TR Forums

i have a q6600 and a 4850. right now im running the gpu2 client, but i was looking at all that smp stuff and it just looks like a mess to a guy like me. i got the windows client working, but it was taking at least an hour for every percent and that didn't seem right. all that virtual os stuff also has me confused. i have vista 64 now, so what should i be doing?

Well, you could run one instance of the SMP client and one instance of the regular single-core version. This would leave one full core available to the GPU2 client as I believe it puts about 30% on one core to do it's GPU folding.

Or you could just run two instances of SMP with the GPU client, but that might slow down one of the SMP clients a bit more. Though it would ensure you're using all the folding power you've got.

The Linux virtual machine stuff is pretty confusing if you don't know what you're doing. I believe there was a guide on here for doing so, and I know there's a guide on the overclock.net forums with a pre-made Ubuntu VM you can download and use in VMware.

onlycodered wrote:

Or you could just run two instances of SMP with the GPU client, but that might slow down one of the SMP clients a bit more. Though it would ensure you're using all the folding power you've got.

Doing this on my Q9300+4850 took like an hour a step in the SMP clients. Not worth it.

Vista probably makes it worth trying since it uses so much less of the CPU to run the GPU client.

But those who have both GPU and SMP clients running on the same machine will have to show you how to do it so that it behaves well. There's talk of it in this forum somerwhere. Basically you want to give certain CPU cores to particular SMP clients, sharing 3 of them among 2 SMP clients, but maybe that was with Windows XP where the GPU client takes so much more of the CPU core's time.

Ragnar Dan wrote:

Vista probably makes it worth trying since it uses so much less of the CPU to run the GPU client.

Still maxes one core for me.

skuj wrote:

i have a q6600 and a 4850. right now im running the gpu2 client, but i was looking at all that smp stuff and it just looks like a mess to a guy like me. i got the windows client working, but it was taking at least an hour for every percent and that didn't seem right. all that virtual os stuff also has me confused. i have vista 64 now, so what should i be doing?

I'd like to see a good discussion about this issue, too. I've posed a similar question here:

http://techreport.com/discussions.x/15315

Maybe we'll get some more definitive ideas out of that discussion.

As a point of reference as to what the Q6600 is capable of while running on Linux on Vmware on WinXp.

A STOCK Q6600 running @ 2.4Ghz processes 1% in ~17minutes using 2 Cores under Linux which nets about 1500 Points Per Day (PPD).
2 Clients would net ~ 3000PPD.

Another OVERCLOCKED Q6600 running @ 3.2Ghz processes 1% in ~12.5 minutes using 2 cores: 2200PPD
2 Clients would net ~4400PPD

Add on the GPU client which will take processing away from the above.

JPinTO wrote:

A STOCK Q6600 running @ 2.4Ghz processes 1% in ~17minutes using 2 Cores under Linux which nets about 1500 Points Per Day (PPD).
2 Clients would net ~ 3000PPD.

Sheesh...with the Windows SMP client running on Vista x64, I only average about 1700 PPD on all 4 cores. I just started using the GPUv2 client, though and that is pumping out 5000 PPD from my 8800GT -- though I have turned the fan on that way up.

[quote]Sheesh...with the Windows SMP client running on Vista x64, I only average about 1700 PPD on all 4 cores. [/quote]

Exactly. If one is at all tech-oriented, It's worth spending an evening installing Vmware, and Linux. Plus you get to expand your knowledge base.

The install isn't particularly challenging, and the steps are well laid out: viewtopic.php?f=9&t=47827

The Linux SMP client is not only more efficient, it is also a great deal more reliable. I have never had a problem with my Linux SMP client except when the thing expired and I wasn't aware that had happened. After figuring that out (a problem Stanford has in getting its announcements spread about properly because they don't announce things where they ought to, in my opinion), updating the client or at worst stopping the current client, deleting a core, and then restarting it is all one has to do. Easy.

The big push to using the Linux SMP client came from a user here named Hotdog. He wrote a thread called "Hotdog's Experiences" and got me and a few others interested in it. After that he faded away, though, and I haven't seen word from him in quite a while. After that thread, Tarx's stickied thread became the place to go to get people started running a virtualized Linux SMP client.

The change with the new folding core (software), though, has made it a much bigger gain to run the Linux SMP client. I used to get 1180-some PPD on my Opteron and ~2100-some PPD on my C2D, but with the new core they're both producing a heck of a lot more now. The Opteron is ~1600 PPD now, and the C2D is doing above 2700 PPD. Until they produce a new folding core for Windows, it's a huge waste to run it at all, now. The new folding core, FahCore_a2.exe, is kicking ass all over the place.

[quote="Ragnar Dan"]The Linux SMP client is not only more efficient, it is also a great deal more reliable. I have never had a problem with my Linux SMP client except when the thing expired and I wasn't aware that had happened.[/quote]

I agree on the efficiency. The LinSMP reliability is MUCH better than WinSMP. I've been running into the situation whereby after a Work Unit has completed, it stalls out getting the next one. Probably happens 25% of the time for me, so perhaps it's my build(s). This necessitates a reboot of the Linux guest, or a shutdown and restart of the client. Not a huge deal, particularly if you are monitoring their status.

I should probably update my last post since I finally saw the same problems others have had.

My main machine finally had one of the a2 core WU's fail to allow the next one to be downloaded, or some such thing. Anyway, except when the error happens, the Linux SMP client is still faster and usually more reliable. If you disable -advmethods it may receive fewer or none of the new a2 core WU's, I forget off hand now.

At some point I'll fix this post with more accurate information.