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

Discussion relating to Socket 7 hardware.
Jim
K6'er Elite
Posts: 1745
Joined: Wed Jan 21, 2004 7:10 pm
Location: Toronto

Post by Jim »

That is an interesting observation Peter; and one that might apply to the Rev 104 also. The reason I say that is that the closest I came to getting OS Mark to run on Superpuppy 2 was a run that got through about 2/3 of the second iteration when it crapped out; but I was getting better than previous numbers up to that point. - Hot though.

And another thought for Stedman. I remember you saying that your Rev 104 board had both the internal and external tag RAM disabled in the WPCredit settings. "Q" : Is that with the 3rd level cache enabled ? If so it shouldn't be. The external tag RAM should be enabled when the 3rd level cache is enabled; and if it isn't that will adversely affect your numbers.
Superpuppy 3
K6-3+ 450 ACZ (6x100)
DFI K6BV3+/66 Rev B2 (2 Meg) w/ 2x28mm Chipset Fans
2x256 Meg PC 133 Hynix SDRAM
1x 20G Maxtor (7200)
2x 80G Maxtor (7200) Ducted w/ 2x486 Fans Mount
52/24/52/16 LG CDR/RW/DVD
8/4/3/12/24/16/32 LG Super Multi
ATI 9000 aiw Radeon AGP
SB Audigy 1 MP3 Sound
CMD 649 IDE Controller
NEC USB 2 Card
DonPedro
K6'er Elite
Posts: 578
Joined: Wed Jul 27, 2005 2:11 pm

Post by DonPedro »

@Jim: I have "edited" my former post to shed more light onto this issue.
Jim
K6'er Elite
Posts: 1745
Joined: Wed Jan 21, 2004 7:10 pm
Location: Toronto

Post by Jim »

K, I still don't have Superpuppy-3 back up properly yet, so for the moment Superpuppy-2 is my only fully functional machine. Until SP3 is fully operational, w/ backups once again, I will not be doing any more testing. However, here are the OS Mark results that I got before it went down. Peter, you should be interested in these. They include the false results generated by running OS Mark without reinstalling it after a system crash.

The tests were run w/ varying kinds and amounts of RAM, varying FSB speeds, varying Multipliers, varying Bit Colour, from varying Drives. The colour did not seem to affect results; but everything else did.
Attachments
OS Mark.zip
(24.6 KiB) Downloaded 489 times
Superpuppy 3
K6-3+ 450 ACZ (6x100)
DFI K6BV3+/66 Rev B2 (2 Meg) w/ 2x28mm Chipset Fans
2x256 Meg PC 133 Hynix SDRAM
1x 20G Maxtor (7200)
2x 80G Maxtor (7200) Ducted w/ 2x486 Fans Mount
52/24/52/16 LG CDR/RW/DVD
8/4/3/12/24/16/32 LG Super Multi
ATI 9000 aiw Radeon AGP
SB Audigy 1 MP3 Sound
CMD 649 IDE Controller
NEC USB 2 Card
DonPedro
K6'er Elite
Posts: 578
Joined: Wed Jul 27, 2005 2:11 pm

Post by DonPedro »

hi jim,

thanks for the numbers.

I appreciate very very much the work you are doing, but I regret that the format of the numbers you collected is not very user-friendly ..... :(

some time ago I asked you if you want to join me in collecting all these bench-numbers in a more "scientific" ;) way, so that they can a) easily be identified under what circumstances they were get and b) easily compared to each other.

it is not more work to do here, in fact once the guidelines of how to gain and collect the data are agreed upon and the system is setup to do so, it is a relief and makes much more fun to run the benchmarks. you just need a tool called spreadsheet-programm, setup the headers for each column of data and there you go.

here is a sample screenshot of what it might look ..... again, this is only a sample shot (very simplified and condensed) and illustrates a comparison of systems derived from my real collection of bench-data. but I think you might get the idea ....

