The same Digital Venturis Socket-3 system but with the Pentium Overdrive running at 100 MHz.

POD@100, no L2 cache.

Athlon 1200, 512MB PC133 SDRAM @ default settings, W98SE:

Pentium 200, 64 MB 60ns EDO ram @ default settings, W98SE:

The Digital Venturis Socket-3 system was rather slow as it lacked L2 cache. I thought I would test a Pentium Overdrive at 100 MHz with my PC Chips M921 UMC Chipset motherboard, the 256KB cache on this board is amazingly not fake.

If I upgraded the cache on this PC-Chips M921 board to 512KB and used Windows 2000 I think I would get a very competitive score. Using Windows 95B and with only 256KB L2 cache this is still my best SuperPi 1M score with a Socket-3 system.

POD83 @100, PC-Chips M921 UMC chipset 256KB Cache, 16MB FPM. SuperPi 1M: 33m 16s.

A normal Pentium with double the memory bus width and 66MHz FSB is much faster than the POD @100 when it comes to SuperPi 1M!

P100, 5DHX v1.2 512KB cache, 32MB FPM. SuperPi 1M: 22m 26s (W95B)

lazibayer wrote:

Tillamook 133x3 on SiS530
Can't turn on L2 cache and the poor memory bandwidth of SiS530 may hurt a lot.

mmx_5smm_133x3_noL2_superpi.jpg

I have a Gigabyte GA5 SMM board running at 133MHz fsb with a K6 3plus 450 at 600MHZ running with no on L2 board cache which is rubbish anyway at only 512k. I can get in Superpi 1.1e a time of 4 mins 49 seconds, which isn't bad for a SS7 board with poor memory bandwidth. System as follows

Gigabyte GA5 SMM v1.2
K6-3+ @ 600MHz (4.5x133)
2x256Mb Kingston pc133 valueram @ 2-3-2-6
Voodoo 3 2000 pci (onboard vga disabled)
Windows 98se

WPCREDIT tweaks applied.

Stedman 😀

Stedman5040 wrote:

I have a Gigabyte GA5 SMM board running at 133MHz fsb with a K6 3plus 450 at 600MHZ running with no on L2 board cache which is r […]
Show full quote

lazibayer wrote:

Tillamook 133x3 on SiS530
Can't turn on L2 cache and the poor memory bandwidth of SiS530 may hurt a lot.

mmx_5smm_133x3_noL2_superpi.jpg

I have a Gigabyte GA5 SMM board running at 133MHz fsb with a K6 3plus 450 at 600MHZ running with no on L2 board cache which is rubbish anyway at only 512k. I can get in Superpi 1.1e a time of 4 mins 49 seconds, which isn't bad for a SS7 board with poor memory bandwidth. System as follows

Gigabyte GA5 SMM v1.2
K6-3+ @ 600MHz (4.5x133)
2x256Mb Kingston pc133 valueram @ 2-3-2-6
Voodoo 3 2000 pci (onboard vga disabled)
Windows 98se

WPCREDIT tweaks applied.

Stedman 😀

I have one of those motherboards on my test bench at this very moment!

My best SuperPi 1M score so far with a K6-3+ in Windows 98se is 4m 44s using a Gigabyte GA-5AX with the CPU at 5x120 MHz, I will try to beat that score with the GA-5SMM.

My board comes from Compaq Presario 5420 but I have crossflashed the latest Gigabyte BIOS.

Hi Skyscraper,

To get good results in superpi you will have to apply the WPCREDIT tweaks as it improves memory bandwidth considerably. Can you currently run your board with 133MHz fsb with your K6+ chip as some seem to be flakey at this speed. I have beaten the 4m 44s barrier with using an Epox EP MVP3 G2 board and an Asus P-5A board with a K6+ chip running with overclocked fsb. Scores are closer to 4m 30s.

Stedman

Stedman5040 wrote:

Hi Skyscraper,

To get good results in superpi you will have to apply the WPCREDIT tweaks as it improves memory bandwidth considerably. Can you currently run your board with 133MHz fsb with your K6+ chip as some seem to be flakey at this speed. I have beaten the 4m 44s barrier with using an Epox EP MVP3 G2 board and an Asus P-5A board with a K6+ chip running with overclocked fsb. Scores are closer to 4m 30s.

Stedman

I think many k6-2+ and and K6-3+ CPUs start to get stability problems at 133/135 MHz FSB, especially when near the avarge max clock of ~600 MHz. At least the GA-5SMM has a PCI/4 divider so getting a high FSB stable should be less trublesome than with the GA-5AX. I have not had the time to tinker anymore with the GA-5SMM after installing Windows 98se late last night, I will start with benching the board as it was when I got it with a K6-2-400 to get a performance baseline.

I do not think tweaking with WPCREDIT would have helped very much with my Gigabyte GA-5AX, the memory speed was already decent at 200 MB/s in Sandra 99. I do actually not think I have seen anyone get a really good SuperPi score with the GA-5AX, at leat not in Windows 9x.

Upgraded my Socket A rig from a Thunderbird 1200 (SuperPi results here) to a Thoroughbred 2400+. Not the performance boost I expected, but still quite an improvement:

Stedman5040 wrote:

Hi Skyscraper,

To get good results in superpi you will have to apply the WPCREDIT tweaks as it improves memory bandwidth considerably. Can you currently run your board with 133MHz fsb with your K6+ chip as some seem to be flakey at this speed. I have beaten the 4m 44s barrier with using an Epox EP MVP3 G2 board and an Asus P-5A board with a K6+ chip running with overclocked fsb. Scores are closer to 4m 30s.

