Timeline for Power apps for S5R4 Run?

Richard Haselgrove
Richard Haselgrove
Joined: 10 Dec 05
Posts: 1,933
Credit: 198,496,307
RAC: 445,162

RE: I'm not sure this is

Message 83482 in response to message 83480

Quote:

I'm not sure this is the best thread, but I have a question regarding S5R4 runtime on my Xeon E5440. It is quad core and the benchmarks are:

2903 Whetstone
5671 Dhrystone

for each core. My S5R4 WUs are taking 50 hours to complete. This is far longer than I would expect given my dual core E8500 cpu (with slightly better per core benchmarks) finishes the same S5R4 WUs in under 8 hours.

Does anyone know why the performance is so slow on the Xeon?


Your computers are hidden, so we can't see any other details of your system, or of the work you have completed - there might be some clues hidden in there.

But as someone who runs a pair of dual Xeon E5320s (host 831490), I would agree that 50 hours sounds much too long.

I've given you the link to my machine, so that you can see how much (and how little) information is available: I hope that will persuade you that it's safe to allow similar details of your computers to be visible to others (all of the private information, such as names and IP addresses, that you can see for your own machines, is hidden from the rest of us).

My Xeons run under Windows - XP when I first purchased them, Vista now. You don't mention your OS - one known problem is Linux (particularly Ubuntu) running CPUs at reduced speed under BOINC.

Another problem could be memory bottlenecks. I found that there was a significant performance problem in the E5xxx architecture if you don't install memory in matched sets of four FB DIMMs, in the right slots to enable quad channel memory access. It might help if you could give us some information about your system (motherboard, RAM etc.).

Bikeman (Heinz-Bernd Eggenstein)
Bikeman (Heinz-...
Moderator
Joined: 28 Aug 06
Posts: 3,515
Credit: 451,137,498
RAC: 110,415

You are absolutely right, the

You are absolutely right, the performance is quite catastrophic for this kind of CPU.

The only thing that comes to my mind right now is "thermal trip", that is, the CPU thinks (right or wrong) that it's temperature has reached a certain threshold and will automatically reduce frequency and voltage, resulting in a drastic loss of performance. E@H is much more demanding than the BOINC benchmark, so it's not impossible that the threshold temperature is only reached during E@H operation and not during the BOINC benchmark.

But anyway, in this case there would be a malfunction of the cooling system or a malfunction of the temperature sensor, because even the most demanding application should not trigger this in a well designed PC. But it's worth to check the cooling.

CU
Bikeman

Randall  McPherson
Randall McPherson
Joined: 14 Mar 05
Posts: 6
Credit: 8,341,604
RAC: 0

RE: RE: I'm not sure this

Message 83484 in response to message 83482

Quote:
Quote:

I'm not sure this is the best thread, but I have a question regarding S5R4 runtime on my Xeon E5440. It is quad core and the benchmarks are:

2903 Whetstone
5671 Dhrystone

for each core. My S5R4 WUs are taking 50 hours to complete. This is far longer than I would expect given my dual core E8500 cpu (with slightly better per core benchmarks) finishes the same S5R4 WUs in under 8 hours.

Does anyone know why the performance is so slow on the Xeon?


Your computers are hidden, so we can't see any other details of your system, or of the work you have completed - there might be some clues hidden in there.

But as someone who runs a pair of dual Xeon E5320s (host 831490), I would agree that 50 hours sounds much too long.

I've given you the link to my machine, so that you can see how much (and how little) information is available: I hope that will persuade you that it's safe to allow similar details of your computers to be visible to others (all of the private information, such as names and IP addresses, that you can see for your own machines, is hidden from the rest of us).

My Xeons run under Windows - XP when I first purchased them, Vista now. You don't mention your OS - one known problem is Linux (particularly Ubuntu) running CPUs at reduced speed under BOINC.

Another problem could be memory bottlenecks. I found that there was a significant performance problem in the E5xxx architecture if you don't install memory in matched sets of four FB DIMMs, in the right slots to enable quad channel memory access. It might help if you could give us some information about your system (motherboard, RAM etc.).

