TL;DR - you might be running into an issue similar to the one I had with the MSI K8T NEO-V motherboard: MSI K8T NEO-V + Athlon 64 - very slow 3D performance in Windows 98 (works great in Windows XP)
For some reason, that motherboard has an AGP related bug that causes it to run at a third its normal speed in Windows 98 (and, boy, I sure tried all possible driver and BIOS versions!).
However, it works perfectly in Windows XP. And it's not something related to the chipset, I know for a fact that the VIA K8T800 is very fast in Windows 98, since I have other boards with the same chipset that don't exhibit this problem.
After finding the right RAID driver to slipstream into the installer (VIA RAID drvdisk 5.1.6 2008), and finding the right USB imager (nLite + WInSetupFromUSB), XP SP3 installed with zero issues. As I suspected, 3DMark01 in XP is double my score in 98.
Some folks were saying, I think on the ISA Card USB thread that under win 98 USB cards cause the system to slow down immensely, dunno if it's the polling or what, driver conflicts? Anyway, if you've got PS/2 keyboard and mouse, might be worth turning off onboard USB in CMOS setup for a trial run.
Unicorn herding operations are proceeding, but all the totes of hens teeth and barrels of rocking horse poop give them plenty of hiding spots.
Some folks were saying, I think on the ISA Card USB thread that under win 98 USB cards cause the system to slow down immensely, dunno if it's the polling or what, driver conflicts? Anyway, if you've got PS/2 keyboard and mouse, might be worth turning off onboard USB in CMOS setup for a trial run.
I disabled the USB controller in the BIOS but only saw ~1.5% increase in score. (One USB port even shared an IRQ with the GPU)
The plot thickens with this graphics performance issue. I ran Final Reality on 98SE and XP with driver version 61.76 (98) and 61.77 (XP) to see what exactly is causing this slowdown.
But first I ran 3DMark01 again to see how 61.xx performs, 17,183 for 98 and 26,533 for XP. Almost a 10,000 point gap.
AGP bandwidth doesn't seem to be the culprit. 3D transfer rate seems almost half of what it is in XP, could that be the smoking gun?
But it gets more complicated. I also ran ArchMark 0.50 for a more broken down benchmark.
ArchMark 0.5.0
198SE XP 2Info:Driver GeForce FX 5950 Ultra/PCI/SSE2 v1.5.1 GeForce FX 5950 Ultra/AGP/SSE2 v1.5.1 3Info:Resolution 1024x768 @ 75.75Hz 1024x768 @ 60.60Hz 4Info:Method Flush Flush 5Fillrate:32:Mode R8G8B8A8 Z24 S8 R8G8B8A8 Z24 S8 6Fillrate:32:Col 1888.441 M Pix/s 1888.472 M Pix/s 7Fillrate:32:Z 3612.770 M Pix/s 3628.780 M Pix/s 8Fillrate:32:ColZ 1883.435 M Pix/s 1884.141 M Pix/s 9Fillrate:32:ZPassColZ 1879.256 M Pix/s 1880.669 M Pix/s 10Fillrate:32:ZCullLEqual 6976.077 M Pix/s 7097.200 M Pix/s 11Fillrate:32:ZCullGEqual 6976.150 M Pix/s 7104.064 M Pix/s 12Fillrate:32:ZCullEqual 6976.305 M Pix/s 7104.018 M Pix/s 13Fillrate:32:S 3767.064 M Pix/s 3765.215 M Pix/s 14Fillrate:32:SCull 3767.046 M Pix/s 