Athlon X2 4400 / FC4 slower when only 1 CPU under load

bluebright · #1 12th October 2005, 04:40 AM

Running FC4 64 bits, I found that performance is worse when 1 processor is under load than when both are under load. For example, a "bzip2 somefile" takes 310 sec to complete when 1 processor is under load, but it takes only 165 sec to complete if the other processor is busy at the same time. This problem persists for the default FC4 kernel and the most recent one, and also for a 2.6.14-rc3 kernel. Based on various hints I found through google, I have tried kernels with and without powernow, various settings for power management, and different settings for ACPI (either compuled into the kernel or as kernel options) but so far I have not been able to get the same performance when only 1 CPU is under load. Does anyone have an idea what I could try next, or, better yet, has anyone had this problem as well and managed to fix it?
I have a GA-K8N Ultra-9 motherboard with an AMD Athlon 64 X2 4400 processer, 2 Gb memory.

Thanks!

Shadow Skill · #2 12th October 2005, 04:47 AM

I'm not trying to insult you but are you running an smp kernel? It also depends on the programs being used not every program is made to really use a dual core system to its true potential.

bluebright · #3 12th October 2005, 05:08 AM

I'm only mildy offended :o) Yes, all the kernels I have tried are SMP kernels, and the dmesg output for the recent kernels that I tried all nicely report that two CPUs have been detected. I agree that not all programs are optimized to use dual core. In fact, I have also tried a non-SMP kernel (so I get only one CPU), and there the same test takes 165 sec. Moreover, bzip2 does not take advantage of the dual core, it only runs on one CPU, so I would expect the same test to take 165 sec, not twice as long...

bluebright · #4 25th October 2005, 05:40 PM

OK, I found the solution: the problem was the test itselt. See
http://forums.amd.com/lofiversion/index.php/t58301.html
for more details.

bluebright · #5 10th December 2005, 05:30 PM

The problem was actually not in the test itself (The details are in the thread given in my previous post). The problem was real and manifested itself when using only 1 core (other one idle) with a process that communicated with the memory. In that situation, processing was about a factor of two slower than it should be. This problem has finally been resolved, after emailing back and forth with giga-byte for more than a month. Upgrade to BIOS version F8 and make use of your CPU to the fullest!