K6-3+ 450ACZ under / over clocking results.

Jim · Post by **Jim** » Wed Nov 23, 2005 5:06 pm

@ Stedman : Please do. @ KachiWachi : As far as the PCR Files are concerned, I got mine from the K6Plus downloads section; and assume he did the same. In which case they should be the same.

Guest · Post by **Guest** » Mon Nov 28, 2005 4:58 pm

-Jim,

Sorry for the delay I've been a bit busy. Here is my register dump booting up at 550 and using powertweak and ctu. Again no other tweaks. The cache is on and the page count disabled.

---------------------------------------------------------
WPCREDIT Version 1.4 / Copyright (c) 1998-2004 By H.Oda!
----------------------------------------------------------
PCI Bus : 0 Vendor ID: 10B9 ALi Aladdin V
Device : 0 Device ID: 1541 Host Bridge
Function: 0 Revision : 04 Host Bridge
----------------------------------------------------------
00 01 02 03 04 05 06 07 08 09 0A 0B 0C 0D 0E 0F
+ -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
00 | B9 10 41 15 06 00 10 24 04 00 00 06 00 40 00 00
10 | 00 00 00 D8 00 00 00 00 00 00 00 00 00 00 00 00
20 | 00 00 00 00 00 00 00 00 00 00 00 00 B9 10 41 15
30 | 00 00 00 00 B0 00 00 00 00 00 00 00 00 00 00 00
40 | 13 04 C1 7D 00 00 00 06 1D E7 A0 5C 00 00 FF FF
50 | 00 F0 00 CC 00 05 07 FF 00 00 00 00 00 00 08 00
60 | 3F 70 7F 70 BF 70 FF 70 FF 00 FF 00 FF 00 FF 00
70 | 00 0F C6 02 00 00 00 00 00 00 00 00 00 00 00 00
80 | 00 00 00 00 EC F4 69 1F EA 20 20 00 00 4B 42 32
90 | 24 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
A0 | 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
B0 | 02 00 10 00 03 02 00 1C 02 03 00 07 09 00 00 00
C0 | 90 80 FD DF 00 00 00 00 BF 4A 00 00 00 00 00 08
D0 | 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
E0 | 01 00 01 00 00 00 00 00 30 EC 01 80 00 11 00 29
F0 | 00 00 00 08 00 88 96 03 00 00 00 00 00 00 00 00
----------------------------------------------------------

Copy of my register settings with P5A1.04 board with K6-2+/550. Running at 550 (100fsb)the following readings were given by Everest. Sorry about the messy post but I guess you'll sort it out.

Mem read 338MB/s
Mem write 172MB/s
Mem Latency 199.2ns

Everest reports my chipset as version 4 and interestingly also reports the north and south bridges as a ULI/ALI chips. Chipset manufacturer is reported as ULI.

As you can see there is a marked similarity with the register settings of your earlier post, but with the odd bit here and there altered.

Stedman

Jim · Post by **Jim** » Mon Nov 28, 2005 7:49 pm

@ Stedman: Judging by your Everest results, you are getting much better numbers than I am when the difference in speed is taken into account. Also I edited your "Settings Post" then stacked it up against mine; and found them to be radically different. To some extent this must be caused by the fact that you are running 1 stick of RAM as opposed to 3. I also noted that you say your chipset is identified as the ULi chipset by Everest. Mine is identified as the ALi chipset by Everest. Right now my ASUS Machine is down, but when I bring it back up, I will be checking your settings against what I have been using. Also checking to see if there are any ways in which the ULi chipset differs from the ALi chipset in terms of what is where in the settings.

Edit: Also noticed that your numbers have shot up quite a bit. Was that Power Tweak's doing ?

KachiWachi · Post by **KachiWachi** » Mon Nov 28, 2005 10:53 pm

"12/2002 - ULi Electronics Inc was established as an independent subsidiary of ALi Corporation that focuses in designing core logic, multimedia, and host-based peripheral chipsets."

