SiS530 @133MHz (GA-5SMM/Soyo 5SSM) setup, benching, tweaking

Discussion relating to Socket 7 hardware.
DonPedro
K6'er Elite
Posts: 578
Joined: Wed Jul 27, 2005 2:11 pm

Post by DonPedro »

informed soyo about the 80gb issue, maybe they have an idea ...

after having finished installing win98se (netcard still not available) I would like to let you know the first bench numbers I get without ANY tweaks at all:

Soyo 5SSM
512 ram @ 100mhz, 3-3-3-7
K6-3-450
onboard video (unified memory architecture)
onboard sound (ess-solo 1)

everest:
======
read 136
write 72
latency 472,1

I am the biggest guy on the block !! :)

I will continue my quest in some hours, time for a sleep.
User avatar
Stedman5040
Veteran K6'er
Posts: 271
Joined: Mon Sep 05, 2005 4:22 pm

Post by Stedman5040 »

@Donpedro

I did some work on a pcr file for the SIS530 chipset. Its not 100% complete but a lot of the relevant features are there. Maybe you could refine it a bit. Are you interested? if so pm me with your email address and I'll send it to you.

Stedman.
DonPedro
K6'er Elite
Posts: 578
Joined: Wed Jul 27, 2005 2:11 pm

Post by DonPedro »

the story continues ....

the netcard that did not work is a surecom 320xr. it is a card that is pci 2.2 compatible and works flawlessly in an asus p5a which featutres only pci 2.1! why the card does not work with the soyo 5ssm I don't know. it does not even produce an "unknown device" in the device manager. tools that scan the pci-bus for connected devices also can't find it, not a trace.

I replaced the card with a d-link dfe 550tx. works. but I am a little afraid that the surecom won't run. according to tests on my p5a the surecom netcard needs the lowest system resources and hence does not cut that much into memory performance as the d-link or even the highly praised 3com cards.

the bench results mirror my expectations, performance drops significantly.

below see my comparison table which I will continue to work on by step by step enhancing the system. so we will see where the greatest performance gains can be achieved.

col 1 was tested with 512mb. in order to have win98 use ALL the ram I entered the line "conservativeswapfileusage=1" in the system.ini - file under the section 386enhanced. col 2: I reduced the amount to 256mb ram to make my results readily comparable to most of the other bench-runs I have collected over time. going down to 256 ram does not eat into bench numbers here because ram usage during the benches is always well under 200 mb (so no swaping occurs). further difference apart from installing a netcard is that I downloaded and installed some updates (security) for win98 which should not have an influence on system speed, I assume.

osmark version 1beta7.
Attachments
sy5ssm 1.png
sy5ssm 1.png (15.52 KiB) Viewed 15039 times
Last edited by DonPedro on Mon Nov 13, 2006 5:05 pm, edited 1 time in total.
DonPedro
K6'er Elite
Posts: 578
Joined: Wed Jul 27, 2005 2:11 pm

Post by DonPedro »

as you can see, memory peformance drops significantly in the synthetic benches as well give lower numbers in real world benches (osmark).

interesting to note is that contrary to the overall performance degradation, superpi-1m-bench for no reason I can't even guess of runs 37 seconds faster!!!! :?:

another outstanding performance gain (and also to my surprise) is shown by the webpage-load bench which goes up about 40%!! but I have to admit that the osmark numbers in col2 are the best of 3 runs where col1 osmark-results are taken from a single run. I know this is a flaw but finally the overall negative impact of adding a device (netcard) is obvious.

EDIT: forgot to mention that the reason why I left out both the nBodyOpenGL bench and also LorenzAttractor bench is that it makes no sense to run these benches on a sis-onboard graphics device. these two benches alone would need half an hour each at least to complete. hence there is also no overall osmark-score, which will only be produced when one chooses the "official run" (i.e. best of 3 complete runs).
Last edited by DonPedro on Mon Nov 13, 2006 5:23 pm, edited 3 times in total.
Jim
K6'er Elite
Posts: 1745
Joined: Wed Jan 21, 2004 7:10 pm
Location: Toronto

Post by Jim »

In your last post on page 1 of this thread you said you were using a D Link netcard that works perfectly in all your other systems; but was not even showing up in Device Manager on the Soyo. Now you say that the netcard that didn't work was a Surecom 320xr. Got me confused. Did you try both unsuccessfully? Sometimes you have to manually use the "Add New Hardware" feature in Windoze to load a driver. This is particularly true if the driver install that you are using operates from an "inf" file as opposed to a "Setup.exe" file.

