Overclocking the BP6: What is the best setup??

Batch codes, RAM specs, BIOS settings, etc..
Post Reply
purrkur
Linux Guru
Posts: 687
Joined: Fri Dec 12, 2003 5:57 pm
Location: Sweden
Contact:

Overclocking the BP6: What is the best setup??

Post by purrkur »

If there is one reoccuring theme here at BP6.com then it is how can be squeeze more performance out of our BP6's and what setups are the fastest? I set out to answer the latter question once and for all in a test I did on two different BP6 motherboards (version 1.0 and 1.1). The processors used were a couple of 366MHz Celerons and a couple of 533MHz Celerons. I tested the 366's at two speeds (366 & 550) and the 533 got tested at three speeds (533, 576 and 600(!!) MHz). To make it real interesting I also tested my newly modified BP6 version 1.1 with a pair of Pentium III's running at 650MHz and to put creaming on the cake I also tested a system from Dell using dual 550MHz Xeon processors and an all-SCSI setup.

I have finally completed a 10 page webreport on the results that I have posted on my own website. I would of course love to hear your opinion on the piece so criticism, comments, questions and flames are all welcomed! (yes, I am wearing my asbestos underwear...). The testing was done under Linux so for you running Linux on your BP6's out there, you actually get to test your own systems and compare your results to mine because I explain exactly how I do my testing.

So what were the results?? Find out over here...

http://davidt.homeip.net/extra/bp6-bench/page1.html
2x533MHz@544MHz, 2.0V
640MB PC100 memory
Realtek RTL-8139 NIC
Maxtor 6Y080L0 80GB hdd
Debian Linux stable with 2.4.8 kernel
Billl
Posts: 211
Joined: Fri Jul 25, 2003 1:52 pm
Location: USA

Post by Billl »

Made for an interesting read Purrkur. I think you may have skewed the results a bit though by not keeping the ram the same. You might have tried switching the ram in the two baords and then rerun the tests just to be sure. But your results pretty much prove out, whats been said many times, that in an unmodified board the 366's at 550 are the true kings. Your last test which is probably the closest to anything we do real world showed that. Also one test you might have considered could have been SETI with the calibrated workunit. This would really give a true cpu test. And since it takes so long to complete a true stability test. Anytime you have to raise the FSB over 100 you have no choice but to actually OC the entire system. So then your weakest link becomes the card that won't take it. One of the reasons I've never been in favor of the 533's overclocked. It's not suprising at all that the PIII's would be the over all winners since they are a much better core. And of course the 100 MHZ FSB jump doesn't hurt either. Big disadvantage though is all the work you need to do to make them work stable. Thanks for adding more evidence to what I believed all along.


Billl
davd_bob
Confused
Posts: 1043
Joined: Fri Feb 13, 2004 2:30 am
Location: Houston, TX

Post by davd_bob »

EXCELLENT WORK !!!!! :bounce:

It looks like you did a pretty comprehensive bench and tried to give pointed then general results. I like it. A lot.

A couple points/questions...
1a) Did any of the CPU intensive tests take advantage of the multiplier ie. 5.5 vs 8.0 vs 6.5?
1b) Or the XEON or the Piii's cache size diff? Does the XEON have half speed cache?
1c) You might list the cache differences between the Mendocino, Piii and Xeon, for those who are to lazy to research it for themselfs.(like me)

2) HD results...I didn't grasp if when they were mechanical driven vs cache driven. (Maybe im just not that smart.)

3) You made the coment:
purrkur wrote: The very interesting part is that the Xeon system is a lot faster than the Celeron at the same speed, which is of course due to the fast SCSI drive it has.
I disagree that the results prove the SCSI proformance since the XEON was also part of this equation. I agree that you are probably correct...but Im just saying there are 2 main componant differences and you cant claim the improvement is resulting exclusively due to SCSI.

4) Can you track the hits you get just on searches for that COOL benchmark?

Lastly, Im limited to 800x600 graphics and the format of the pages forced me to scroll left and right for each line of text I read. There are no-doubt others with the same limit. Can you make an adjustment allowing wordwrap to screen or something?

Again, as is it is GREAT work.
David
purrkur
Linux Guru
Posts: 687
Joined: Fri Dec 12, 2003 5:57 pm
Location: Sweden
Contact:

Post by purrkur »

Only two comments so far but lots of questions and comments to respond to!
Billl wrote:I think you may have skewed the results a bit though by not keeping the ram the same.
Billl, I seriously don't think so. The tests I ran are not the type that require massive amounts of memory, nor do they utilise memory (other than for loading the process). The machine that had 256 megs of memory had about 200 megs worth of free memory during the tests. Also, when I tested the Xeon machine, I did shut off X but I did not stop other services that the machine was running, such as Apache, Mysql and other things that it runs on a day to day basis, so it was actually at a disadvantage here. Also, I would consider some of the scores that the BP6 with 256 megs of memory to be better than the one that had 512! Anything specific you had in mind?

