Conroe SuperPI test

[mmj_uk]

Hello,

Just wondering if someone with Conroe could run 2 instances of 1MB superpi at the same time and see if their time decreases much. Curious to know what time it gets when each core is having to share the cache rather than 1 having it all exclusively.

I don't have a motherboard for my Conroe atm so can't run the test myself.

Cheers.

 

[Gashman]

interesting proposal, should in theory be two exact same times

 

[trojan698]

well it should be similar, you have to bear in mind that core 0 will be running all your the background processes which would increase the time slightly.

core 0 on the left, core 1 on the right



For some reason, when I try to run both super pi's at once, I get an error; 'not convergent in sq' although it's 24 hours prime stable so I assume it's a problem with super pi itself.

 

[ns400r]

You have to run the second instance from a copy. Can't run it twice from the same folder.

 

[mmj_uk]

You have to run the second instance from a copy. Can't run it twice from the same folder.

yeah as above just make a duplicate folder and run each EXE separately, I tested with my Opteron and score was about 0.5-1sec slower with both running at same time.

I read the Conroe SuperPI is a bit unrealistic since the single core has the entire 4MB cache to play with and doesn't need to page out to main memory as much, so just wondered what the hit was like with only half the cache available to it.

 

[mmj_uk]

interesting proposal, should in theory be two exact same times

yeah but I'm wondering if the time will be much slower (and closer to athlon) when the cache is being shared.

remember most programs these days aren't multithreaded so the main core in conroe is having 4MB cache to play with, but once multithreading becomes mainstream each core will only have 2MB effectively.

 

[mmj_uk]

well it should be similar, you have to bear in mind that core 0 will be running all your the background processes which would increase the time slightly.

core 0 on the left, core 1 on the right

...

For some reason, when I try to run both super pi's at once, I get an error; 'not convergent in sq' although it's 24 hours prime stable so I assume it's a problem with super pi itself.

what is your time when you run it on its own? (a single instance)

 

[Boogle]

Both cores have to use the same system memory, so performance will be somewhat degraded.

Additionally Windows DOES NOT use CPUs from a high-level. It doesn't go 'oh these processes are obviously 'background processes', so I'll dump them on some random CPU and keep them there'. Each CPU is merely a unit for executing threads, and the first available CPU gets the next thread. Simple as that. OK, its actually a lot more complicated since you have scheduling, various forms of thread interrupting, etc. etc.

But in essence threads go on any CPU at any time, that is why task manager shows ~50% load on each CPU when you run a single-threaded app.

 

[Rroff]

I got exact same times from each... however it is slower than running 1 copy of superPi...

at slightly over 3gig on my E6600 I got 15seconds with just one copy and 17seconds on each when running 2 at the same time.

EDIT: Just did a 4Mb test... with 2 copies running it takes 9 seconds longer than just 1 copy, and there is a 1 second difference between the 2 copies.