Dual-core stress testing w/ Prime95 - Should I run two simultaneous instances?

[CZroe]

OK, I'm lazy so I just copied the settings that got Jared Walton to 2400MHz here and will be stress-testing it (Unlike Jared, I left chipset voltages at stock like many said to do). I downloaded the 64-bit version of Prime95 and set the affinity for core 0 and left it running all night (Well, that was only 6hrs ago but you know how it is ). I planned on letting it do this all day on one core and then another 24hrs on core 1. Of course, this is only stress testing one core at a time but it is useful for isolating the OC limit of one core (and therefore the entire CPU). Because I'm not planning on trying to find my maximum OC (I just want to see if 2400MHz is stable so I can just leave it), I don't really need to test individual cores seperately to see if 2400MHz is stable; however when I remove affinity for one core, it doesn't really stress anything! Both cores hover a little over 50% CPU utilization.

So, should I just go back to testing one core and then the other? I can imagine this getting very tedious in the future when quad+ multiple cores are the norm! Would it put too much stress to run two processes of Prime95 simultaneously each with affinity set for a different core?

Also, I'm using the retail HSF + AS5 in an Antec P180 w/ two large 120mm fans exhausting air from directly around the CPU (Standard P180 design... PSU is relocated to the bottom). Is it just me who thinks so or is that truely PLENTY of airflow considering that any air the HSF blows out is immediately evacuated? I feel like I don't need a premium HSF and I shouldn't even turn the fan speeds up (The switch is still set to low/quiet). Temps got up to 45C after 6hrs on one core.

Also, FSB [I mean external memory clock] is still a full 200MHz but I did turn down the HyperTransport multi from 5x to 4x as per the guide. What exactly am I doing here? Is it just the communication bandwidth between cores or between the whole CPU and the chipset? I know that lowering an FSB via divider to achieve a higher overclock can be counterintuitive to the resulting performance. Can lowering HT to acheive an overclock actually slow me down?

 

[stevty2889]

Run 2 instances of prime95 at the same time, one for each core, not one core at a time. I don't think you'll have any trouble running at 2.4ghz with the stock heatsink.

How can you be running at 2.4ghz if the FSB is still at 200mhz?? You have to OC the FSB to OC the cpu..your FSB would have to be at least 240mhz to have your CPU at 2400mhz because it has a max 10x multiplier.

And it's been shown that lower HTT speeds, have little to no efffect on performance at all. Most motherboards will become unstable if the HTT goes higher than 1000mhz, so thats why you need to lower the multiplier to OC.

 

[CZroe]

Thanks! That'll save me some time. I'm sure the results are comparable to single-core stability testing.

Well, FSB kinda doesn't mean FSB anymore so I was kinda picking & choosing. FSB has traditionally been the bandwidth/clock speed to memory through the chipset; but now the memory controller is on the CPU and they are calling the internal HT bus between cores the FSB (not even the external HT bus to the chipset!). Because it is the bottleneck I still say that 200MHz memory is 200MHz FSB, "400" FSB in DDR-speak but doubled-up with dual-channel (To feed an "800MHz" FSB in Intel-speak). Because FSB is 240MHz but the maximum bandwidth it can pull from memory is 200MHz-worth (x 2 for Dual-channel, x 2 for DDR), I still say 200MHz. There is no performance increase from the FSB OC without faster memory so I still treat the memory clock as the FSB clock. Even the article calls the 240MHz FSB a 240MHz "CPU Bus" instead of FSB so the lexicon is changing.

Well, even if there were a small performance hit, a resulting 960MHz HT for an additional 400MHz on each core is close enough for me! Easily justifies the loss of 40MHz of HT bandwidth. Thanks for the piece of mind.

Also, I'm wondering: I just copied the settings in his guide. If it proves stable, should I aim for lower voltages or is the writer's resulting 1.40v (An increase of 0.1v) generally safe enough to not care? Like I said, I'm lazy and I don't want to do any additional stress testing if I've reached the point of diminishing returns. I have Cool 'n Quiet disabled but I'd like to enable it again once I install Microsoft's hotfix. If I don't need 1.4v, will the additional 0.1v cause any CnC issues? I'm tempted to try 1.35v for lower temps but I don't want to waste the time if it's not needed at all.

 

[stevty2889]

I wouldn't worry about 1.4v at all, like I said, I'm at 1.475v, and using the stock heatsink(although the one that comes with the 4200+ is the heatpiped version), but your temps are lower than mine. I wouldn't worry unless you start reaching 60c.

 

[CZroe]

Thanks again! Puts my mind at ease. Time to torture this puppy...

 

[pookguy88]

Does prime95 let you pick which processor to test when you have dual cores?
Like, how do you know each instance of Prime95 is testing each core?

 

[biostud]

Originally posted by: pookguy88
Does prime95 let you pick which processor to test when you have dual cores?
Like, how do you know each instance of Prime95 is testing each core?

run prime95 -A0 (CPU 0)
run prime95 -A1 (CPU 1)

voila

 

[CZroe]

Originally posted by: pookguy88
Does prime95 let you pick which processor to test when you have dual cores?
Like, how do you know each instance of Prime95 is testing each core?

Yes, it has "Affinity" in the "Advanced" menu where I can choose core 0 or core 1 for instance. I can then watch in the Task Manager as one core spikes to 100% CPU useage and the other remains relatively untouched. It's also possible to set affinity in the Task Manager for other processes that aren't dual-core/SMP aware.

Edit: What he said! Works with commandline switched too... neat!

 

[CZroe]

Question! Should starting the second instance of Prime95 cause a LIGHTER workload? That's what I'm experiencing. If I start one instance on core 0 it will ramp up to 100% CPU usage. As soon as I start the second instance overall CPU usage drops to 6-7% and the Task Manager graph shows near-bottom worloads on both cores. What's going on? I seem to get a better stress test with one instance set with no affinity. Pagefile usage is still high (doubles with the second instance).

 

[CZroe]

Well, I guess I won't be using Prime95 as a stability test any time soon: It crashes with a rounding error every time shortly after I use the -a1 switch... even if I only launch one instance. It does it at stock speeds and voltages too. It's gotta be an X64 bug or something even though I'm using the 64-bit version.

 

[TheRyuu]

An easier way to choose the core would be to get SP2004 and just uncheck the No Affinity box and then choose which CPU to do the priming on (thats what I do at least)

 

[CZroe]

Originally posted by: wizboy11
An easier way to choose the core would be to get SP2004 and just uncheck the No Affinity box and then choose which CPU to do the priming on (thats what I do at least)

Choosing it within the Prime95 application works perfectly. It when I use the commandline switch to specify core 2 (-a1). I can see that CPU usage is higher on core 2 than core 1 when I do this but it's nothing like selecting it from within the application where it will be 100% with no load on the other core (Using the switch puts a lot of the load on the other core but about 65% on the core you specify).