SiS530 @133MHz (GA-5SMM/Soyo 5SSM) setup, benching, tweaking
*cough* *cough* http://img215.imageshack.us/img215/7788/550my2.jpg
Addendum
Everest 2.20.405 benchmarks: 418 Read, 133 Write, 207.1 Latency.
SuperPi: 256K = 101s, 1M = 537s
Sandra 2005 sr2a Mem: MMX 195 / FPU 193
Sandra 2005 sr2a CPU: Dry ALU 1340 / Whet FPU 677
I do not apologize for using the "updated" program to bench, nor retract my initial bench's. I also use it to test for rig stability. The only diparity I see is the the 67% in the Latency bench, which is very dubious and highly questionable. (Do not diminish the value of the Copy benchmark in the 3.5 version of everest either).
WinXP Pro, 110x5 K6-2, 1x 128MB Infenion PC100 C2-2-2-6, 1x128MB Generic PC100 Cas 2-2-2-7, HP CD-RW 8000+, Quantum Fireball 30lct, ATi Rage 128 GL 32MB AGP, Turtle Beach Aureal 1 PCI, IBM etherjet 10/100 PCI, AZT3011 ISA Modem. (Wpcredit tweak, write allocation enabled in bios, enhanced page mode).
Addendum
Everest 2.20.405 benchmarks: 418 Read, 133 Write, 207.1 Latency.
SuperPi: 256K = 101s, 1M = 537s
Sandra 2005 sr2a Mem: MMX 195 / FPU 193
Sandra 2005 sr2a CPU: Dry ALU 1340 / Whet FPU 677
I do not apologize for using the "updated" program to bench, nor retract my initial bench's. I also use it to test for rig stability. The only diparity I see is the the 67% in the Latency bench, which is very dubious and highly questionable. (Do not diminish the value of the Copy benchmark in the 3.5 version of everest either).
WinXP Pro, 110x5 K6-2, 1x 128MB Infenion PC100 C2-2-2-6, 1x128MB Generic PC100 Cas 2-2-2-7, HP CD-RW 8000+, Quantum Fireball 30lct, ATi Rage 128 GL 32MB AGP, Turtle Beach Aureal 1 PCI, IBM etherjet 10/100 PCI, AZT3011 ISA Modem. (Wpcredit tweak, write allocation enabled in bios, enhanced page mode).
Last edited by KGB on Thu Nov 30, 2006 9:25 am, edited 1 time in total.
-K6-2 550 Gigagyte GA-5AX (5.2) 1x512MB GeForce 2 Ti500 64MB
-Pentium III 933 Asus P3C-L 2x128MB GeForce 4 Ti4200 128MB DDR
-Pentium III 750 Asus P3B-F ATi Rage128 GL 32MB
-Celeron 1.2 Asus TUSI-M 2x128MB
-Pentium IV 1.5 DELL 2350 1x128MB
-Athlon 900 Asus K7M 1x256MB 2x128MB GeForce 3 64MB
-AthlonXP-M 2400+ Asus K7V880 2x512MB HIS Radeon HD3850 512MB DDR
-Pentium III 933 Asus P3C-L 2x128MB GeForce 4 Ti4200 128MB DDR
-Pentium III 750 Asus P3B-F ATi Rage128 GL 32MB
-Celeron 1.2 Asus TUSI-M 2x128MB
-Pentium IV 1.5 DELL 2350 1x128MB
-Athlon 900 Asus K7M 1x256MB 2x128MB GeForce 3 64MB
-AthlonXP-M 2400+ Asus K7V880 2x512MB HIS Radeon HD3850 512MB DDR
nice to read from you again!
it seems you got a cold or are you just coughing because your hide-out is a dusty place?
get yourself an aspirin and drink some hot tea over a fairly long time. or straighten your shoulders, adjust your cochones and run the whole bench-suite we use (same program's versions please) and then come back with your results. both will cure your coughing
btw, this thread is about sis530 chipset's performance, but we will be glad to see if a k6-2 without onboard cache can do some harm. for better comparison choose a speed setting that allows good comparison (fsb / multi = 450, 550).