I must have hidden my computers years ago and forgotten about it. You should be able to see them now. The computer is 1594781.

I ran CPU-Z which can monitor clock speed in real time and the CPU seems to be running consistently at its rated speed under full E@H load. As you can see from the computer specs, I'm running 4GB (shows up as slightly less in XP) of PC2 5300 RAM in 4 1GB sticks. The machine is actually an HP XW6600 with something close to this configuration, I think:

HP xw6600 Workstation
Genuine Windows Vista® 32 Business w/downgrade to Windows XP Professional custom installed
650W 80 PLUS Energy Efficient Chassis
HP xw6600 Localization kit
Intel® Xeon® E5440 2.83 12M/1333 QC 1st CPU
NVIDIA Quadro FX3700 512MB PCIe Graphics
HP 4GB (4x1GB) DDR2-667 ECC FBD RAM
HP 160GB WD Raptor SATA NCQ 10K 1st HDD
HP 16X DVD+-RW SuperMulti SATA 1st Drive

Is the CPU-Z reading an adequate indication that the problem is not thermal or is there some other tool I could run to check if that is the problem? Does the computer information shared show any problems?

Gundolf Jahn
Gundolf Jahn
Joined: 1 Mar 05
Posts: 1,079
Credit: 341,280
RAC: 0

RE: ...one known problem is

Message 83485 in response to message 83482

Quote:
...one known problem is Linux (particularly Ubuntu) running CPUs at reduced speed under BOINC...


I'm not sure if Richard meant the BOINC "cpu throttling", but if your settings for

Use at most XX percent of CPU time
(Can be used to reduce CPU heat)
Enforced by version 5.6+

is not 100%, then you can get problems with restarting tasks under any OS (as far as I know). Check also your local preferences!

Gruß,
Gundolf

Computer sind nicht alles im Leben. (Kleiner Scherz)

Bikeman (Heinz-Bernd Eggenstein)
Bikeman (Heinz-...
Moderator
Joined: 28 Aug 06
Posts: 3,515
Credit: 451,137,498
RAC: 110,415

RE: Is the CPU-Z reading

Message 83486 in response to message 83484

Quote:

Is the CPU-Z reading an adequate indication that the problem is not thermal or is there some other tool I could run to check if that is the problem? Does the computer information shared show any problems?

I'm not that familiar with Windows tools but I'm sure others can recommend a tool that also monitors temperature in real time. That would be interesting.

I don't see any hint in the log information of your unhidden computer, but it confirms that there is a real problem, for example, the wall clock time elapsed is consistent with the unusually high CPU time. The processing also seems to be uniformly slow, even the initial setup phase of the app which is not very CPU intensive is remarkably slow.

To exclude BOINC as a factor here, it might also be instructiove to run teh science app in a standalone benchmark test, a reference workunit was posted here, the Windows scripts for tis test can be found here.

CU
Bikeman

johannes
johannes
Joined: 6 Jun 05
Posts: 6
Credit: 19,562,388
RAC: 0

more food for our

Message 83487 in response to message 83484

more food for our guesswork:
in Randall's task details some lines are logged which I haven't seen in any other task details:

Unrecognized XML in parse_init_data_file: computation_deadline
Skipping: 1221510941.000000
Skipping: /computation_deadline

to make matters even more puzzling, these lines also appear in the task details of his Core E8500, which completes its task in the usual 7 hours.

John Clark
John Clark
Joined: 4 May 07
Posts: 1,087
Credit: 3,143,193
RAC: 0

Temperature monitoring on

Temperature monitoring on Intel Core 2 architecture, like your rig, can be done with RealTemp 2.60 which can be found here. This is one I use for my Quads, but not the older Xeon Prestonia or P3 Coppermines I run for Einstein.

Shih-Tzu are clever, cuddly, playful and rule!! Jack Russell are feisty!

