Still haven't found a solution. Anybody got any suggestions? 🙁

if you have some PCI cards like sound cards, try changing slots? sometimes this can help.

Unfortunately I didn't use any sound cards. I usually add sound cards later, after the initial performance and stability tests.
Also, the first thing I do is to always disable the onboard sound and other unused devices like serial / parallel ports

So... I finally gave up on this board. It's a great board for early Windows XP, but something is killing its performance in Windows 98 and as much as I tried, I could not find any workaround. 🙁

adi88 wrote on 2020-11-09, 07:26:

So... I finally gave up on this board. It's a great board for early Windows XP, but something is killing its performance in Windows 98 and as much as I tried, I could not find any workaround. 🙁

I have encountered the exact same issue on an Asus K8V-MX motherboard.

Performance under Win9x is roughly half of what I get under WinXP. Exact same hardware during both tests, identical BIOS settings, same driver versions (VIA 5.24A and Nvidia 40.72 WHQL). For some reason, the GPU runs much worse under Win9x. I tried a few different 4-in-1 driver versions, but that didn't help either. Not sure what's going on here.

Update: with some tips from @bloodem I was able to get proper GPU performance under Win9x on the K8V-MX motherboard. I downloaded the earliest available BIOS version from Asus' website (v0112) and flashing that completely fixed the problem.

As the original thread starter suggests, there seems to be some kind of bug in the later BIOS versions for certain socket 754 motherboards which severely reduces Win9x GPU performance.

Glad to have helped, @Joseph_Joestar!
It would be interesting if you could compare the old and new BIOS versions, maybe there's some obvious setting which is changed (unlikely - this seems like a lower level AGP related bug, but still worth a shot). 😀

This board is fairly sparse with its settings, and I don't see any obvious differences on the AGP sub screen.

For what it's worth, I always loaded the BIOS defaults after flashing and only changed a few unrelated things like boot order and SATA configuration. Since both the new and the old BIOS used the same settings, I'm assuming it's some deeper level stuff that got changed in the later revisions.

Yeah, figured that would be the case...

bloodem wrote on 2022-03-04, 13:47:

Yeah, figured that would be the case...

I did notice one difference under Win9x. With the newest BIOS, dxdiag reported "AGP Texture Acceleration" as "Not Available". But with the oldest BIOS, it is now correctly reported as "Enabled".

Under WinXP, this setting was reported as "Enabled" using both the old and the new BIOS.

Interesting. Although, I'm pretty sure that's an effect, not the actual cause (since lack of AGP texturing would not cause such a dramatic hit in performance).
So it's probably related to the BIOS bug, but something else more sinister is happening behind the scenes. 😁

I am interested in this thread as I have a system in my arsenal which is almost identical to the OP's. K8T-Neo/A64-3400+/FX5900/ESS Solo1 which I use for Win98 and DOS. I do not remember any under performance in Win98, certainly it is faster than my AthlonXP/Geforce4 rig. I will take a look at bios version when I return to my home in about 1 week, I have been traveling for a few weeks now.

Joseph_Joestar wrote on 2022-03-04, 12:36:

Update: with some tips from @bloodem I was able to get proper GPU performance under Win9x on the K8V-MX motherboard. I downloaded the earliest available BIOS version from Asus' website (v0112) and flashing that completely fixed the problem.

As the original thread starter suggests, there seems to be some kind of bug in the later BIOS versions for certain socket 754 motherboards which severely reduces Win9x GPU performance.

Man you were so lucky...
I tried the same thing with the MSI K8T Neo-V, but unfortunately even the earliest BIOS that I could find had the same issue. 🙁
Glad that it worked in your case, though.

.info wrote on 2022-03-04, 22:43:

I am interested in this thread as I have a system in my arsenal which is almost identical to the OP's. K8T-Neo/A64-3400+/FX5900/ESS Solo1 which I use for Win98 and DOS. I do not remember any under performance in Win98, certainly it is faster than my AthlonXP/Geforce4 rig. I will take a look at bios version when I return to my home in about 1 week, I have been traveling for a few weeks now.