EDIT regarding same versions: according to my tests everest bench results are the same with version 1.52.211, 2.01.347, 2.20.405 versions above show "better" results, but I stick with the old, but "free" versions because a system just does not run faster because of a change in the benchmark-program. also sticking with one version is a matter of data consistency within the collection of bench scores.
it seems you got a cold or are you just coughing because your hide-out is a dusty place?
get yourself an aspirin and drink some hot tea over a fairly long time. or straighten your shoulders, adjust your cochones and run the whole bench-suite we use (same program's versions please) and then come back with your results. both will cure your coughing
btw, this thread is about sis530 chipset's performance, but we will be glad to see if a k6-2 without onboard cache can do some harm. for better comparison choose a speed setting that allows good comparison (fsb / multi = 450, 550).
EDIT regarding same versions: according to my tests everest bench results are the same with version 1.52.211, 2.01.347, 2.20.405 versions above show "better" results, but I stick with the old, but "free" versions because a system just does not run faster because of a change in the benchmark-program. also sticking with one version is a matter of data consistency within the collection of bench scores.
- Stedman5040
- Veteran K6'er
- Posts: 271
- Joined: Mon Sep 05, 2005 4:22 pm
Peter,
Just a quick update to let you know that I got the GA-5SMM board working at 133fsb with 4x multiplier (533MHz) with one stick of Samsung pc133 CL2 ram 128Mb (actually 256mb but high density). I'll post some benchmarks when I get time to run them. Perhaps the 3.5x133 benchmark would be interesting for you as this gives 466MHz. I will, of course take it up to 600MHz as and when.
Stedman
Just a quick update to let you know that I got the GA-5SMM board working at 133fsb with 4x multiplier (533MHz) with one stick of Samsung pc133 CL2 ram 128Mb (actually 256mb but high density). I'll post some benchmarks when I get time to run them. Perhaps the 3.5x133 benchmark would be interesting for you as this gives 466MHz. I will, of course take it up to 600MHz as and when.
Stedman
stedman,
just in case it went unnoticed: on the previous page of that tread I asked you for some information regarding your setup (need that for completing the data table) and also suggested a tool that might help with your matrox mystique card.
anyway, good news that someone gets the 530 chipset running with 133 sdram speed on the ga-5smm. how you achieved this makes me wonder! how does the onboard vga survives this?
in the meantime I am not lazy. I equipped the soyo 5ssm with a k6-2+ cpu and I plan to run tests at 4.5x100, 4x112, 5.5x100, 5x112, 6x100 and 5.5x112. this will give a quite good survey on what is possible. I wish I had a gf2-mx pci card around, I think this would give a real boost.
just in case it went unnoticed: on the previous page of that tread I asked you for some information regarding your setup (need that for completing the data table) and also suggested a tool that might help with your matrox mystique card.
anyway, good news that someone gets the 530 chipset running with 133 sdram speed on the ga-5smm. how you achieved this makes me wonder! how does the onboard vga survives this?
in the meantime I am not lazy. I equipped the soyo 5ssm with a k6-2+ cpu and I plan to run tests at 4.5x100, 4x112, 5.5x100, 5x112, 6x100 and 5.5x112. this will give a quite good survey on what is possible. I wish I had a gf2-mx pci card around, I think this would give a real boost.
- Stedman5040
- Veteran K6'er
- Posts: 271
- Joined: Mon Sep 05, 2005 4:22 pm
Peter,
I will pass on the info when I get home tonight. The operating system is ME and the driver for the matrox card is the standard ME driver for the card. I will give you the Voodoo 2 driver as it is the one from the 3dfx Italian site.
The northbridge of the chipset does get a little warm. I would guess at about 40C and I was thinking of putting a small fan on it. The southbridge seems to be OK and I would say is 25 to 30C.