Bikeman (Heinz-Bernd Eggenstein)
Bikeman (Heinz-...
Moderator
Joined: 28 Aug 06
Posts: 3,515
Credit: 451,137,498
RAC: 110,415

RE: more food for our

Message 83489 in response to message 83487

Quote:

more food for our guesswork:
in Randall's task details some lines are logged which I haven't seen in any other task details:

Unrecognized XML in parse_init_data_file: computation_deadline
Skipping: 1221510941.000000
Skipping: /computation_deadline

to make matters even more puzzling, these lines also appear in the task details of his Core E8500, which completes its task in the usual 7 hours.


I also get those lines, seems to be related to the version of the BOINC core client? Anyway, nothing to be concerned about.

CU
Bikeman

Randall  McPherson
Randall McPherson
Joined: 14 Mar 05
Posts: 6
Credit: 8,341,604
RAC: 0

I ran Realtemp. On idle, the

I ran Realtemp. On idle, the cpus are all in the 38-39 C range. On full load running E@H they are in the 59-60 C range. This does not seem too hot.

Realtemp also monitors the CPU frequency in realtime. The clock under full load hovers around 2833.3-2833.4 MHz, so that would seem to indicate no clocking problem.

To make sure it wasn't an issue with the Einstein app, I downloaded and ran some WUs of Seti@Home on both the quad core (problem computer) and dual core (reference). The Seti WUs also ran consistently slower on the Quad core with about the same time ratios.

As recommended above, I tried to run the reference E@H WU using the windows script. When I try to run one of the S5R3 reference WUs with the script it gives the error "The system can not execute the specified program." and immediately exits. I'm not sure if this is indicative of the problem with my machine or simply the script failing for some unknown reason. The logs generated by the script did not contain any clues.

This problem is strange. The CPUs show up fully loaded in windows task manager, CPU temperatures rise on all the cores, the clock seems to be stable at the rated speed, and yet the machine is painfully slow. Any ideas where else I can look?

Bikeman (Heinz-Bernd Eggenstein)
Bikeman (Heinz-...
Moderator
Joined: 28 Aug 06
Posts: 3,515
Credit: 451,137,498
RAC: 110,415

Very interesting problem

Message 83491 in response to message 83490

Very interesting problem indeed.

Concerning the failure of the reference unit to execute: The original reference unit was designed for linux and OSX, The add-on script for Windows needs the windows science app executable, that would be the one that ends in *_1.exe and can be copied from the BOINC/projects/einstein.phys.uwm.edu directory of your BOINC installation.

If the reference unit in standalone gives normal runtime, then this has probably something to do with either your BOINC installation or BOINC settings.

If the reference unit in standalone runs also too slow, well...I think then it would make sense to use a benchmarking suite like "Sandra" to try to detect WHAT is so slow (memory??? integer benchmark? Floating point benchmark?) . The memory benchmark would be most interesting. Please also check the BIOS settings, inadvertently disabling RAM caching (L1 and L2 cache) could probably slow down the system rather drastically, and in a way consistent with the observations made.(The Cache size display in BOINC is too unreliable to draw conclusions).

Any other ideas anybody? The temperature readings suggest that we do not have "thermal trip", and that the CPU is indeed doing "something" all the time, but it still isn't more productive per core than my 866 MHz Pentium III...go figure.

Quote:

I ran Realtemp. On idle, the cpus are all in the 38-39 C range. On full load running E@H they are in the 59-60 C range. This does not seem too hot.

Realtemp also monitors the CPU frequency in realtime. The clock under full load hovers around 2833.3-2833.4 MHz, so that would seem to indicate no clocking problem.

To make sure it wasn't an issue with the Einstein app, I downloaded and ran some WUs of Seti@Home on both the quad core (problem computer) and dual core (reference). The Seti WUs also ran consistently slower on the Quad core with about the same time ratios.

As recommended above, I tried to run the reference E@H WU using the windows script. When I try to run one of the S5R3 reference WUs with the script it gives the error "The system can not execute the specified program." and immediately exits. I'm not sure if this is indicative of the problem with my machine or simply the script failing for some unknown reason. The logs generated by the script did not contain any clues.

This problem is strange. The CPUs show up fully loaded in windows task manager, CPU temperatures rise on all the cores, the clock seems to be stable at the rated speed, and yet the machine is painfully slow. Any ideas where else I can look?

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.