3767.152 M Pix/s 15Fillrate:32:SPass:S 3767.021 M Pix/s 3767.114 M Pix/s 16Fillrate:32:SPass:ZFailS 3371.856 M Pix/s 3389.083 M Pix/s 17Fillrate:32:SPass:ZPass:S 3370.500 M Pix/s 3389.052 M Pix/s 18Fillrate:32:SPass:ZPass:ZS 3372.136 M Pix/s 3388.964 M Pix/s 19Fillrate:32:SPass:ZPass:Col 1887.920 M Pix/s 1887.533 M Pix/s 20Fillrate:32:SPass:ZPass:ColZ 1887.922 M Pix/s 1888.038 M Pix/s 21Fillrate:32:SPass:ZPass:ColS 1881.136 M Pix/s 1882.033 M Pix/s 22Fillrate:32:SPass:ZPass:ColZS 1881.574 M Pix/s 1881.585 M Pix/s 23Fillrate:16:Mode R5G6B5A0 Z16 S0 R5G6B5A0 Z16 S0 24Fillrate:16:Col 1888.436 M Pix/s 1888.470 M Pix/s 25Fillrate:16:Z 3672.530 M Pix/s 3573.453 M Pix/s 26Fillrate:16:ColZ 1888.420 M Pix/s 1888.422 M Pix/s 27Fillrate:16:ZPassColZ 1888.363 M Pix/s 1888.348 M Pix/s 28Fillrate:16:ZCullLEqual 7037.183 M Pix/s 7248.024 M Pix/s 29Fillrate:16:ZCullGEqual 7030.350 M Pix/s 7248.180 M Pix/s 30Fillrate:16:ZCullEqual 7037.310 M Pix/s 7248.319 M Pix/s 31Bandwidth:Mode R8G8B8A8 Z24 S8 R8G8B8A8 Z24 S8 32Bandwidth:Clear:All 36941.690 M B/s 37557.642 M B/s 33Bandwidth:Clear:Color 28623.741 M B/s 28819.173 M B/s 34Bandwidth:Clear:ZAndStencil 50395.408 M B/s 52314.222 M B/s 35Bandwidth:Clear:Z 37795.338 M B/s 39236.413 M B/s 36Bandwidth:Clear:Stencil 18088.840 M B/s 18217.156 M B/s 37Bandwidth:Draw 28608.411 M B/s 28727.120 M B/s 38Bandwidth:BurnedByRAMDAC 238.291 M B/s 190.641 M B/s 39Bandwidth:Physical 28846.701 M B/s 28917.761 M B/s 40Geometry:Mode R5G6B5A0 Z0 S0 R5G6B5A0 Z0 S0 41Geometry:Plain:Fan 157.691 M Tris/s 157.694 M Tris/s 42Geometry:Plain:List 52.593 M Tris/s 52.594 M Tris/s 43Geometry:Plain:Clip 52.576 M Tris/s 52.594 M Tris/s 44Geometry:LightD1 39.445 M Tris/s 39.436 M Tris/s 45Geometry:LightP1 15.435 M Tris/s 15.435 M Tris/s 46Geometry:LightP8 4.780 M Tris/s 4.781 M Tris/s 47Texturing:Mode R5G6B5A0 Z0 S0 R5G6B5A0 Z0 S0 48Texturing:Bilinear:1 1883.308 M Pix/s 1888.095 M Pix/s 49Texturing:Bilinear:2 1887.695 M Pix/s 1888.139 M Pix/s 50Texturing:Bilinear:3 892.600 M Pix/s 892.733 M Pix/s 51Texturing:Bilinear:4 892.606 M Pix/s 892.732 M Pix/s 52Texturing:Trilinear:1 1887.723 M Pix/s 1886.186 M Pix/s 53Texturing:Trilinear:2 1153.558 M Pix/s 1153.731 M Pix/s 54Texturing:Trilinear:3 696.118 M Pix/s 696.195 M Pix/s 55Texturing:Trilinear:4 566.866 M Pix/s 566.919 M Pix/s 56Readback:Mode R8G8B8A8 Z24 S8 R8G8B8A8 Z24 S8 57Readback:Whole32:R8G8B8A8 52.598 M Pix/s 50.087 M Pix/s 58Readback:Whole32:B8G8R8A8 52.787 M Pix/s 50.350 M Pix/s 59Readback:Whole32:R8G8B8 52.829 M Pix/s 50.349 M Pix/s 60Readback:Whole32:B8G8R8 52.852 M Pix/s 50.348 M Pix/s
…Show last 24 lines
