Rebuild of main PC - what OS ?

TheKnat

TheKnat

TheKnat

Associate
Joined
13 May 2007
Posts
1,832
Location
Lancashire, UK
My main PC has been running like a dog recently and its caused me to have to stop using the F@H CPU client because it causes the system to become unstable.

Sooo if I'm going to rebuild this system, I might as well make it get me as many points as possible. Considering the following spec, what would be the best setup for me to go for? :

Intel QX6800 @ stock (needs overclocking, but lack of time has stopped me)
2 x Nvidia 8800 GTX
2GB RAM

Its on XP at the moment, so from what I can gather, either of the 2 below options would be the best:

Option 1:
PC built using Vista. Running 2 GPU clients each assigned to 1 core of the processer.
Install VMWare and run x64 version of Unbuntu and then run a Linux SMP client for the CPU

Option 2:
Same as above, but using XP instead of Vista.

As far as I know, there isn't a linux GPU client and it would cost me around 9000PPD to lose both those cards.

Ideas?
 
If you use vista (certainly the case for me) the GPU client takes only a fraction of 1 core to run (in my case on the Q6600, only a few % of the cpu total). I would imagine that you could therefore run vista and dedicate three cores to VMWare, and have one core dedicated to running the GPU clients. I must admit i quite like vista, so would opt for option 2 - i ran vista on 2gb or RAM for a while and i didnt really notice much of a difference in speed over xp (plus the 64bit version of vista would allow you to have more RAM if needed - 4gb makes vista really fly). This would also boost your ppd, but at a cost!

There is a fudge using WINE to get the gpu clients working, although it is not offcially supported by stanford as far as i know, and it does look quite complicated. I think there is a post about it on the stanford folding forum, but i read about it first on xtremesystems.org/forums - they have a folding section and some more details on running the GPU client under linux.
 
I'd get 2Gb more RAM regardless of what you do, the Linux SMP client will take up 1.3Gb (well, that's what I have mine set to atm). You can also setup two linux SMP clients on one VM for a decent boost of PPD.

As for OS? If you have it, Vista 64 with the 4Gb of RAM.
 
So looks like I need to get 2GB more RAM, then install Vista x64. I'll assign the 2 GPU clients to 1 core and then use the remaining 3 to power VMWare.

The top of the producers chart will be mine!
 
If you're overclocking consider what effect 4x1GB modules will have versus 2x2GB.
On my Asus board the 4 RAM slots are split across two channels. I'm assuming it's easier to OC on a single channel matched pair than two channels with matched pairs.

Someone on the overclocking sub-forum will know for sure, or someone here will set me straight if this is just guff.
 
I've found that limiting my Ubuntu VMs to only 500MB means I get a nice steady stream of 2605s a rare 2665 :( and sometimes lovely 2662s :D More memory means more 2665s which I don't want.
 
2x2Gb makes life easier - but more volts should acheive 4x1 - limited mine (when i used it to 600mb and also never got 2665 unlike winsmp which is all i ever get :()
 
Last edited:
My Ubuntu Server VMs are set to 320MB. It's lower than most recommend but it works OK for me. I'd go for 512 if you don't mind losing the RAM, though.

I hear you can run the GPU client in Linux through WINE for no points loss. I've been thinking that this would probably be the absolute fastest option as you could run SMP natively, but it sounds like hassle and personally I don't want to lose Windows.

I'd go for Vista x64. The two GPU clients combined should use no more than 10% of the CPU, which leaves lots for the SMP clients. Set up two Linux VMs under VMWare (if under x64 you'll need to either use the Server 2.0 beta, as I do, or disable driver signing enforcement on every boot.) If you feel brave enable -advmethods on the SMPs. You'll get the new a2 core WUs which are brilliant PPD.

I'll assign the 2 GPU clients to 1 core and then use the remaining 3 to power VMWare.
Click to expand...
Rather than assigning the cores, I set the VMs to 'low' priority and the GPU client to 'slightly higher'. This is a better option because it allows VMWare to suck up the unused CPU cycles from the core your GPU clients run on (with both the GPU clients this core will be ~40% utilised.)
 
Getting a bit confused now. Is it better to have 1 VMWare running 2 clients, or 2 VMWare's and 1 client on each ?
 
One VM, 2 clients. Though saying that I've not done that before, its always been 2 VMs x 2 clients each for me :) The reason why I have the RAM set to 1.3Gb is because I've had some problems in the past with large WUs.
 
TheKnat said:
Getting a bit confused now. Is it better to have 1 VMWare running 2 clients, or 2 VMWare's and 1 client on each ?
Click to expand...

You definitely need two VMs running, because each VM can only use two cores. Whether you run one or two SMPs in each VM is your choice. Two might give slightly better points but it'll use more memory.
 
You must log in or register to reply here.
Back
Top Bottom