Virtual Machine and CUDA Questions???

Message boards : Number crunching : Virtual Machine and CUDA Questions???
Message board moderation

To post messages, you must log in.

AuthorMessage
Terror Australis
Volunteer tester

Send message
Joined: 14 Feb 04
Posts: 1817
Credit: 262,693,308
RAC: 44
Australia
Message 1070747 - Posted: 26 Jan 2011, 8:58:57 UTC

Q1) Just wondering. Provided you have the right hardware, is it possible to run CUDA from a virtual machine ?

Q2) If you run a 32bit OS can you run a 64bit VM (on 64 bit hardware) and get the full benefit of the 64bit system ?

I tried looking in the VM Ware forums but these questions weren't answered

T.A.
ID: 1070747 · Report as offensive
Profile -= Vyper =-
Volunteer tester
Avatar

Send message
Joined: 5 Sep 99
Posts: 1652
Credit: 1,065,191,981
RAC: 2,537
Sweden
Message 1070750 - Posted: 26 Jan 2011, 9:14:10 UTC - in response to Message 1070747.  
Last modified: 26 Jan 2011, 9:16:03 UTC

Q1: I'm not really sure but i doubt it. Don't think noone has invented graphics bypass directly to PCI-e adresses.

Q2: No! Again if someone/company hasn't invented a way of running 64bit hosts on a 32bit system. It is possible if you have 64bit all the way.
example:

HOST-HYPERVISOR-VIRTUAL MACHINE:

You need to have a:
64-64-64 configuration to make it work.

In your case you have:
64-32-64 and that is not possible from what i know.

Kind regards Vyper

P.S Could you please answer personal messages that i have wrote to you? D.S

_________________________________________________________________________
Addicted to SETI crunching!
Founder of GPU Users Group
ID: 1070750 · Report as offensive
-BeNt-
Avatar

Send message
Joined: 17 Oct 99
Posts: 1234
Credit: 10,116,112
RAC: 0
United States
Message 1070751 - Posted: 26 Jan 2011, 9:18:44 UTC
Last modified: 26 Jan 2011, 9:27:29 UTC

I don't believe you would be able to run CUDA in the VM as even on Hyper-V or VM with hardware virtualization as video cards are installed as generic devices. Parallels Workstation 4.0 Extreme does support hardware virtualization on the video card but don't expect great performance and for that matter I'm not even sure you would actually be using CUDA.

You can not run a 64bit VM on top of a 32bit O/S and cpu. However with that being said you can run a 64bit VM on top of a 32bit host O/S if you have a 64bit cpu with the proper extensions. It looks like they are saying that running it on a 32bit host will limit your VM to 32bit memory limits. If you are running VMware use their tool to figure out if you have one that supports it. But mostly if you have a newer processor it will support it. That and you need to make sure the VM software you are using has the ability to do hardware virtualization and not just software virtualization.

With all that being said, don't run Boinc inside a VM, performance will be horrible. Simply run it on the host OS and be done with it.

*Edit*
On your 32/64 question here is another link to check out.
http://communities.vmware.com/thread/149794
Traveling through space at ~67,000mph!
ID: 1070751 · Report as offensive
Profile RWZeitgeist

Send message
Joined: 20 Feb 07
Posts: 10
Credit: 31,131,540
RAC: 28
United States
Message 1070778 - Posted: 26 Jan 2011, 13:12:39 UTC - in response to Message 1070747.  

Q1) Just wondering. Provided you have the right hardware, is it possible to run CUDA from a virtual machine ?

Q2) If you run a 32bit OS can you run a 64bit VM (on 64 bit hardware) and get the full benefit of the 64bit system ?

I tried looking in the VM Ware forums but these questions weren't answered

T.A.


From the VirtualBox User Manual (see http://www.virtualbox.org/manual/ch03.htm) an answer to your second question:

VirtualBox supports 64-bit guest operating systems, even on 32-bit host operating systems, provided that the following conditions are met:

1. You need a 64-bit processor with hardware virtualization support (see the section called “Hardware vs. software virtualization”).
2. You must enable hardware virtualization for the particular VM for which you want 64-bit support; software virtualization is not supported for 64-bit VMs.
3. If you want to use 64-bit guest support on a 32-bit host operating system, you must also select a 64-bit operating system for the particular VM. Since supporting 64 bits on 32-bit hosts incurs additional overhead, VirtualBox only enables this support upon explicit request.


