mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > Hardware

Reply
 
Thread Tools
Old 2007-01-13, 00:56   #12
Prime95
P90 years forever!
 
Prime95's Avatar
 
Aug 2002
Yeehaw, FL

11101111111102 Posts
Default

Quote:
Originally Posted by sgrupp View Post
What is your guess on memory contention or bus limitations for 4 LL tests running on a quad core with a total of 8M of L2 cache, George?
I have absolutely no idea.
Prime95 is offline   Reply With Quote
Old 2007-01-14, 18:23   #13
S485122
 
S485122's Avatar
 
"Jacob"
Sep 2006
Brussels, Belgium

1,753 Posts
Default

Quote:
Originally Posted by Cruelty View Post
Maybe it is FSB that's holding back this CPU? I would lower the multiplier to x8 and increase the FSB from 266 to 333 MHz to see if that helps.
Also setting an affinity for every instance would be advisable for quad-core since all the communication between the cores happens through FSB.
BTW: does running 2 or 3 instances instead of 4 make any difference?
I will have to see how I can increase the memory speed with my current hardware (ASUS P5B-E, QX6700, 4GB Corsair XMS 6400 cl4). But I do not have much time these days. I tried to change the memory parameters but is is not so easy on the motherboard I use since there are so many parameters to be fiddled with when altering the latencies. My first tries where not stable. I am going to do some research to find relevant data. Modifying the multiplier and FSB will be next.

But concerning your second suggestion, I did some tests running one, two, three and four instances of prime95.

Running one instance or four benchmarks together : 31 ms/ It
Running two instances on different dies (core 1 and 3 for instance.) : 33,5 ms/It
Running two instances on the same die : 35 ms/It
Running four instances : 41 ms/It
All in all, running four instances still gets more work done per time unit than running three, two or one.

It is definetely a memory bottleneck. Intel seems to have gone over the edge of what is possible with their memory architecture and the processing power of their core 2 duo architecture.

It is possible that the new architecture of the Core 2 Duo needs other optimisations in Prime95, but the results of the dual cores do not seem to point in that direction.

A temporary conclusion may be : don' rush to get the Intel QuadCores for prime95.
S485122 is offline   Reply With Quote
Old 2007-01-15, 22:25   #14
Cruelty
 
Cruelty's Avatar
 
May 2005

22·11·37 Posts
Default

I've just noticed that you have 4GB RAM on-board. From what I see on Corsair www, there is no 2GB module of this kind of RAM - this means that you are using 4x1GB combination. If you are using 32-bit Windows XP, then it doesn't make sense to put more than 2GB RAM since OS will utilize only ~3GB. Besides, Asus advises to use less than 3GB of memory with 32-bit Windows XP...
I am also wondering if populating all 4 DIMM slots has an impact on the performance of the system...
Cruelty is offline   Reply With Quote
Old 2007-01-16, 13:54   #15
S485122
 
S485122's Avatar
 
"Jacob"
Sep 2006
Brussels, Belgium

1,753 Posts
Default

The memory I use was intended for a second quadcore that would run in linux of XP64.

I also tried to lower the multiplier to 9 increasing the FSB speed to 300. The results were even worse : iteration times of 47.

This means that the bottleneck is not in the interface between the processor and the north-bridge, but between the north-bridge and the memory.

My conclusion (at the moment) is that for Prime95 one must use the quickest memory available and try to run the memory at a 2:1 rate. From what I read previously a 1:1 rate was perfect for the core2 duos. :-(
S485122 is offline   Reply With Quote
Old 2007-01-20, 23:52   #16
Cruelty
 
Cruelty's Avatar
 
May 2005

162810 Posts
Question

Attached you'll find benchmarks for C2Q and C2D CPUs. I have tested each CPU in 3 scenarios:

1. 3x large FFT test + benchmark (C2Q) / large FFT + benchmark (C2D)
2. 3x small FFT test + benchmark (C2Q) / small FFT + benchmark (C2D)
3. only benchmark

Running scenario 2 has only slight performance hit on benchmark results (2-3%). However when running scenario 1, there is a huge performance hit on C2Q CPU and max. ~10% hit on C2D.

I have verified it with different memory settings, and from what I have seen, memory speed and timings have no impact on the results. FSB speed also does not change anything...

So, couple of ideas:

1. Perhaps this is OS related? Has anyone tested Core2 CPUs under linux?
2. Perhaps this is some flaw of Core2 architecture (e.g. unified L2 cache)?