61Readback:Whole32:Zuint 52.755 M Pix/s 50.342 M Pix/s 62Readback:Whole32:Zfloat 52.717 M Pix/s 50.347 M Pix/s 63Readback:Whole32:S8 52.985 M Pix/s 50.350 M Pix/s 64Readback:Small32:R8G8B8A8 34.674 M Pix/s 72285 Pix/s 65Readback:Small32:B8G8R8A8 35.405 M Pix/s 68227 Pix/s 66Readback:Small32:R8G8B8 34.673 M Pix/s 73578 Pix/s 67Readback:Small32:B8G8R8 34.752 M Pix/s 86063 Pix/s 68Readback:Small32:Zuint 34.839 M Pix/s 69565 Pix/s 69Readback:Small32:Zfloat 34.843 M Pix/s 69564 Pix/s 70Readback:Small32:S8 35.117 M Pix/s 85585 Pix/s 71TexCache:Mode R8G8B8A8 Z24 S8 R8G8B8A8 Z24 S8 72TexCache:RGBA 4096 B 4096 B 73TexCache:DXT1 4096 B 4096 B 74TexCache:DXT5 1024 B 1024 B 75TexCache:Pal 1024 B 1024 B 76Tiling:Mode R8G8B8A8 Z24 S8 R8G8B8A8 Z24 S8 77Tiling:Combined:Width 4 4 78Tiling:Combined:Height 2 2 79Tiling:Color:Width 4 4 80Tiling:Color:Height 2 2 81Tiling:Z:Width 2 2 82Tiling:Z:Height 2 2 83Tiling:Stencil:Width 2 2 84Tiling:Stencil:Height 2 2
Bit of a text vomit from my spreadsheet but the values I'm interested in are:
Left is 98, right is XP:
1Bandwidth:Draw 28608.411 M B/s 28727.120 M B/s 2Bandwidth:BurnedByRAMDAC 238.291 M B/s 190.641 M B/s 3Bandwidth:Physical 28846.701 M B/s 28917.761 M B/s
It seems 98 is equal in this benchmark, which is OpenGL based. Could it be API specific? The "Readback: Small" stats are even weirder as 98 is much faster there. Even stranger is that ArchMark is reporting that the GPU is in PCI mode, even though I'm clearly getting AGP bandwidth, and both Nvidia CP and Powerstrip report I'm in AGP mode. Maybe it's related though?
But before I aligned the drivers to 61.xx, I ran Final Reality on 98 with 53.04 drivers and got this for my bus benchmark (XP was running the MS included 56.xx driver):
Maybe the newer driver isn't optimized for that particular benchmark? Was it misreported? I'm still at a loss here. I'm trying to find a benchmark that has an OpenGL and DirectX mode to compare. I'm trying to benchmark Half Life, since that has an OpenGL and Direct3D mode, but now it's bluescreening on launch in 98.
Its very likely that the Hybrid AGP port on this board and others like it is the culprit in Win98, its not 100% AGP but rather AGP emulated via PCIe.
I have a very similar board and it too has odd results under 98, took me a while to find out that the AGP slot isnt true AGP, it explained all the odd issues and results. I eventually just installed XP on that system as it has full PCIe support and handles the fake AGP port better.
Its very likely that the Hybrid AGP port on this board and others like it is the culprit in Win98, its not 100% AGP but rather AGP emulated via PCIe.
I have a very similar board and it too has odd results under 98, took me a while to find out that the AGP slot isnt true AGP, it explained all the odd issues and results. I eventually just installed XP on that system as it has full PCIe support and handles the fake AGP port better.
This makes me wonder if its viable to sacrifice the CPU (Wolfdale) in order to have native AGP interface...
VIA PT880 has completely normal AGP 8x slot, no hybrid. Also Windows 98 works with PCI-E just fine, at least on some boards. From personal experience at least Intel 965, G31 and X58.
In this particular case there seems to be some really weird software compatibility problem and I guess not much can be done here.
