more than 1 smp client with a quad?

Associate
Joined
21 Jan 2005
Posts
762
Location
Manchester
Is it worth it or will the clients just share the recources so, therefore, give the same ppd?

btw quad system running at 3.1 with 4gb ram using vista.
 
In my experience, two instances of SMP under linux gives the best ppd. Running anymore and the ppd will be very similar as with running two.
 
ITR said:
In my experience, two instances of SMP under linux gives the best ppd. Running anymore and the ppd will be very similar as with running two.


I was specifically talking about winsmp
 
From what I have read winSMP is a pain in the old buttocks just running one instance so running more than one might be setting yourself up for some future headaches.
 
Take a look at the resources being used by one WinSMP client, specifically the amount of CPU being used. If it is below 90% for much of the time it may well be worth running two clients.

A dual core CPU can run 2 SMP clients so I would be surprised if a quaddie couldn't.
 
I'm running WinSMP on a dual core CPU and it uses 100% of free cycles at all times.

The only advantage I can see to having two clients is, if one client dies, you don't lose all your work! ;)
 
I don't think I'm going to bother running more than one client as its running at 3100 ppd with a 2610 and there's no point risking stability. Theres mixed messages re any performance gain anyway.
 
In Linux there is a definite gain to be made. Cob's insane output is proof of that.

However, the WinSMP client is clunky by comparison so any gain is likely minimal.
 
SiriusB said:
In Linux there is a definite gain to be made. Cob's insane output is proof of that.

However, the WinSMP client is clunky by comparison so any gain is likely minimal.

Not going to bother, like you say I could run ubuntu and run multiple clients but as its an office machine I cannot be bothered. I am, however, rebuilding my home machine and taking the watercooled setup from my office back home. Because of the insane price decreases of the q6600 I bought another one.

I'm going to strip the server out and place mb and drives another case with decent aircooling with the q600, then Ive decided to take that watercooling rig and case and fit the new go q6600 at home and clock the bejesus out of it, then I'll run some instances of vmware for the unix client.

Thats when I get the time that is, speaking of which, drop me an email Sirius. Ta
 
I posted a comparison of average ppd for one, two, three and four linux SMP clients running on my Quad a while back, but it's been deleted now.

In native Ubuntu, one client used something like 65% of the CPU, two used 72%, three used 85% and four used 100%. It was odd, tho some of the Quad-only units use a lot of cache.

Running Ubuntu in VMWare needs two VM's with two clients in each, as each VM can only access two cores.

As for WinSMP, it was far too unstable for me to carry out any testing.
 
Does this apply under VMWare on a Dual (i.e. running 2 clients). I'm showing only 4% idle on the host side so not too sure and I don't want to abuse the deadlines (it's for the science, innit :))
 
You can run two Linux SMPs on a dual-core - my 4400+ managed it just fine and I achieved between 5-15% gain. Not a huge amount but I was meeting the deadlines [with time to spare] so I thought it was worth it.

I tried again on my E6600 but Stanford had released the evil, evil 1385 pointers. These use quite a lot of cache and my E6600 struggled. I saw a negative impact so now I only run a single client.

From what I have read on here the WinSMP client has some horrible WUs, and from experience is very quick to crash/lock up. Not sure I could take babysitting two of them, even if there was a gain.
 
I've had no issues with the WinSMP client other than a tendency to delete WUs in progress when rebooting (more than likely caused by the way I run the WinSMP client). It's actually the Linux SMP client that's given me the odd headache.

Biggest headache of all was when I upgraded the host's Linux kernel and turned on Intel's Virtualisation (VT) support. Started the VM and it promptly locked the PC solid and trashed the VM image. Fortunately I still had the Ubuntu server ISO handy (server is way lighter on memory and install time than the desktop version). That new image got backed up sharpish. :D
 
SiriusB said:
I tried again on my E6600 but Stanford had released the evil, evil 1385 pointers. These use quite a lot of cache and my E6600 struggled. I saw a negative impact so now I only run a single client.
Yeah some of the cache-hungry WU's could do with having only another two clients running alongside them on a Quad. My frame times can rise from my 18min average to as high as ~26mins if I've a cache-gobbler running.
 
Back
Top Bottom