But tell you what, I think your point is very valid. I should test one of my boards with 256 megs of memory and again with 512 megs of memory to make sure that memory isn't a restricting factor. It also gives me an idea that there should be a "page 11" where I bring up questions and comments like this one. I knew my benchmarking presentation could only get better with help from you guys!
Billl wrote:But your results pretty much prove out, whats been said many times, that in an unmodified board the 366's at 550 are the true kings.
Yep. And now we got proof!
Billl wrote:Also one test you might have considered could have been SETI with the calibrated workunit. This would really give a true cpu test. And since it takes so long to complete a true stability test.
Yeah, we have had this discussion before actually, on SETI. Because I knew little of it I did a fair amount of searching for sites using SETI as a true benchmark but I found only information saying why it should not be used! The reasons being that the different versions of SETI may work differently and also because I found SETI results that were so contradicting. I therefore came to the conclusion that SETI probably isn't a good idea. As for the openssl test, it is a true, pure CPU test and it gives reliable results which is why I like it. As for testing stability, my testing wasn't about stability this time around. If I do write about stability then I can tell you that Linux has some software that is specifically written to stress test hardware (specifically memory and cpu) and it probably does it better than SETI, but that is for another day!
Billl wrote:Big disadvantage though is all the work you need to do to make them [PIII's] work stable.
Yeah, I am experiencing that right now. I would love to sit down and tinker with my BP6 for quite some time to get everything right but I rarely have free time for extended periods in order to do something like the mods needed. Pain in the ass really.
davd_bob wrote:EXCELLENT WORK !!!!!
Thanks! Glad you liked it :)
davd_bob wrote:1a) Did any of the CPU intensive tests take advantage of the multiplier ie. 5.5 vs 8.0 vs 6.5?
Hmm. I am not sure I understand your question. The multiplier is reflected in the frequencies the CPU's were working at so I guess it does. Care to elaborate?
davd_bob wrote:1b) Or the XEON or the Piii's cache size diff? Does the XEON have half speed cache?
1c) You might list the cache differences between the Mendocino, Piii and Xeon, for those who are to lazy to research it for themselfs
Good one. It is a weakness in my documentation and I was actually going to do something about it but I have been/am home sick so once I am back at work I will collect that info and add it. IIRC then the XEON has 256 kb of cache. I don't know what speed they are running at so I need to check that.
davd_bob wrote:2) HD results...I didn't grasp if when they were mechanical driven vs cache driven.
OK. Good question there since it may not be understandable to all. On the page with the two pics showing harddrive performance, the top one is really a test of reading from the harddrive cache (write-back cache performance) while the other shows speeds when reading from the disk itself.
davd_bob wrote:I disagree that the results prove the SCSI proformance since the XEON was also part of this equation. I agree that you are probably correct...but Im just saying there are 2 main componant differences and you cant claim the improvement is resulting exclusively due to SCSI.
I think if you look at the tests again that the tests carried out, specifically the CPU tests, that the XEON processor has absolutely no advantage whatsoever, and it was on numerious times bested by a small amount by the Celeron running at 550MHz. I have been working with the high-end Intel CPU's for quite some time and if you ask me, the XEON stuff is usually marketing and nothing else. Sure, they are better at some tasks because of increased internal cache (although mine don't have that much) but not many applications use that advantage. So paying 2-3x the price of a regular Pentium III or IV to get that small advantage is plain stupid in my opinion. People should spend that money on buying more memory or faster disks. The overall gain will be much greater than buying a XEON processor.

The harddrive in the Dell is a 10000 rpm unit that has a seek time of 5.4ms . It is fast, but it is also as old! I can check when the Dell was built and post it but i would say that it is from 1999 or so. It is also connected to an ancient SCSI adapter. The Maxtor runs at 7200 rpm and has a seek time of around 9ms or so. It is a new drive, also connected to an old (but very good) IDE adapter.

I got a fair amount of experience with SCSI drives and adapters and my feeling has been that they are faster, more reliable and better adapted to multitasking and heavy loads than IDE/ATA drives (especially because of the adapters which are far more advanced than the IDE/ATA adapters). I feel that this test I did reflects what I have felt all along so I am sticking to what I wrote. Maybe someone with a BP6 and SCSI setup can verify my testing or proove me wrong?
davd_bob wrote:4) Can you track the hits you get just on searches for that COOL benchmark?