Note carefully the comment that this configuration "incurs additional overhead", suggesting the guest operating system won't run as fast as it would if the host operating system was a 64 bit system.

FWIW, I've been using VirtualBox to run Windows 7 Ultimate 64 bit as a guest on a 64 bit Fedora Linux host for many months without issues.

Bob
ID: 1070778 · Report as offensive
Profile -= Vyper =-
Volunteer tester
Avatar

Send message
Joined: 5 Sep 99
Posts: 1652
Credit: 1,065,191,981
RAC: 2,537
Sweden
Message 1070784 - Posted: 26 Jan 2011, 13:23:24 UTC - in response to Message 1070750.  

I'm sending a "Shutting Up" to myself.
Thanks for digging BeNt..

//Vyper

_________________________________________________________________________
Addicted to SETI crunching!
Founder of GPU Users Group
ID: 1070784 · Report as offensive
Profile tullio
Volunteer tester

Send message
Joined: 9 Apr 04
Posts: 8797
Credit: 2,930,782
RAC: 1
Italy
Message 1070787 - Posted: 26 Jan 2011, 13:26:01 UTC

I am running a Solaris VM via VirtualBox on my 32-bit Linux host, and a BOINC client and a SETI@home app by Dotsch on it (BOINC 6.2.19 and SETI 6.03) and performance is awful, but my dual core Opteron 1210 with SSE3 capabilities is seen only as pentium_pro+mmx by the BOINC client. Solaris sees it correctly and performance is not that bad.
Tullio
ID: 1070787 · Report as offensive
MarkJ Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 08
Posts: 1139
Credit: 80,854,192
RAC: 5
Australia
Message 1070790 - Posted: 26 Jan 2011, 13:36:01 UTC

I would add that the BOINC developers have been adding some Virtual Box hooks into later (6.12.7) versions.
ID: 1070790 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14650
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1070833 - Posted: 26 Jan 2011, 16:09:55 UTC - in response to Message 1070790.  

I would add that the BOINC developers have been adding some Virtual Box hooks into later (6.12.7) versions.

There are some very interesting ideas coming out of CERN for the (eventual) re-launch of LHC@Home using virtualisation technology - BOINC-VM and Volunteer Cloud Computing (powerpoint slides).
ID: 1070833 · Report as offensive
Profile tullio
Volunteer tester

Send message
Joined: 9 Apr 04
Posts: 8797
Credit: 2,930,782
RAC: 1
Italy
Message 1070849 - Posted: 26 Jan 2011, 16:55:37 UTC - in response to Message 1070833.  

I would add that the BOINC developers have been adding some Virtual Box hooks into later (6.12.7) versions.

There are some very interesting ideas coming out of CERN for the (eventual) re-launch of LHC@Home using virtualisation technology - BOINC-VM and Volunteer Cloud Computing (powerpoint slides).

That's why I am running a Virtual Machine. I want to be prepared.
Tullio
ID: 1070849 · Report as offensive
-BeNt-
Avatar

Send message
Joined: 17 Oct 99
Posts: 1234
Credit: 10,116,112
RAC: 0
United States
Message 1070910 - Posted: 26 Jan 2011, 20:40:40 UTC

My question to you guys though, what's would be the advantage of wanting to run it in the VM? Even if performance is on par with running it from the host OS? I mean it isn't like you would see an improvement from running a host boinc with a VM boinc and getting double work units, it just doesn't work that way. The only main advantage I could see for something like that is if for instance you are running a windows box because you have too, but boinc runs more efficiently in another operating system like Linux. But for that to be beneficial you would have to have a 1/1 hardware VM which isn't going to happen unless you have full VT capable hardware and pay-ware VM, and even then you don't know or can't really expect what kind of efficiency between the host and guest OS you will get without trying it. In some cases you could end up with better performance than the host can achieve, but I'm not so sure that boinc or more specifically Seti@Home and it's apps are going to agree with that since they are pretty low level on the cpu and gpu.

