Oldschool: K6-2 faster at SETI than a K6-3??

[Jeff7]

I've heard and seen that the K6-x procs aren't the greatest at SETI, however, this makes no sense. Sandra benchmarks the K6-3 higher than the K6-2 (duh) but SETI takes at least 5hrs longer on the K6-3. Both are 400MHz. I've ran several work units with each, and the results are consistent. What gives here?

 

[Baldy18]

I've heard that the K6-3 is slower because the extra cache increases latencies and seti hates that.

 

[ViRGE]

I assume we're talking about the L3 cache latencies? I can't see the lantency on the L2 cache being higher than what the mobo-based cache is. If that's the case, wouldn't disabling the L3 cache get performance right back up there?

 

[Marrkks]

wow! a dicussion about k6's and seti, whood'a thunk it. are the mb's the same? my second question would be are the fsb's the same.?

The k6's use a "cache" (L2 or L3)on the mobo, this can range from 512K(256?) to 2MB. that in itself could make a difference. ie: see Xenon w/2mb on board cache.

please post some info on the set ups. I am very interested, I am running my seti queue and seti on a k-6/3.

 

[BadThad]

The only valid comparison would be to run the exact same wu's on both machines. I think you got a stinker(s) on the K6-3 machine. Cache is always faster than ram, especially on-die cache which the K6-3 has. It doesn't make sense that a K6-2 would be faster than a K6-3 in SETI.

 

[Migroo]

This isnt exactly technical, but from what I remember from the times - the K6-3 was 'almost' an Athlon so to speak (in the performance tests) which was why the K6-3 did not last long after the Athlon's introduction.

I would therefore err towards agreeing with badthad. There is a test WU around somewhere, supposed to represent an 'avarage'. Is it from ARS site?

For more information, visit the SETI Timer site here. Although its not 100% accurate, it gives you some idea.

Results for a 0.417 angle range WU were:
K6-2 - 24.xx hours
K6-!!! - 21.xx hours
Both running at 400mhz and running a 100 FSB (not sure if that is correct...)

Sounds about right to me

 

[Baldy18]

Originally posted by: ViRGE
I assume we're talking about the L3 cache latencies? I can't see the lantency on the L2 cache being higher than what the mobo-based cache is. If that's the case, wouldn't disabling the L3 cache get performance right back up there?

Sorry about that. Yes I did mean the L3 cache. But that is all just a guess on my part.

 

[BadThad]

The L3 cache latency would still be much less than RAM. After thinking about this I've decided the comparison of the seti results is definately invalid. There's no way ADDING cache would slow seti...impossible.

 

[Jeff7]

It's the same motherboard, same FSB, same RAM...the ONLY setting that I need to change is the voltage. I'll see if I can find that "test" WU - I've heard it mentioned a few times on this forum.

 

[Joe O]

What motherboard is it? Does the BIOS recognize the K6-III?

Here is the page with the benchmark WU,

and this is the page with reported results for various K6-2 and K6-III systems.

The times reported are 20:46 for a K6-III (DFI K6BV3+ RevA1, RAM 2-2-2.) and 25:12 for a K6-2 (DFI P5BV3+ (VIA MVP3 chipset) 1MB cache.) both running at 100MHZ FSB with a multiplier of 4 (400).

What I found interesting was that a K6-III (Tyan Trinity S1598MoBo, RAM 2-2-2, 2MB L3. at 100MHZ FSB with a multiplier of 4.5 (450) was 20:44, only 2 minutes less!

 

[Jeff7]

Yup, it recognizes it fine; it also recognizes K6-3+ processors just fine. It's an Epox MVP3C2 motherboard.

 

[Marrkks]

well to get to the bottom of this you are going to need to find out some info about the WU's. specifically the angle ratio(AR). SetiSpy is great for this and is very easy to set up if you are useing SetiDriver.

I have a k6-3+ 450(@600) mb is a Epox mvp3g5, running 100fsb, mem cas 2, 4-way interleave.

a WU with an AR of .415 to .950 takes roughly 15hr 30min.
a WU with an AR greater than 1.500 take just over 13hrs.
a WU with an AR less than .415(.390 and .402) both took almost 17hrs.(16:50 and 16:46)

I imagine a slower cpu would spread the times out even a little further. just wondering were you useing the computer when you ran these times? this can change the wu times also.

btw- that k6+ proc should be good for another 100-150mhz.

 

[Jeff7]

Ok, the results are in. I used the WU that Joe O linked to. Data is formated as, Time/WU, Floating point Operations/sec, Cycles per Floating point operation:

K6-2 400
26:36:33
41.92Mflops
10.27 CpF

K6-3 400
22:30:16
50.70Mflops
7.91CpF


So yes, the K6-3 is actually a wee bit quicker and more efficient than the K6-2.

 

[BadThad]

ahhhh.....vindicated!