Code-Morphing CPUs and Driver Compatibility

[prosaic]

This is a borderline troubleshooting question, so, if it's outside the purview of this section I can post elsewhere. But the question involves a pretty complex situation, so I thought I'd give it a shot here first.

I have a Toshiba Libretto L5 (subnotebook) that uses a Transmeta Crusoe TM5800 processor. Ostensibly, this processor alters its internal routines to adapt the way it runs operating systems and software. This "code morphing" is supposed to result in gradual improvements in system efficiency over time. If I am in error in these statements I hope that someone with some familiarity with these chips will provide the needed correction.

This little computer works reasonably well, though it certainly won't set any speed records. But it has shown a tendency, unlike all of the other WinXP machines of my acquaintance, to suffer degradation in stability after a couple of months of operation on a fresh installation. I have been through this cycle three times since purchasing the computer in January of this year.

A typical cycle starts with a fresh OS installation / applications and utilities software installation. Everything works fine and continues to work fine for several weeks. Then, without installation of any new software or drivers, I start seeing infrequent blue screen errors. The errors gradually increase in frequency until I get annoyed and reformat the the drive to start over.

Now, we all know that there is some new software (in the way of Windows updates and firewall and anti-virus updates) being added to the system. But it is hard to point the finger at any of these changes as being the source of the errors because I have never seen a blue screen immediately after or closely following any of these changes in the system. The errors are more likely to occur when the system is being heavily "exercised", as in burning CDs while using the mail client, a browser, a spreadsheet, and a text editor. (Hey, it ain't no Pentium 4.)

Now I know you're saying to yourself that the dope hasn't said what blue screen errors he's been seeing. There's the rub. The messages are ALWAYS different. AFAIK not one repeated error code in the whole history. The only thing the messages have in common is that, when the system starts up again, I see the popup that tells me that the system has recovered from a serious error. The link to the Microsoft error analyzer says that it can't tell me what is at fault, but that it is a driver. Suggestion is to remove anti-virus software and to disable AV and other special features in the BIOS.

What I want to know is this -- how can the system behave perfectly for a couple of months and then suddenly start going wacko like this? I have tried removing individual items and using different AV and software firewalls after fresh installations to see if the behavior is altered. It is not.

Is it possible that code morphing is doing this? I'm running the same OS, apps, and utilities on several other Pentium-based systems with no similar issues whatsoever. If the behavior was consistent I'd be thinking I had a memory (silicon or disk) problem, but this thing runs flawlessly for weeks after a new isntallation and then slowly starts going toes up.

Now for the icing on the cake. I met a sysadmin from New York on a flight the other day. He has the same computer, and he has had exactly the same problems with his. He doesn't use the same apps as me, but his anti-virus is NOD32 (which is what I was using on the immediately previous installation). Other than that his system is configured pretty differently from mine. But he is running Windows 2000 Pro.

System design flaw? Coincidence? Bad karma?

- prosaic

 

[CTho9305]

Interesting. I would think that software shouldn't be able to tell that it isn't running on a native x86 processor - properly done code morphing should produce identical output to natively running the code. I guess it's possible that the transmeta guys are optimizing too agressively and breaking something, but it seems like they would have done thorough verification of their designs (like every other processor manufacturer).

Out of curiosity - does the code morphing software "remember" what it learns between boots? If so, does it store information somewhere on the hard drive? If so, wouldn't that require a transmeta-aware OS? If not, why would the problem take time to develop?

edit: Does transmeta, or your laptop manufacturer offer support?

 

[Peter]

Code morphing forgets what it learnt across a poweroff cycle. That's not it, certainly.
Drivers that rely on a certain kind of execution speed, or even the being constant of the execution speed, should all be burnt and buried by now, but every once in a while one turns up.
More likely, you got a bad component in there that corrupts your OS installation gradually.
Prime suspect: The RAM. www.memtest86.com

 

[prosaic]

Thanks for the replies!

CTho9305 --

I thought that code morphing (at least for the Transmeta processors) did remember across power off cycles, but Peter says not so. It seemed to me that Transmeta was touting this technology as a performance improver. That doesn't seem to make sense if they mean that the optimization occurs only within one operating session only to have the gains wiped out on reboot. Well, I guess maybe that's market-speak for you.

In matters of technical support Transmeta refers users to the vendors who integrate the processor into systems. Unfortunately this notebook isn't marketed here by Toshiba. I bought it through Dynamism.com, and their resources for dealing with an issue like this seem considerably more limited than those that Toshiba might be able to bring to bear on the subject.

Peter --

Thank you for the information on that limitation of code-morphing. Hardly seems worth the effort. I can tell you for certain that this puppy doesn't set speed records.

Of course you are right about the RAM. I guess if it was a virtual memory issue the hard drive would have failed or at least worsened considerably by now. I don't think I can run memtest on the thing. No floppy drive. If I recall, I have to boot from a floppy to use memtest, right? I guess I have to locate a memory tester somewhere, or wait until I get back home.

