Page 2 of 2

Posted: Mon Jul 31, 2006 8:34 pm
by l0new0lf
Don the pie bench for 1m finished in 4m 59s . I dont know if this is good or bad but im guessing this is mostly dependent on your cpu speed and I suppose fsb plays a role as well but I could be wrong. Anyone out there with a 600 Mhz k6 want to give it a go?

everest 2.20.405 results:


CPU CPU Clock Motherboard Chipset Memory Read Speed
K6-III+ 600 MHz Tekram P5M4-M+ MVP4 Int. 341 MB/s

CPU CPU Clock Motherboard Chipset Memory Write Speed
K6-III+ 600 MHz Tekram P5M4-M+ MVP4 Int. 156 MB/s

CPU CPU Clock Motherboard Chipset CL-RCD-RP-RAS Latency
K6-III+ 600 MHz Tekram P5M4-M+ MVP4 Int. 203.4 ns

Hey now you got me curious. :) Ill be happy to upload my version of everest 2.80.534 so you can get a more complete analysis. Do you have msn messenger or yahoo?

Posted: Mon Jul 31, 2006 11:24 pm
by Uranium235
lonewolf, your results are very impressive. This is from a new build of mine, a Soyo SY-5EMA+ V1.1 with a K6-3+ 400ATZ @600 (6x100) and 256MB RAM (2,2,2,5). I haven't used any tweaks on this machine, hence the results.

Everest v2.20.405 Benchmarks
Memory Read 258 MB/s
Memory Write 120 MB/s
Memory Latency 262.6 ns

Hot CPU Tester Pro 4
Total Score 1104
MetaMark to Mhz ratio 1.84

Super PI
[1M] 320 seconds

Posted: Wed Aug 02, 2006 12:19 am
by l0new0lf
sounds like a sweet system uranium. My other system which is quite a bit quicker than this one but hardly used :) has a soyo mobo (Socket A) and it's not nearly as stable as my k6 system. Anyhow when you get a chance run the benches again once u get the system tweaked. Im guessing from your numbers already youll be close if not pass me up. Hey what chipset does that soyo use anyway? just curious

Posted: Thu Aug 03, 2006 9:40 am
by Uranium235
Thanks, it's got an MVP3 chipset even though it's labeled as being an "Eteq" chipset.
I suppose you use WPCredit to tweak your board? Do you have any recommendations for what offsets may yield the best results?
I've got a socket A board also that's in my main machine but I don't mess with it anymore. I still enjoy tweaking these old socket 7 boards and they still do the job. :)

Posted: Thu Aug 03, 2006 11:55 am
by l0new0lf
Uranium try these out and let me know if they help. Some obviously may already be turned on on your board. For instance the default setting on this mobo for the L3 cache was set to write-thru for some reason so I set it to write-back. And you probably already know this but be careful with 64-67, actually be careful with them all :). Make sure you have a way to restore your computer just in case.

Device 0

53 f0
64 16
65 16
66 16
68 45
6c 28 <---gives a nice boost in memory bandwith
50 88
52 86
6d 5f
71 98
70 dc

device 7

41/09
42/09

Posted: Thu Aug 03, 2006 8:06 pm
by Jim
You got a gain by setting it to "Writeback" on an MVP3 chipset? Usually that gives you about a 50% hit on your memory performance on an MVP3 board. Halves your cacheable ram.

Posted: Thu Aug 03, 2006 8:21 pm
by Guest
Sorry I didnt explain it clearly. The setting which gave me a substantial increase in memory bandwith was the dram start cycle (6c 28) . Try it with either sandra or sysid.

Posted: Thu Aug 03, 2006 8:22 pm
by l0new0lf
rather 6c/28

Posted: Thu Aug 03, 2006 8:43 pm
by l0new0lf
one other correction... I was using a 1 ns wait state for the "read latch" for stability since my fsb is 120. You could try no wait states and in theory it should be quicker ala 6D/1F instead of 6D/5f. And I dont know whether all of you noticed or not but i am using an MVP4 chipset and although they are similar there are subtle differences and what may work on one system/board may not work on another

Posted: Sun Aug 06, 2006 12:14 am
by Uranium235
Thanks for the pointers. :) I finally got around to applying some tweaks with some good results. There did appear to be some subtle differences between the MVP3 and MVP4 PCR files but they do share many similar settings.
Here's the offsets that worked for my board:

DEVICE 0
53/F0
68/45
6D/1F
6C/48
This did give a good boost. 28 made my machine crash but 48 [MD-to-HD pop set to normal] did the trick for me.
70/FD
71/DF
74/CO


DEVICE 1
40/F3
41/6E


I already had loaded George E. Breese's Memory Interleave Enabler so it wasn't necessary to change those associated offsets.
The MVP3 PCR file I had loaded did not show a Device 7. What settings were these for?
>device 7
>41/09
>42/09


Does anyone see anything else I may have missed?

Here's the result's after the tweaks I made:
Everest v2.20.405 Benchmarks
Memory Read 282 MB/s
Memory Write 132 MB/s
Memory Latency 230.2 ns

Hot CPU Tester Pro 4
Total Score: 1129
MetaMark to Mhz ratio: 1.88

Super PI
[1M] 311 seconds

Here's the earlier results again for comparison:
Everest v2.20.405 Benchmarks
Memory Read 258 MB/s
Memory Write 120 MB/s
Memory Latency 262.6 ns

Hot CPU Tester Pro 4
Total Score 1104
MetaMark to Mhz ratio 1.84

Super PI
[1M] 320 seconds

I can't seem to squeeze anything else from this board at this point. I'm going to bump the FSB to 112 and see what results I get later on.

Posted: Sun Aug 06, 2006 8:56 am
by Jim
If you check out a post titled "WPCredit as applied to a DFI K6BV3+/66" in the software and tweaking section you will find the tweaks I used to get mine up to : Everest Memory Read = 303 MB\s; Memory Write = 145 MB\s; Memory Latency = 213.5 ns @ 6x100 MHz FSB.

Not sure which version of Everest I used for the tests, would have been the latest available at that time. Note also that WIN-XP will give better results than WIN98-SE. -- About 10 or 12 higher Memory Read.