Had a hell of a time trying to build a Athlon 64 bit system,.. you know when things just don't go right, you start swapping hardware all over the place desperately trying to just make it work. Video of the saga below

https://www.youtube.com/watch?v=-Y-17mPxBGo

Last time I messed with socket 939 I found it was tweaky as all hell and only liked certain RAM sticks installed during the right lunar alignments.

Ozzuneoj wrote on 2026-01-23, 03:27:

I was testing some video cards today and I've been running into a really annoying problem when using Wintune97. Sometimes it lists the wrong Video Card as being in the system and I can't seem to find any way to make it list the right one. When this happens it will list some different card that I have had in the system at some point. Right now it keeps saying I have a Trident Blade3D even though I'm testing an SIS 6326 (all drivers installed and Windows lists it properly). I have tried deleting all references to the one it is detecting from the Windows registry, and I have even tried starting Wintune with the database file deleted\moved so it has to start a fresh database... it still finds the same wrong video card.

Anyone have any idea where it is finding the video card and what I can do to delete or reset this? I wouldn't even care that it gets it wrong once in a while if I could just correct it manually or have it detect it again... but I don't want benchmark runs to be tainted with incorrect specs that can't be fixed.

This is all under Windows 98SE.

The ghost in the machine.

I may have just found the slowest AGP video card I've ever come across.

It is a generic S3 Trio3D/2X that is only 4MB... and I think it is configured for only half the usual memory bus width. The card's memory is also clocked at 83Mhz, vs 100Mhz for better cards (like the Diamond Speedstar A55).

All of the other Trio3D/2X cards I have found were 8MB and had memory chips on both sides (either two on each or four on each). This one just has four on the front side with nothing on the back. Performance is ridiculously bad... Even in Wintune97, it manages only 20mp vs the 32mp I get with the Speedstar A55 or another generic 3D/2X card, both of which have two chips on each side. So I am inclined to say this one has half the memory bus.

The memory chips on the card are shorter than most SDRAM chips (not sure what this package is called) and are super generic looking. They say 9915 USA 616181-8.

Anyway, in some fairly light games it is an unusable slide show at 640x480, where as the other 3D/2X cards are not great but are totally playable.

I am going to set this gem aside, just in case I find any other contenders for the slowest AGP card. If I find a few, I may have to do some more benchmarks... 😅

(I'm not counting pro-oriented cards that are obviously way more powerful but are only barely able to run games due drivers not being meant for it.)

EDIT: Definitely something extra slow about this card. SIV was able to tell me that it has a core clock of 66Mhz and a memory clock of 83Mhz. Another generic card (8MB with 4 chips on each side) I have also has those same clocks and manages 25mp in Wintune97. So, something is making this thing 25% slower. The Speedstar A55 is clocked at 80Mhz core, 100Mhz memory, and scores 32mp by comparison... which seems about right for the difference over the 25mp card, but is nowhere near enough to account for a 60% performance improvement over the slow card.

Not really a retro activity, but something that made me smile. In out local newspaper (https://www.lenouvelliste.ch/suisse/pfas-dans … romande-1484222) the image used for this article's illustration published today is an AMD K7 😉

Minutemanqvs wrote on 2026-01-23, 06:43:

Not really a retro activity, but something that made me smile. In out local newspaper (https://www.lenouvelliste.ch/suisse/pfas- […]
Show full quote

Not really a retro activity, but something that made me smile. In out local newspaper (https://www.lenouvelliste.ch/suisse/pfas-dans … romande-1484222) the image used for this article's illustration published today is an AMD K7 😉

Haha! I love it!

Hey, if they can reduce pollutants by keeping all the K7 processors out of landfills, that's great. Not sure how much of an impact that will have, but more Slot A retro PCs and lest waste is always good. Maybe they are planning to produce Slot A motherboards to satisfy demand? (I'm kidding btw.) 😁

CharlieFoxtrot wrote on 2026-01-22, 18:16:

AppleSauce wrote on 2026-01-22, 01:43:

I've mostly been mucking about with my FM Towns computer that i acquired about a year ago, I went through a whole journey of acq […]
Show full quote

I've mostly been mucking about with my FM Towns computer that i acquired about a year ago, I went through a whole journey of acquiring parts , doing lots of repair work , especially on the Towns crt that got damaged in transit , and dealing with OS stuff.

The attachment 20260115_201120.jpg is no longer available

The attachment 20250910_162103222.jpg is no longer available

The attachment 20251103_2150482.jpg is no longer available

The attachment 20251117_211440.jpg is no longer available

The attachment 20260107_215546.jpg is no longer available

These ase seriously cool computers, it is awesome that you got it fixed. Looks stunning!

