Computational errors with Athlon XP 2000+. Got AMD RMA, but I lapped my heatsink - am I screwed?

[Blackhole Forever]

I just transferred 8GB of files (over the network) from my Pentium 233 box to my Athlon XP 2000+ box. Ran MD5 checksums on both sets of files and the Athlon skipped a beat on 2 of 308 files. Reran the MD5s for the 2 files and got a match this time to the checksums generated on the P233.

Anyway, that was the last straw... I've had problems with this CPU giving different results on MP3 encodes a month or so earlier.

Called AMD and told them what was going on; told them that on my other box (a Barton 2500+), I never get inconsistent results like this, no matter how many times I rerun the checksums, encodes, etc. This is true - the Barton is solid in this respect.

So, I've got an RMA. They want the CPU, heatsink, fan, etc.

Now, I had lapped the heatsink some time ago. The thermal pad is gone too, in lieu of some Artic Silver 3. I can erase all evidence of the compound (given AMD's policy on thermal compounds....voids warranty), but I'm wondering about the other alterations.

Has anyone ever submitted a claim under these conditions? In short:
. original CPU
. original heatsink + fan
. no thermal pad
. heatsink lapped

Just wondering - thanks...

 

[flexy]

whatever test you ran, its rather bizarre.

If ypu want to know whether your memory works (it could be memory fault, too)...get memtest86 and let it run for a few hours to check whether the ram is ok. (Create bootable CD)

Then get prime95 and let it run a few hours in torture-mode.

If both programs dont give any errors...then your CPU/Mem is ok. Whatever checksum program xyz will give afterwards regardless.

 

[Markfw]

Networks and your other system could also be the cause.... Chill... Try again. Run more tests, too many variables.

 

[saechaka]

tell them you changed motherboards, so it won't have the original thermal pad. so you decided to lap the heatsink whatever that means.

 

[Blackhole Forever]

flexy...

I will run memtest86 and maybe prime95 - thanks. I think it's the CPU. The system had been getting unstable at around 54 degrees Celsius at full load. Unstable as in black screen, fully locked up; and that's too low of a temp for it to be doing that. I have an SLK-900 heatsink on it now and it's running cooler at around 45 degrees Celsius max at full load.

But, I get these computational aberrations like that documented above. The MD5 utility is md5summer.exe. The bad encodes I had gotten earlier (pre SLK-900 heatsink) were from lame.exe (v3.90.3).



Markfw900...

The network is fine. All 8GB of transferred data was identical between the two machines. The P233 generated valid MD5s, so it's okay. The Athlon XP 2000+ did not for 2 of 308 files transferred. That's not okay, and something my Barton 2500+ never does.

I will run more tests...

 

[Blackhole Forever]

saechaka...

I suppose I could tell them I switched motherboards and changed out the thermal pad, if they'll buy that. I'm not sure they'll accept my lapping the heatsink, however.

Lapping a heatsink means to sand the base of the heatsink as flat and smooth as possible for the purpose of providing better contact between it and the cpu core.

E.g. my brother lapped an SLK-900A heatsink for me for my Barton 2500+ (overclocked to 3200+ at stock 1.65 voltage). Prior to this, I had seen it get as high at 59 degrees Celsius using the stock retail heatsink and fan. With the unlapped SLK-900A heatsink and a Sunon medium-speed 92mm fan, it was running around 52 degrees Celsius at load. With the heatsink lapped, it's been hitting maybe 42-44 degrees Celsius at load.

 

[mechBgon]

Are the 2000+ and 2500+ getting different results with the same motherboard? You presume a great deal by blaming the CPU for a corruption of a file that has been handled first by a PHY, a southbridge, a northbridge, memory controllers, memory modules, the northbridge and southbridge again, and a hard drive. I mean sure, you could be correct, but...

And the warranty's terms are pretty clear. If you want warranty coverage despite having voided the warranty, then just tell them the whole truth ("I am a tweaker and I lapped my heatsink") and trust that they will extend you some slack for that. I was a bicycle mechanic for a long time, and I can tell you that if the customer told me the straight truth, then I always valued that and cut them slack. But if they told me tales, then I usually went by the book. :evil:

 

[Blackhole Forever]

mechBgon...

