VOGONS


First post, by iulianv

User metadata
Rank Member
Rank
Member

I've just noticed that I have several 400 MHz CPUs around (K6-2, K6-III, Celeron and Pentium II), so I'm thinking about comparing several setups at that speed (I have an ATX s370 board and spotted an AT one - both VIA-based - for the Celeron, an AT slot1 board - also VIA-based - for the Pentium II, two MVP3 boards for the K6-2 and K6-III and a 430TX board for the K6-2 running as 6*66MHz).

I guess Win98SE would be just about right for this, but I have no experience in serious benchmarking, so I don't know what software to use; I've downloaded several versions of Sandra (from 99 to 2002 - the newer the better?), but I'm pretty sure it's not enough for some relevant testing.

Also, probably a fresh reinstall of the OS is in order before testing each setup, so the whole thing might take a while...

Reply 2 of 10, by noshutdown

User metadata
Rank Oldbie
Rank
Oldbie
leileilol wrote:

spoiler: pentium wins

except for special 3dnow game versions coupled with a voodoo2/3

You can try MDK2, Final Reality, Q3A, 3DMark2001 should also work fine

thats no doubt...
and i have celeron-coppermine and p3-s engineer samples that can run at 4x clock. 🤣

Reply 3 of 10, by ux-3

User metadata
Rank Oldbie
Rank
Oldbie

Indeed. Depending on what you do, a P2-266 can beat a K6-2 500. Run a turbo Pascal game eith RT200 error. The p2 is fast enough to cause the error, the K6-2 is not.

Retro PC warning: The things you own end up owning you.

Reply 4 of 10, by elianda

User metadata
Rank l33t
Rank
l33t

Your interpretation is wrong, since the differences in the original code of the CRT unit in TP7 are due to the execution times on the LOOP command. Nowadays it is replaced by DEC JNZ because it does the same but executes much faster.

So what you basically see there is the difference in LOOP execution time on different CPUs, but it does not tell you about average performance. Especially because LOOP was used in a lot of early timing loops there were cores where this command was artificially slowed. For more modern software this doesn't matter since compiler use always DEC JNZ.

Retronn.de - Vintage Hardware Gallery, Drivers, Guides, Videos. Now with file search
Youtube Channel
FTP Server - Driver Archive and more
DVI2PCIe alignment and 2D image quality measurement tool

Reply 5 of 10, by ux-3

User metadata
Rank Oldbie
Rank
Oldbie

Sure you see the difference in loop times. Just look at my post again: "Depending on what you do..." And when it comes to loop counting, the p2 wins. So where am I wrong?

Retro PC warning: The things you own end up owning you.

Reply 6 of 10, by elianda

User metadata
Rank l33t
Rank
l33t

ok I reread again and I guess I interpreted a bit too much into your statement.
In the sense of that specific RT200 TP behaviour with unexpected result, this is right.

I just wanted to add that this behaviour is due to the cycle count of the LOOP instruction used in this specific loop (crt unit initialization). So it seems that the LOOP instruction takes a lot more time on a K6-2 than on a P2, but since it is never used in newer compiler this tells nothing about real world performance you probably want to get from a benchmark. If you would do the same loop with DEC CX, JNZ target the result would shift.
So you are right, results can strongly depend on what you do.

Retronn.de - Vintage Hardware Gallery, Drivers, Guides, Videos. Now with file search
Youtube Channel
FTP Server - Driver Archive and more
DVI2PCIe alignment and 2D image quality measurement tool

Reply 8 of 10, by SarahWalker

User metadata
Rank Member
Rank
Member

LOOP is definitely fast on early K6-2s (single cycle), much quicker than on P2. That's why the K6 has issues with Windows 95a >= 300MHz. Maybe they slowed it down for the CXT core?

Reply 10 of 10, by ux-3

User metadata
Rank Oldbie
Rank
Oldbie
leileilol wrote:

is 500mhz early? I couldn't run One Must Fall on that one.

No, they were the among the latest. They had already relabled the lower multipliers for high values. I have a K6-2 300 AFR, which was one of the fastest K6-2 CPUs which would still clock at lower speeds.

I can't drag myself to try if it does RT200 though...

Retro PC warning: The things you own end up owning you.