HWbot wPrime bug? Athlon faster that AXP!

[trodas]

I wonder, if anyone can explain me this - wPrime 32M test on the very same hardware - a ancient MSI 6340 mobo, same ram settings, same FSB, same everything: except for a CPUs!

 

Athlon 900MHz: 182sec

 

Athlon XP 1100MHz: 228 sec

 

 

...now could anyone explain me, how 200MHz slower CPU with half the cache could be 46sec faster?! WTF!

 

http://hwbot.org/submission/2436983_trodas_wprime___32m_athlon_900mhz_%28socket%29_182sec_972ms

 

Barton core suxx so much, compared to Thunderbird A4 core or wPrime have probem with Barton CPU's, killing their score somehow?!

 

 

 

...

 

Similary weird is the MaxxMem score, even there is the difference minimal:

 

Athlon 900MHz: 17.1 Marks

 

Athlon XP 1100MHz: 16.4 Marks

 

 

 

Lack of detection from the mainboard cannot (IMHO) answer this, not to mention the slow-down is minimal, not 46sec more on 182sec score...!

Edited October 25, 2013 by trodas
fixed typo

[trodas]

wPrime 1024M

 

Athlon 900MHz: 5867.795 sec

 

Athlon XP 1100MHz: 7321.237 sec

[Turrican]

the kt133 chipset doesn't offically support barton cpus. so no wonder some boards may have problems with those cpus.

your cpu is even recognized as "unknown cpu type".

 

i would better use a prober board instead of this crappy one.

[Massman]

Moved to benchmark software

[trodas]

Turrican - no question that the board is crappy thing Yet when the CPU obviously work, then I have no idea why the results should differ so much. It just feels weird.

It is like - when the CPU is unknown, then we better slow the results down

 

I know it is probably not like that, but... it just make me think, if the VIA C3 was not so slow, just because it is "unknown CPU". (not to the present board at the time, tough)

 

Clearly I can only hope that someone can inject the AXP cpu support into the bios to get somewhat more respectable scores with AXP cpu. This is just nuts:

 

 

Also it is interesting that SuperPi do show a speed up on the AXP, unlike wPrime. Maybe it use other instructions/way to access mainboard? Less dependent on the ram access?

[ObscureParadox]

Over here the price of a NF7-S really isn't that much, about £30 if you are lucky and I don't really know what that is in your currency.

 

I'm sure that would solve your problems and then boost your scores a little more to make you more competitive

[Antinomy]

Also it is interesting that SuperPi do show a speed up on the AXP, unlike wPrime. Maybe it use other instructions/way to access mainboard? Less dependent on the ram access?

I might know the answer. wPrime greatly improves from SSE making a Thoroughbred rip a Thunderbird more than 50% IIRC. But your board might not initialize SSE on Barton (yes, SSE on A-XP was an optional feature that required enabling by BIOS).

 

You can run WCPUID to check whether SSE is enabled (CPU-Z might only show that it's supported by CPU, not actually working). And is you find WCPUID with plugins, there's a plugin to enable SSE on A-XP.

 

Try to turn it on and check again.

[trodas]

ObscureParadox - I did not want and different board, and Abit NF7-S is not actually worth that much It is not worth a damn, because it is too old... I was just playing with this mobo. That it is.

 

 

Antinomy - this SSE thing makes some sense! I would like to see that if the SSE is not used, then still the Barton beat Thunderbird, but maybe disabling it off is not good at all and... Yep! I definitively try this.

Now I battling with the Sapphire, but when I got tired of the terrible bios issues with the PI-A9RX480, then I take a look at this and report back...

[trodas]

Damn, no, that is not it. The SSE seems to be enabled by default after start:

 

 

And only one thing that is not enabled is the "System Call Extension" (what is this, anyway?) and re-enabling did not solve anything. Still same slow speed.

 

So I read there:

http://devforums.amd.com/forum/messageview.cfm?catid=36&threadid=19897&STARTPAGE=2&FTVAR_FORUMVIEWTMP=Linear

That maybe a Windows need to be reinstalled to utilize the SSE. Hmmm. Tried that now and same results. Still slow.

 

I got to 226 sec in wPrime 32M, but still a 200MHz slower Athlon with half the L2 cache do it at 182sec... Maybe some of the chipset calls between KT133 and the CPU is up to the blame? Also MaxxMem did not speed up, still 16.4 Marks, while Athlon does 17.1 ...