I did get this board running in the past at 600 with a K6-3+ for a while.
Regards,
Stedman
I will pass on the info when I get home tonight. The operating system is ME and the driver for the matrox card is the standard ME driver for the card. I will give you the Voodoo 2 driver as it is the one from the 3dfx Italian site.
The northbridge of the chipset does get a little warm. I would guess at about 40C and I was thinking of putting a small fan on it. The southbridge seems to be OK and I would say is 25 to 30C.
I did get this board running in the past at 600 with a K6-3+ for a while.
Regards,
Stedman
- Stedman5040
- Veteran K6'er
- Posts: 271
- Joined: Mon Sep 05, 2005 4:22 pm
Peter,
Here is more info. The Matrox driver is dated 5/11/1998 version 4-10-1682. The tests were done at screen res 1024x768 at 16bit high color. The voodoo2 driver is Fastvoodoo2 4.0 gold edition downloaede from 3dfxzone.it
Here are some benches to give you a taster of 133fsb. The first run was at 533 MHz and memory at 3-3-3-6 using 128mb of samsung ram cl2. External cache disabled.
everest scores with no wpcredit tweaks.
MR/MW/ML 238/139/267
Register 50 set to F0h
MR/MW/ML 261/143/243
Register 52 set to 50h
MR/MW/ML 270/144/235
register 55 set to 0B
MR/MW/ML 347/142/193
With the memory set to 2-3-2-6 with the above wpcredit tweaks
MR/MW/ML 381/144/177
With the above memory set to 2-3-2-6 and wpcredit tweaks and cpu set to 600MHz (4.5x133)
MR/MW/ML 395/153/169
For some reason the board does not like my Hynix CL2 ram. It just beeps loudly at me and refuses to boot.
I did run some Hotcpu tester pro lite 422 benches on the 533 speed and 600 speed. Results as below.
cpu speed---------------------------------533-----------------600
total score--------------------------------945----------------1065
metamark to mhz ratio------------------1.83---------------1.78
integer instructions-----------------------170----------------191
floating point instuctions-----------------63------------------73
mmx instuctions--------------------------359----------------404
3dnow instructions------------------------326---------------367
memory fill rate memcpy method-------188---------------193
memory fill rate c++ method------------183----------------187
memory rep mousd-----------------------188----------------193
memory fpu 64-bit------------------------186----------------189
memory assembly 32bit------------------174----------------180
memory fillrate assembly 64bit----------183----------------187
memory fillrate mmx 32bit---------------190----------------191
memory fillrate mmx 128 bit-------------190----------------193
insertion sort-------------------------------7-------------------8
shell sort------------------------------------254---------------274
merge sort----------------------------------444---------------511
heap sort-----------------------------------256---------------264
quick sort-----------------------------------1003-------------1070
large objects sort quick sort method-----231---------------239
fast fourier transform----------------------350--------------401
fast fourier transform optimised-----------386-------------421
Any good?
Stedman
Here is more info. The Matrox driver is dated 5/11/1998 version 4-10-1682. The tests were done at screen res 1024x768 at 16bit high color. The voodoo2 driver is Fastvoodoo2 4.0 gold edition downloaede from 3dfxzone.it
Here are some benches to give you a taster of 133fsb. The first run was at 533 MHz and memory at 3-3-3-6 using 128mb of samsung ram cl2. External cache disabled.
everest scores with no wpcredit tweaks.
MR/MW/ML 238/139/267
Register 50 set to F0h
MR/MW/ML 261/143/243
Register 52 set to 50h
MR/MW/ML 270/144/235
register 55 set to 0B
MR/MW/ML 347/142/193
With the memory set to 2-3-2-6 with the above wpcredit tweaks
MR/MW/ML 381/144/177
With the above memory set to 2-3-2-6 and wpcredit tweaks and cpu set to 600MHz (4.5x133)
MR/MW/ML 395/153/169
For some reason the board does not like my Hynix CL2 ram. It just beeps loudly at me and refuses to boot.