Stedman

I just realized that I already got a better score with the Gigabyte GA-5AX using a higher FSB, I just did not post it in this thread.

This is without WPCREDIT and without using the Sandra memory tweak (I had forgot all about that one).

There is not a snowballs chance in hell I will beat this score with the GA-5SMM but I sure will try. So far 4m 51s is my best score with the GA-5SMM but I have not used WPCREDIT yet.

K6-3+ @607 135*4.5 with 256MB PC135 CL2

This thread has alot of information and more benchmarks and screenshorts.
K6-III+ 450 fastest overclock

following results all run on mvp3 board with 2mb cache, and 100fsb:
k6-2-550, write allocation enabled: 6:50
k6-2-550, write allocation disabled: 5:56(yes its 5:56, no typo)
k6-2-550, onboard cache off, write allocation enabled: 10:15
k6-2-550, onboard cache off, write allocation disabled: 10:17
k6-2-300cxt, write allocation enabled: 9:05
k6-2-300old, write allocation unsupported: 8:32
k6-3+500 oc 600: 4:49. i guess this is the fastest 100fsb socket7 record without using the "low memory bug". 😒

and i wonder if sis530 is the slowest socket7 chipset when compared at same fsb? i mean chipsets with support for dual-voltage cpus.

noshutdown wrote:

following results all run on mvp3 board with 2mb cache, and 100fsb: k6-2-550, write allocation enabled: 6:50 k6-2-550, write all […]
Show full quote

following results all run on mvp3 board with 2mb cache, and 100fsb:
k6-2-550, write allocation enabled: 6:50
k6-2-550, write allocation disabled: 5:56(yes its 5:56, no typo)
k6-2-550, onboard cache off, write allocation enabled: 10:15
k6-2-550, onboard cache off, write allocation disabled: 10:17
k6-2-300cxt, write allocation enabled: 9:05
k6-2-300old, write allocation unsupported: 8:32
k6-3+500 oc 600: 4:49. i guess this is the fastest 100fsb socket7 record without using the "low memory bug". 😒

and i wonder if sis530 is the slowest socket7 chipset when compared at same fsb? i mean chipsets with support for dual-voltage cpus.

I would imagine that you are right as regards performance at 100fsb. Using the best wpcredit tweaks I can find and the 5.5x multiplier with a K6 3+ cpu running at 550MHz I can get around 5m 38secs. Increasing speed to 600 MHz reduces this to about 5m 17 secs, which is about what I get with the fsb at 133 and the cpu running at 533MHz with onboard cache off. Using a tweaked MVP3 chipset on a Epox EP-MVP3 G2 or an tweaked Aladdin5 chipset on an Asus P5A with a K6 3+ cpu I can get around 4m 50 secs 4m 55secs for Superpi 1M

Stedman

Aopen Fortress 5000 Mainboard with 2x Xeon 550Mhz (Pentium 3 with 2Mbyte Fullspeed L2 Cache).
GeForce 6200 AGP 128bit 256Mb DDR.
512Mb ECC SDRAM PC100

SuperPi on One CPU.

System: Pentium III 933Mhz/VC820/256MB RDRAM PC-800/WinME
3m 9.189s

Wow only almost 10 seconds faster, Cache really counts in SuperPi 😀

matze79 wrote:

xeon.jpg

Wow only almost 10 seconds faster, Cache really counts in SuperPi 😀

i would say that p3-933 time is a bad run, a well tuned p3-1g(133fsb and 256k) is capable of around 2:10.
but your time is still pretty impressive, i would estimate the reasonable time of a celeron550 to be around 4:10.

Does anyone own a P3 550 Katmai and can give it a try ?

Here is a Katmai 550 on a dog slow Gigabyte BX 2000+ board.

And here is a Katmai 600 on a HP/Asus P2B98-XV board, I do not think this board is very speedy but it's at least faster than the Gigabyte board.

Notice that these scores are with Windows 98SE, in Windows XP the scores would have been better. The scores are at least a year old.

kaputnik wrote:

Athlon 1200, 512MB PC133 SDRAM @ default settings, W98SE:

My rig was about 4.5s slower than yours running same speed (12x100MHz) and memory timings in Windows 98SE.

But with Windows XP and some tuning the story is a bit different (I thing the original speed of the CPU is 1.33GHz):

How long does it take to run 32m? Out of blind curiosity I just started it on my AMD Duron 650, is that an impossible feat of computing power?

I posted this in another thread but I might aswell post the result here in the correct thread aswell.

Cyrix 686 33 MHz (CPU=FSB=PCI=33)
FIC VA-502 (VIA VPX)
64MB 168pin SDRAM
S3 Trio64V+ 2MB PCI.

256KB L2 Cache in WB mode.
All BIOS settings at the fastest values.
Linar Burst supported and activated.

A few results with my Athlon XP Barton equipped KT7-RAID at speeds of 500mhz, 2000mhz, and 2400mhz. I was hoping to crack the 1 minute mark
but don't think that's possible with Cas2 133mhz sdram as opposed to the DDR RAM that most chipsets with a Barton would be running.