for the "insiders" and "experts" :) : the data shown here is taken from an asus P5A, 1.06, and if you look close you will find some very interesting and astonishing numbers! in a word: "jaw-dropping" ....
Attachments
k6-2 300 bis k6-3-400 bresults.png
k6-2 300 bis k6-3-400 bresults.png (21.7 KiB) Viewed 13694 times
User avatar
Stedman5040
Veteran K6'er
Posts: 271
Joined: Mon Sep 05, 2005 4:22 pm

Post by Stedman5040 »

Just changed my set up from the K6-3+/400ATZ back to the K6-3+/450ACZ. Could not get stable at 600 with the 400ATZ. This 450 is quite new and before putting it back online could not get it to 600 stable.

It is now up at 600 nice and stable and runs prime95 until the cows come home. I have started reducing the applied core voltage and am now at 1.9V (1.93V shows on the hardware monitor).

What is the lowest voltage anybody has run stable at 600??

Stedman.

--------------------------------------------------------------

AX59pro rev 1.32 (1Mb L2)
K6-3+/450ACZ @600 (1.9V, 100fsb) (Globalwin Cak38 cooler)
384Mb (3x128)Hynix pc133 CL2 ram 2-2-2-5
Creative Geforce2Ti 64Mb (not overclocked)
WD204BB HDD
Creative SB live 5.1
Samsung CD/RW
Jim
K6'er Elite
Posts: 1745
Joined: Wed Jan 21, 2004 7:10 pm
Location: Toronto

Post by Jim »

Guess it depends on what you mean by "stable". If you mean "Rock Solid", you probably have the record. Mine would run Everest @ 1.9; and boot into Windows @ 1.85; but I would not call it "stable" enough to generally run at either voltage. Also NOTE : Your test results in memory read and write speed, will take a hit by running it @ barely enough voltage to keep it running.

Edit : Don't recall what OS you are running; but using "Rain" (w/98SE) has helped some people get 600Mhz stable w/ ATZs.
Superpuppy 3
K6-3+ 450 ACZ (6x100)
DFI K6BV3+/66 Rev B2 (2 Meg) w/ 2x28mm Chipset Fans
2x256 Meg PC 133 Hynix SDRAM
1x 20G Maxtor (7200)
2x 80G Maxtor (7200) Ducted w/ 2x486 Fans Mount
52/24/52/16 LG CDR/RW/DVD
8/4/3/12/24/16/32 LG Super Multi
ATI 9000 aiw Radeon AGP
SB Audigy 1 MP3 Sound
CMD 649 IDE Controller
NEC USB 2 Card
User avatar
Stedman5040
Veteran K6'er
Posts: 271
Joined: Mon Sep 05, 2005 4:22 pm

Post by Stedman5040 »

Jim,

I am running Win ME and have just run through the OSMARK suite of benchmarks. Ran through in just over two hours with no signs of a crash at 1.9V.

Everest memory benchmarks are not brilliant but do not appear to change whether I am at 2.05v or at 1.9v. Memory timings are 2-2-2-5. The board has the Via chipset. Would be interested to hear about some of the tweaks that you use on wpcredit for your DFI board.

Scores at 600 (6x100) are

Memory Read 235
Memory write 111
Memory Latency 259

Quite poor compared to the ASUS P5A

Get slightly better at 550 (5.5x100)

Memory read 240
Memory write 115
Memory latency 262

Powertweak not running
Stedman
Jim
K6'er Elite
Posts: 1745
Joined: Wed Jan 21, 2004 7:10 pm
Location: Toronto

Post by Jim »

Not sure about ME. 98 PowerTweak helps, XP Don't use PowerTweak. For improvements via WPCredit, go to the sotware and tweaking forum; - see "WPCredit as applied to a DFI K6BV3+/66" Be prepared to be very bored before you have finished reading it; - but you will find how I got mine up to 312 Memory Read, 164 Memory Write, & 206 Memory Latency on my 1 Meg cache board if you read that and also "Overclocking the DFI K6BV3+" in the same forum. Best Everest results were obtained @ 5.5x103, though better results in some Sandra tests were obtained @ 6x103.