Any other ideas?
Attached Files
File Type: zip core2-bench.zip (1.8 KB, 131 views)
Cruelty is offline   Reply With Quote
Old 2007-01-21, 10:44   #17
S485122
 
S485122's Avatar
 
"Jacob"
Sep 2006
Brussels, Belgium

110110110012 Posts
Default

Cruelty,

What memory do you use ? Which were the speeds you configured ?

When I have time (in short supply at the moment for me :-( I will swap 8500 C5 memory from my D945 machine to my QX6700. I hope it will give better results.
S485122 is offline   Reply With Quote
Old 2007-01-21, 21:58   #18
Cruelty
 
Cruelty's Avatar
 
May 2005

110010111002 Posts
Default

In C2Q config I have OCZ Platinum DDR2-800 rev.2. I've ran it @ 533 (3-3-3-12-1T), 666 (4-4-4-15-1T), 800 (4-4-4-15-1T). Unfortunately motherboard I have (Asus P5N-E SLI) is still not stable to test all the memory configurations at different FSB speeds. The above tests were done at default FSB=1066, at 1333 the only stable memory frequency is 666.
Cruelty is offline   Reply With Quote
Old 2007-01-22, 06:39   #19
S485122
 
S485122's Avatar
 
"Jacob"
Sep 2006
Brussels, Belgium

33318 Posts
Default

At the suggestion of George I do not use the benchmark to test the speed of Prime95 on the quadcore. I run the assigned LL tests all in the same range at the moment. To have results in an acceptable time I decreased the iterations per screen output to 2048.

My board is based on the P965 chipset. I did not see an impact as huge as you did running 4 LL test together, but perhaps the 27M exponents I currently test, with a 1536K FFT is not what you consider "large".

When running 3 LL tests and one P-1 stage 2, there is a little more slowing down : about one or two miliseconds per iteration.

If all goes well I will test my quickest memory on that board next week-end. I'll keep you posted.

Last fiddled with by S485122 on 2007-01-22 at 06:44
S485122 is offline   Reply With Quote
Old 2007-01-22, 08:00   #20
Cruelty
 
Cruelty's Avatar
 
May 2005

22×11×37 Posts
Default

I have also been running other benchmarks and under e.g. Everest Ultimate 3.5, the only "strange" results I receive is when running memory write test and "CPU-photoworx" test. Both indicate that the memory subsystem of nVidia 650i chipset is apparently not that great. This "photoworx" test shows that a 3GHz C2Q is slower than 2.66GHz C2D (according to software producer, this test relies mostly on fast RAM)...

BTW: what temperatures do you have under full stress, and what utility do you use to check temps?
Cruelty is offline   Reply With Quote
Old 2007-01-22, 11:29   #21
Cruelty
 
Cruelty's Avatar
 
May 2005

162810 Posts
Default

Forgot to mention it myself
Running @ 3GHz @ 1.2V I have 73C on one pair of cores and 71C on the other pair of cores (4 instances of small FFT test) . I am using Coretemp 0.94 utility (verified with Everest Ultimate 3.5).
Asus PC-probe shows 59-60C CPU temperature.

And my CPU cooler is

Last fiddled with by Cruelty on 2007-01-22 at 11:33
Cruelty is offline   Reply With Quote
Old 2007-01-22, 12:31   #22
S485122
 
S485122's Avatar
 
"Jacob"
Sep 2006
Brussels, Belgium

1,753 Posts
Default

I have somewhat lower temperatures : 53°C, 52°C, 47°C and 47°C, as previously stated I use a Zalman 9700 for cooling, there are three case ventilators. I use coretemp to monitor temperatures. The Asus utility gives a temperature some 20-25°C lower.
S485122 is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Dual Core to Quad Core Upgrade Rodrigo Hardware 6 2010-11-29 18:48
exclude single core from quad core cpu for gimps jippie Information & Answers 7 2009-12-14 22:04
Quad Core Questions... TomYosho Information & Answers 2 2009-09-14 13:01
Quad Core R.D. Silverman Hardware 76 2007-11-19 21:57
Optimising work for Intel Core 2 Duo or Quad Core S485122 Software 0 2007-05-13 09:15

All times are UTC. The time now is 22:59.


Tue Nov 30 22:59:21 UTC 2021 up 130 days, 17:28, 0 users, load averages: 2.20, 1.91, 1.66

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, Jelsoft Enterprises Ltd.

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.