VOGONS


First post, by Sphere478

User metadata
Rank l33t++
Rank
l33t++

I have been looking for one of these cards since I think probably like 2005 and I finally found one. I’m really curious if it ends up being faster than a radeon 7500. Or radeon 9250

Stay tuned. 😀 I’m gonna find out on my all pci /isa systems. 😀

In the meantime post your thoughts and expectations

Also, think I might be able to flash it to a fire gl of some type?

Sphere's PCB projects.
-
Sphere’s socket 5/7 cpu collection.
-
SUCCESSFUL K6-2+ to K6-3+ Full Cache Enable Mod
-
Tyan S1564S to S1564D single to dual processor conversion (also s1563 and s1562)

Reply 1 of 21, by douglar

User metadata
Rank l33t
Rank
l33t

I never tested a radeon 9100, but I had difficulties getting either a PCI radeon 7500 or a 9250 to work reliably with my K6-III board, probably because I was using the AT Power connectors P8 & P9, so it was potentially a 3.3V power issue.

I'd say my best "non-geforce" PCI benchmark scores came from a PCI Voodoo 3 with SGRAM, although my PCI Banshee pulled a head by a bit in 3dMark 99. Maybe 3dMark 99 weights single-texturing higher than multi texturing?

Here are my benchmarks--
download/file.php?id=229021&mode=view

The 9100 should perform pretty close to the 8500le, yes? It should do pretty well. About 2 months ago, I tested my K6-III with an AGP Radeon 8500 and it did quite well.

Here are some other fish to look for--

  • I have in my notes that ATI made a Radeon 9650 PCI. That's quite the white whale. Might be "apple only", but maybe it could be flashed to PC? Or get one of the PCI-e versions and put it in an adapter?
  • Visiontek made a Radeon X1300PRO & X1550 PCI. That's easier to find if you are willing to move to XP.
  • The AGP-PCI bridged Matrox G450 card I got was disappointing. Not worth hunting for unless you just want to double check my results.

Reply 2 of 21, by douglar

User metadata
Rank l33t
Rank
l33t

I’d add that while the 9100 should do pretty well, it probably needs a faster cpu to make much of a difference. A 2gh P4 was entry level gaming when that gpu came out. With a K6-3, it might be a bit slower than a 3dfx or a geforce mx card just because the driver was optimized for a system with more cpu.

Reply 3 of 21, by BitWrangler

User metadata
Rank l33t++
Rank
l33t++
douglar wrote on 2026-01-08, 18:25:
[…]
Show full quote
  • I have in my notes that ATI made a Radeon 9650 PCI. That's quite the white whale. Might be "apple only", but maybe it could be flashed to PC? Or get one of the PCI-e versions and put it in an adapter?
  • Visiontek made a Radeon X1300PRO & X1550 PCI. That's easier to find if you are willing to move to XP.
  • The AGP-PCI bridged Matrox G450 card I got was disappointing. Not worth hunting for unless you just want to double check my results.

I may have heard whispers of a 9550 PCI but not a 9650, as well it doesn't seem to fit the budget card numbering used at the time.

Unicorn herding operations are proceeding, but all the totes of hens teeth and barrels of rocking horse poop give them plenty of hiding spots.

Reply 4 of 21, by gonzo

User metadata
Rank Member
Rank
Member
douglar wrote on 2026-01-10, 12:23:

With a K6-3, it might be a bit slower than a 3dfx or a geforce mx card just because the driver was optimized for a system with more cpu.

Exactly what I mean, too.

I LOVE CPUs RUNNING IN [GonzoHz]

Reply 5 of 21, by douglar

User metadata
Rank l33t
Rank
l33t
BitWrangler wrote on 2026-01-10, 14:41:
douglar wrote on 2026-01-08, 18:25:
[…]
Show full quote
  • I have in my notes that ATI made a Radeon 9650 PCI. That's quite the white whale. Might be "apple only", but maybe it could be flashed to PC? Or get one of the PCI-e versions and put it in an adapter?
  • Visiontek made a Radeon X1300PRO & X1550 PCI. That's easier to find if you are willing to move to XP.
  • The AGP-PCI bridged Matrox G450 card I got was disappointing. Not worth hunting for unless you just want to double check my results.

