You are running in ACPI mode. Double click on "PCI Bus”, I think, and see if "IRQ steering" is working correctly. If it isn't, the problem may be due to an IRQ conflict. Disabling a controller on an IRQ would free it, and moving cards to different slots might assign them to different IRQs that are not conflicting.

USB 2.0 devices are also infamous for gobbling up IRQ resources and not playing nice with other devices in Windows 98. Would you happen to be using a USB 2.0 card or the onboard 1.1 ports?

Kahenraz wrote on 2024-03-13, 01:45:

I disagree. This shouldn't be a problem with disk access if DMA is functioning correctly. How system is a K2 not a 386.

Admittedly my computers may have just been configured incorrectly at the time 😂

I just realized that I had some typos that you quoted. I fixed them in my original post.

I am currently using onboard 1.1 usb ports with the nusb36.exe usb 2.0 drivers so i can access flash drives. I don't see any errors in device manager. The new bios or ALi AGP/IDE drivers must have setup ACPI because now it shuts off without the "It is now safe to turn of your computer" warning. I don't see any ACPI options in the bios and didn't set it up when I installed 98SE. I could probably get better performance with DX7 or 8 but I am trying to stay as modern as possible with this build.

I enabled k6 write allocate in the bios and got a substantial increase in cpu performance. I still get the best performance with only 256mb ram with external cache enabled. I would like to get a 128mb stick to try since that is all the 512k l2 cache can effectively use.

Isn't the maximum of 512K only 64MB without a write-through option in the BIOS?

Did you confirm if IRQ Steering is working?

I upped the fsb from 95 to 105x5.5 @ 577mhz and bumped the core voltage from 2.0 to 2.1. Unfortunately I wasn’t able to run 110 or 115 fsb even with good pc133 ram. Disabling the external cache still didn’t allow me to use more ram without a 5% performance decrease.

I wasn’t able to get coolbits, rivatuner, or power strip working to overlock the gpu but the asus v9560 video suit edition is basically an overclocked fx5500.

The IRQ operation looks normal. I have all IRQ settings in the bios set to auto.

Glad to hear that it's running smoothly now!

As already mentioned by another user, you could be getting more FPS by using an older Nvidia card - newer cards require newer drivers and these do not run well on a K6-2 system. Driver 7.76 is the fastest, particularly for OpenGL games. Driver 21.83 has the best compatibility while still being reasonably fast. You can get around 3500 points with a K6-2+ 550 in 3DMark99 with a Geforce 2 MX using the older drivers.

Have you tried using only 128 MB of RAM? I believe that the ALi chipset boards have only 512 KB of cache and can only cache up to 128 MB, so you're losing speed when you go above that limit. Test 3Dmark99 with 128 MB and 256 MB of RAM and see if there's a difference.

Poor memory performance can often be a bottleneck on Super Socket 7 systems. A good way to test it is the benchmark function in Everest 2.20. You can often get better performance by tweaking the memory settings of the chipset with a tool like WPCREDIT. Here's an example from a VIA MVP3 board (K6-2+ 550 MHz, 100 MHz FSB):

Default settings: Read: 244, Write: 122, Access time 266, 3DMark99: 3093
After tweaking: Read: 307, Write: 160, Access time: 210, 3DMark99: 3402

These results will be less dramatic for ALi boards because they have better read speed by default, but it is still worth the effort.

I saw that even the voodoo3 runs faster than an fx5500 and the 5200 is just trash. What would be the fastest agp 2x, 128bit card that uses the older drivers? I was looking for a Geforce 4 MX400 for my older system that is more period correct but for this build i wanted the most modern 02-03 components that could possibly work. An end of era build. Im sure the USB 2.0 adapter ,WIFI card,and excess RAM take a hit on the system too. However the sata ssd seems to be running completely lag free.

Its an Asus board with an asus card so I like the matching pair. And the fact that is supports DX9c, has double DVI output, and even supports my monitors 144hz refresh rate. I am getting better performance than others I have seen with FX cards and it seems pretty stable. I runs cool and should be plenty enough for my quakeII and counter-strike needs. 53.04 is supposed to be the fastest while 44.03 is supposed to be the most stable for the FX series in 98 supposedly. I currently have DX9b. I would like to try 8.1 or 9c.