Am looking forward to seeing what the 2Meg board does for results. Right now am waiting on trying to get a pair of Thermaltake 1414 copper ram coolers. - Not readily obtained locally.
Superpuppy 3
K6-3+ 450 ACZ (6x100)
DFI K6BV3+/66 Rev B2 (2 Meg) w/ 2x28mm Chipset Fans
2x256 Meg PC 133 Hynix SDRAM
1x 20G Maxtor (7200)
2x 80G Maxtor (7200) Ducted w/ 2x486 Fans Mount
52/24/52/16 LG CDR/RW/DVD
8/4/3/12/24/16/32 LG Super Multi
ATI 9000 aiw Radeon AGP
SB Audigy 1 MP3 Sound
CMD 649 IDE Controller
NEC USB 2 Card
User avatar
Stedman5040
Veteran K6'er
Posts: 271
Joined: Mon Sep 05, 2005 4:22 pm

K6-III+/450 Undervolting @ 600

Post by Stedman5040 »

Thought I would reply to DonPedro here as it seems more appropriate.

I got this K6-III+/450ACZ late last year along with three others from the US. Thought I might sell 2 of them after picking the best of them. Anyway first of all after receiving them I installed them to make sure they were all working. No problems there. For my set up I have used an AX59PRO board that can get me all the way down to 1.35V.

Initially after the first install of the cpu I set up the voltage to 2.1V and tried upping the multiplier on the board to 5.5x and it booted into Windows fine. Tried ctu to increase the muliplier to 6x and got a lock up in Windows explorer.

Before installation of the K6-III+ I was running a K6-2+/550 @ 1.8V so I thought about reducing the voltage to see how low I could go and remain stable. The K6-III+ ran fine at 550 @ 1.8V. Ran for 2 months like this and thought about having another go at 600.

Raised voltage to 2.05V and booted up at 550. Upped the multiplier to 6 and hey presto had no stability issues. Prime95 ran for six hours without a glitch. Gradually reduced the voltage and even used 6x for booting and I am now down at 1.8v. May even try less for fun.

Seems like an anti-burn in. I suppose it is really a slow burn in.

I am now trying the second cpu and I am at the moment at 600 at 2.05v.

I'll let you know how it goes on. :wink:


Stedman
Jim
K6'er Elite
Posts: 1745
Joined: Wed Jan 21, 2004 7:10 pm
Location: Toronto

Post by Jim »

Something you guys are apparently missing is the effect of having a lot of RAM installed. Superpuppy-3 on its DFI is running 512 Meg. Superpuppy-2 on its ASUS is running 768 Meg. Both machines have a considerable amount of RAM uncached by the 3rd Level Cache. That is why I get rising results in Sandra's "Memory Bandwidth" test by rerunning it several times. Cache hits increase w/ each succeeding run up to a point. W/ Superpuppy-2 the phenomena continues longer, and the results go higher, because it has double the uncached RAM @ 3rd level that Superpuppy-3 has.

I very much doubt if I could get 241, 237 scores in Sandra's Memory Bandwidth test w/ my cache disabled. You guys should consider installing more RAM. Look at the results KGB is getting in Everest Memory Read.
Superpuppy 3
K6-3+ 450 ACZ (6x100)
DFI K6BV3+/66 Rev B2 (2 Meg) w/ 2x28mm Chipset Fans
2x256 Meg PC 133 Hynix SDRAM
1x 20G Maxtor (7200)
2x 80G Maxtor (7200) Ducted w/ 2x486 Fans Mount
52/24/52/16 LG CDR/RW/DVD
8/4/3/12/24/16/32 LG Super Multi
ATI 9000 aiw Radeon AGP
SB Audigy 1 MP3 Sound
CMD 649 IDE Controller
NEC USB 2 Card
DonPedro
K6'er Elite
Posts: 578
Joined: Wed Jul 27, 2005 2:11 pm