I may have heard whispers of a 9550 PCI but not a 9650, as well it doesn't seem to fit the budget card numbering used at the time.

There may have been a PCI version of this card---
https://theretroweb.com/expansioncards/s/unknown-radeon-9650

Reply 6 of 21, by BitWrangler

User metadata
Rank l33t++
Rank
l33t++

That's a 9600 though, says so on it, no idea why it would be titled as 9650

Edit: likewise the apple land 9650s come back as 9600 Pros when you compare the board image. Some Apple or 3rd party nonsense that 9650 designation not an ATI one.

By the way I was trying to dig out possible 9550 PCI and came across this on Deb's site (anyone seen her lately?)
https://www.gpuzoo.com/Compare/ATI_Radeon_925 … TI_Radeon_9550/

Which would seem to imply there was little point in a 9550 PCI when it was about level with a 9250 PCI in AGP form, except the memory, and I don't tend to see good/fast/wide memory going on PCI versions of cards, so it probably would have got the same handicap as 9250 PCI. So 9250 PCI might be as good as it gets in that timeframe.

Unicorn herding operations are proceeding, but all the totes of hens teeth and barrels of rocking horse poop give them plenty of hiding spots.

Reply 7 of 21, by douglar

User metadata
Rank l33t
Rank
l33t
BitWrangler wrote on 2026-01-10, 15:19:

That's a 9600 though, says so on it, no idea why it would be titled as 9650

Hard to say. I’ll see what I can find out. I suspect the card has the r351 110nm die shrink of the 9600 chip.

I could be mistaken about the whole pci 9650 thing. I found in my notes an annotation that I saw an oem 9650 pci when making this list last summer. http://www.vogonswiki.com/index.php/ATI#R300 Sometimes work requires that I stay up late at night and I just start scanning through lists of video cards to stay awake. I take notes, but my recollection of all the cards I looked at isn’t always so clear the next day. So it could have been a typo. I didn’t record a pcb number.

Reply 8 of 21, by gonzo

User metadata
Rank Member
Rank
Member

@ Sphere478: just an idea how to make the test (the comparision) more easy - you can use an AGP-VGA instead of an PCI-model. They are lots of Super-Socket-7-mainboards having an AGP-slot, and the Radeon 9100 AGP was by far much more on the market compared to the PCI-version. The same applies for the AGP-versions of the Geforce -VGAs compared to their PCI-models.
By the way - a Radeon 9800 AGP (having a 3,3 Volt-key for the AGP-voltage) can be used with a Super-Socket-7-system, too. Of course, like all other AGP-VGAs onto a socket-7-board with AGP, this 9800 will run at a maximum of AGP 2x-speed.

I LOVE CPUs RUNNING IN [GonzoHz]

Reply 9 of 21, by Ozzuneoj

User metadata
Rank l33t
Rank
l33t
douglar wrote on 2026-01-10, 15:11:

There may have been a PCI version of this card---
https://theretroweb.com/expansioncards/s/unknown-radeon-9650

The label seems to pretty clearly say 9600, even if it is smudged a bit.

BitWrangler wrote on 2026-01-10, 15:19:
That's a 9600 though, says so on it, no idea why it would be titled as 9650 […]
Show full quote

That's a 9600 though, says so on it, no idea why it would be titled as 9650

Edit: likewise the apple land 9650s come back as 9600 Pros when you compare the board image. Some Apple or 3rd party nonsense that 9650 designation not an ATI one.

By the way I was trying to dig out possible 9550 PCI and came across this on Deb's site (anyone seen her lately?)
https://www.gpuzoo.com/Compare/ATI_Radeon_925 … TI_Radeon_9550/

Which would seem to imply there was little point in a 9550 PCI when it was about level with a 9250 PCI in AGP form, except the memory, and I don't tend to see good/fast/wide memory going on PCI versions of cards, so it probably would have got the same handicap as 9250 PCI. So 9250 PCI might be as good as it gets in that timeframe.