I did run some Hotcpu tester pro lite 422 benches on the 533 speed and 600 speed. Results as below.
cpu speed---------------------------------533-----------------600
total score--------------------------------945----------------1065
metamark to mhz ratio------------------1.83---------------1.78
integer instructions-----------------------170----------------191
floating point instuctions-----------------63------------------73
mmx instuctions--------------------------359----------------404
3dnow instructions------------------------326---------------367
memory fill rate memcpy method-------188---------------193
memory fill rate c++ method------------183----------------187
memory rep mousd-----------------------188----------------193
memory fpu 64-bit------------------------186----------------189
memory assembly 32bit------------------174----------------180
memory fillrate assembly 64bit----------183----------------187
memory fillrate mmx 32bit---------------190----------------191
memory fillrate mmx 128 bit-------------190----------------193
insertion sort-------------------------------7-------------------8
shell sort------------------------------------254---------------274
merge sort----------------------------------444---------------511
heap sort-----------------------------------256---------------264
quick sort-----------------------------------1003-------------1070
large objects sort quick sort method-----231---------------239
fast fourier transform----------------------350--------------401
fast fourier transform optimised-----------386-------------421
Any good?
Stedman
very good numbers! I think nobody before scored that high everest numbers on a sis530 board!
thank you for the information on the "ram-issue". I had the same experience when I installed my 256mb infineon chip (rated 222 at 133). the system "screamed" loudly but did not even show the boot screen, just stayed black. I will give some 128mb stick a try. I never tried it with only 128mb. will report on that.
also what I find interesting is how well you did with tweaking via wpcredit. I never come even to close to your speed-jumps while going through the registers and turning the bits. PLEASE could you give more detail on what bits in the respective registers you are turning on/off. just by looking at the byte value of a register I can not deduct what you did.
also please don't forget to run that ITHREAD bench again (with k6-2+550). the score of 7xxx is totally wrecked and can not be explained by whatever setting. it needs "adjustment" before I can post the comparison table
thank you for the information on the "ram-issue". I had the same experience when I installed my 256mb infineon chip (rated 222 at 133). the system "screamed" loudly but did not even show the boot screen, just stayed black. I will give some 128mb stick a try. I never tried it with only 128mb. will report on that.
also what I find interesting is how well you did with tweaking via wpcredit. I never come even to close to your speed-jumps while going through the registers and turning the bits. PLEASE could you give more detail on what bits in the respective registers you are turning on/off. just by looking at the byte value of a register I can not deduct what you did.
also please don't forget to run that ITHREAD bench again (with k6-2+550). the score of 7xxx is totally wrecked and can not be explained by whatever setting. it needs "adjustment" before I can post the comparison table
salvete,
the really good thing about the aladdin 7 is, that by doing some tricks it succeeds in featuring 128bit memory access!!!
aem .. what does xxxbit access mean? is there a normal access bitwidth for socket 7 boards supported by chipset? and which kind of tricks or tools do change it, donpedro? thx for solving the fog!
@ stedman: try to change the 6c offset to 00. it gave the best speedup for my mvp3c2 board in mem write - maybe it will work for you.
the really good thing about the aladdin 7 is, that by doing some tricks it succeeds in featuring 128bit memory access!!!
aem .. what does xxxbit access mean? is there a normal access bitwidth for socket 7 boards supported by chipset? and which kind of tricks or tools do change it, donpedro? thx for solving the fog!
@ stedman: try to change the 6c offset to 00. it gave the best speedup for my mvp3c2 board in mem write - maybe it will work for you.
Last edited by kalabok on Thu Nov 30, 2006 8:10 pm, edited 2 times in total.