Post by DonPedro »

@jim:

I think you are mixing up something here. I am no expert on the technical issues with 2nd and 3rd level cache. but I do know that once you install a K6-III or K6+ cpu, ram is cached up to 4gb. so the 3rd level cache will then loose a great deal of its cachability trait. also it is questionable and therefor of interest whether what is good for the ali-chipset is good for the via-chipset? I urge you, for the better for all of us, to dig this out for the via-board owners. please run a set of bench-tools with and without cache and let us all know the numbers you get.

may I humbly propose following tools (all free and available on the internet). If you could use the versions I indicate below would be nice because I already have my collection of bench-numbers on these to compare with. if not it is no tragedy. except with osmark all these tools run together in less than 20 minutes. but osmark reveals what our tweaking is really worth!

I am really interested how enabling/disabling 3rd level cache on the via-chipset influences performance. I also would like to suggest that you then open a new thread on this issue ....

(half)synthetic benchmarks:
-everest (1.5.,2.01,2.20)
-cpumark99 1.0 (crap but ha, that will please all with 2mb cache!) http://cpumark.fw.hu/
-hotcpu tester 4.2.2 http://7byte.com/index.php?page=download
-sandra 2001-750 (found no link on that version, newer available)
-clibench 0.7.10 (found no link on that version, newer available)
-neumeier4 http://www.planet3dnow.de/files/benchma ... charc4.exe

(fake)realword benchmark:
osmark 107ba http://www.vanshardware.com/files/OSMark.zip

*******

re KGB's fantastic read-results: he does so by DISabling 3rd level cache! so your argument is beyond my apprehension :)
Jim
K6'er Elite
Posts: 1745
Joined: Wed Jan 21, 2004 7:10 pm
Location: Toronto

Post by Jim »

He also gets very poor memory write results. I am not sure exactly what to attribute his memory read results to, except that that particular board does give better than average results there usually. Additionally he has that stick of 512 Meg of RAM in there. Having a lot of RAM does push the results up. Having it all on 1 stick, I don't know the effects of; but comparing his numbers to Cake's, (no putz there), who was running the same board, (albeit w/ less RAM), @ 5x115, he is getting better numbers than Cake. So part of that may be the 5.5 Multiplier, and part of it may be that big stick of RAM. I don't think that disabling the cache gives any major performance gain if any. When I tried it on one of my machines the Everest Memory Read went up by 3 points. The thing to check I suppose is if the Sandra numbers continue to rise w/ the cache disabled the way they do w/ it enabled. If they do; and to the same degree that they do w/ the cache enabled, then my theory is wrong. But if they don't then my theory is right. TA152H had all the RAM on his machine cached @ 3rd level. And his Sandra scores stayed flat, w/ very little variation.
Superpuppy 3
K6-3+ 450 ACZ (6x100)
DFI K6BV3+/66 Rev B2 (2 Meg) w/ 2x28mm Chipset Fans
2x256 Meg PC 133 Hynix SDRAM
1x 20G Maxtor (7200)
2x 80G Maxtor (7200) Ducted w/ 2x486 Fans Mount
52/24/52/16 LG CDR/RW/DVD
8/4/3/12/24/16/32 LG Super Multi
ATI 9000 aiw Radeon AGP
SB Audigy 1 MP3 Sound
CMD 649 IDE Controller
NEC USB 2 Card
DonPedro
K6'er Elite
Posts: 578
Joined: Wed Jul 27, 2005 2:11 pm

Post by DonPedro »

@Jim
I understand that tony used 3x128mb sticks, not one 512mb.

re better sandra results with more ram by repeatedly running it: I think your concept about cached/uncached ram (above) is faulty. my understanding is that as long as an application fits completely into ram (does not have to be swapped partly to the harddisk) AND chipset/tagram/cpu-on-die cache are able to cache the installed ram then I don't see why more ram (which then actually is NOT USED at all, because the programm already fits into available ram) should improve performance.