I have one of these Apple 109-a58503-20 cards and I just pulled the heatsink off to see what it says. It says 9600 Pro on the die. Also, I used digital calipers and measured the die at ~8.86x8.86mm with is roughly the same as the 76mm^2 reported for the RV350 ... for what that's worth.

The attachment 20260112_125636.jpg is no longer available

This particular card doesn't have the extra pins for the Apple Display power (pretty sure that's what they are for), but I have not tried it on a PC yet. I have a feeling it will show up as a 9600 Pro though if it works.

It seems that in Apple land, specificity of hardware specs and naming is\was not normally a thing. It'd basically be a "Radeon Ninety-six-whatever.... mid range model." So, it could be a 9650, 9600, Pro, XT, whatever and so few people cared or bothered to dig into it any further that there are effectively no records about the hardware online... you find the same card listed as all of those things (some with or without the extra pins) and as far as I can tell, not a single die shot of any of them to confirm what it is. I found one discussion here, but again, no one seems to have cared enough to document any more than what is said there.

On the other hand, in the land of PCs it is recognized that there is a pretty substantial difference between a "9600" and a "9600 XT", and the existence of a 9650 at all would warrant serious discussion among enthusiasts and collectors (see this thread as proof, 🤣 ).

All that said, a Radeon 9500\9600 series of any kind in PCI would be pretty interesting, though the bottleneck gets significant at that point and if it ended up being a 64bit card like a 9600SE or 9550, it may or may not actually be faster than the 9100 in all situations. I personally find the 9100 interesting because it was the last time anyone bothered sticking a flagship GPU of any generation on a PCI card (even if R200 was previous gen when the 9100 came out).

(More pics of the Apple card.)

The attachment 20260112_142346 (Custom).jpg is no longer available
The attachment 20260112_142401 (Custom).jpg is no longer available
The attachment 20260112_142427 (Custom).jpg is no longer available

Now for some blitting from the back buffer.

Reply 10 of 21, by douglar

User metadata
Rank l33t
Rank
l33t

Yours sure looks like an rv350.

Here’s the pics collected so far—-

https://theretroweb.com/chips?chipManufacture … ge=24&name=9600

There was an rv351

https://www.techpowerup.com/1036/rv351-info

I have a mac 9600xt at home that looks a little different 109-A58503-20 vs 109-A13600-10

https://theretroweb.com/expansioncards/s/ati- … n-9600xt-mac-g5

Mean while… when do we get to see the radeon 9100 benchmarks?

Reply 11 of 21, by Sphere478

User metadata
Rank l33t++
Rank
l33t++
douglar wrote on 2026-01-08, 18:25:
I never tested a radeon 9100, but I had difficulties getting either a PCI radeon 7500 or a 9250 to work reliably with my K6-III […]
Show full quote

I never tested a radeon 9100, but I had difficulties getting either a PCI radeon 7500 or a 9250 to work reliably with my K6-III board, probably because I was using the AT Power connectors P8 & P9, so it was potentially a 3.3V power issue.

I'd say my best "non-geforce" PCI benchmark scores came from a PCI Voodoo 3 with SGRAM, although my PCI Banshee pulled a head by a bit in 3dMark 99. Maybe 3dMark 99 weights single-texturing higher than multi texturing?

Here are my benchmarks--
download/file.php?id=229021&mode=view

The 9100 should perform pretty close to the 8500le, yes? It should do pretty well. About 2 months ago, I tested my K6-III with an AGP Radeon 8500 and it did quite well.

Here are some other fish to look for--

  • I have in my notes that ATI made a Radeon 9650 PCI. That's quite the white whale. Might be "apple only", but maybe it could be flashed to PC? Or get one of the PCI-e versions and put it in an adapter?
  • Visiontek made a Radeon X1300PRO & X1550 PCI. That's easier to find if you are willing to move to XP.
  • The AGP-PCI bridged Matrox G450 card I got was disappointing. Not worth hunting for unless you just want to double check my results.