----

Thanks again to both of you.

- prosaic

 

[CTho9305]

Originally posted by: prosaic
Thanks for the replies!

CTho9305 --

I thought that code morphing (at least for the Transmeta processors) did remember across power off cycles, but Peter says not so. It seemed to me that Transmeta was touting this technology as a performance improver. That doesn't seem to make sense if they mean that the optimization occurs only within one operating session only to have the gains wiped out on reboot. Well, I guess maybe that's market-speak for you.

In matters of technical support Transmeta refers users to the vendors who integrate the processor into systems. Unfortunately this notebook isn't marketed here by Toshiba. I bought it through Dynamism.com, and their resources for dealing with an issue like this seem considerably more limited than those that Toshiba might be able to bring to bear on the subject.

Peter --

Thank you for the information on that limitation of code-morphing. Hardly seems worth the effort. I can tell you for certain that this puppy doesn't set speed records.

Of course you are right about the RAM. I guess if it was a virtual memory issue the hard drive would have failed or at least worsened considerably by now. I don't think I can run memtest on the thing. No floppy drive. If I recall, I have to boot from a floppy to use memtest, right? I guess I have to locate a memory tester somewhere, or wait until I get back home.

----

Thanks again to both of you.

- prosaic

Nope, it can be burned to a CD. There's an ISO available on the website.

 

[Peter]

You can make a bootable CD w/ memtest86 on, right. You can also boot it off a USB floppy drive.

Code morphing optimizes itself for the software you're currently running. That happens quickly enough to eliminate the need to store results across a power cycle.

That's also not why you perceive the machine as slow - it's because it was made to provide reasonable speed for the intended task, while maximizing on battery life. That's an entirely different target than making a fast machine with a just-so-bareable battery life.

 

[prosaic]

Ah, yes. That will teach me to post a reply without first checking the link. I went there immediately after replying, found the CD image download, burned it, and used it. After 1 full pass there were no errors reported. I had to interrupt the testing in order to get on with my business today. I thought I would post back to let you know. I intend to let memtest86 run overnight to see if anything shows up. I suspect that, if there is a problem with memory, it will be difficult to detect.

A couple of years ago we had a workstation with some faulty memory that didn't show even during an extended run with memtest. The memory did show up as being bad when it was plugged into a memory tester. I would think (or hope) that such exceptions would be rare.

Thanks to both of you for your help. I'll report my findings, whichever way it turns out.

- prosaic

 

[prosaic]

Well, memtest86 is up to the 8th pass with no errors reported. I'll let it run overnight.

If this and, possibly, a subsequent check of the memory in a tester show no fault in the memory I'm going to consider trying to open a dialog with someone at Transmeta. My new-found friend from the red-eye flight the other night and I are going to pursue this until we find out what's happening.

Thanks again for your help. I was hoping that the problem was going to turn out to be memory, and even in the face of negative error reports I'm still willing to believe that. I guess that I could try replacing the memory as an experiment. Anyone know of a better supplier of these modules than top-of-the-line Mushkin? I don't want to flounder too much on this as it takes 2-3 months for adverse effects to make themselves known as it is!

- prosaic

 

[Peter]

To my own experience from mainboard prototyping, I must say if memtest86 says it's OK then you're 99.9 percent safe that your RAM actually is OK.

 

[aka1nas]

Also run the extended memtest tests and not just the standard ones if you are really paranoid about the RAM being bad.

 

[prosaic]

Thanks for your help. Yeah, I know that the memory is probably okay, but I won't be able to lay this to rest (or, likely, get much response from Transmeta without being armed with some solid data) without pursuing this as far as it can be pursued.

You've been very patient. As I said, my real question for this forum has already been answered by Peter. I don't wish to foster an ongoing troubleshooting thread in the Highly Technical forum. I really just wanted to know if it was possible that the code-morphing process itself could be responsible for this behavior. I think that, even though it goes back to scratch between sessions, it's still possible that the code morphing might combine with the WinXP optimization routines to produce some undesirable effects. This time around I have noticed a change over time in the order of loading of some services. Interesting, no? I suggested this to Ralph, the sysadmin I met on the plane, by phone today, and he said that he noticed a change in launch order over time on his Windows 2000 Pro system without having made any changes in its configuration other than normal updates. I'm wondering if it could be significant that both of us use Diskeeper.

But I must do everything in my power to prove that I have no hardware issues and to prove that I don't have any errant drivers (including pseudo-drivers loaded by AV and firewall software) that are causing the ruckus. The problem with proving the latter is that I have to operate the system for two or three months without protection. Makes it kind of useless for my daily routine in which I love this little thing for its portability and its unobtrusiveness in meetings. I guess I'll be hauling the Tecra around for a while.

Thanks again, everyone, for the information and suggestions!

- prosaic

 

[sparkyclarky]

Stress test it with Prime 95 for 24 hours too, just to make sure it's not some sort of random heat buildup.