The 2000+ and 2500+ are getting different results on different motherboards. Also, so it's clear, there has never been any corruption of a file that was transferred between PCs on my home network. But, I do a verify for good measure. This consists of either running a DIFF utility between the two PCs, or running MD5 checksums individually on each PC (after the files are transferred, of course) and comparing the MD5 output file. You probably know this, but running checksums pushes the CPU to 100%.

Today, I was annoyed by the fact that my Athlon XP 2000+ did not give consistent results while running checksums.

Last month, I was checking some MP3 encodes on my Athlon XP 2000+ and found corruption. I also know that at the time, the CPU was getting hot, and the system was getting unstable.

So, I copied some 250 .WAV files from the 2000+ to the 2500+ (oc'd to 3200+). The files were identical between both machines. I ran the same batch lame.exe encoding process several times on each computer. The 2500+ gave consistent results every time. The 2000+ (which I was now cooling via an open case and turbo floor fan) gave inconsistent results amongst 3-4 of the resultant 250 .MP3 files. Sometimes, it was the same file, sometimes it wasn't. The fact that it was sometimes the same file seemed awfully strange...I don't know what to make of that.

As for AMD, I could tell them the whole truth but I'd feel more confident doing that knowing how it's worked for others. No offense but this is a big corporation we're talking about here; not some local businessperson.

 

[flexy]

Originally posted by: Blackhole Forever
flexy...
I will run memtest86 and maybe prime95 - thanks. I think it's the CPU.

don't think and don't guess

these programs are here for the reason you'll exactly KNOW whats working (and what's not)...and avoid RMAing your CPU by just guessing 'it could be the CPU'. YES, it COULD be the CPU, but as mbg said, it could also be a lot more.

What about bios-settings, ram-timings etc. ?


Edit:
bhf, i have never heard of lame.exe or md5summer.exe as valid/common testing tools, at least not to my knowledge. I'd really be interested what memtest/prime95 say.

 

[mechBgon]

The 2000+ and 2500+ are getting different results on different motherboards.

Exactly. So how do you know it isn't the motherboard that's got the problem? Also, why do you feel the CPU is getting too hot? A retail-boxed 2500+ can handle 100% duty cycle for any amount of time, so with your lapped heatsink and thermal grease, you should be golden on the cooling, unless it's in some cramped, hot-running HTPC case or something

I used to use a 2500+ down at work in a combined server/workstation role on an Asus A7N8X Deluxe with 1.5GB of RAM. Sometimes network transfers did corrupt, with the typical situation being that I'd download a large file from the Internet and it would mysteriously be a very wrong size when it was complete. Have you tried a different NIC, out of curiosity?

 

[VirtualLarry]

I have to agree with the others, that it seems far more likely that the issue might be with the network (flaky PHY? Buggy DMA?) or the mobo (bad caps?). Post some details about the mobos, your FSB speeds, voltages, etc.

You might just need a tad more vcore, if the mobo's regulation is a tad low, or the onboard LAN might be flaky and corrupt frames en-route (I have that problem on my MSI KT4V-L, btw - cheapo LAN), or if you are running an NF2-based system at 200Mhz FSB, make sure that your northbridge chipset is adequately cooled. (Also if SiS chipset.)

Buggy drivers can also cause problems. If you are running W2K, did you remember to run the LargePageMinimum.reg registry patch? Otherwise the hardware pre-fetch of the Athlon will interfere with the default usage of 4MB pages by the OS, and can cause cache-coherency violations, which can lead to data-corruption, usually seen more with AGP systems.

 

[Blackhole Forever]

Well, an update...

I have not been able to get Prime95 to run on this box for more than a few hours without hanging...

MemTest86 will not complete either - hangs or screen goes garbled after a few hours...

My brother has the same motherboard, same CPU, same memory, same BIOS settings. He ran md5summer.exe against 50GB of data twice on his system and got identical checksums on both runs. I couldn't get identical checksums on just 8GB of data.

I'll keep banging away at this and let ya know what happens...