X1300 pci I have. It posts but drivers dont work.

I think I found some obscure x300/x500 something based pci card that was like a firefl or workstation thing. Tried it, couldn’t get it working. And the video connectors were terrible also. So no go.

7000,8000,9000 seem to be the series cards that work the best.

A 9650 pci is very intriguing

I have a parhelia, it is about on par with a 7000/7500 if I recall

G450 is nice, but nothing special

douglar wrote on 2026-01-10, 12:23:

I’d add that while the 9100 should do pretty well, it probably needs a faster cpu to make much of a difference. A 2gh P4 was entry level gaming when that gpu came out. With a K6-3, it might be a bit slower than a 3dfx or a geforce mx card just because the driver was optimized for a system with more cpu.

I have ran a 9800pro in k6-3 and cyrix mii systems. It made a difference. There is still performance on the table it seems above 9250

gonzo wrote on 2026-01-12, 16:59:

@ Sphere478: just an idea how to make the test (the comparision) more easy - you can use an AGP-VGA instead of an PCI-model. They are lots of Super-Socket-7-mainboards having an AGP-slot, and the Radeon 9100 AGP was by far much more on the market compared to the PCI-version. The same applies for the AGP-versions of the Geforce -VGAs compared to their PCI-models.
By the way - a Radeon 9800 AGP (having a 3,3 Volt-key for the AGP-voltage) can be used with a Super-Socket-7-system, too. Of course, like all other AGP-VGAs onto a socket-7-board with AGP, this 9800 will run at a maximum of AGP 2x-speed.

I have other builds for the agp stuff. This is trying to find the fastest for my pci only socket 7 builds

douglar wrote on 2026-01-13, 00:53:

Mean while… when do we get to see the radeon 9100 benchmarks?

Card arrived today. Computer however is in storage rofl. And there are some other priorities. So not any time soon 🤣.

But in theory, eventually.

I should be able to post better pics of it in the next week.

Its at my friends house and I think he may have a cyrix computer we might be able to bench it in. Will see if that’s possible. But my retro collection is currently in disarray because of moving activities haha.

Sphere's PCB projects.
-
Sphere’s socket 5/7 cpu collection.
-
SUCCESSFUL K6-2+ to K6-3+ Full Cache Enable Mod
-
Tyan S1564S to S1564D single to dual processor conversion (also s1563 and s1562)

Reply 12 of 21, by dm-

User metadata
Rank Member
Rank
Member

the worst thing about radeon cards - they does not like to work on non/standard agp/pci bus freq.

this is why i prefer to stay with geforce pci cards.

btw r350 has same pinout like 9250 chips. one day i'll do the chip swap on some random radeon 9250 pci, but really not soon.

Reply 13 of 21, by tehsiggi

User metadata
Rank Member
Rank
Member
dm- wrote on 2026-01-13, 05:56:

the worst thing about radeon cards - they does not like to work on non/standard agp/pci bus freq.

this is why i prefer to stay with geforce pci cards.

btw r350 has same pinout like 9250 chips. one day i'll do the chip swap on some random radeon 9250 pci, but really not soon.

It has the same ball arrangement, but not the same pinout.

RV250 (Radeon 9200 & 9250)

The attachment Screenshot 2026-01-13 at 11.05.46.png is no longer available

RV350 (Radeon 9600)

The attachment Screenshot 2026-01-13 at 11.05.58.png is no longer available

AGP Card Real Power Consumption
AGP Power monitor - diagnostic hardware tool
Graphics card repair collection

Reply 14 of 21, by NeoG_

User metadata
Rank Member
Rank
Member
dm- wrote on 2026-01-13, 05:56:

the worst thing about radeon cards - they does not like to work on non/standard agp/pci bus freq.