Next, this is gonna show my ignorance but that's life. Most of your test results are meaningless to me because I have no idea what the tests are testing. e.g. WTF are "Maze Threads"? (The same is true of most of the others.)

What I do find interesting is the fact that Surecom cards require less system resources to operate, as compared to others. I would like to see a group of straight comparison test results, (including Sandra & Everest), run on a system where the only thing changed from one test set to another is the netcard. That would more accurately pin down the effect on system performance of various netcards.

I had always operated on the assumption that a netcard was a netcard. Six of one half a dozen of another. Now you inform me "Not so"; that is nice to know; and kind of you to share. (performance GAINS!!)

Any other tricks up your sleeve that enable you to get the numbers you post?
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 »

first a note: I have edited my last post above to make my point as clear as possible

@jim

yes, you are right about me first complaining about the d-link netcard. thats why I corrected myself, but obviously not in an understandable manner. I will edit that first mention to make my mistake clear.

also I tried what you are suggesting, to manually invoke the "new hardware" detection routine, but that did not help. the card was plugged in but for the system the card was just not there. nada.

regarding osmark:
I have opened a new thread "benching with osmark" which should answer all questions.

regarding influence on speed of hardware devices used:
:) yes, I have exactly what you are asking for. ready for posting since months, a meticulously carried out investigation on the impact of using

- an ide-adapter / or not
- turning off / on usb
- sb16isa vs sb128pci vs hercules fortissimo II
- d-link netcard vs surecom vs 3com

to make it even more timeconsuming ;) I took the pain of sysyhos on my shoulders and enhanced this cross comparison by further pimping up each benchrun (with some exception) by a stepwise tweaking with wpcredit. each benchrun consisted - if I remember correctly - of everest, the directx-game mbtr (mercedes benz truck racing) and the opengl game quake3arena. since I wanted it "perfect" I thought it would be nice to do it twice, so I run the whole bench-complex with 550mhz and 600mhz. no joke! :)

furthermore I chose hot cpu tester for a more thorough synthetic bench (compared to everest) and ran it against some combinations of hw-on-off-combinations/550/600 and also using the stepwise tweaking by wpcredit.

I just did not have the time to make it into a post :) because it would need a lot of textual explanation and analysis to make it sound and appealing to the readers. but I promise, time is near, I did it not only for me but because it should make it to the public.
Jim
K6'er Elite
Posts: 1745
Joined: Wed Jan 21, 2004 7:10 pm
Location: Toronto

Post by Jim »

Interesting. In terms of LAN Cards, (Local Area Network), I have two different kinds of Olicom cards, (2325 & 2327 as best I remember), a Link Sys card, a D Link card, 2 kinds of 3 Coms, a Diamond card, and a Davicom card, (that only has drivers for Win98).

You have just given me a lot of unexpected work to do.

EDIT : Re the faster "Web Page Load " results, maybe that is just the result of "More system resources" being used for the task. So it becomes a sawoff. Slower overall machine w/ better internet performance, because more system resources are occupied running the LAN card, vs. Faster overall machine but w/ poorer internet browsing performance because less system resources are used by the LAN card.

Which is just another way of saying there is no such thing as a free lunch.
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 »

"Which is just another way of saying there is no such thing as a free lunch." :)

the webpage-bench is run without internet access. pages are loaded locally from within the osmark-folder.
Jim
K6'er Elite
Posts: 1745
Joined: Wed Jan 21, 2004 7:10 pm
Location: Toronto

Post by Jim »

Doesn't answer the issue. If performance varies w/ nothing but the LAN card changed, then the LAN card is the source of the variation. If a LAN card uses more system resources then by extension it should generally slow down the system. At the same time if a LAN card uses more system resources then it should perform its function better. (Unless it is a bad design). That is what your test results seem to show.

The fact that the pages are stored locally in the OSMark folder is irrelevent if they are routed through the LAN card for the purpose of testing "Web Page Loading Speed". How that would be done, I have no idea; but it would be consistant w/ a program designed to test "Web Page Loading Speed" to include the LAN card's functions within the test to come up w/ an accurate result.