Here's a pretty good article kind of articulating what I'm talking about. It's on Linux with benchmark results comparing the host system versus KVM, Virtual Box 3.2, and 4 beta. In some tests it does worse in some tests it does better. Weird, interesting, and makes me want to play around.

Traveling through space at ~67,000mph!
ID: 1070910 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14650
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1070923 - Posted: 26 Jan 2011, 21:26:44 UTC - in response to Message 1070910.  

My question to you guys though, what's would be the advantage of wanting to run it in the VM? ...

In Cern's case, they're going to supply the VM, with the internal environment set up to their specification. That way, they can control and correct for the OS envionment, and only have one version of their application. To quote from slide 4:

most clients run Windows, but CERN runs Scientific Linux and porting to Windows is impractical.

They're expecting to change the application code frequently as well, and having a single operating platform makes it much quicker and easier to test updates.
ID: 1070923 · Report as offensive
-BeNt-
Avatar

Send message
Joined: 17 Oct 99
Posts: 1234
Credit: 10,116,112
RAC: 0
United States
Message 1070992 - Posted: 27 Jan 2011, 0:56:45 UTC - in response to Message 1070923.  

My question to you guys though, what's would be the advantage of wanting to run it in the VM? ...

In Cern's case, they're going to supply the VM, with the internal environment set up to their specification. That way, they can control and correct for the OS envionment, and only have one version of their application. To quote from slide 4:

most clients run Windows, but CERN runs Scientific Linux and porting to Windows is impractical.

They're expecting to change the application code frequently as well, and having a single operating platform makes it much quicker and easier to test updates.


That makes perfect sense I suppose, having a controlled environment. The biggest leap would be however, how many people helping you understand how to get a VM up and going? I know everything should be forward looking and that fits the bill some what I suppose. But in the end even in a VM environment, each machine acts differently with that VM. So if you tune it to match the machine you are running the next persons will not be the same. Will be interesting to see what comes out of it for sure.
Traveling through space at ~67,000mph!
ID: 1070992 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14650
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1070998 - Posted: 27 Jan 2011, 1:13:29 UTC - in response to Message 1070992.  

My question to you guys though, what's would be the advantage of wanting to run it in the VM? ...

In Cern's case, they're going to supply the VM, with the internal environment set up to their specification. That way, they can control and correct for the OS envionment, and only have one version of their application. To quote from slide 4:

most clients run Windows, but CERN runs Scientific Linux and porting to Windows is impractical.

They're expecting to change the application code frequently as well, and having a single operating platform makes it much quicker and easier to test updates.

That makes perfect sense I suppose, having a controlled environment. The biggest leap would be however, how many people helping you understand how to get a VM up and going? I know everything should be forward looking and that fits the bill some what I suppose. But in the end even in a VM environment, each machine acts differently with that VM. So if you tune it to match the machine you are running the next persons will not be the same. Will be interesting to see what comes out of it for sure.

That's why they call it BOINC-VM. The idea is that the user doesn't have to set up the VM themselves, but BOINC does it for them. CERN are also developing their own tools to enable a (native OS) BOINC client to communicate with and receive information about the state of a science application running in the VM - something I understand is difficult with traditional VMs.
ID: 1070998 · Report as offensive
-BeNt-
Avatar

Send message
Joined: 17 Oct 99
Posts: 1234
Credit: 10,116,112
RAC: 0
United States
Message 1071027 - Posted: 27 Jan 2011, 3:35:30 UTC - in response to Message 1070998.  

Ah Richard, good information indeed. So instead of it being a VM image it's going to have it's own software VM suite with the VM built into it. That would be nice. I was thinking something along the lines of shipping a VM images setup for users which would still leave you needing to setup the software to control the VMs. Pretty cool sounding, but I think I'll stick with doing it the way I always have. ;) Of course if I had multiple machine to run boinc only on......then it would be handy!
Traveling through space at ~67,000mph!
ID: 1071027 · Report as offensive

Message boards : Number crunching : Virtual Machine and CUDA Questions???


 
©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.