@kalabok
sdrams features 64 bit memory access by design. the designers of the aladdin 7 chipset however managed to accomplish a 128 bit memory access (pseudo) by whatever tricky technique. it works only when 2 identical ram-sticks are used. you don't need any additional tools. I think they intertwined read/write access to both ram sticks. since this tricky technique needs some managing overhead performance is not par with real 128 bit access but it still gives a boost of up to around 30%.
off thread, did you have a look on the comparison chart I posted in the "latency timer...." thread?
sdrams features 64 bit memory access by design. the designers of the aladdin 7 chipset however managed to accomplish a 128 bit memory access (pseudo) by whatever tricky technique. it works only when 2 identical ram-sticks are used. you don't need any additional tools. I think they intertwined read/write access to both ram sticks. since this tricky technique needs some managing overhead performance is not par with real 128 bit access but it still gives a boost of up to around 30%.
off thread, did you have a look on the comparison chart I posted in the "latency timer...." thread?
I am sorry to tell you but you won't have any luck with trying to accomplish 128 bit memory access on any socket7 chipset except the aladdin7 chipset. it is the only one I know of that provides that feature. it needs more than using 2 sticks of ram, the feature's technology comes with the chipset. it certainly is not available on via socket7 boards ....
- Uranium235
- Senior K6'er
- Posts: 183
- Joined: Mon Sep 12, 2005 9:59 pm
Stedman, those are excellent Everest scores at 133FSB. Higher than what I had scored on the P5S-VM. Could you post a complete Wpcredit dump for me to compare to? Thanks.
DonPedro, thanks for the info on the ALi 7. The 128bit memory access was something I was unaware of. Sounds almost like an early dual-channel interface. This gives me some ideas to test out....
DonPedro, thanks for the info on the ALi 7. The 128bit memory access was something I was unaware of. Sounds almost like an early dual-channel interface. This gives me some ideas to test out....
- Stedman5040
- Veteran K6'er
- Posts: 271
- Joined: Mon Sep 05, 2005 4:22 pm
Peter,
I ran the ithreads test again and got a score of 139
as regards the register changes they are as follows
Register 50 from 20h to F0h
Register 51 from DAh to 9Ah (probably of use only when external cache is enabled)
Register 52 from D0h to 50h
Register 55 from 0Fh to 0Bh (This bit change really shoots up the score for memory read)
Just going to run off the OSmark on the 600MHz.
Stedman
I ran the ithreads test again and got a score of 139
as regards the register changes they are as follows
Register 50 from 20h to F0h
Register 51 from DAh to 9Ah (probably of use only when external cache is enabled)
Register 52 from D0h to 50h
Register 55 from 0Fh to 0Bh (This bit change really shoots up the score for memory read)
Just going to run off the OSmark on the 600MHz.
Stedman
stedman,
thank you for your commitment!
your everest scores clearly show that the soyo board has a far weaker implementation of the memory subsystem. I can not set cas latency from 3 to 2, my system immediately freezes. although I use the same wpcredit-tweak in all other aspects the soyo board is just is a world away from the ga-5smm
I will post the new chart tomorrow.
it will give numbers for:
- K6-3-450 vs K6-2+450 (showing the difference between what can be expected from going from 128kb on-die cache to 256)
- K6-2+550 with onboard cache on/off, revealing how much onboard cache helps if so at all
- for comparison your results on the ga-5smm, K6-2+550
btw, did you receive the tool for the matrox card I sent to you via e-mail?
thank you for your commitment!
your everest scores clearly show that the soyo board has a far weaker implementation of the memory subsystem. I can not set cas latency from 3 to 2, my system immediately freezes. although I use the same wpcredit-tweak in all other aspects the soyo board is just is a world away from the ga-5smm
I will post the new chart tomorrow.
it will give numbers for:
- K6-3-450 vs K6-2+450 (showing the difference between what can be expected from going from 128kb on-die cache to 256)
- K6-2+550 with onboard cache on/off, revealing how much onboard cache helps if so at all
- for comparison your results on the ga-5smm, K6-2+550
btw, did you receive the tool for the matrox card I sent to you via e-mail?