There are Intel chipset based boards that can support Core2 (even 45nm in some cases), so maybe getting one of those could help.
VIA PT880 has completely normal AGP 8x slot, no hybrid. Also Windows 98 works with PCI-E just fine, at least on some boards. From personal experience at least Intel 965, G31 and X58.
In this particular case there seems to be some really weird software compatibility problem and I guess not much can be done here.
There are Intel chipset based boards that can support Core2 (even 45nm in some cases), so maybe getting one of those could help.
98 and pcie is hit or miss, at least from my experience in getting 98 to play nice with it. Gpu has a lot to do with 98 compatibility here too with newer pcie 2/3 gpus having more problems with stability.
If I was going to do 98 and pcie again it would be hardware close in age to 98, early pcie 1 with first gen pcie gpus. I would also stick to common hardware made by the better supported fabs, gives a good mix of drivers to work with.
Not saying that 98 can’t be 100 stable with pcie I just feel it’s better suited to agp and has the most support there.
Initially when building my own system i also had a significant gap between 98 and XP. That driver i modified to detect some of the devices the new installer leaves out and this slightly older one had no erroring.
henk717wrote on 2021-12-18, 01:35:Could you try my modded driver available here? https://storage.henk.tech/Drivers/VIA_HyperionPro_V514HM.exe […] Show full quote
Initially when building my own system i also had a significant gap between 98 and XP. That driver i modified to detect some of the devices the new installer leaves out and this slightly older one had no erroring.
henk717wrote on 2021-12-18, 01:35:Could you try my modded driver available here? https://storage.henk.tech/Drivers/VIA_HyperionPro_V514HM.exe […] Show full quote
Initially when building my own system i also had a significant gap between 98 and XP. That driver i modified to detect some of the devices the new installer leaves out and this slightly older one had no erroring.
Another fun benefit for VIA fans is that the Vinyl audio driver is also integrated in the package.
Would love to know if it makes a difference or at least installs nicer.
Yes I definitely will after I get back home from the holidays. What did you mod with this driver exactly?
The Vinyl audio driver has been integrated and on Windows 98 i modded the inf files to include devices that they left out in this release. Otherwise on the T5710 it leaves missing devices that only the 4in1 driver would pick up.
While i don't expect miracles you mentioned an install error on 5.24 which i also got, 5.24 just isn't a nice driver on 98. This one should be a nice one stop shop for VIA. If it helps with the speed awesome! If it doesn't then you will at least have an installer for your VIA that doesn't bug out during the installation and that picks up all devices.
If you use this in XP the only benefit is the integrated Vinyl sound. The XP inf's did not need any modding since XP already ships with all the drivers.
What does surprice me is how high your 3Dmark2001 score is in XP. My last one is 22230 on XP with a 6900GT and without overclocking i am stuck at 1700. That CPU is really helping you out in that benchmark or my RAM is destroying my performance since i haven't figured out how to get my board to run the ram at 400mhz yet.
henk717wrote on 2021-12-20, 02:29:The Vinyl audio driver has been integrated and on Windows 98 i modded the inf files to include devices that they left out in thi […] Show full quote
henk717wrote on 2021-12-18, 01:35:Could you try my modded driver available here? https://storage.henk.tech/Drivers/VIA_HyperionPro_V514HM.exe […] Show full quote
Initially when building my own system i also had a significant gap between 98 and XP. That driver i modified to detect some of the devices the new installer leaves out and this slightly older one had no erroring.
Another fun benefit for VIA fans is that the Vinyl audio driver is also integrated in the package.
Would love to know if it makes a difference or at least installs nicer.
Yes I definitely will after I get back home from the holidays. What did you mod with this driver exactly?
The Vinyl audio driver has been integrated and on Windows 98 i modded the inf files to include devices that they left out in this release. Otherwise on the T5710 it leaves missing devices that only the 4in1 driver would pick up.
While i don't expect miracles you mentioned an install error on 5.24 which i also got, 5.24 just isn't a nice driver on 98. This one should be a nice one stop shop for VIA. If it helps with the speed awesome! If it doesn't then you will at least have an installer for your VIA that doesn't bug out during the installation and that picks up all devices.