Hehehe! To a degree. I got monitoring software on my server that allows me to track the traffic on the server. I will be able to see some information on how much traffic the page generates.
davd_bob wrote:Lastly, Im limited to 800x600 graphics....
Doh! I am using a table that has a width of 800 pixels. I guess I am spoilt with my 20.1 inch TFT screen running at 1600x1200 :) I made a quick check into the polls that we did on what resolution we were using and I found that four had answered 800x600. I have therefore changed the max size down to 700 so it should fit on your screen. Please check again and let me know if it looks better.
2x533MHz@544MHz, 2.0V
640MB PC100 memory
Realtek RTL-8139 NIC
Maxtor 6Y080L0 80GB hdd
Debian Linux stable with 2.4.8 kernel
Billl
Posts: 211
Joined: Fri Jul 25, 2003 1:52 pm
Location: USA

Post by Billl »

purrkur wrote:Only two comments so far but lots of questions and comments to respond to!
Billl wrote:I think you may have skewed the results a bit though by not keeping the ram the same.
purrkur wrote:Billl, I seriously don't think so. The tests I ran are not the type that require massive amounts of memory, nor do they utilise memory (other than for loading the process). The machine that had 256 megs of memory had about 200 megs worth of free memory during the tests. Also, when I tested the Xeon machine, I did shut off X but I did not stop other services that the machine was running, such as Apache, Mysql and other things that it runs on a day to day basis, so it was actually at a disadvantage here. Also, I would consider some of the scores that the BP6 with 256 megs of memory to be better than the one that had 512! Anything specific you had in mind?
Yea it was pretty simple, just swap the ram in the two boards and run the tests again. That eliminates any chance that ram is a factor. Of course all this assumes you have the timings set exactly on both boards?
purrkur wrote:But tell you what, I think your point is very valid. I should test one of my boards with 256 megs of memory and again with 512 megs of memory to make sure that memory isn't a restricting factor. It also gives me an idea that there should be a "page 11" where I bring up questions and comments like this one. I knew my benchmarking presentation could only get better with help from you guys!
Glad to help.
Billl wrote:But your results pretty much prove out, whats been said many times, that in an unmodified board the 366's at 550 are the true kings.
purrkur wrote:Yep. And now we got proof!
Amen. :)
Billl wrote:Also one test you might have considered could have been SETI with the calibrated workunit. This would really give a true cpu test. And since it takes so long to complete a true stability test.
purrkur wrote:Yeah, we have had this discussion before actually, on SETI. Because I knew little of it I did a fair amount of searching for sites using SETI as a true benchmark but I found only information saying why it should not be used! The reasons being that the different versions of SETI may work differently and also because I found SETI results that were so contradicting. I therefore came to the conclusion that SETI probably isn't a good idea. As for the openssl test, it is a true, pure CPU test and it gives reliable results which is why I like it. As for testing stability, my testing wasn't about stability this time around. If I do write about stability then I can tell you that Linux has some software that is specifically written to stress test hardware (specifically memory and cpu) and it probably does it better than SETI, but that is for another day!
Well I would agree with you if you just run any work unit. But their is a standard work unit used for bechmarking. And assuming you used the same version of the command line client for Linux, your rusults would be very fair between CPU's. One of the advantages of SETI is it really stresses the CPU for an extended period of time and you can't artificially shorten the test. Since it's not done until it's done. That's all I was getting at.
Billl wrote:Big disadvantage though is all the work you need to do to make them [PIII's] work stable.
purrkur wrote:Yeah, I am experiencing that right now. I would love to sit down and tinker with my BP6 for quite some time to get everything right but I rarely have free time for extended periods in order to do something like the mods needed. Pain in the ass really.
Yup.



Billl
davd_bob
Confused
Posts: 1043
Joined: Fri Feb 13, 2004 2:30 am
Location: Houston, TX

Post by davd_bob »

davd_bob wrote:1a) Did any of the CPU intensive tests take advantage of the multiplier ie. 5.5 vs 8.0 vs 6.5?
purrkur wrote:Hmm. I am not sure I understand your question. The multiplier is reflected in the frequencies the CPU's were working at so I guess it does. Care to elaborate?
Im talking about the ratio of processor calculation steps to system accesses. Here is an example: a UNI-processor board testing 2 processor at 500Mhz. A 333@500 uses 5 clicks(calculation steps) between ram accesses but a 500@500 is getting 8 clicks. I think that is where SETI results go wonky.
purrkur wrote:I have changed the max size down to 700 so it should fit on your screen. Please check again and let me know if it looks better.
looks great now. THANKS :D
There are *almost* no bad BP6s. There are mostly bad caps.

