The official Old School is Best School Round 5 thread.

[trodas]

You are welcome. I stumbled upon this error in SuperPi, when I tried these cache off runs and I could not believe that thousads overclockers use that SuperPi program, yet I have to the the one to spot the bug there

http://www.vogons.org/viewtopic.php?f=46&t=44726

 

Back to the work: there is the promised screen from the too fast run - 200MHz Duron w/o caches is still way too fast:

 

 

Now bad and good news. Bad: the re-run with anoter SuperPi in high priority ended up bad, because my failure. SuperPi use program directory for cache files and when I run two instanced from SAME location, the cache files go mixed. The 1M SuperPi reported error in calculation. The machine was THAT slow, it does not even recognized the PrtScr press (repeatedly done), so I have nothing to show. Sadly. 3.5h wait just for preparation is IMHO quite slow and in the 32M test, I already saw 1st loop - 16h...!!! )

 

However the good news are, that I'm aware of the error and now I run two SuperPi instanced from two different directories (even drives), so it should be good now. Running from now... sorry.

 

As even more good news, the WinNT does not object to running at 10.7MHz, so I run the SuperPi 1M on the 10.7MHz Pentium 1 now. If it could be any reference, the 16k run took 1min and 24sec (screenshot will come later, at 10.7MHz the floppy and even CDROM does not work, dunno why :-) ). Also dunno, what the 1M run will took. These WinNT are optimized reasonably well, so... maybe it will be too fast and it will need some slow-down too? Dunno. We see.

(the 10.7MHz Pentium feels faster that the slowed-down Duron, lol ... cache rules! (yes, the Pentium 1 have both caches enabled and 512k L2 is not all that all )

[trodas]

More bad news: Pentium 1 @ 10.7MHz is still too fast, believe it or not. 3h 15min for SuperPi 1M:

 

 

That suxx. I need something to slow down. What could run well on P1 & WinNT? Suggestions welcome.

 

There is the 16k screen promised:

 

(and no, the time is running correctly, I checked)

 

...

 

I thought that I could try disable the caches on the 10.7MHz Pentium, but that did not went well. When the graphic card resolution & driver is inicializig, it trash display and crash.

With disabling just L2 it works, but that is not slowing down things significantly, so there is no way to get even close to 24h run on SuperPi 1M w/o some sort of slow-down app

 

...

 

On the other hand, the Duron 600 is going strong 3h 50min for preparation on the SuperPi 32M "dummy load" run, no info in the 1M run, not even the preparation yet And I reduced the priority of the load run to just "Above average" ...

[trodas]

Pentium 1, 10.7MHz w/o L2 cahe - SuperPi 1M 3h 43min:

 

 

Too fast. I need slow down... The Duron slow-down is loking good, but it did not register the PrtScr as of it yet

[trodas]

Report on progress:

 

When I attempted to save the PrtScr from the running Duron, the machine crashed. That is something this machine like to do, when run caches off That deter me a lot from more Duron testings for now. I hate when machine is stable to do SuperPi 32M run or almost month run Aquamark 3 test, but not stable to survive pasting the copy of Win screen into MSpaint

 

...

 

So meanwhile, while I look out for the lanbonden suggested CPU-slow-down program (thanks!) I did the tests Dead Things suggested - comparing the SuperPi 1.5 and 1.6 in terms of speed. Since I was managed to sucesfully attach an SSD to the Asus TXP4-X, I was wonder, even w/o having the DMA on, if this make the machine faster for SuperPi:

 

 

After all, 14h 3min 27.128sec was my record: http://hwbot.org/submission/2978646_

 

Now I get 13h 52min 18.820sec for v1.5: http://hwbot.org/submission/3009521_

And then I get 13h 52min 4.296sec for v1.6: http://hwbot.org/submission/3009526_

 

 

So, v1.6 is 12sec faster on Pentium 1, when the test took almost 14h. I would say that this is minimal difference and it might be the other way around when I re-run these tests.

Both started after userinit ended, realtime priority, closed Explorer and after couple or 16k runs, till I get fastest 7.020sec time for the 16k run. It took me a plenty of runs with the v1.6, yet only 2 runs on v1.5 to get "there". Then 32M run. No waza ram optimizing, because 1) I did not yet learned to do it manually 2) WinNT does not support the tools that do it 3) I hear that it do no help on WinNT anyway.

 

WinNT optimized as much, as I can and mainly - except a reboot, there is no difference for both runs. Sure, 12sec might look plenty, but not on ~14h run. I'm quite happy to get under 14h anyway. That was what I have it it for me :banana:

 

Next - get DMA working for more SPEEEEEEEEEEEED!

 

...and slow-down 1M test enought to get over 24h... Okay, okay.

[ludek]

I keep my fingers crossed, keep doing this trodas!