If you use this in XP the only benefit is the integrated Vinyl sound. The XP inf's did not need any modding since XP already ships with all the drivers.
What does surprice me is how high your 3Dmark2001 score is in XP. My last one is 22230 on XP with a 6900GT and without overclocking i am stuck at 1700. That CPU is really helping you out in that benchmark or my RAM is destroying my performance since i haven't figured out how to get my board to run the ram at 400mhz yet.
Sadly this doesn't fix the speed issue but it does install correctly, and I don't get a memory conflict for a random Host Bridge device anymore. Something I discovered recently is that System Information is broken after installing drivers. Whenever I click to view any hardware information, it causes an illegal operation and closes.
Edit: I think the System Info crashes are an issue with imaging my partition with Clonezilla. Requires further investigation.
Edit 2: It was unstable timings.
Last edited by VDNKh on 2022-10-08, 04:12. Edited 2 times in total.
After a few months of on and off troubleshooting this AGP issue I've definitely narrowed the problem down to the AGP aperture/GART. For whatever reason there seems to be no AGP aperture and/or GART exposed to the display driver. When 3DMark01 polls the display driver for its aperture size (AGP Memory), it reports 0 bytes where it should report whatever it's set to in BIOS. This also fits in with the amount of performance I've lost. Because my GPU is essentially running in PCI mode with extra bandwidth, I'm getting slightly better than PCI mode performance. If I run it in true PCI mode the performance is a little worse but not the huge drop off one would expect (Nature test: 30 FPS vs 27 FPS at the beginning). This also fits in with Archmark reporting that the GPU is in fact in PCI memory mode, it was right all along. I've also verified this by swapping the VIA AGP driver with the generic MS PCI-PCI Bridge driver, there's no difference in behavior.
Inspecting the device registers, went over bit by bit and compared it to the Intel AGP 3.0 specification and VIA's PT880 Pro technical documentation. Everything is set properly, aperture is enabled. Not sure if the memory pointer is set correctly.
AGP35Compat,ViaCompat,Via4x registries in the Nvidia driver.
Replicating the registers in the Host Bridge device (where the AGP registers reside) to the PCI-PCI Bridge device (where the VIA AGP driver is installed and the GPU is connected)
Installing the VIA AGP driver on the Host Bridge device.
Installing the VxD version of the VIA AGP driver.
Forcing the chipset into AGP 2.0 mode with registers (even though VIA says it's not supported).
Comparing registers/hardware configurations to XP, no useful information as it runs in AGP 3.5 mode with Microsoft's universal AGP driver (though there are some weird differences)
Setting the register to report that the AGP registers are located in the PCI-PCI Bridge device and to ignore the Host Bridge device AGP registers. Causes Windows to fail to start up when loading the VIA AGP driver (so the driver is doing something, hmm..)
So where does that leave this? Well I want Vogon's help. I want to see what a working AGP aperture looks like on Windows 98. From the registries in the Host Bridge, PCI-PCI Bridge, and GPU, the memory resources allocated to each device by Windows, and what is polled by 3DMark01. By default, Windows 98 creates a resource conflict with the Host Bridge and the PCI-PCI Bridge. I didn't pay it much attention since just manually setting it seemed to fix it, but now I am very suspect of it. If you'd like to help please PM me or post here.
I used WPCREDIT to look at the registers, pictured is a fast way to copy it to text.
Do you use moded (beta) BIOS for your board (LINK) ?
IIRC, in RivaTuner there is VIA AGP support section, check what you can do from there.
Decrease AGP Aperture Size.
agent_x007wrote on 2022-03-19, 19:18:Do you use moded (beta) BIOS for your board (LINK) ?
IIRC, in RivaTuner there is VIA AGP support section, check what you can do […] Show full quote
Do you use moded (beta) BIOS for your board (LINK) ?
IIRC, in RivaTuner there is VIA AGP support section, check what you can do from there.
Decrease AGP Aperture Size.
