New MMX-optimized App crashing on K6-2+ & CPU wrongly identified

[talaktalan]

Hello,

yesterday I did some benchmarking on my Athlon 64 3200+ with the new apps (and the older 1.x apps) and the results were just great (I used the first of the test-WUs from this site):



So I decided to give a try on an older Box with a K6-2+. I installed the MMX-optimized app, but apparently I get only erroneous results. Please see the results and the stderr out here:
http://setiathome.berkeley.edu/result.php?resultid=416750545
http://setiathome.berkeley.edu/result.php?resultid=416801342
http://setiathome.berkeley.edu/result.php?resultid=416826594

Is the MMX-optimized app compatible with 586 generation chips like the K6? Or what could be the reason for the errors?

I also noticed a wrong cpu identification in the stderr out (which can also be seen when running the CPUID 10-04-A.exe on this system). The cpu is wrongly identified as a 'AMD K7 Athlon' but it is in fact a K6-2+ (family-model 5-D-4). I will attach the CPUID 10-04-A output to this post.

Could you give me an idea what's wrong here?

Thx in advance, best regards

Alexander

[attachment deleted by admin]

[BenHer]

So many CPU types...so little time.

Will correct the ID table.  Probably should work on that standalone text file for CPU ID table which program can read in to identify newer CPUs (or older ones with mistakes).

[Pepo]

Probably should work on that standalone text file for CPU ID table which program can read in to identify newer CPUs.

But then you should maybe add some fingerprint at the end of each list (if you want to disribute updated files)? Although, those who really want to could be anyway able to change the file as wild as possible.

Peter

[BenHer]

Pepo,

If you really wanted to, you could download the boinc source, edit it, and have your PC called whatever you like. 

However, I was thinking of using the built in MD5 sum functions (already part of boinc) for keeping simple twiddlers from modifying the file.

[talaktalan]

Hello BenHer,

thx a lot for your reply.

I was in no way 'complaining' about the wrong cpu identification, only included it because it might be related to the problem with the crashing app. Of course you can not include every exotic cpu (and the K6-2+ is pretty exotic now, besides it never was mainstream) into the identification table. At least not, until someone has provided the necessary data.

Back to the problem: I have activated my K6-III system and started the MMX-optimized app v2.0 there. Surprisingly (well at least to me ) it is running without a problem so far. The workunit is not ready until now (well, you know, it's not the fastest computer nowadays ). I think this is good news, because it could mean that there is no general problem with K6 type cpus and their instruction set.

It also points to a possible reason for the failure on the K6-2+ (thx to Ananas for the hint), please see an excerpt from the stderr file on the K6-III (where the MMX-app is running currently):


  Version: Windows MMX 32-bit 'Ni!' based on seti V5.15  'Chicken Good!'
      Rev: (R-2.0|MMX|FFT:IPP_MMX|Ben-Joe)
    CPUID: 'AMD K6! K6-III 3D (Sharptooth)'
     cpus: 1 cores: 1 threads: 1   cache: L1=32K  L2=256K L3=0K
 features: mmx 3Dnow 
    speed: 400 MHz  -- read megs/sec: L1=1343, L2=974, RAM=340


And now for comparison the same excerpt from the result file of one of the failed workunits from the K6-2+ computer:


  Version: Windows MMX 32-bit 'Ni!' based on seti V5.15  'Chicken Good!'
      Rev: (R-2.0|MMX|FFT:IPP_MMX|Ben-Joe)
    CPUID: 'AMD K7 Athlon (Argon)'
     cpus: 1 cores: 1 threads: 1   cache: L1=32K  L2=128K L3=0K
 features: mmx 3Dnow 3Dnow+ 


Unhandled Exception Detected...

- Unhandled Exception Record -
Reason: Illegal Instruction (0xc000001d) at address 0x0055066E


As you can see, the application is apparently crashing when it comes to the clock measurement. Well, I'm only reading this from the file, I'm not a software developer, but maybe you will agree?

Do you rely somehow on the mainboard bios for clock measurement? Because on this mainboard the K6-2+ is not fully supported, it is a nearly 10 years old socket 7 board with Intel i430TX chipset (that's the one which was developed for the Pentium MMX mainly, Intels last socket 7 cpu).

Any possible workaround (besides not to run the optimized application )?

Best regards

Alex

[BenHer]

Good identifcation of the spot where error might be occuring.

There are a few things going on with Speed measurement..but they are all inside the program (except a few calls to the Operating system, not the BIOS).

I suspect the program is trying to see how fast memory can be read using MMX instructions, but something about MMX isn't working the way it does on other systems.  Thats what is occuring during CPU speed testing.

The oldest K6 I have access to is a K6-3D+, and it rarely even boots (laptop).  Not the place to really do serious testing.

[talaktalan]

Ok, thx. I guess I wil now wait until the K6-III is finished (to see if everything is fine and validated). I also have a SETI beta workunit running on the K6-2+ currently (with the standard app from Berkeley). Let's see what comes out there.

As a next step (in a few days) I might swap both cpus. If I underclock the K6-III a bit, it should also work on the i430TX board. The K6-2+ will go into a slightly more modern super 7 board with full support. And then I'll check what happens.

I give an update, when ready with that. That is, if you don't have any other suggestion.

Regards

Alex