Stress Testing Old Computer

[storm5510]

I've set my old computer back up and I am running a stress/self test. No errors so far. My question is, are any of the LL tests repeated if I let it sit and run for a day or so?

The computer is a 2004 model HP Pavilion with a 2.8 GHz P4. I took the heat sink off, cleaned both surfaces and applied some high-conductivity thermal paste I got to use on my primary computer. It's running around 55°C. I've seen it get a lot hotter than that. I was using the older type thermal paste which is not much more than white lithium grease.

Note: I did a search for my question and didn't come up with anything relative.

[Prime95]

Yes, it will eventually repeat tests. But repeated tests are still valuable for stress testing.

[Uncwilly]

My understanding is that the stress test does not test whole LL's. My understanding is that it test known bits, using different sizes of FFT's.George pre-posted me.

If you want to resurrect that machine for "useful" work look at the LMH forum, espcially the LMH>100 section, espcially the 332... thread. Or try Operation Billion Digits.

[Kevin]

Originally Posted by storm5510

I've set my old computer back up and I am running a stress/self test. No errors so far. My question is, are any of the LL tests repeated if I let it sit and run for a day or so?

The computer is a 2004 model HP Pavilion with a 2.8 GHz P4. I took the heat sink off, cleaned both surfaces and applied some high-conductivity thermal paste I got to use on my primary computer. It's running around 55°C. I've seen it get a lot hotter than that. I was using the older type thermal paste which is not much more than white lithium grease.

Note: I did a search for my question and didn't come up with anything relative.

Yes, the stress/self test will eventually repeat. You're not really running a full LL test when you stress/self test. You're just running the first 1000 iterations of an LL test for a select set of exponents, and the program has a database of what the result should be for these exponents after 1000 iterations. There's only so many exponents in the database, so eventually you'll be doing computations on the same exponents again.

[markr]

Your P4 may be five years old, but it is still very capable at any work type in GIMPS. Just be aware that a first-time LL test - currently around 47M - may take around 2.5 - 3 months, if it runs 24/7. Running a double-check or two first is a good stress-test that is also productive.

If you'd prefer quicker tasks, GIMPS is short of people to do double-checking, or P-1 if you can give it enough memory.

Regardless, it's your machine. Welcome to GIMPS!

[storm5510]

I was considering using it for trial factoring only. I think that's what you're calling P-1. I can understand why there is a shortage in trial factoring or double checking. Most want to work on a possible record. That's human nature. I thought about doing this with my primary computer. It's a 3 GHz Core 2 Duo. However, I let the server assign it what it wanted.

I still have an earlier "worktodo" file on an external hard drive. It contains some factoring that has already been done. The old computer is not internet connected. I will take one of those numbers and see how it deals with it. To connect it, I would have to run a lot of cable, dig out my router and so on. I would rather not. Forgot to mention, the old computer has 1 GB RAM. That's all it can take.

Originally Posted by Uncwilly

If you want to resurrect that machine for "useful" work look at the LMH forum, espcially the LMH>100 section, espcially the 332... thread. Or try Operation Billion Digits.

Operation Billion Digits!

I found the LMH section. I'll have to do some reading to figure out how to create a "worktodo."

[Mini-Geek]

Originally Posted by storm5510

I was considering using it for trial factoring only. I think that's what you're calling P-1. ... Forgot to mention, the old computer has 1 GB RAM. That's all it can take.

Trial factoring and P-1 are related, (they both are much faster than LLs and are used to find factors, but the math behind the two and the factors you find are quite different: TF finds factors based on the size of the factor alone, P-1 finds factors based on the smoothness of the k in 2kp+1, which is the form of all Mersenne factors; of course most factors smooth enough to be found with P-1 are relatively small, but they can reach impressive sizes at times) but they are separate options in Prime95/GIMPS. GIMPS requires that you have 300MB assigned to Prime95 in order to do P-1 factoring. The more memory you can give it, the better. I'd guess you could assign 800MB on a dedicated Windows machine with 1 GB of total RAM without problems (take with a grain of salt, I have no little to no related experience). I recommend checking how much memory is normally free and subtracting 25-50MB. Prime95 won't let you assign over 90% of total RAM (900MB in this case). Disable all unneeded background tasks to help performance and free up memory.
To learn more about the math behind TF and P-1, see http://www.mersenne.org/various/math.php

Originally Posted by storm5510

Operation Billion Digits!

The LMH Forum? I take it that is seperate from the GIMPS site?

Yep, a billion digits. They work on doing some early factoring of Mersenne numbers around p=3.3 billion, which are just over 1 billion decimal digits. Note that it's not possible to LL test such large numbers with Prime95 right now. If anything could LL test it, it'd take WAY too long to be practical with current hardware, math, and software.
It's separate from GIMPS, but is part of mersenneforum.org.
http://www.mersenneforum.org/forumdisplay.php?f=50
http://home.earthlink.net/~elevensmooth/Billion.html

[storm5510]

The only thing on the old computer is Windows XP Pro. There is nothing else installed other than the video drivers. It would run on Windows 2000. It is just old enough to have all the drivers for it. XP has a lot of "baggage" that is sometimes not needed.

P95 is using 780 MB or RAM for the stress tests, according to the task manager. It also shows 20 processes total. including P95. 77 MB is in the system cache and just over 50 MB is free. I have a little application called SpeedFan running so I can watch the CPU core temperature. It's hovering around 55°C, more or less. The CPU is a Prescott and they were known for running hot.

I imagine I could probably give P95 800 MB if it needed it. That's more than I've got allocated on my primary machine, 512 MB for two cores. I'm sort of cautious about assigning a lot of RAM because there is probably a point where it becomes counterproductive.

[lfm]

Originally Posted by storm5510

The only thing on the old computer is Windows XP Pro. There is nothing else installed other than the video drivers. It would run on Windows 2000. It is just old enough to have all the drivers for it. XP has a lot of "baggage" that is sometimes not needed.

P95 is using 780 MB or RAM for the stress tests, according to the task manager. It also shows 20 processes total. including P95. 77 MB is in the system cache and just over 50 MB is free. I have a little application called SpeedFan running so I can watch the CPU core temperature. It's hovering around 55°C, more or less. The CPU is a Prescott and they were known for running hot.

I imagine I could probably give P95 800 MB if it needed it. That's more than I've got allocated on my primary machine, 512 MB for two cores. I'm sort of cautious about assigning a lot of RAM because there is probably a point where it becomes counterproductive.

This is not really an old machine by GIMPS standards. You could run first time LL checks on it if you wanted.

I don't think p95 needs more than 300MB on a single core and that is only for P-1 factoring.

55C is pretty cool. Nothing to worry about there either. You maybe even have a little overclocking headroom if you want to get into that.

It doesn't matter if the stress tests repeat. They mainly just have to try to keep things hot to see what "boils over". If they have run without error messages for a day or more you should be fine to try some actual work.

It would probably be much better if you were to get it online. It can run for long periods without attention then. Even if you do small TF work units.

[storm5510]

I dug out the router and a cable during the wee hours of the morning. So, it's connected now.

I have it set on the server to do trial factoring only. It appears that it will be able to do one exponent every 17 hours, and that's not bad.

I set the RAM to 512 MB even though it won't use in for factoring.

It went through 10 hours of self tests and passed every one.

When I was using to older thermal compound, it was nothing to see it get into the mid 60's on a moderate load even in a cool room. What I used was a product called OCZ Freeze. It's really thick and gray in color.