ULi Home -> http://www.uli.com.tw/eng/index_eng.php

M1541 -> http://www.uli.com.tw/eng/products/prod ... ?pserno=71

Jim · Post by **Jim** » Mon Nov 28, 2005 11:57 pm

That may mean something to you; but to me the question of whether the ULi chipset differs in any substantiative way from the ALi version remains unanswered. It could be they have different settings and capabilities, or they could have the same settings; but in different places; and finally they could be identical but for their names (IDs). Any of these will cause the problem that I have been having w/ WPCredit, but only in the third case is it safe to ignore the warning.

Stedman5040 · Post by **Stedman5040** » Tue Nov 29, 2005 2:17 am

Jim,

I am running two sticks of ram both 128Mb Hynix pc133 cl2. Looking at the two sets of data most of the settings that are different tend to be 'reserved settings'. Most of these settings show a difference too between my 1.06 board and my 1.04 board. For instance as mentioned before some of them are for enabling internal tag ram etc.

The settings that can be changed show some data that are different, but these are mainly from toggling one bit of data within the register.

I enabled the fast ram read setting again with wpcredit and my memory read went up to 355MB/s but memory write went down to 152MB/s and latency reduced to 182.3ns. Sandra gave results of 184/182 on memory bandwidth benchmark.

Putting the fast ram setting back to disabled in wpcredit did not recover the memory write figure to its previous high of 172. It stayed down at 152MB/s.

I'll try again to see if if sets up fast again at boot and also try increasing fsb to 105 and 110 ifthe cpu will take it.

Stedman.

DonPedro · Post by **DonPedro** » Tue Nov 29, 2005 6:20 am

@jim:

some years back ALi and UMC established ULi, a joint venture. each of the founding firms transferred some business units to ULi. ALi brought in their chipset-unit. the days of manufacturing ali-5 chipsets were long over, but from the day of the merger, the new company associated with it was ULi. so it is just a matter of changing the name. the product in question (1541c for example) was not altered in any way by ULi. the production lines did not exist anymore.

KachiWachi · Post by **KachiWachi** » Tue Nov 29, 2005 6:21 am

@Jim - WPCREDIT will still dump the available registers correctly. The only thing that *might* be in error is the text string descriptions of the registers...that's it.

I sent for a chipset datsheet...we'll see if I get a response.

@Stedman5040 - How are you determining the "Reserved" registers...from WPCREDIT?

@DonPedro - Thanks for the history lesson! I'm sure that is mentioned on the ULi site someplace...

DonPedro · Post by **DonPedro** » Tue Nov 29, 2005 6:44 am

just for comparison and for the people who need something to investigate, here are 2 wpcredit- screenshots of 1541 and and 5243 chipset components

they are taken from a Gigabyte GA-5AX, rev. 5.2, ALi-5 chipset, K6-2+550@600, mem-timing 2-2-4-5-8.
external cache=disabled,
host dram read command mode = bypass (it is reported for the ga-5ax this setting "equals" the setting of "enhanced page count=16T" on a p5a.

everest reports:
mem-read 342
mem-write 129
latency 217

Jim · Post by **Jim** » Tue Nov 29, 2005 4:37 pm

Very well gents, - What you are saying suggests that there is in fact no difference between the ULi & ALi chipsets other than the name. My question now becomes : Why, (according to Stedman), are so many of the "Reserved" settings changed if the two chipsets are the same? Got to go to work now will look into this in greater depth tonight. and Thanks all of you, - much appreciated.

Also @ Stedman : The higher Everest "Memory Write" numbers are probably the result of running Sandra's "Memory Benchmark" before you ran Everest. In that case, I am getting slightly better Everest numbers than you; but very slightly. See my last post of "Superpuppy 2" Everest numbers in this thread.

Edit : Found this still sitting unposted when I got home from work.