It was running incredibly fast at 112 fsb but would only make it half way threw 3dmark99 before random reboots. If I could control the pci/agp fsb it may have worked. I was hoping disabling external cache and only using agp 1x mode would help. Its a shame most k6 boards can't run 133 fsb or more than 1 stick of ram.

The FX 5200 is not trash. You can't run driver-heavy graphics cards pre-Coppermine without major performance penalties. That's why asdf53 was recommending older graphics cards that used older drivers. An FX 5200 will beat the pants off of a TNT2, unless you're on a K6. Then sometime like the TNT2 or Rage 128 wins.

The FX 5200 just isn't optimal for this build.

Here's an example how newer driver versions have a negative impact on performance, these are some 3dmark99 scores on a K6-2+ at 600 MHz:

Geforce 3 21.83: 3525
Geforce 3 7.76: 4453
Voodoo 3: 3615

I believe Geforce 3 is the fastest card that still supports driver 7.76, Geforce 4 already needs 29.xx. You absolutely don't need a GF3 though, a Geforce 2 MX is also very good and it would reach around 3800 points here.

Regarding FSB overclocking, you have to remember that you need to disable the external cache because this is what mostly limits the overclocking stability. Few boards can reach FSB speeds of 110 or more with the cache enabled. And when you disable the cache, you lose quite a bit of performance. The difference between FSB 100 with cache enabled and FSB 133 with cache disabled isn't very large. If you can run it with the cache enabled at 105 or 110 MHz that's pretty good.

Here are some benchmarks to see the impact:

gf3, fsb100, 600mhz, ext. cache on: 3dmark99 3360
gf3, fsb133, 600mhz, ext. cache off: 3dmark99 3525
gf3, fsb133, 600mhz, ext. cache on: 3dmark99 3976

So with the FSB at 110 it would match or even surpass FSB 133 with disabled cache. This is on a VIA board, I don't know how an ALi board would behave.

I don't know what it's like on VIA chipsets, but I was told that overclocking the bus on a 430TX Socket 7 does not play well with the IDE controller and can cause data corruption.

As a general rule, I try not to tax the bus too heavily beyond spec in this case. The 430TX does not support ECC memory, so there is even less opportunity to catch any fault that might appear.

asdf53 wrote on 2024-03-17, 06:19:

....
Have you tried using only 128 MB of RAM? I believe that the ALi chipset boards have only 512 KB of cache and can only cache up to 128 MB, so you're losing speed when you go above that limit. Test 3Dmark99 with 128 MB and 256 MB of RAM and see if there's a difference.
....

This.
Cacheable area is 128MB!
Except you have the very late revision G of the chipset, then there is no external tag ram and no restriction for the cachable area.

I tweaked the ram settings, enabled x111-2111 mode and passive release. I also increased the aperture size to 256mb which performed the best instead of 128 or 64 like everyone said. This eeked out even more performance. Finally I increased the multiplier to 6 for 630mhz for a final speed run and got the best performance out of dozens of different combinations.

I probably plan on just leaving it at 577mhz because I can’t actively monitor the cpu temp and this board is limited to a 50x50mm cooler. I did purchase an asus v7700ti (GeForce 2 GTS) to try with older drivers but will be limited to vga monitors. I look forward to trying 128mb ram.

Did you ever try with 64MB of memory? My 430TX with 512KB of cache can only cache a maximum of 64MB.

adam731432 wrote on 2024-03-17, 21:45:

I tweaked the ram settings, enabled x111-2111 mode and passive release. I also increased the aperture size to 256mb which perfor […]
Show full quote

I tweaked the ram settings, enabled x111-2111 mode and passive release. I also increased the aperture size to 256mb which performed the best instead of 128 or 64 like everyone said. This eeked out even more performance. Finally I increased the multiplier to 6 for 630mhz for a final speed run and got the best performance out of dozens of different combinations.

I probably plan on just leaving it at 577mhz because I can’t actively monitor the cpu temp and this board is limited to a 50x50mm cooler. I did purchase an asus v7700ti (GeForce 2 GTS) to try with older drivers but will be limited to vga monitors. I look forward to trying 128mb ram.