Cheers , I'm just glad its all up and running.

BitWrangler wrote on 2026-01-23, 04:43:

Last time I messed with socket 939 I found it was tweaky as all hell and only liked certain RAM sticks installed during the right lunar alignments.

HEHEHEH indeed it is that finicky! it seems the planet were not aligned correctly for me

Ozzuneoj wrote on 2026-01-23, 05:28:

I think it is configured for only half the usual memory bus width. The card's memory is also clocked at 83Mhz... I am going to set this gem aside, just in case I find any other contenders for the slowest AGP card. If I find a few, I may have to do some more benchmarks... 😅

Congratulations.

That is a challenge. If you play fair and install actual S3 Trio3D drivers, are those drivers going to push 3D rendering the CPU? Would S3 ViRGE 3D divers inflict even more pain by directing that workload to the VPU?

I do have a ViRGE DX with removable VRAM, but alas it's a PCI card so I'm out. Best of luck! 😀

Ozzuneoj wrote on 2026-01-23, 05:28:

I may have just found the slowest AGP video card I've ever come across. […]
Show full quote

I may have just found the slowest AGP video card I've ever come across.

It is a generic S3 Trio3D/2X that is only 4MB... and I think it is configured for only half the usual memory bus width. The card's memory is also clocked at 83Mhz, vs 100Mhz for better cards (like the Diamond Speedstar A55).

All of the other Trio3D/2X cards I have found were 8MB and had memory chips on both sides (either two on each or four on each). This one just has four on the front side with nothing on the back. Performance is ridiculously bad... Even in Wintune97, it manages only 20mp vs the 32mp I get with the Speedstar A55 or another generic 3D/2X card, both of which have two chips on each side. So I am inclined to say this one has half the memory bus.

The memory chips on the card are shorter than most SDRAM chips (not sure what this package is called) and are super generic looking. They say 9915 USA 616181-8.

Anyway, in some fairly light games it is an unusable slide show at 640x480, where as the other 3D/2X cards are not great but are totally playable.

I am going to set this gem aside, just in case I find any other contenders for the slowest AGP card. If I find a few, I may have to do some more benchmarks... 😅

(I'm not counting pro-oriented cards that are obviously way more powerful but are only barely able to run games due drivers not being meant for it.)

EDIT: Definitely something extra slow about this card. SIV was able to tell me that it has a core clock of 66Mhz and a memory clock of 83Mhz. Another generic card (8MB with 4 chips on each side) I have also has those same clocks and manages 25mp in Wintune97. So, something is making this thing 25% slower. The Speedstar A55 is clocked at 80Mhz core, 100Mhz memory, and scores 32mp by comparison... which seems about right for the difference over the 25mp card, but is nowhere near enough to account for a 60% performance improvement over the slow card.

While typically the maligned SiS 6326 is actually faster than most Virge class, I did find one a couple of decades ago, that for some reason was set up to run at only 40% of the nominal clock frequency. So that one was quite a dog. I did find that either rivatuner or the Rage3D tool (forgot name) did have SiS 6326 clock modding built into it by that time and it ran fine at the standard frequency. It also coped with 120% until it heat soaked, no sink on it, so maybe could have gone another 10% with a sink on it. Not sure if that one is going to turn up again. I had some half dozen 6326 "pass through" back in the early noughts, but some of them went in basic surfer boxes I was knocking out for fam and friends.

BitWrangler wrote on 2026-01-23, 15:09:

Ozzuneoj wrote on 2026-01-23, 05:28:

I may have just found the slowest AGP video card I've ever come across. […]
Show full quote

I may have just found the slowest AGP video card I've ever come across.

It is a generic S3 Trio3D/2X that is only 4MB... and I think it is configured for only half the usual memory bus width. The card's memory is also clocked at 83Mhz, vs 100Mhz for better cards (like the Diamond Speedstar A55).

All of the other Trio3D/2X cards I have found were 8MB and had memory chips on both sides (either two on each or four on each). This one just has four on the front side with nothing on the back. Performance is ridiculously bad... Even in Wintune97, it manages only 20mp vs the 32mp I get with the Speedstar A55 or another generic 3D/2X card, both of which have two chips on each side. So I am inclined to say this one has half the memory bus.

The memory chips on the card are shorter than most SDRAM chips (not sure what this package is called) and are super generic looking. They say 9915 USA 616181-8.

Anyway, in some fairly light games it is an unusable slide show at 640x480, where as the other 3D/2X cards are not great but are totally playable.

I am going to set this gem aside, just in case I find any other contenders for the slowest AGP card. If I find a few, I may have to do some more benchmarks... 😅

(I'm not counting pro-oriented cards that are obviously way more powerful but are only barely able to run games due drivers not being meant for it.)

