There was a similar problem here: https://www.vcfed.org/forum/forum/genres/pent … -win98se-driver that remained unresolved.
Maybe it is actually a resource / IRQ conflict of some sort... Seeing how this revision of cards is a fair bit newer, and maybe a little less compatible with the 98 drivers ment for the 1st generation of G450.
That would certainly seem to tie in with the whole Bus Mastering thing

Also, any suggestions would be most welcome, on how to make Win98 work with Bus Mastering enabled, so there won't be a performance hit

By changing VIA motherboard to Nvidia or SIS.

The Serpent Rider wrote on 2022-02-01, 14:51:

Also, any suggestions would be most welcome, on how to make Win98 work with Bus Mastering enabled, so there won't be a performance hit

By changing VIA motherboard to Nvidia or SIS.

Noted!

Thanks! 😁

I can confirm that this card is useless on a VIA chipset. I had to disable Bus Mastering to get it to work without corruption in Windows XP and the performance tanked. I know this, because I was able to get 3DMark 99 to run once or twice before the corruption started and was getting around 50fps. After changing this setting, performance dropped to about 25fps. This is on a fast 3Ghz Pentium 4.

After disabling Bus Mastering, OpenGL also broke completely with Quake, but seems to still work with Quake 2.

I'm curious if anyone has any experience with these cards on Socket7 PCs.
As these cards could be a good candidate for a a lower spec system with no AGP, like a Pentium MMX ....Especially if you want DVI output!

A Pentium 233 is what I was testing it with; I had the same idea. While it did work, driver overhead negatively affected performance. Performance was sometimes significantly worse than as Rage XL in OpenGL, such as in Quake 2. DirectX performance was "okay" in Incoming, but lagged quite terribly in Unreal, so it can't take the place of, say, a Banshee or Voodoo 3. It's the CPU which was definitely holding it back. Maybe it would be fine for 2D, but 3D was disappointing. On the positive side, it does not require 3.3V on the PCI bus to work.

The card is also completely useless on a VIA chipset, where it fails to behave properly with Bus Mastering enabled. Turning it off causes performance to half.

It also does not support drivers for Windows 3.11, unlike the G200 and G400.

@Kahenraz
So you mean to say you tested with via AND other chipsets? And on the other chipsets, although Bus Mastering was enabled, and performed twice as fast as on the VIA, ...still, performance was bad in 3D ????

If so, did you also play around with different drivers for the card? (on the non-VIA chipsets i mean)

Was the problem consistent when trying the card in all of the pci slots?

I tested it on an Intel 430TX where I confirmed that it worked in a PCI slot with no 3.3V. It was Windows 98 and ran without problems but was too slow when paired with a weak 233 Mhz processor after driver overhead.

I then tested it on a fast socket 775 Pentium 4 with some VIA chipset and Windows XP. I could get drivers to install and sometimes I could get one benchmark or game to open before the video would go corrupt, that's how I saw OpenGL working in Quake and 50 fps in 3DMark 99.

Eventually I couldn't even boot without the system locking up when it reached the desktop. Changing PCI slots didn't help. I finally got back to the desktop again without locking up by removing all other addon cards. I then disabled Bus Mastering, which made everything stable. But performance was halved in 3DMark 99 and OpenGL was broken in Quake 1 (black screen); although OpenGL did work in Quake 2.

VIA chipset drivers were installed. I did not try testing other BIOS options to see if that made a difference.

I also updated the card's BIOS to the latest 2.1 from the stock 1.6. I also tried 2.0 from an IBM update. The lockups were consistent among all of these.

I have had other Matroxen, Matroxii, Matroxuses, G400 and G200 running on Palamino/VIA/DDR platform in past so not sure there is any great antagonism between socket A VIA and Matrox. Though I have no idea what driver I ran, current ones off site in 2003 probably.

The G450 seems to be a PCIe internally. You can see the large PLX bridge chip near the PCI edge connector and the installation of a PCI-to-PCI bridge driver when it's first detected. This is most likely the cause of the Bus Mastering issue.

If it was just a faster G400 on a PCI card with DVI then it would be amazing. Sadly, this is not the case. And since it is not reliable on VIA chipsets, its usefulness is limited.

Thank you @Kahenraz for all this invaluable information!
At least now we/I know more about this card, and what not to bother with...
vbios was definitely on my list 😀

I dug around and found some benchmark results that I did a few years back on some MATROX cards.
So just for a comparison, on my AthlonXP 2000+ with 256MB DDR, and WinXP;
my results in 3DMark2000 for AGP vs PCI versions, are:
G400 AGP - card#1 = 3149
G400 AGP - card#2 = 2872
G450 AGP - card#1 = 2494
G450 AGP - card#2 = 2257
G450 PCI - card = 1719 (& 1254 in Win98SE with Bus Mastering disabled in the driver)

Yeah.. Not a great performer even on a fast CPU. Maaaybe(?) almost on par with a TNT2 M64 PCI...

But at least it may be decent for some of those office types Pentium II/III machines, with no AGP... Or for a (non-VIA) Sk7 mobo with a K6/2 400mhz on FSB66.
Maybe paired with a voodoo2 with no pass-through cable and use a dual input LCD (DVI in matrox & VGA in voodoo2). That could be cool!..

