Message boards :
Number crunching :
LHC on GPU ?
Message board moderation
Author | Message |
---|---|
Send message Joined: 10 Jul 07 Posts: 2 Credit: 1,593,021 RAC: 0 |
Is the LHC sixtrack team going to offer a gpu based work unit or run with the traditional cpu wu only? |
Send message Joined: 12 Jul 11 Posts: 857 Credit: 1,619,050 RAC: 0 |
Looking at it; but will be several months at least. See my next status report coming soon. Eric. |
Send message Joined: 4 May 07 Posts: 250 Credit: 826,541 RAC: 0 |
Does the Project have enough work (WU) to justify the effort to develop GPU for LHC? Are the people at CERN pushing for a faster turn-around of the work they send to LHC? |
Send message Joined: 12 Mar 12 Posts: 128 Credit: 20,013,377 RAC: 0 |
I have good ATI Radeon card, so don't forget to support it too! |
Send message Joined: 9 Aug 05 Posts: 36 Credit: 7,698,293 RAC: 0 |
So much tasks available. 900.000!!! |
Send message Joined: 29 Nov 13 Posts: 59 Credit: 4,012,100 RAC: 0 |
Yea, I've never seen anywhere close to that many before! :) Team AnandTech - WCG, Uni@H, F@H, MW@H, Ast@H, LHC@H, R@H, CPDN, E@H. Main rig - Ryzen 3600, MSI B450 Gm Pro C AC, 32GB DDR4 3200, RTX 3060 Ti 8GB, Win10 64bit 2nd rig - i7 4930k @4.1 GHz, 16 GB DDR3 1866, HD 7870XT 3GB(DS), Win7 64 |
Send message Joined: 13 Sep 14 Posts: 6 Credit: 444,724 RAC: 0 |
Is code available for porting to GPU ? What should be the policy : - Specialized but efficient (CUDA for NVIDIA GPUs only) - Standard but less efficient (OpenCL for ATI, Intel and NVIDIA GPUs) WU's on first |
Send message Joined: 16 Oct 13 Posts: 59 Credit: 342,408 RAC: 0 |
Yes, the code is available. You are free to modify it if you have the expertise to do so. http://sixtrack.web.cern.ch/SixTrack/ |
Send message Joined: 19 Feb 08 Posts: 708 Credit: 4,336,250 RAC: 0 |
I am using my first GPU board on SETI@home, with programs written by Lunatics for Windows only, and in Einstein@home. My CPU is an AMD A10-6700 with four cores. While 4 different CPU programs can run on it, only one GPU program can run on the GPU, so at least 3 cores remain idle. It is true that GPU programs use very little CPU (0.5 in Einstein, 0.04 in SETI), so leaving space to other programs, but 4 GPU programs will never run together on my GPU board. Tullio |
Send message Joined: 13 Sep 14 Posts: 6 Credit: 444,724 RAC: 0 |
If your GPU is crunching less than an average of 70% of utilization, you can run several programs on 1 GPU. Just add the file app_config.xml in the BOINC Data \projects\<project> directory with the following content. Example for seti@home to run 2 programs/GPU: <app_config> <app> <name>astropulse_v6</name> <gpu_versions> <gpu_usage>0.5</gpu_usage> <cpu_usage>0.08</cpu_usage> </gpu_versions> </app> <app> <name>astropulse_v7</name> <gpu_versions> <gpu_usage>0.5</gpu_usage> <cpu_usage>0.08</cpu_usage> </gpu_versions> </app> <app> <name>setiathome_v7</name> <gpu_versions> <gpu_usage>0.5</gpu_usage> <cpu_usage>0.08</cpu_usage> </gpu_versions> </app> </app_config> For einstein@home, apps name are: einsteinbinary_BRP4, einsteinbinary_BRP4G, einsteinbinary_BRP5 and hsgamma_FGRP4 WU's on first |
Send message Joined: 19 Feb 08 Posts: 708 Credit: 4,336,250 RAC: 0 |
I believe the trend in CERN is to go the Virtual way, like vLHC@home and Atlas@home. More CERN projects are being prepared. See the talk by Pete Jones at the Budapest BOINC workshop. Tullio I cannot advice to read the slides of Rom Walton because I am unable to read then on my Linux boxes, there is too much ligh shining. I tried to read them on my new Windows 8.1 PC but Microsoft asked me to buy its Office and I am not going to do this just to see a few slides. But see also the Dave Anderson slides on the history of BOINC. |
Send message Joined: 4 May 07 Posts: 250 Credit: 826,541 RAC: 0 |
I am using my GPU to run SETI (cuda50) Tasks. These seem to be well suited for a GPU as they are relatively short in duration (usually about 20 min.) so when a Task grabs the GPU it runs until done. As far as I know, BOINC has no way of managing/doing Task switching on a GPU (but I may be wrong as it has been a long time since I've looked at running other Projects with long GPU Tasks). The last time I tried running a Project with long Tasks on a GPU the Project hogged the GPU. Now if LHC is going to be pushing a lot more Tasks out to the "crunchers" then maybe it would be worth being able to run them on GPU. My preference would be to see the effort put into running LHC Tasks on a Virtual machine. (Just my opinion.) Tom |