Did you patch Windows 98 with any r_loew stuff ?
I have tried all of that as well. This post offers a possible solution by downgrading my BIOS, but my CPU is not compatible with that version. Plus I'm not even sure that would fix my problem, as it is a different chipset with a different AGP issue (no AGP texture support). RivaTuner reports everything is normal with AGP, which is correct, the physical interface is working as intended. I've also tried every aperture size available, and even verified that the chipset is correctly reporting it in its registers.
r_loew's VRAM patch is meant for NVidia's 82.69 driver, I use 45.23 for compatibility with the games I play. His VBIOS patcher that limits VRAM did not work either. I have used his 4GB RAM patch as well and it didn't help this issue at all.
I have tried all of that as well. This post offers a possible solution by downgrading my BIOS, but my CPU is not compatible with that version.
Yeah, one of the reasons why I stayed with a NewCastle Athlon64 was because it's supported by the oldest BIOS for my K8V-MX.
BTW, I can say with certainty that, in my particular case, it was the BIOS downgrade that solved the Win98 performance issues. How do I know? Because all the BIOS settings were the same between the two tests. And nothing was changed on the software side either (same drivers, same OS etc).
As the original poster speculates in that thread, a bug which cripples Win98 performance was likely introduced in one of the later BIOS revisions.
I can confirm that my 4coredual-sata2 also suffers from this problem. I didn't really mess with the settings much... but after simple installation of chipset drivers, directX and GPU drivers it behaves just like described earlier by VDNKh. In my case I tried Radeon 8500 and X800 XT AGP - both behaves exactly the same.
With the help of Joseph_Joestar, I've been comparing a working GART configuration on their K8V-MX with mine.
One thing that stood out to me was the xE0000000 - xEFFFFFFF allocated to Motherboard resources when it seems it should be given to the VIA Standard Host Bridge. That Host Bridge is the device that configures and exposes the AGP and GART. I tried forcing the VIA driver on that device but it did not make a difference. Considering that on a clean install of 98 this device throws a memory resource conflict, I think this is the trouble maker. I'm just not sure how.
I've also been trying to make heads and tails of the memory pointers in the registers but I'm not a computer scientist. Relevant ones are in red. The top two are listed as the "VIA Standard Host Bridge" in Windows and control AGP and GART. In red is the Base GART Low Address. The bottom 2 are PCI/PCI bridges that show up as "VIA CPU to AGP2.0/AGP3.0 Controller" in Windows (confused yet?). They don't expose any AGP configuration but the registers in red are also partially responsible for mapping the GART. I think... I hope that someone who really knows what they're looking at can help out. I've searched Google far and wide for this problem and while many report it, no solutions are provided.
A new old stock Artic NV Silencer 4 popped up on eBay which I swiftly bought. Installing it on my 5950 soon.
Loosened my memory timing a little to 3-3-3-5 after I discovered it was ever so slightly unstable in Prime95. That also fixed Microsoft's System Information from crashing. If only it fixed my AGP issues.
Last edited by VDNKh on 2022-10-08, 04:18. Edited 1 time in total.
I've also been trying to make heads and tails of the memory pointers in the registers but I'm not a computer scientist. Relevant ones are in red. The top two are listed as the "VIA Standard Host Bridge" in Windows and control AGP and GART. In red is the Base GART Low Address. The bottom 2 are PCI/PCI bridges that show up as "VIA CPU to AGP2.0/AGP3.0 Controller" in Windows (confused yet?). They don't expose any AGP configuration but the registers in red are also partially responsible for mapping the GART. I think... I hope that someone who really knows what they're looking at can help out. I've searched Google far and wide for this problem and while many report it, no solutions are provided.
Glad you managed to narrow it down. Hopefully, someone else who can make some sense of those memory mappings will come along. I'm just a simple guy with an interest in retro hardware, so troubleshooting that is a bit above my IT skill level. 😁
One thing does come to mind though. Assuming that you get proper performance under WinXP, it might be worth checking how these devices are configured under that OS in comparison to Win98.