Also because of it requiring only 5V, it could work even on a 486!.. so that's also cool!
That may make it the ONLY option for pulling digital image out of a 486.

If used as a replacement for something like an S3 or basic 2D card, it might be worth considering as an upgrade, even if the 3D features are too taxing on the CPU and are ignored.

A major downside for me in a 486 is that there are no Windows 3.1 drivers. I feel like that's a definite hallmark of the era. Although Windows 95 would work just as well.

Another thing worth mentioning is that,while I did not experience any driver issues with my Pentium MMX, there's always a chance that it requires instructions that are not present on a 486. Although I do seem to recall reading elsewhere of someone using it in their 486.

There is actually a G200 PCI which can have an optional flat panel connector attached to it. I have one of these but found it to not be DVI but something else. I think it's some kind of VESA connector; I've researched it before but don't recall the details offhand. I do have an adapter for it, but my LCD can't sync to it. It might require either a CRT or something designed to work with this early standard.

Unfortunately, it seems to only output through the flat panel connector when attached, so I can't get to the desktop to install drivers to try a different resolution or refresh rate.

I have been experimenting with the G450 for a while and cannot find any changes in the BIOS to make it stable on a VIA chipset. Changing the resolution also has the risk of the frequency going out of range and unable to sync with my LCD, especially when trying to make it work in dual-head mode.

So far, it has been very disappointing.

Kahenraz wrote on 2024-03-15, 01:56:

The G450 seems to be a PCIe internally. You can see the large PLX bridge chip near the PCI edge connector and the installation of a PCI-to-PCI bridge driver when it's first detected. This is most likely the cause of the Bus Mastering issue.

It's not PCIe, you skipped a generation. The G450 is AGP internally, and the bridge chip (by HiNT, not by PLX) indeed causes various compatibility issues. For example, that bridge seems to not correctly forward PCI configuration cycles at a bus clock of 20MHz. I recently tried a G450 in an SiS5571 motherboard (a cheap Pentium chipset that is able to run with single memory modules by making all memory accesses 32-bit accesses). It crashes in Windows 2000 if the board runs at 32 bit memory width, but it worked OK with completely filled banks (64 bit memory bus).

My mistake. It's the G550 where Matrox started using PCIe. However, my G450 does use a PLX branded chip.

mkarcher wrote on 2024-03-15, 17:47:

I recently tried a G450 in an SiS5571 motherboard (a cheap Pentium chipset that is able to run with single memory modules by making all memory accesses 32-bit accesses). It crashes in Windows 2000 if the board runs at 32 bit memory width, but it worked OK with completely filled banks (64 bit memory bus).

I dunno if I forgot I heard that before or it's news... anyway, another thing to add to the "Weird stuff to try" list, run 6x86 or P54 on 2x50 on SiS5571 with single module and see if performance compares to 5x86-100 or POD-100.

Kahenraz wrote on 2024-03-15, 16:12:

There is actually a G200 PCI which can have an optional flat panel connector attached to it. I have one of these but found it to […]
Show full quote

There is actually a G200 PCI which can have an optional flat panel connector attached to it. I have one of these but found it to not be DVI but something else. I think it's some kind of VESA connector; I've researched it before but don't recall the details offhand. I do have an adapter for it, but my LCD can't sync to it. It might require either a CRT or something designed to work with this early standard.

Unfortunately, it seems to only output through the flat panel connector when attached, so I can't get to the desktop to install drivers to try a different resolution or refresh rate.

20240315_121403.jpg

20240315_121424.jpg

20240315_121440.jpg

If i'm not mistaken, that's not normal DVI... It's longer, and looks to be the same as the "DVI" from the Voodoo3 3500.
On the v3 3500, I was able to output VGA signal with this DVI M1-DA (30+5) pin cable, that's used for video projectors:

I've no clue how to pull digital signal in the case of the G200 though..

Kahenraz wrote on 2024-03-15, 17:57:

My mistake. It's the G550 where Matrox started using PCIe. However, my G450 does use a PLX branded chip.

20240315_135449.jpg

So, that settles the HiNT vs. PLX question. In fact, HiNT got bought by PLX in 2003. The G450 PCI DVI I recently had at hand uses a HiNT branded PCI/PCI bridge. To get it run stable at 40MHz (e.g. for an 5x86 at 160MHz), I needed to add a heatsink to that bridge chip.

On the other hand, you are definitely correct that there are G550 cards with PCI Express. The G550 is prominently known to exist as PCIe x1 card, so you don't require a big PCIe x16 slot for that kind of card, which makes it a great "secondary display" card. See for example this announcement https://www.hartware.de/2005/07/13/matrox-ann … -graphics-card/ .

The PCIe version of the G550 is using a bridge chip, though. The G550 is an AGP chip, just as the G450. This is a high quality image of the low-profile version of the G550 PCIe at vgamuseum.info. You can clearly identify the TI XIO2000A x1 PCI Express® to PCI bus translation bridge. I found vendor photos for the non-low-profile variant, which have the earlier XIO2000 instead of the XIO2000A, so it seem Matrox switched to the successor chip at some time.