IMG_2188.jpeg

IMG_2187.jpeg

IMG_2192.jpeg

IMG_2189.jpeg

IMG_2191.jpeg

Maybe you are right. When you have both options 128MB or 256MB, then you can tell (us) exactly....
According to this old post, Asus used a 10 bit tag ram:
https://groups.google.com/g/alt.comp.periphs. … s/c/elf8JcqEKuE

Kahenraz wrote on 2024-03-17, 21:57:

Did you ever try with 64MB of memory? My 430TX with 512KB of cache can only cache a maximum of 64MB.

I don’t have any 64 or 128mb yet. I have read conflicting reports about the p5ab. But most say 128mb for early versions and 512 for later. One user reported that 768mb did create a slight decrease in performance but he was able to have multiple browser tabs open at the same time which shows that it was being used.

I have also read that many k6 boards have a problem with adding a third stick of ram no matter what size it is. Something to do with the chipset limitations and ram timing going crazy when you add a third stick.

http://www.k6plus.com/phpBB3/viewtopic.php?t=1292

asdf53 wrote on 2024-03-17, 06:19:

Glad to hear that it's running smoothly now! […]
Show full quote

Glad to hear that it's running smoothly now!

As already mentioned by another user, you could be getting more FPS by using an older Nvidia card - newer cards require newer drivers and these do not run well on a K6-2 system. Driver 7.76 is the fastest, particularly for OpenGL games. Driver 21.83 has the best compatibility while still being reasonably fast. You can get around 3500 points with a K6-2+ 550 in 3DMark99 with a Geforce 2 MX using the older drivers.

Have you tried using only 128 MB of RAM? I believe that the ALi chipset boards have only 512 KB of cache and can only cache up to 128 MB, so you're losing speed when you go above that limit. Test 3Dmark99 with 128 MB and 256 MB of RAM and see if there's a difference.

Poor memory performance can often be a bottleneck on Super Socket 7 systems. A good way to test it is the benchmark function in Everest 2.20. You can often get better performance by tweaking the memory settings of the chipset with a tool like WPCREDIT. Here's an example from a VIA MVP3 board (K6-2+ 550 MHz, 100 MHz FSB):

Default settings: Read: 244, Write: 122, Access time 266, 3DMark99: 3093
After tweaking: Read: 307, Write: 160, Access time: 210, 3DMark99: 3402

These results will be less dramatic for ALi boards because they have better read speed by default, but it is still worth the effort.

With my bios tweeks I am getting 305 read 127 write and 231 latency.

That's pretty good already! This is what I'm getting on a Gigabyte GA-5AX with 128 MB RAM and a K6-2+ at 550 MHz:

326 136 202.6

Here are some WPCREDIT settings that you can try:

48 and 72: This is where you can set the RAM timings, best I could do was 2-2-2-4-7. Not all boards will run at these timings, so you'll need to experiment.
49: SDRAM Internal Page detection = 1, SDRAM Enhanced Page Mode = 0, SDRAM precharge all insert 1 wait = 0
72: Delay SDRAM Controller 1 clk = 1

41: Fast DRAM Read = 1 (supposed to be faster, makes system freeze on my board)
43: Fast NAJ asserted in single write cycle = 1 (supposed to be faster, had no effect on my board)

The number refers to the memory offset where the setting can be found. One address I did not find was the setting for 4-bank memory interleaving, maybe this setting is the same as "SDRAM internal page detection" on ALI boards. I remember this having a large impact on speed on VIA boards.

If you want to compare your benchmark result to the one I posted above, remember to benchmark at 550 MHz and FSB 100 to have the same baseline.

Kahenraz wrote on 2024-03-17, 14:51:

The FX 5200 is not trash. You can't run driver-heavy graphics cards pre-Coppermine without major performance penalties. That's why asdf53 was recommending older graphics cards that used older drivers. An FX 5200 will beat the pants off of a TNT2, unless you're on a K6. Then sometime like the TNT2 or Rage 128 wins.

The FX 5200 just isn't optimal for this build.

I must have been thinking about the PCI version of the FX5500. Allot of value cards only used 64-bit ram to cut down on cost. The ATI Radeon 9250 would have been a good 98secard if it had 128-bit ram.

I have read that you can get data corruption when using a SSD if not set up right on an older system like this. Something about trying to overwrite data when it gets full but I should never have that problem with 120gb.

The Ali IDE drivers seemed to help set everything up but I am still not sure where the IDE channels come into play when you are using a PCI SATA controller.