Guest · Post by **Guest** » Tue Nov 29, 2005 4:52 pm

Jim,

This is really really odd as I am getting absolutely the opposite effect from you when running the Sandra memory benchmark.

I booted the P5A at 550 as yesterday and ran powertweak. Then I ran Everest with the following scores

Memory read 337
Memory write 174
memory latency 193

I then applied a couple of tweaks with wpcredit and reran Everest and got as follows

Memory read 358
Memory write 179
Memory latency 182.6

Not bad I thought. I then went on to run Sandras memory benchmark and after a few runs got

Int MMX 192
FPU 192

which lies about halfway between the Ali5 chipset and the Ali7 chipset results. Good I thought I'll now run Everest and get those better results Jim says you should get. Whoops. As happened yesterday the memory write score reduced by a full 18 points down to 161. I couldn't get the high 179 score back again without a reboot.

Anyway now for the 105 fsb. I'll post as soon as I get the results.

Stedman.

Jim · Post by **Jim** » Tue Nov 29, 2005 5:07 pm

Stedman : Is that a P5A or a P5A-B you are using ? As for the Sandra tests affecting the Everest numbers, see page 2 of this thread. The rising Sandra numbers shown there, I think is a consequence of repeated running of the test, - which results in a higher percentage of cache hits when you have, as I do, a lot of uncached RAM at the 3rd Level cache level. Note : Prior running of the Sandra tests having an effect on the Everest numbers probably only occurs when the 3rd level cache is enabled.

Guest · Post by **Guest** » Tue Nov 29, 2005 5:26 pm

Jim,

This board is a P5A (ATX) ver1.04 at the moment with a K6-2+/550.

Just upped fsb to 105 and using no tweaks after firing up at 577 and using powertweak I get as follows.

Memory read 351
Memory write 132

Memory latency 190

But now after running Sandra I get

Memory read 352
Memory write 157

Memory latency184.5

The write figure however is still worse than recorded at 100fsb.

Stedman

Jim · Post by **Jim** » Tue Nov 29, 2005 5:48 pm

You have to keep track of exactly what you are doing. Cache enabled ? Bios settings ? Which Power Tweak settings are you using ? etc. - Or it doesn't mean anything. Have to record every detail, cause even you, much less others, won't know how you achieved a given set of numbers a week later. You
could try the PowerTweak settings that I listed in this thread somewhere near the bottom of page 1.

@ 577 MHz they gave me :
Everest :
Memory Read = 340 MB/s
Memory Write = 165 MB/s
Memory Latency = 194.7 ns

Which you will find about 1/3 of the way down page 1 of this thread. Running Sandra before Everest resulted in :
Everest :
Memory Read = 335 MB/s
Memory Write = 185 MB/s
Memory Latency = 202.1 ns

Anyway I am using a P5A-B which explains why the reserved settings are different.

DonPedro · Post by **DonPedro** » Tue Nov 29, 2005 6:09 pm

thank you everybody for your results.

here I have another screenshot. it shows the results of a P5A, "notoriously (in)famous" version 1.06

you might ask what "seti client running" means. well, in order to achieve high performance with this board given a higher than 400mhz cpu, I found out that it needs a well cooked processor. so I installed the seti-boinc client http://setiathome.berkeley.edu/ and had it run until temperature was well above 40 degrees. the client was still running in the background when everest measured memory performance.

EDIT: the bigger story behind that is that with another setting of fsb=100 and overclocked to 500 (from 450 original) (picture shows fsb=105, multiplier 4.5) I brought temperature up to 49 degrees, then I stopped the seti-client. while temperature went down I benched with everest (every 2 to 3 degrees lower). the performance of memory write started to drop when temperature reached 41 degrees. the performance for that was 340/139/202. at 41 degrees memory-write dropped to 129, at 37 degrees to 108, at 32 degrees memory -write 63 !!!!

no tweaks were used for these results (no usage of wcpredit, powertweak, k6speed, etc.)