Two Antiques Vastly Different

[EdH]

I have two nearly identical machines that are not acting at all that way. The following is identical for both:

$ lscpu
Architecture:          x86_64
CPU op-mode(s):        32-bit, 64-bit
Byte Order:            Little Endian
CPU(s):                4
On-line CPU(s) list:   0-3
Thread(s) per core:    1
Core(s) per socket:    4
Socket(s):             1
NUMA node(s):          1
Vendor ID:             AuthenticAMD
CPU family:            16
Model:                 4
Model name:            AMD Phenom(tm) II X4 B95 Processor
Stepping:              3
CPU MHz:               3000.000
CPU max MHz:           3000.0000
CPU min MHz:           800.0000
BogoMIPS:              5984.57
Virtualization:        AMD-V
L1d cache:             64K
L1i cache:             64K
L2 cache:              512K
L3 cache:              6144K
NUMA node0 CPU(s):     0-3
Flags:                 fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm 3dnowext 3dnow constant_tsc rep_good nopl nonstop_tsc extd_apicid pni monitor cx16 popcnt lahf_lm cmp_legacy svm extapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ibs skinit wdt hw_pstate vmmcall npt lbrv svm_lock nrip_save

Machine 32 has 6GB memory while machine 51 has 8GB. All memory is:

DIMM DDR3 Synchronous 1333 MHz (0.8 ns)

and, shows:

clock: 1333MHz (0.8ns)

math32 has two different RAM modules, while math51 modules are the same.


top for math32 running las:

$ top

top - 14:53:27 up 12 days,  6:34,  2 users,  load average: 4.83, 4.17, 3.90
Tasks: 205 total,   2 running, 203 sleeping,   0 stopped,   0 zombie
%Cpu(s): 90.6 us,  1.1 sy,  0.0 ni,  8.3 id,  0.0 wa,  0.0 hi,  0.0 si,  0.0 st
KiB Mem :  5594564 total,   454540 free,  3437892 used,  1702132 buff/cache
KiB Swap:  2084860 total,  2084860 free,        0 used.  1823424 avail Mem 

  PID USER      PR  NI    VIRT    RES    SHR S  %CPU %MEM     TIME+ COMMAND     
17084 math32    20   0 3067396 2.403g   7424 S 354.5 45.0  51:46.30 las

top for math51 running las:

$ top

top - 14:53:25 up 14 days,  6:44,  2 users,  load average: 3.37, 3.65, 3.71
Tasks: 208 total,   1 running, 207 sleeping,   0 stopped,   0 zombie
%Cpu(s): 82.4 us,  0.8 sy,  0.0 ni, 16.8 id,  0.0 wa,  0.0 hi,  0.0 si,  0.0 st
KiB Mem :  7658948 total,  2775356 free,  3346428 used,  1537164 buff/cache
KiB Swap:  4024316 total,  4024316 free,        0 used.  3983772 avail Mem 

  PID USER      PR  NI    VIRT    RES    SHR S  %CPU %MEM     TIME+ COMMAND     
 6922 math51    20   0 2819596 2.393g   7280 S 331.5 32.8 344:45.53 las

The return time for math32 is around 44 minutes per las run, while math51 takes around 2 hours and 40 minutes to complete its las run. Math32 is in a warm environment, while math51 is cooler.

So, should I assume a warmer climate and less memory is far better than a cooler climate with more?

[sdbardwick]

Originally Posted by EdH

So, should I assume a warmer climate and less memory is far better than a cooler climate with more?

Heh. Since we are making assumptions, I'd assume that the slower box is misconfigured or has cpu thermal problems (bad HSF mounting or bad HSF).

[VBCurtis]

I'd install the "sensors" package, which reports a variety of motherboard/CPU temp readings. Or find a package that provides live updates of CPU actual speed; you're getting results at 1/4 speed, and min CPU speed is listed as 800/3000 = 28% or so of regular speed; this also suggests math51 is running as slowly as it is able to.

[EdH]

I had tried "xsensors" which I use regularly with my other machines, but for these CPUs, I get something called k10temp which seems OK and nothing else available. I have realized a thicker plot, though. I swapped the locations of the two machines and in starting the "faulty" one, I find that the case fan is at super-rpm at idle, even though the BIOS says to idle slow. Indeed, more study of this is warranted. . .

[EdH]

Note: What I thought was a case fan (mentioned earlier) is actually the CPU fan, which is ducted to the CPU heatsink.

I got sensors to work, but it gave me the same k10temp, as xsensors, which is apparently just a GUI sensors wrapper. It just shows a continuous reading of the static one sensors gives me.

I did remount the heat sink and of note, the CPU fan seemed to take longer to run up to full speed, but the temp was still 19 C while running ECM on all four cores. I even blocked the flow of air from the fan to the heat sink and the temp only rose to 25 C.

There is a heatsink on another chip within the CPU exit air stream, which was warm to the touch. I remounted that one as well, to no noticed improvement. The CPU temperature seems to indicate that it is stepped down (since at full use it's generating very little heat), but the fan seems to indicate that it "thinks" it is overheating.

Oddly, if my memory serves me, which is often not the case, this was a top performing machine until recently. Maybe it's tired.

[EdH]

Well, testing shows the CPU is humming along at 3000 MHz, with occasional dips during idle times.

I thought I had discovered "A" bad RAM module and with a single different one, had the machine running where it should be.

But, I couldn't leave things alone, could I?

I now seem to have "FOUR" bad RAM modules, or I just made the machine behave worse!

The modules all survive basic memory tests, but I will need to check them in a different system to see anything further.

[EdH]

The study continues, but is not advancing!

The good news is that all the modules passed a more thorough test on a different machine.

I then installed different modules into the poorly behaving one and, what do you know, it worked as it should. I reset and retested and all went well. So, I shut it down and moved it to its new location and WTF, back to misbehaving! I changed nothing but the power cable, net cable and location.


On another note, somehow, I have "fixed" the full speed fan issue. It "appears" to be idling quietly, like it should.

[Viliam Furik]

Originally Posted by EdH

The study continues, but is not advancing!

The good news is that all the modules passed a more thorough test on a different machine.

I then installed different modules into the poorly behaving one and, what do you know, it worked as it should. I reset and retested and all went well. So, I shut it down and moved it to its new location and WTF, back to misbehaving! I changed nothing but the power cable, net cable and location.


On another note, somehow, I have "fixed" the full speed fan issue. It "appears" to be idling quietly, like it should.

Perhaps the power cable is too long, or of bad quality?

[EdH]

Originally Posted by Viliam Furik

Perhaps the power cable is too long, or of bad quality?

I tried it in place this morning and no go. I moved it back to the bench and did extensive testing. For a moment it seemed the onboard video chip was the culprit, but further testing proved it was only affecting the fan speed, but not the performance. I've set it aside for now.

Perhaps when I get enough "round tuits" saved up, I'll do some swapping with the other one and see if I can isolate the component.

[EdH]

At the risk of jinxing it, I am posting that I believe I have it fixed, using the term in a loose fashion. I have not repaired the ailing trouble, but I have the machine working in the fast manner I was seeking.

I believe the onboard video was, as earlier thought and dismissed, the problem. I have added a video card to the system and it seems to be running at full speed again. I will be testing further, but it has continued through testing that failed during prior sessions.

This machine is run headless with ssh/vnc access and I have noted a significant decrease in that regards, but the terminal operations appear to be running as they should.

[EdH]

How disappointing! It ran fine for several hours and then gradually fell off with the last iterations of (relatively) the same task going from 37 minutes to 55 minutes to 83 minutes to 162 minutes.