I have made 3 screenshots of system status. the interesting part of the pics is the box with the statistics about the "Physical Memory". in the last pic it is the value of "peak working set".

the first pic shows system state before sandra (2001-750) is loaded, then when it is started (not the test), 2nd pic when mem-test is running. 3 rd pic reports sandra perf-properties specifically.

the system reports that sandra uses about 4mb without doing anything and 138mb when running the mem-test. it fitted completely into available ram.
Attachments
sandra2001-750 load during mem-test.png
sandra2001-750 load during mem-test.png (6.44 KiB) Viewed 13425 times
system load during mem-test of sandra2001-750.png
system load during mem-test of sandra2001-750.png (12.85 KiB) Viewed 13425 times
sandra before and when started.png
sandra before and when started.png (29.66 KiB) Viewed 13425 times
DonPedro
K6'er Elite
Posts: 578
Joined: Wed Jul 27, 2005 2:11 pm

Post by DonPedro »

@steadman

you mentioned that you successfully ran osmark (on the ax59pro). did you record the results and could you post it?
Jim
K6'er Elite
Posts: 1745
Joined: Wed Jan 21, 2004 7:10 pm
Location: Toronto

Post by Jim »

OK Peter,
1) I get rising Sandra Memory Bandwidth results on Both Superpuppy-2 and Superpuyppy-3
2) Superpuppy-3 Had a 1 Meg cache board & 512 Meg of RAM, of which 256 Meg is cached @ 3rd level.
Sandra Memory Bandwidth scores for Superpuppy-3 rise as follows :
INT MMX : 206; 212; 215; 216 after which they start to fall off.
Float FPU : 209; 212; 214; 214 after which they start to fall off.
3)Superpuppy-2 Has 768 Meg of RAM, of which 640 Meg is uncached @ 3rd level.
Sandra Memory Bandwidth scores for Superpuppy-2 rise as follows :
INT MMX : 210; 213; 229; 231; 238; 241 after which they start to fall off.
Float FPU : 210; 216; 228; 234; 237; 237 after which they start to fall off.
4) TA152H Has a 2Meg cache Tyan board, (capable of caching 512 Meg of RAM) on which he had 384 Meg of RAM installed.
His Sandra Memory Bandwidth results were flat. Never varied by more than 1 point.
5) KGB has a board wherein the mobo cache is defective and disabled; on which he has 512 Meg of RAM installed.
His Sandra Memory Bandwidth results are also flat. Do not rise.

Conclusion : The more RAM that is uncached at the 3rd Level, on a given board, the more the Sandra Memory Bandwidth results will rise with repeated running. The more RAM that is cached @ the 3rd level, (all else being equal, which in the case of SP2 & SP3 they are NOT), the higher the initial results will be. TA152H reached the same conclusion. If you send me a PM w/ your e-mail address, I will forward to you a number of e-mails I received from Him when we were discussing the subject. NOTE : TA152H, has spent his entire working life in the computer field; and collects computers owning over 80 of them.
Superpuppy 3
K6-3+ 450 ACZ (6x100)
DFI K6BV3+/66 Rev B2 (2 Meg) w/ 2x28mm Chipset Fans
2x256 Meg PC 133 Hynix SDRAM
1x 20G Maxtor (7200)
2x 80G Maxtor (7200) Ducted w/ 2x486 Fans Mount
52/24/52/16 LG CDR/RW/DVD
8/4/3/12/24/16/32 LG Super Multi
ATI 9000 aiw Radeon AGP
SB Audigy 1 MP3 Sound
CMD 649 IDE Controller
NEC USB 2 Card
Post Reply