EDIT: Definitely something extra slow about this card. SIV was able to tell me that it has a core clock of 66Mhz and a memory clock of 83Mhz. Another generic card (8MB with 4 chips on each side) I have also has those same clocks and manages 25mp in Wintune97. So, something is making this thing 25% slower. The Speedstar A55 is clocked at 80Mhz core, 100Mhz memory, and scores 32mp by comparison... which seems about right for the difference over the 25mp card, but is nowhere near enough to account for a 60% performance improvement over the slow card.

While typically the maligned SiS 6326 is actually faster than most Virge class, I did find one a couple of decades ago, that for some reason was set up to run at only 40% of the nominal clock frequency. So that one was quite a dog. I did find that either rivatuner or the Rage3D tool (forgot name) did have SiS 6326 clock modding built into it by that time and it ran fine at the standard frequency. It also coped with 120% until it heat soaked, no sink on it, so maybe could have gone another 10% with a sink on it. Not sure if that one is going to turn up again. I had some half dozen 6326 "pass through" back in the early noughts, but some of them went in basic surfer boxes I was knocking out for fam and friends.

Oh man... yeah, that would be a real dog.

Speaking of SiS 6326, I had a really crazy experience with one several years ago. I won't spoil it... it's a funny story. 🤣

Also, since we're on the topic... I tested four SiS 6326 cards and I gotta say, what the heck is up with those things? Every single one of them displays very obvious "ghost" lines that are visible, for example, when moving a window around with a flat desktop background. Not sure if there is an exact name for this, but it isn't like a graphical artifact... it is an output issue, like there is poor filtering or something bleeding across to where it shouldn't be. Any dark horizontal lines\stripes (like the title bar on a window or a line of black text) cause a sort of shadow that is parallel to it, spans the entire width of the display and moves with the window.

The cards work "fine"... though there was a lot of locking up when trying to start or exit games... it's just the video output is really bad. I thought it was just one or two, but all four of the cards I tested do it to some degree. Oddly, the cheapest looking, most generic cards have much less visible ghosting, while my brand name Diamond cards have it the worst.

This also isn't an issue of the RAMDAC being pushed too hard. It does it at any resolution and any refresh rate. Switching to any other card makes it go away, but all four SiS 6326 cards exhibit it. Even the most generic Trio3D/2X cards had basically flawless output in comparison, which is really saying something.

I installed Vista 32-bit in Virtual PC 2007 so I could run r3dfox within a Windows 2000 host. I started by installing SP1. I tried a different ISO which had updates integrated but that didn't work. These huge 5+GB ISOs don't work in VPC07. It will mount, but if you actually examine file contents you'll see gibberish. I extracted all of the files out of an offline updates ISO on the host side, and after I installed SP1 in the VM I accessed the updates through a network drive. The first install was about 6GB, after updates it exploded to ~13GB, 🤣.

Having fun with my ST6-RAID. Man, this board REALLY is the cream of the crop of Tualatin boards. Yes, maybe the SiS 635 exists but I genuinely care less when this thing runs on steam on stock voltage alone.

Well, another day, another bunch of old video cards to test. This time I'm working on a pile of Intel i740 cards, and I have run into something strange. Several cards, several different manufacturers, all of them work perfectly... except that at 1280x1024@85Hz I get these lines on the display that move around with whatever window or GUI element they are connected to (notice the placement of them next to windows):

They all do this at this specific res\refresh rate. There are no artifacts of any kind at any other resolution or refresh rate. 1280x1024@75Hz is fine, 1024x768@85Hz is fine, 1600x1200(256color)@75Hz is fine. Oddly enough, the Intel branded card I have (like this one) seemingly didn't have the issue at 85Hz, but it's because the card would not actually switch to 85Hz. When I select that refresh rate it just goes to 75Hz.

Is this just my wonky test system? Can anyone else replicate this? I am using the PV 40 drivers from this package under Windows 98SE:
https://vogonsdrivers.com/getfile.php?fileid=330&menustate=0

Seems like an odd issue either way. Everything I can find online says that the i740 should do 1280x1024@85Hz, and the drivers seem to let it happen. I do not intend to use that res since I have 4:3 monitors and it is a 5:4 res (bleh!) but I wouldn't mind some validation if others can replicate this.

Also, I just want to throw out there that these cards are really underappreciated these days. Going from the SiS 6326 and Trio3D/2X to the i740 is an insanely huge improvement in gaming performance. The old Jedi Knight demo I was using for testing runs better on the i740 at 1024x768 with colored lighting on than it did on the other cards at 640x480 with colored lighting off. i740 may not have been a Voodoo 2 killer, but it is absolutely nowhere near the bottom of the 3D accelerator stack.