Well, the K8T-Neo is a different motherboard than my K8T Neo-V, so I wouldn't be surprised if that one works OK.

Well you are most correct. My K8T-Neo with A64-3200+, FX5900 and bios v3.31a is scoring 12207 on 3DMark2001 and 9640 in 3DMark2000 default benchmarks (win98se). Seems like Neo should be similar to Neo-V but apparently not.

Joseph_Joestar wrote on 2022-03-04, 14:54:

I did notice one difference under Win9x. With the newest BIOS, dxdiag reported "AGP Texture Acceleration" as "Not Available". Bu […]
Show full quote

bloodem wrote on 2022-03-04, 13:47:

Yeah, figured that would be the case...

I did notice one difference under Win9x. With the newest BIOS, dxdiag reported "AGP Texture Acceleration" as "Not Available". But with the oldest BIOS, it is now correctly reported as "Enabled".

DX9_Win98.jpg

Under WinXP, this setting was reported as "Enabled" using both the old and the new BIOS.

Very interesting. I have a similar issue with my PT880 based motherboard but AGP texture acceleration is working. My issue is there is no AGP memory exposed to the display driver. In 3DMark01, if you look in the system info, under the display driver, does your AGP memory match what the AGP aperture is set to in BIOS?

VDNKh wrote on 2022-03-17, 12:05:

In 3DMark01, if you look in the system info, under the display driver, does your AGP memory match what the AGP aperture is set to in BIOS?

Yes, it's 128 MB in my case, which is what I'm using in the BIOS.

This is with the oldest BIOS version though. I don't think I've checked this using the latest BIOS which had the aforementioned performance problems. And I'm not too keen on flashing that back on my rig.

I noticed the same performance problem on my MSI K8T Neo V when using 45.23 driver on Windows 98. The cause seems to be no AGP memory exposed to display driver. 3DMark01 reports 0MB AGP memory on display adapter in system info. This confirms findings of VDNKh in Re: Not Another Ultimate Windows 98 Build.

This problem partially goes away with 61.76 driver when 32MB AGP memory is reported (it seems to be a workaround in NVidia driver), however that driver version is not useful in Windows 98 due to many compatibility issues. It doesn't respect AGP aperture size set in BIOS setup and is fixed to 32MB.

3D Mark 2001 results (1024x768 32bit), FX5600:

• Windows 98, driver 45.23 - 5187
• Windows 98, driver 61.76 - 7189
• Windows XP, driver 45.23 - 8548

For comparison PIII 900E scores 4981 on Windows 98, driver 45.23.

I'm afraid fixing this problem will require someone with a 754 board not experiencing this problem doing a dump of chipset settings using wpcredit when AGP memory is working and when it isn't. This will need upgrading BIOS on purpose to break it temporarily. The problem could be having exactly the same BIOS settings.

Most likely VIA AGP driver is the culprit here, getting confused by chipset settings set by BIOS.

My Athlon 64 3400+ scores about 18k points in 3D Mark 2001 with GT7600 in comparison to 8548 with FX5600. A dual OS rig with anything less than FX5900 is going to be totally inferior in Windows XP, but anything faster than that is not going to work in Windows 98 well due to requiring driver newer than 45.23.

Issues with AGP memory size are not unique to VIA chipsets and drivers. I checked my 440BX rig with another FX5600 to see what 3DMark01 reports. Despite BIOS being configured for 64MB AGP aperture size, 3DMark01 reports 32MB AGP memory (same as Sisoft Sandra 2003). In a separate section Motherboard Info -> AGP capabilities I see 64MB aperture size in 3DMark01 though. MSI K8T Neo V sees 128MB AGP memory in Windows XP when AGP aperture is set to 128MB, not just half of it.

There is also a chance this issue affects synthetic benchmarks only, not period correct games as 128MB of video memory is plenty and AGP memory may not be needed.