Chip testing

[Tullphan]

This may be a stupid question, but is there a way to test the entire functionality of a chip to make sure everything about it is working as it should? Making sure nothing is disabled or has gone bad? Something that wouldn't show up on OCCT or Prime95?
Thanks.

 

[Idontcare]

There is really no straightforward of doing this.

What you need is a method of getting your CPU hot (e.g. running prime95 small FFT in background, or linpack) and then run EVEREST Ultimate Edition for Instruction Latency testing in parallel.

The instruction latency test will test every instruction your processor is capable of executing (there are more than 1,000 of them) and will do so in the presence of a heated CPU if you have prime95 running in the background.

Thus while prime95 won't be checking the stability of all the instructions, you can get a feeling for whether they all work in that environment by observing whether the Everest test completes, causes BSOD or if the latency for any given instruction (you need spreadsheet here to easily check for changes) is different when testing with Prime95 versus without Prime95.

Ideally you would loop the instruction latency tester and combine a few hundred repeat runs...obviously a labor intensive task.

 

[somethingsketchy]

Asides from doing what Idontcare just recommended, there's not much you can do unless you run something like Folding@home for a few weeks straight or IntelBurnTest (I.B.T.). As far as I know of IntelBurnTest will allow you to set the number of "runs" the program will go through to thoroughly test your CPU.

Some warnings I will give:

1) Be sure you have plenty of cooling power for the CPU. I.B.T. will make your temps skyrocket if you aren't too careful. I would recommend opening a window while you perform the tests.

2) Be sure to have a lot of RAM. Depending on the test you tell I.B.T., the software will consume a lot of RAM (if you choose the most extreme option). I had 4 GB of RAM go down to 2-300MB by the time the I.B.T. was running.

3) It will take a while to run through one "step". A friend's overclocked q9650 took over 40 seconds to run just one step and about 20 minutes to complete a 20 step run. He figured if his CPU could run through that with no changes in the numbers (there are numbers to verify your CPU is running very stable), then his overclock was stable.

Other than that you will have to play around with some benchmarking software, games, general applications, etc. before you can make that determination, although it will be very apparent once something is wrong with CPU.

 

[Idontcare]

I should have linked to the following for instructions on how to get Everest to do an instruction latency dump: (its not intuitively obvious from driving thru the standard menu options)

These lists were created by Lavalys EVEREST Instruction Latency dump feature. If you do not believe in software measurements, wait for the official Intel/AMD/etc. guide and hope it will be more detailed and accurate than the current one.

You can create such dump in EVEREST by right-clicking on the bottom status bar of EVEREST main window -> CPU Debug -> Instruction Latency Dump.

It fully works on trial version, too. x64 dumps are based on an unreleased developer x64 EVEREST version.

http://instlatx64.fw.hu/

 

[magreen]

Don't forget that old tetrahedron screensaver dmens was talking about, too

 

[techmanc]

How reliable are software like Micro-Scope Diagnostic Suite in testing motherboards and CPU. A lot of the system tests run for seconds?

 

[dmens]

but is there a way to test the entire functionality of a chip to make sure everything about it is working as it should

i assume you mean functional testing? exceedingly difficult for the end user. even the quick test after sort is gathered over decades of x86 experience, which includes all the different eccentricities of x86 and each processor family, and so forth. the comprehensive silicon testing after tape-out to market often takes over a year.

as for speed test, no program can guarantee anything, even that one particular program is stable for a year. for the end-user, the best option is to get the machine hot, then guardband heavily. i would advise against a light guardband since speed outliers are often require unique scenarios.

in addition, a program which limits one processor family might be fine for another family, so run a few different programs. say something which misses the cache a lot and branches (like a compiler or something), just for completeness.

 

[Tullphan]

I read in a forum, I don't remember where, that someone had trouble with something not functioning properly in his chip. Can't remember exactly what it was.
I just want to make sure the chip is working as it should...whether at stock speeds or overclocked.