Question for the Quad Gods

[Tobit]

So, I finally appear to have my first Quad working pretty efficiently. I threw away Vista and went back to XP. Unfortunately, I only have a 32-bits license so I've lost 1GB of RAM in the process. However, the system is much more stable and efficient.

I m currently running:

1 x VM with SMP Linux client for 1950 PPD
1 x GPU2 for 2025 PPD

This is essentially using only 3 cores and task manager reports 70% CPU usage.

My question is, do you think I can add a Windows console client to pick up the remaining slack?

 

[aldamon]

My question is, do you think I can add a Windows console client to pick up the remaining slack?

Absolutely. Just assign it to the unused core with the proper switch (-A0 or -A1 or -A2 or -A3) on the shortcut.

 

[Xilikon]

At what speed your Q6600 is running ?

If the speed is high enough (like 3.0-3.2 GHz), you can assign the SMP client to core 0 and 1 only, assign the GPU2 client to core 2 then install a console client and assign it to core 3 only. You would be looking at about 4500-4700 ppd if thing is configured right.

Assigning cores is essential because if a client is hopping around the cores, it lose efficiency. Locking to a single core (or 2 for the SMP) will minimize this and it will run at peak efficiency.

 

[Kendrak]

2nd client!

If you never reboot it great!

Just never reboot Have to do a hard kill at the PSU

 

[Tobit]

If the speed is high enough (like 3.0-3.2 GHz), you can assign the SMP client to core 0 and 1 only, assign the GPU2 client to core 2 then install a console client and assign it to core 3 only. You would be looking at about 4500-4700 ppd if thing is configured right.

Yeah, the Q6600 is currently running at 333x9. I just added the additional client and FahMon is showing about 4350 PPD but I haven't assigned anything to particular cores yet. How do I do this properly? The switches aldamon suggested don't work so I must be doing something wrong.

Thanks

 

[Xilikon]

Just open the task manager. You will see the processes related to the clients. There will be a GPU2 process, a console process and 4 FahCore_A1 processes. On each process, right-click on it and select Define affinity. In the window, you will see 4 tick boxes so just leave the desired core ticket and remove the rest.

Since you seems to run WinSMP, it's hard to adjust the affinity so just leave as is (if you do, it will reset when it get a new WU). Adjust for the GPU2 and console clients. For the SMP, you can assign affinity but you need to setup VMWare and run 1 instance then adjust the VMWare instance affinity.

 

[Tobit]

I'm not running the Windows SMP client. I am running VMWare and using the Linux SMP client.

On the console and GPU clients, wouldn't I have to reset the affinity each time it starts a new WU?

 

[Tobit]

Ok, I just set:

vmware-vmx.exe - Cores 0 & 1
FahCore_11.exe - Core 2
FahCore_82.exe - Core 3

My only concern is that the FahCore* processes restart each time a new WU is downloaded, won't affinity get reset again even if the next WU still uses the same core?

 

[aldamon]

Tobit, like I said, use a switch on the shortcut for the standard client. Don't use affinity.

 

[Tobit]

Tobit, like I said, use a switch on the shortcut for the standard client. Don't use affinity.

I did, but it didn't like that switch as it returned the help screen. This is with the 6.10Beta3 console client.

Bad argument: -A2
Bad Argument(s) received

 

[aldamon]

I did, but it didn't like that switch as it returned the help screen.

Huh? Use it on the Windows shortcut not in the console.

 

[Tobit]

Huh? Use it on the Windows shortcut not in the console.

Same thing, is there some special type of syntax I need to be using other than the obvious?

 

[aldamon]

Weird. You're right. It looks like the switch doesn't work with v6. Try the 5.04 console client instead.

 

[Xilikon]

put it outside the quote.

 

[Tobit]

Weird. You're right. It looks like the switch doesn't work with v6. Try the 5.04 console client instead.

Will do but now need to wait til this WU is finished. heh.

 

[Tigerbiten]

I tend to use SMP-Seesaw-Pro to set affinities in Windoze XP.

Luck ..........

 

[BillR]

Did everyone forget about affinity changer or have I missed something?

The only guy I know who couldn't make it work is relic, go figure

 

[=ACID RAIN=]

Yes. I am helping the enemy. I'm being a good sport

In this case you could try 2 windows SMP clients in separate folders, with affinity changer handling their loads per core. @ 3.6GHz, I get about 4500PPD when the machine remains idle with 2 SMP clients and affinity changer.

Now load the GPU client and use the program you're using now to set its affinity to one CPU. This will drop your PPD on one instance of SMP, but your overall PPD may approach 5K. Just a guess, of course.

You could also let the program float around, but I'm not sure if it would matter or not.

 

[aldamon]

Now load the GPU client and use the program you're using now to set its affinity to one CPU. This will drop your PPD on one instance of SMP, but your overall PPD may approach 5K. Just a guess, of course.

We like to hit preferred deadlines at Team 33.

 

[=ACID RAIN=]

If I hit completion a full 24 hours before preferred deadlines, I think there's some room for additional work. 1760pts in 19 hours.

Q6600 @ 3.6GHz
2GB ram
2 SMP clients

I've even hosed a WU with a reboot 24 hours in and STILL made the preferred deadline. Pretty sure on a quad that's overclocked, there is room for a GPU client to make preferred deadlines.

We prefer to squeeze every last bit out of our rigs on Team 32, all while making stanford consider shortening deadlines.

 

[aldamon]

We prefer to squeeze every last bit out of our rigs on Team 32, all while making stanford consider shortening deadlines.

No, what you suggested was running SMP on two cores, GPU on the third and another instance of SMP on the forth. Here at Team 33 we understand a single core running SMP is not going to hit any preferred deadlines. Here's the quote in case I misunderstood:

In this case you could try 2 windows SMP clients in separate folders, with affinity changer handling their loads per core. @ 3.6GHz,

Now load the GPU client and use the program you're using now to set its affinity to one CPU. This will drop your PPD on one instance of SMP, but your overall PPD may approach 5K. Just a guess, of course.

It would drop the PPD all right. It would also miss preferred deadlines on the SMP instance running on one core.

 

[Tigerbiten]

I was thinking 2x VM's with each client only running at 75-80%.
Now both VM's are asigned to cores 0, 1 & 2.
That should load the 3 cores to 100% without overloading them.
Then run the GPU client on core 4.

The its a question of how overloaded the memory bus is.
If its not overloaded then all 3 clients should be putting out max points.

Luck ................

 

[delvryboy]

I was thinking 2x VM's with each client only running at 75-80%.
Now both VM's are asigned to cores 0, 1 & 2.
That should load the 3 cores to 100% without overloading them.
Then run the GPU client on core 4.

The its a question of how overloaded the memory bus is.
If its not overloaded then all 3 clients should be putting out max points.

Luck ................

This is exactly what I have been thinking.