Fastest OS

[Stonecold]

Assuming they are all running on the exact same hardware, which major operating systems process cores the fastest on an SMP client? The major operating systems I was thinking of are Windows 7, Mac OS (whatever the most recent version is), and Linux (Ubuntu 11.10). And would running the fastest OS in VirtualBox in Windows (assuming the fastest OS isn't Windows) be good or would Folding@home be too slow on a virtual machine?

[Ravage7779]

Pick one you like and go with it. Whatever has a slight edge today, will be behind the curve tomorrow. The one thing about this project, is nothing stays the same. Virtual machines do incur an overhead hit, altho not a lethal one. But native on any platform will be better than virtual whatever running on top of whatever. The only virtual situation i would get into at this point, is gpu under wine running on linux.

if you plan on gpu folding at all, i would go windows. The PG's plans for gpu on anything but windows is in the "long term" range. So figure 5+ years to never for that to happen, so by the time some miracle happens and they come out with a non windows gpu client, you will be buying new hardware anyways.

If you are going to use the box while folding, go with whatever OS you like to use. If you are going to stick the box in a corner, and all its going to do is fold, go linux. If you are using the box, you will be checking on it periodically anyways, if not, linux is pretty much set it up, and never have to touch it again...ever.

*clink clink*

[Stonecold]

I'm not wondering about the speed for the GPU client. All I'm running is the SMP one. I've heard that Linux is the fastest for a dedicated Folding@home machine when using only the CPU, but I'm not sure if it's entirely true.

[Jesse_V]

I'd say that it depends on the cores available. See: viewtopic.php?f=44&t=20319&p=202272#p202272

[jimerickson]

ravage7779 is right . fastest today means nothing. tomorrow it could be different. pick what you like so it will be enjoyable long term.

[Stonecold]

Alright. I'll use Windows then.

[ChasR]

Right now, the Linux SMP client is 20% faster than the Windows client on the same WU. Linux in a VM running under Windows is 15% faster than than native Windows SMP folding

[Stonecold]

20%!? That's a lot! Does using Folding@home in a virtual machine use much more CPU percentage then Folding@home running normally?

[bruce]

I have no data, but if ChasR's numbers are right, he's saying that running in a VM has a 5% overhead, reducing the +20% native Linux to +15% with Linux under Windows.

In the past, some people have suggested setting up a dual-boot system, but that's a bad idea (compared to running a VM continuously) because with dual-boot, you'll end up running one or the other almost all of the time.

[ChasR]

I'm confident my numbers are correct when you compare performance on the same WU. However, I do have one caveat. The mix of WUs you receive is different running Windows SMP and Linux SMP and between v7 and v6.34 and that will affect production due to project variability. As an example on p11040, Windows SMP has very high production, while Linux SMP doesn't receive that WU. In the end, the observed differrence over weeks or months will be higher or lower than 20% depending on WU assignment patterns. My testing was done on C2Qs and 2600Ks.

As for using Linux in a VM, the VM will show 99% CPU usage in Windows task manager. You have to be sure that the VM runs at idle priority when ungrabbed else it will interfere with windows apps and the GPU client.

I know everyone loves to see numbers to back up performance assertions, so here are some from a semi dedicated 2600K @ 4.8 GHz running Windows SMP, Linux in a VM, and Native Linux on p6901:

Code: Select all

Project ID: 6901
 Core: GRO-A5
 Credit: 7164
 Frames: 100


 Name: ChasR Big VM (2600 K @ 4.8)
 Path: \\CHASR-Desktop\fah\
 Number of Frames Observed: 300

 Min. Time / Frame : 00:20:44 - 52,188 PPD
 Avg. Time / Frame : 00:22:40 - 45,656 PPD


 Name: HTPC 10.10 (2600 K @ 4.8)
 Path: \\HTPC\FAH\
 Number of Frames Observed: 300

 Min. Time / Frame : 00:20:49 - 51,875 PPD
 Avg. Time / Frame : 00:21:12 - 50,475 PPD


 Name: HT-PC SMP (2600 K @ 4.8)
 Path: \\Tmp-pc\fah\SMP\
 Number of Frames Observed: 300

 Min. Time / Frame : 00:24:27 - 40,753 PPD
 Avg. Time / Frame : 00:24:44 - 40,055 PPD


 Name: HTPC VM (2600 K @ 4.8)
 Path: \\HTPC-UBUNTU\fah\
 Number of Frames Observed: 300

 Min. Time / Frame : 00:21:49 - 48,350 PPD
 Avg. Time / Frame : 00:22:56 - 44,862 PPD

[7im]

6901 was bigadv. Got numbers for regular SMP?

[ChasR]

I will have to make clean numbers or find a post where I published them as current results from my C2Qs are polluted by GPU folding. If I'm lucky, I'll have three similar rigs running different OSes on the same WU.

[Stonecold]

If I set the virtual machine to low priority, then can I change the priority of Folding@home in the virtual machine assuming all it'll be doing is folding?

[ChasR]

THe priority setting of the Client in the VM won't make a discernible difference in production if FAH is the only thing running in the VM.

[ChasR]

Here are some comparissons on Q6600s @ either 3.2 or 3.0 Ghz:

Code: Select all

Project ID: 7506
 Core: GRO-A3
 Credit: 706
 Frames: 100


 Name: CCH VM  (q6600@3.0, + 2 GPU instances)
 Path: \\Cch-ubuntu\cch fah\
 Number of Frames Observed: 300

 Min. Time / Frame : 00:06:29 - 9,648 PPD
 Avg. Time / Frame : 00:06:32 - 9,537 PPD


 Name: Cindi Win SMP  (q6600@3.0 + 2 GPU instances, same machine as above)
 Path: \\Cprcc-ws03\FAH\FAH SMP\
 Number of Frames Observed: 100

 Min. Time / Frame : 00:07:34 - 7,652 PPD
 Avg. Time / Frame : 00:07:47 - 7,335 PPD


 Name: COINS SRV4 (Q6600 @ 3.2, Native Ubuntu 10.04)
 Path: \\Coins-srv4\fah\
 Number of Frames Observed: 200

 Min. Time / Frame : 00:05:54 - 11,113 PPD
 Avg. Time / Frame : 00:05:56 - 11,020 PPD


 Name: CR L 1 (q6600 @ 3.2, Native Ubuntu 10.10)
 Path: \\Cr-l1\fah\
 Number of Frames Observed: 300

 Min. Time / Frame : 00:05:51 - 11,256 PPD
 Avg. Time / Frame : 00:05:59 - 10,882 PPD


Name: Marie BO VM (q6600@3.2, + GPU)
 Path: \\Mcbo-ubuntu\mcbo-fah\
 Number of Frames Observed: 300

 Min. Time / Frame : 00:06:02 - 10,747 PPD
 Avg. Time / Frame : 00:06:20 - 9,992 PPD
 Cur. Time / Frame : 00:06:44 - 9,087 PPD
 R3F. Time / Frame : 00:06:43 - 9,118 PPD
 All  Time / Frame : 00:06:45 - 9,055 PPD
 Eff. Time / Frame : 00:06:56 - 8,722 PPD


 Name: Marie BO Win SMP (q6600@3.2+GPU)
 Path: \\Abco-cahill\Marie2 FAH SMP\
 Number of Frames Observed: 280

 Min. Time / Frame : 00:07:06 - 8,418 PPD
 Avg. Time / Frame : 00:07:13 - 8,215 PPD