No BP6s remaining
Athlon 2800
Sempron 2000
ViaCPU laptop with Vista.(Works great after bumping ram to 2Gig)
P-III 850@100
Derek
Site Admin
Posts: 2489
Joined: Tue Jul 23, 2002 3:55 pm
Location: Canada
Contact:

Post by Derek »

Phenomenal work Purrkur! Your article would be a great addition to the BP6.Com database. Can I add it with your permission?
-Derek
purrkur
Linux Guru
Posts: 687
Joined: Fri Dec 12, 2003 5:57 pm
Location: Sweden
Contact:

Post by purrkur »

davd_bob wrote:Im talking about the ratio of processor calculation steps to system accesses. Here is an example: a UNI-processor board testing 2 processor at 500Mhz. A 333@500 uses 5 clicks(calculation steps) between ram accesses but a 500@500 is getting 8 clicks. I think that is where SETI results go wonky.
OK. I think I know what you are getting at and I think the honest answer is no. You might want to consider what you are saying though. The multiplier is a very simple hardware mechanism built into the cpu and it is in no way tied to the external buses on the motherboard. I have never seen any proof that a lower multiplier should give you a more responsive system at least. Also, none of the tests I did were CPU/memory throughput related so ultimately I can't say for sure.
Derek wrote:Phenomenal work Purrkur! Your article would be a great addition to the BP6.Com database. Can I add it with your permission?
Thanks, and sure :) However, before you add it, let me gather up a bit more information first and fix the article with the pointers I have received from davd_bob and Billl. I will let you know once I feel it is complete and then you can add it to your database.

As for others, if you have any comments, questions or flames, speak up now if you feel that something should be changed, added or removed!
2x533MHz@544MHz, 2.0V
640MB PC100 memory
Realtek RTL-8139 NIC
Maxtor 6Y080L0 80GB hdd
Debian Linux stable with 2.4.8 kernel
Wolfram
Posts: 401
Joined: Tue Jul 30, 2002 3:19 am
Location: Germany

Post by Wolfram »

Great work :) Just had a quick look at the graphs, will read the whole thing when I got time. Looks like the 366 Celeron is really the best standard CPU for the BP6.

One note on the hdparm tests on page5: I think you should let the second diagramm start at zero, like you did with all other diagrams.
BP6, RU BIOS, XP SP3, ACPI, 2x366@523(1,95V), Pentalpha HS + 1x 12cm fan @5V, 768MB, Powercolor Geforce 3, RTL8139D NIC, Terratec EWS64L, Samsung M40 80GB (2,5''), LiteOn CDRW
davd_bob
Confused
Posts: 1043
Joined: Fri Feb 13, 2004 2:30 am
Location: Houston, TX

Post by davd_bob »

My uncle Yoda said, "Do! Or do not! There is no try."

Im pretty sure most of us that use this website have thought about "Trying" to do what you have "done" Purrkur.

Again, I gota say, Kuddos to purrkur.
There are *almost* no bad BP6s. There are mostly bad caps.

No BP6s remaining
Athlon 2800
Sempron 2000
ViaCPU laptop with Vista.(Works great after bumping ram to 2Gig)
P-III 850@100
purrkur
Linux Guru
Posts: 687
Joined: Fri Dec 12, 2003 5:57 pm
Location: Sweden
Contact:

Post by purrkur »

Thanks for your thumbs up guys. I still haven't found the time to set up the changes even though I got the material ready since this weekend has been eaten up by locusts. I am not sure how this week will turn out timewise for me since I am flying out to Zurich to work on Tuesday. I will be back late Thursday night. I am taking the web pages material with me and who knows, maybe I will find the time to redo the material when I am in Switzerland. However, I never know how these gigs turn out. I sometimes end up doing 16 hour workdays if I am unlucky...

I'll keep you posted once the changes are up (and you too Derek!)...
2x533MHz@544MHz, 2.0V
640MB PC100 memory
Realtek RTL-8139 NIC
Maxtor 6Y080L0 80GB hdd
Debian Linux stable with 2.4.8 kernel
purrkur
Linux Guru
Posts: 687
Joined: Fri Dec 12, 2003 5:57 pm
Location: Sweden
Contact:

Post by purrkur »

Alright. I have finally made some changes as follows:

page3: Contains the biggest change. I have added more information about the Dell system I used as a comparison. I have also added specification information in table form about the three types of processors used in the test as well specifications for the two harddrives used. Thanks davd_bob for the ideas there.

Page 5: Changed the second diagram picture to make them look the same (Thanks Wolfram).

Page 8: Changed the second picture to make it look the same as the first.

Page 10: Added an "afterthought" section to reflect some thought about points made about the amount of memory on the two BP6 boards (Thanks billl).

OK Derek, how do you want to move all of this over to your bp6.com database? :D
2x533MHz@544MHz, 2.0V
640MB PC100 memory
Realtek RTL-8139 NIC
Maxtor 6Y080L0 80GB hdd
Debian Linux stable with 2.4.8 kernel
Post Reply