Ozzuneoj wrote on 2026-01-26, 16:55:

They all do this at this specific res\refresh rate.

Just an idea after reading about the issues with the 6326s too, were you able to reproduce these with a different monitor?

konc wrote on 2026-01-26, 17:12:

Ozzuneoj wrote on 2026-01-26, 16:55:

They all do this at this specific res\refresh rate.

Just an idea after reading about the issues with the 6326s too, were you able to reproduce these with a different monitor?

Thanks for the input. This is a very different kind of issue even though it sounds similar when I describe it. I will have to get a picture of what the 6326 cards are doing to show the difference.

With the i740 cards it is like pixels are being drawn in the wrong place... a digital issue, I guess you could say... which would happen before the video output since it is a CRT.

With the 6326 cards it is a very analog looking problem... like something you'd expect to see if you had a bad cable, or the cable was too close to a source of EMI or something... but it only does it on specific (crappy) cards that all use the same (crappy) chip, and it doesn't seem to matter what cable is used.

I am currently unable to test the i740 issue on a different display since I don't have any others in my office that will do 1280x1024@85Hz over VGA.

Jeff Chen'd my Voodoo 2 cards today.

Nice prints and nice fit. They add an extra layer of protection around the ICs and still leave room for heatsinks / airflow for cooling.

More Tualatin fun

More video card testing...

Tested a bunch of Virge cards today. Much like the Trio64V+, it seems that there are massive swings in performance from one Virge DX or GX card to another. I'm guessing it is as least partially due to the EDO being configured for 1-cycle or 2-cycle mode. I see that over on the VGA Museum there is mention of cards being in 1-cycle mode, but this is one of the few times I have seen anyone mention this.

I find it kind of funny that this isn't something that is ever talked about. People will plan entire builds around 10% performance differences between graphics chips on a benchmark chart, and then most will apparently just get any card with that chip and toss it in their retro PC, not realizing that there can be a 300% swing (that is not a typo) in benchmark performance from one nearly identical card to another. I bet a huge percentage of purpose-built retro systems running S3 (or other?) cards from this era have massively gimped performance because they happened to get a card that is set up for 2-cycle EDO.

Anyway... after dealing with those, I tested a couple of Virge VX and a Vision 968. All three are Diamond branded cards, and, surprisingly, all of them have problems. Two are cosmetically flawless... so unless there is some minuscule broken solder joint somewhere, all I can think is that they have defective memory? Has anyone else noticed S3 VRAM-based cards being unreliable overall? I know that 3D performance and compatibility is absolutely horrid on the Virge VX due to this card being based on the original Virge (as opposed to the DX and GX which are massively improved), but it seems odd that I'd be having artifacts on the desktop and various other problems with so many similar but fairly expensive (for the time) cards.

One Virge VX (2+2MB) has graphical corruption on the bottom 3rd of the screen at higher resolutions. The other VX (2MB) does not, but it has pixelated artifacts in other areas, including an arrant pixel that follows the mouse cursor around. Weirdly, the Vision 968 (2+2MB) has almost the same corruption as the first VX covering the bottom 3rd of the screen at higher resolutions.

I noticed a post here where kixs says he has had a lot of problems with Vision 968 cards as well. Just wondering if anyone else has had a similar experience. In comparison, I have recently tested a few dozen other S3 cards (Trio64V+, Virge, DX, GX, Trio3D, etc.) that do not use VRAM and I may have had one or two that were defective, but it was due to physical damage. Bad RAM or un-fixable stability\artifacting issues were very rare with other cards from this period.

I am tweaking around my new Pentium III system its nearly finished. It's running windows millennium and DirectX 7. I swapped out the ram and PSU for spares because of system freezing showing display corruption but all the problem was I enabled an AGP function in BIOS that caused it. I learned if I have more than 2 gigabytes of RAM windows 9x will not utilise it so I have decided to just install 2 gigabytes than 4. I intend to mod Windows Me to enable boot menu that includes boot to DOS so I may run late DOS games in Millennium DOS. It will be interesting to see if the motherboard built in creative sound chip supports DOS.
Also I may completely disable Windows Millenniums system restore feature and replace it with the windows 98SE SFC Utility this is done by script and enables the ability to skin the taskbar and give some resources and a slight improvement in performance.
The Motherboard:
https://theretroweb.com/motherboards/s/gigabyte-ga-6rx

The 1,000 MHz CPU:
https://theretroweb.com/family/272

The GPU
https://theretroweb.com/expansioncards/s/prol … vg18a-dvi-128mb