![]() |
![]() |
#1 |
Jan 2003
3138 Posts |
![]()
I've recently put together an Intel Core 2 Duo 2.13GHz machine with 1GB of dual channel 667MHz memory at 4-4-4-12. According to the benchmark page, I should get 27ms for a 1024 FFT - and indeed, I see this when I run the benchmark.
I then got exponents to test from Primenet and started to run stage 1 factoring on cpu 0, and a LL test on cpu 1 (affinity manually set). Prime95 is allowed to use up to 192MB of memory, and there's still a lot of free memory available. Here's what happened: cpu 0: nothing running cpu 1: LL - 27ms cpu 0: stage 1 factoring cpu 1: LL - 32ms cpu 0: stage 2 factoring cpu 1: LL - 34ms Is anyone else experiencing this performance drop when 2 copies of Prime95 are running? I'm not sure if I'll see this when both are running LL. |
![]() |
![]() |
![]() |
#2 |
Undefined
"The unspeakable one"
Jun 2006
My evil lair
41·149 Posts |
![]()
Yup, there are two cores, but still only one memory bus. I expect the effect is even more pronounced with the 4 core chips and 4 instances of prime95. We're talking bottlenecks here, memory being the culprit in this particular case.
|
![]() |
![]() |
![]() |
#3 |
Jan 2003
7·29 Posts |
![]()
Thanks for the reply. The memory bottleneck theory makes good sense to me - though I hadn't expected the bottle neck since it uses a 1066MHz bus, and is running in dual channel mode. I hadn't noticed such a pronounced slow down on my dual core P4. Then again, the rest of the machine isn't the same either - this one uses the integrated graphics of the G945, which might also be accounting for loss of memory bandwidth.
I agree that the situation must be even worse for the quad cores. In fact, you won't even need to go to quad cores, the higher end Core 2 processors are likely to suffer noticably from this bottleneck too (still same bus and memory speeds, but 20% faster cpu). |
![]() |
![]() |
![]() |
#4 |
May 2005
23·7·29 Posts |
![]()
Here are my system's specs - I haven't noticed such a dramatic performance impact (~20%).
I would suggest that you increase your FSB from 266 (x4=1066) to 333 (x4=1333) and run your RAM synchronised with FSB. Also make sure that your RAM works in dual-channel mode - someone already reported significant performance hit when running dual core system in single-channel mode. |
![]() |
![]() |
![]() |
#5 | |
Jan 2003
7×29 Posts |
![]() Quote:
I've tried overclocking, but with little success - the machine is a Shuttle SD32G2 small form factor pc. I can raise the FSB from 266 to about 285 max, before it starts to go unstable. In fact, I'm using Geil 800MHz DDR2 memory, but the motherboard can only handle 667MHz (based on the older Intel G945 chipset). So it's running at an underclocked 4:5 ratio. ![]() |
|
![]() |
![]() |
![]() |
#6 |
May 2005
23×7×29 Posts |
![]()
Check out my tests. I was using LLR in the second instance but that should't really matter.
As for your system, I see there is an integrated VGA on-board. AFAIR there were some performance issues with such systems reported some time ago... if you have any PCI-E VGA, then plug it in and see if that helps. Still I am wondering about this "supports dual-channel memory" - check with CPU-Z if memory really works in dual-channel. |
![]() |
![]() |
![]() |
#7 | |
Jan 2003
7×29 Posts |
![]() Quote:
Yep, cpuz reports dual channel, so that's definately not where the problem lies. I'm wondering about the integrated VGA as well and might buy a PCI-E card. Though for the cost of a low end Geforce 7300 PCI-E card and shipping, I could have gone for the 2.4GHz E6600 instead of the 2.13GHz E6400. Also, the E6600 has extra 2MB of cache. Economically, a dedicated graphics card might not be the best value. |
|
![]() |
![]() |
![]() |
#8 |
May 2005
162410 Posts |
![]()
Firstly I would check if that theory about integrated VGA is valid - do you have someone who can borrow you a PCI-E VGA for an hour or so? Either way, I would not go for a brand new card - rather some second-hand GF-6200 or Radeon x300 - or anything else with PCI-E interface.
Additionally you may also want to check if there are some updated drivers from Intel... |
![]() |
![]() |
![]() |
#9 | |
Jan 2003
7·29 Posts |
![]() Quote:
Already have a PCI-E card in my dual P4 rig, will test that out some time soon. If it works, I'll need to get a new card, since the P4 will be offline until I put it back. ![]() I checked and found that the GF6200 is only about 5-10 bucks cheaper than the 7300. So if I do get something, the 7300 will be the way to go. |
|
![]() |
![]() |
![]() |
#10 |
Dec 2003
Hopefully Near M48
2·3·293 Posts |
![]()
I can confirm that I am having the same problem with a Core 2 Duo T7400 desktop. I'm seeing slowdowns of over 10%.
|
![]() |
![]() |
![]() |
#11 |
Jan 2003
7×29 Posts |
![]() |
![]() |
![]() |
![]() |
Thread Tools | |
![]() |
||||
Thread | Thread Starter | Forum | Replies | Last Post |
running two copies of prime95 in the same machine | ppo | Information & Answers | 25 | 2007-07-30 23:25 |
Dual core: three copies or just two? | JuanTutors | Software | 10 | 2007-04-13 01:43 |
Importance of dual channel memory for dual core processors | patrik | Hardware | 3 | 2007-01-07 09:26 |
Running multiple copies of mprime on Linux | hc_grove | Software | 3 | 2004-10-10 15:34 |
Anyone have hourly copies of summary.txt? | PrimeCruncher | Data | 0 | 2004-05-15 23:32 |