Further, I can see no other reason why a sytem which is slower by virtue of having a LAN card that is a resource hog should suddenly be faster than a system which has a LAN card that is not a resource hog; but only when LAN card functions are being tested.
Last edited by Jim on Mon Nov 13, 2006 7:33 pm, edited 1 time in total.
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 »

IF they are routed through the lan ....

that is the point, the webpages are loaded from disk. so the lan-card has no job to do here. the webpageload-bench measures the speed of displaying/rendering html-files without the overhead of going through the net. this bench is not designed to measure net-speed.

so the lan-card's impact by simply being installed (AND respective drivers loaded and working) is the same here as for all other benches.

the reason why the score in col2 is a) contrary to the global trend and b) that much higher could be attributable to the fact that the col1-value is not the best out of 3 runs, possibly the lowest of imaginated 3 runs while at the same time col2 is actually the best score. so that in this "constructed" scenario the gap is that wide. I will try to rerun the bench under scenario col1 if time allows.
Jim
K6'er Elite
Posts: 1745
Joined: Wed Jan 21, 2004 7:10 pm
Location: Toronto

Post by Jim »

For best comparison, you should run the test on each LAN card w/ the same amount of RAM installed, since a varying quantity of RAM installed will cause varying results even if it is not used. I have always found more RAM is better in terms of test numbers, though I am at a loss to see why. (You would think just keeping it refreshed would be a drag on resources).

You should also run the test in the same manner. This is particularly true of OSMark where results can randomly vary quite widely, (as much as 10% in my experience - and you are postulating a 40% variation), from one test run to another, even when running the same test repeatedly in rapid succession on the same hardware w/ the same settings.

*** Test Starts (30 iterations) ***

Test, Score, Time(s)
==== ===== =======
Maze threads test, 525, 34.163
Maze threads test, 553, 32.446
Maze threads test, 539, 33.247
Maze threads test, 501, 35.772
Maze threads test, 512, 35.056
Maze threads test, 522, 34.359
Maze threads test, 537, 33.402
Maze threads test, 544, 32.960
Maze threads test, 526, 34.108
Maze threads test, 525, 34.134
Maze threads test, 555, 32.333
Maze threads test, 557, 32.211
Maze threads test, 518, 34.650
Maze threads test, 547, 32.806
Maze threads test, 525, 34.186
Maze threads test, 548, 32.736
Maze threads test, 557, 32.179
Maze threads test, 535, 33.492
Maze threads test, 528, 33.965
Maze threads test, 544, 32.964
Maze threads test, 541, 33.133
Maze threads test, 552, 32.460
Maze threads test, 539, 33.257
Maze threads test, 521, 34.453
Maze threads test, 510, 35.172
Maze threads test, 507, 35.351
Maze threads test, 518, 34.638
Maze threads test, 521, 34.446
Maze threads test, 516, 34.756
Maze threads test, 531, 33.792

**** Best Results for 30 iterations ****

Test, Score, Time(s)
==== ===== =======
Maze threads test, 557, 32.2108


*** Test Ends ***

Come down to it before I would say one thing is better than another, as a result of OSMark tests, I would run 20 to 30 iterations of each test and average the results.

Actually when I get Superpuppy 3 back up, one thing I intend to do is run 100 iteration runs of some OSMark tests to see if the variation in results gets even higher. If it does then I will run 200 iteration runs, and so on till I find the results stay within a fixed percentage spread.

All of which leaves unanswered the question of whether a LAN card that uses more system resources performs its function better as a result, or whether it is just a case of bad design consuming more resources than necessary.
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

the lan-card question the way you ask it is none that osmark answers. it does not try or even pretend to do.

