: 6210 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Mr P Hucker
Avatar

Send message
Joined: 30 Oct 16
Posts: 180
Credit: 18,395,933
RAC: 301
Message 6212 - Posted: 22 Jun 2023, 20:50:33 UTC - in response to Message 6210.  

Unless you are running an absolutely ancient card from the Kepler or Maxwell families, you will never be issued a CUDA55 application. Most people will be issued the CUDA102 or CUDA118 applications which are much faster and require much newer drivers.

Those CUDA applications are much faster than running a task on the cpu. Productivity will be much greater also even if running cpu tasks on all cores.
220 seconds on your RTX 3080Ti
2750 seconds on my Ryzen 9 3900XT / 24 for all threads = 114 seconds

So my Ryzen is doing twice the work of your GPU.

Unless you're running more than one at a time. If you run two at a time, you're getting as much work out of that card as my CPU. Which is no good. A GPU should be 10-50 times faster than a CPU, look at the times on Einstein.
ID: 6212 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Keith Myers
Avatar

Send message
Joined: 10 May 20
Posts: 308
Credit: 4,733,484,700
RAC: 36,187
Message 6213 - Posted: 22 Jun 2023, 22:37:59 UTC - in response to Message 6212.  
Last modified: 22 Jun 2023, 22:40:42 UTC

Yes, I am running 0.5 gpu_usage on all my gpu projects. So divide that 220 seconds in half.
I looked at the top participants list and most of the first page have cpu runtimes in the one hour to two hour values. So nowhere near your 2750 second runtimes.

A proud member of the OFA (Old Farts Association)
ID: 6213 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Mr P Hucker
Avatar

Send message
Joined: 30 Oct 16
Posts: 180
Credit: 18,395,933
RAC: 301
Message 6214 - Posted: 22 Jun 2023, 22:45:37 UTC - in response to Message 6213.  

Yes, I am running 0.5 gpu_usage on all my gpu projects. So divide that 220 seconds in half.
I looked at the top participants list and most of the first page have cpu runtimes in the one hour to two hour values. So nowhere near your 2750 second runtimes.
1.5 hours is only twice as long as mine. Not everyone has Ryzen 9s. But I thought it sensible to compare a good CPU with your good GPU.

And my point stands, you're not achieving the huge speed increase with GPU as you can with say Einstein, so you're wasting your GPU.

Compare one of your GPUs and one of your CPUs with each other on Asteroids, then compare the same two on Einstein.
ID: 6214 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Bill F
Avatar

Send message
Joined: 23 Jun 16
Posts: 60
Credit: 7,001,632
RAC: 96
Message 6215 - Posted: 23 Jun 2023, 0:47:21 UTC - in response to Message 6214.  

Yes, I am running 0.5 gpu_usage on all my gpu projects. So divide that 220 seconds in half.
I looked at the top participants list and most of the first page have cpu runtimes in the one hour to two hour values. So nowhere near your 2750 second runtimes.
1.5 hours is only twice as long as mine. Not everyone has Ryzen 9s. But I thought it sensible to compare a good CPU with your good GPU.

And my point stands, you're not achieving the huge speed increase with GPU as you can with say Einstein, so you're wasting your GPU.

Compare one of your GPUs and one of your CPUs with each other on Asteroids, then compare the same two on Einstein.


Well maybe he likes the Asteroids@Home project. If so then until Einstein starts serving up Asteroid tasks on the Einstein project site he will probably leave his GPU's running where they are ... don't you think?
ID: 6215 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Mr P Hucker
Avatar

Send message
Joined: 30 Oct 16
Posts: 180
Credit: 18,395,933
RAC: 301
Message 6216 - Posted: 23 Jun 2023, 1:52:08 UTC - in response to Message 6215.  

Well maybe he likes the Asteroids@Home project. If so then until Einstein starts serving up Asteroid tasks on the Einstein project site he will probably leave his GPU's running where they are ... don't you think?
Look at the table in his sig. Einstein is his favourite right at the top.
ID: 6216 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Keith Myers
Avatar

Send message
Joined: 10 May 20
Posts: 308
Credit: 4,733,484,700
RAC: 36,187
Message 6217 - Posted: 23 Jun 2023, 3:18:39 UTC
Last modified: 23 Jun 2023, 3:28:19 UTC

The only reason why Einstein is at the top of the chart of my projects is because the Gamma-Ray gpu tasks award a very high credit award. And they run quickly with my anonymous application. And I've been doing Einstein the longest of all my existing current projects.

I looked at other fast hosts on the top participants page similar to your Ryzen 9 7900XT and the Threadripper hosts have similar clock speeds but don't achieve the same runtimes as your lighter loaded 32 thread cpu. Your host is not typical of the general populace of hosts. I'm just saying that for most hosts, the cpu tasks take a lot longer to run and even with more tasks running at the same time, they don't achieve the same RAC as my hosts using a gpu.

Also, I don't run Asteroids all by itself. It shares the gpus along with all my other gpu projects. Only 20% resource share. If I was using 100% resource share at Asteroids and exclusively running that project all by itself, like you do with a cpu only project, my RAC would be much, much higher. I only run one Asteroid task for every other 5 gpu tasks from Einstein, WCG and Numberfields.

A proud member of the OFA (Old Farts Association)
ID: 6217 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Mr P Hucker
Avatar

Send message
Joined: 30 Oct 16
Posts: 180
Credit: 18,395,933
RAC: 301
Message 6218 - Posted: 23 Jun 2023, 4:19:29 UTC - in response to Message 6217.  

The only reason why Einstein is at the top of the chart of my projects is because the Gamma-Ray gpu tasks award a very high credit award. And they run quickly with my anonymous application. And I've been doing Einstein the longest of all my existing current projects.
My point to Bill was it's a project you like.

I looked at other fast hosts on the top participants page similar to your Ryzen 9 7900XT and the Threadripper hosts have similar clock speeds but don't achieve the same runtimes as your lighter loaded 32 thread cpu.
I found some Intel CPUs almost as fast as mine according to cpubenchmark.net, and they are considerably slower, which made me think maybe Asteroids is more suited to AMD CPUs. But then you mention the threadripper. Mine is a Ryzen 9 3900XT, not 7900. Can you link to a threadripper so I can have a look? I find it hard to believe they aren't way faster than mine.
ID: 6218 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Grant (SSSF)

Send message
Joined: 23 Apr 22
Posts: 154
Credit: 69,772,000
RAC: 1,613
Message 6219 - Posted: 23 Jun 2023, 5:18:50 UTC
Last modified: 23 Jun 2023, 5:23:29 UTC

Most Threadrippers have a significantly lower base, and somewhat lower boost, clock speeds.
eg-
        Threadripper 3990X vs AMD Ryzen 9 3900X
Base Clock     2.9 GHz             3.8 GHz	
Boost Clock    4.3 GHz             4.6 GHz

Higher clock speed, shorter processing time. But if you have enough slower threads, then you still get more work done per hour than a higher clocked system with less cores/threads.
Grant
Darwin NT
ID: 6219 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Previous · 1 . . . 5 · 6 · 7 · 8 · 9 · Next

Message boards : Number crunching : Gpu App?




Copyright © 2024 Copernicus Astronomical Centre of the Polish Academy of Sciences
Project server and website managed by Krzysztof 'krzyszp' Piszczek