Send message Joined: 19 Feb 08 Posts: 708 Credit: 4,336,250 RAC: 0 |
I agree with you Tom. Since I have a new PC with Windows 8.1 and a Radeon graphic board, I installed the Lunatics optimized apps and they run well, consuming very little CPU. Also Einstein@home gives me some GPU tasks. But the projects I am interested more are vLHC@home where I was one of the Alpha tester, and ATLAS@home, which is a memory hog and I allow only two tasks to run on my 8 GB RAM. I am running LHC@home om my main Linux box, no graphic board, only integrated graphics. I have bought an ASUS GeForce 750 board and downloaded the nVidia driver. But it is not clear to me when to switch from the main board socket to the graphic board socket to connect it to my LCD Samsung Monitor. On the Windows PC I am using my Samsung LED TV as a monitor, since I watch very little TV. Cheers. Tullio |
Send message Joined: 13 Sep 14 Posts: 6 Credit: 444,724 RAC: 0 |
IMHO, going virtual is closing doors. 1) A virtual machine is like a black box system where it is hard to evolve, improve algorithms, and use all the computing ressources of the host machine as the limitation is VirtualBox, 2) Using a virtual machine add a software layer beetween the host hardware and the computing application which is less efficient than no virtual. VirtualBox don't support type 1 hypervisor, 3) Ok, the virtual machine is the same for everybody which is simpler to manage for LHC@home developpers. But challenge it!! 4) You talk about seti@home: let's compare results. For example, computing this WU on GPU is 11 time faster than on CPU. So if you go virtual, a WU currently computed in 10h on CPU, will be computed in, say, 12h-15h (TBD) on a virtual machine. Maybe it is not simple as that, but think about the 55mn on GPU!!! WU's on first |
Send message Joined: 10 Sep 08 Posts: 6 Credit: 6,350,253 RAC: 0 |
Tullio Save the file as a .odp in Impress and then open the .odp. Not perfect but it is readable. |
Send message Joined: 19 Feb 08 Posts: 708 Credit: 4,336,250 RAC: 0 |
I am interest in running scientific codes like those of CERN, written using Scientific Linux, also on a Windows PC and Mac OS X machine.The CERN programs which you see running in the VM console have been ported to SuSE Linux 13.2 which is my next release. I am using now SuSE Linux 13.1 on 2 Linux boxes and Windows 8.1 on another PC. All three run vLHC@home and the Windows PC runs also ATLAS@home 64-bit. All this thanks to Virtual Box. Tullio See also the climatepredictio.net slides at Budapest. They are investigating Virtualized computing too. |
Send message Joined: 19 Feb 08 Posts: 708 Credit: 4,336,250 RAC: 0 |
Thanks. I have exported it to PDF format using LibreOffice and it is more visible. Tullio |
Send message Joined: 10 Sep 08 Posts: 6 Credit: 6,350,253 RAC: 0 |
I tried that first and still couldn't read it. :-) |
Send message Joined: 4 May 07 Posts: 250 Credit: 826,541 RAC: 0 |
Does anyone know BOINC if suspends the GPU when you are running one of the exclusive applications on your system? At one time it didn't. Back in the good old days (before retirement) I was often doing "business" work on a system where I was also running BOINC. When you are running things like Photoshop, Illustrator, Framemaker, etc. it is a real plus to be able to put these in the exclusive applications tab because BOINC suspends the Tasks so you have a "free" CPU. Admittedly, some CPU heavy applications do not have a heavy graphics requirement but others (as I mentioned) do, and poor graphics response is the pits. |
Send message Joined: 16 Oct 13 Posts: 59 Credit: 342,408 RAC: 0 |
The VM is a recurrent topic on LHC@home. It seems that there are many people here that can't / don't want to use virtualbox. Also virtualbox will not remove problems between different architectures (differences in computed values) and their maintenance. |
©2024 CERN