the idea behind my approach to add a lan-card (or any other additional hw) to the system is that any additional hw needs also drivers to be loaded to make the hw work. these drivers must be serviced by the operating system in a timely manner, they must be given time to do their thing, otherwise the system would freeze, become unstable, unresponsive etc and would eventually finally crash. this sharing of resources between drivers, services and programs IS what a multitasking os is about. its task is - beside many others - to give to each of the running programs (drivers, services, applications) the appropriate amount of cpu-time. the more important the "program" is the more time it gets do its thing. the most important ones are the system services and drivers, then come the user's applications. so the system decides WHEN, HOW OFTEN, and for HOW LONG it interrupts a lower class "program" to service the needs of the underlying vitally important system. AND THIS is the reason why a benchmark program like osmark - which runs in "normal importance" context - is interrupted by the higher class "programs" beside the interruption because of concurring other normal programs. the result is that you get DIFFERENT scores for each run because of the different amount of system ressources that are used by all the other tasks and processes running at the same time in the background. nor is it the same this tasks are doing! this environment and what it does in the background is rarely exact the same for every given time span. this is just normal situation of any real world application running at "normal" priority. the longer it runs the bigger might be the difference. so to get back to where I started: how much of a system impact any hw has depends on the hw itself (how clever it is designed to ease efficient use of a driver) and the implementation of the driver itself. even if the lan-card in our case here "does nothing" in terms of transferring user induced actual data over the net it still must operate at a basic level and perform some vital tasks. the worse the drivers are written the more time the cpu is wasting, the pci-bus is locked, before other programs and tasks can continue to do their job. this is what I want to show: that you can get better/worse results by choosing the "wrong" or "right" hw and its appropriate driver.

right now I can't do what I suggest you could do to see for yourself. osmark features the ability to let the benches run in normal, high, or realtime (rt) mode. you can find the checkboxes in the upper right part of the main screen. set it to high and then run the maze thread test again (10x is sufficient to see highest / lowest score). I would say that if my assumptions above are right then with increasing the priority a program is run the narrower will be the gap between highest / lowest score.

regarding " since a varying quantity of RAM installed will cause varying results even if it is not used."
I do not agree. you can probably show this behaviour by choosing some synthetic benchmark, but then you are showing not how fast the system is but you are disclosing a flaw in the benchmark itself because THIS behaviour would render the scores you get "meaningless". how fast is the system now? you can't tell! and it does not make any sense at all, because the synth.bench does nothing "useful" and if it does then how am I supposed to use this evidence? running what application with what job to carry out xx - times just to "make it happen with utmost speed"? somehow ridiculous ....
Jim
K6'er Elite
Posts: 1745
Joined: Wed Jan 21, 2004 7:10 pm
Location: Toronto

Post by Jim »

K, Peter, I just rechecked the OS Mark results that I have and found that in the case of OS Mark, lower amounts of RAM is better. This is contrary to my experience w/ both Everest and Sandra, though the variation was more or less within the "Normal 10% range of result variation" I have found in the maze threads test; and may possibly have been caused by the fact that I was using different kinds of RAM in my tests.

That however does not explain the approximately 40% variation you got in your "Web Page Loading " test results. When the machine was fitted w/ a card using less system resources, the score was 102 and when the machine was fitted w/ a card using more system resources, the score was 141. That is way and beyond the normal variation you should get from a small discrepancy in the number of test iterations run. It strongly suggests that somehow OS Mark is factoring in LAN card performance in the results; and the card using more system resources gives better results, - if only when testing things that normally would entail LAN card usage.
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 »

yes, the reason why webpageload-bench shows such a big difference is until now a mystery ....

the last days I was fighting with installing a pci-graphics card to replace the onboard solution. bad luck, I messed it that up that the utmost possible resolution / colors setting is 640x480 by 16 colors.

I decided to reinstall win98 after I found a promising hint in soyo's online faq-section.

will report back.
DasMan2

Post by DasMan2 »

Just to allow you two to reflect on some discoveries I have come across with using different NIC cards (both ISA + PCI). Intel/DFI/RealTek/3Com/Aopen/SMC/

Time to boot-up and general use with web browsers for speed, changes after you have updated your Win 98 S/E or XP O.S.'es.

Also the TCP Optimizer program at Speedguide really straightens out the registry settings to get far better response with the web browser you choose to use.

I would suggest that any preceptional evaluations derived from early tests be checked when full updates to the OS are completed including also having your internet setup tweaked.

I have two SiS530 systems that run better scores on Everest when less than 220 megs SDRAM is used , like 160 megs only or when of course a PCI video card is used for display than the onboard memory sharing video chipset.
Stedman5040's PCR file for this chipset may well improve scores again.
Post Reply