From the testing I've seen they also have the highest overhead which is a thing on a CPU limited platform like SS7

98/DOS Rig: BabyAT AladdinV, K6-2+/550, V3 2000, 128MB PC100, 20GB HDD, 128GB SD2IDE, SB Live!, SB16-SCSI, PicoGUS, WP32 McCake, iNFRA CD, ZIP100
XP Rig: Lian Li PC-10 ATX, Gigabyte X38-DQ6, Core2Duo E6850, ATi HD5870, 2GB DDR2, 2TB HDD, X-Fi XtremeGamer

Reply 15 of 21, by Sphere478

User metadata
Rank l33t++
Rank
l33t++

Not going to have any time to play with this for a while probably. Lmk if anyone sees any damage.

Excited! 😆

Hope to break some personal pci-only socket 7 records!

Sphere's PCB projects.
-
Sphere’s socket 5/7 cpu collection.
-
SUCCESSFUL K6-2+ to K6-3+ Full Cache Enable Mod
-
Tyan S1564S to S1564D single to dual processor conversion (also s1563 and s1562)

Reply 16 of 21, by Repo Man11

User metadata
Rank l33t
Rank
l33t

Assuming that you'll be testing it on something like a TXP4 I'll be curious to see how it compares to this.

After watching many YouTube videos about older computer hardware, YouTube began recommending videos about trains - are they trying to tell me something?

Reply 17 of 21, by Sphere478

User metadata
Rank l33t++
Rank
l33t++

Not sure, but maybe one or both of these.
tyan s1564 or epox mvp4 when I get around to it.

The mvp4 is a pci only and can do a high FSB

The tyan will be 66mhz fsb only.

MVP4 socket 7 aiming for the stars!

Sphere's PCB projects.
-
Sphere’s socket 5/7 cpu collection.
-
SUCCESSFUL K6-2+ to K6-3+ Full Cache Enable Mod
-
Tyan S1564S to S1564D single to dual processor conversion (also s1563 and s1562)

Reply 18 of 21, by gonzo

User metadata
Rank Member
Rank
Member
Sphere478 wrote on 2026-01-15, 06:03:
Not sure, but maybe one or both of these. tyan s1564 or epox mvp4 when I get around to it. […]
Show full quote

Not sure, but maybe one or both of these.
tyan s1564 or epox mvp4 when I get around to it.

The mvp4 is a pci only and can do a high FSB

The tyan will be 66mhz fsb only.

MVP4 socket 7 aiming for the stars!

My favorite is the ASUS P55T2P4 in the latest Version of it. This is for me the absolutely best No-AGP-board with socket 7 in AT-format. It has undocumented options for a Vcore down to 2,0 V; a FSB of 83 MHz can be adjusted (undocumented, = 500 MHz K6-III+ for example); it works with ALL socket-7-CPUs; it can handle K6-2+/III+ and HDDs up to 120 GB with the latest Patch-BIOS from 2002; and it is rock stable in performance. Using a second TAG-RAM, it can adopt at least 256 MB (4x 64 MB) EDO-RAM (maybe even 512 MB, but I am not sure, as I do not have such RAM-modules).

On the other hand, VIA-chipset-boards (having MVP-chipset) are known for a poor RAM-performance compared to Intel-chipsets, so they are not appropriate for best scoring (even they are stable in performance, too).

I LOVE CPUs RUNNING IN [GonzoHz]

Reply 19 of 21, by douglar

User metadata
Rank l33t
Rank
l33t
Ozzuneoj wrote on 2026-01-12, 19:35:

I have one of these Apple 109-a58503-20 cards and I just pulled the heatsink off to see what it says. It says 9600 Pro on the die. Also, I used digital calipers and measured the die at ~8.86x8.86mm with is roughly the same as the 76mm^2 reported for the RV350 ... for what that's worth.

I tool the heat sink off my Apple Radeon 9600xt and it had an R360 chip-- https://theretroweb.com/chips/4165

Looking at your chip-- The part number ends with an "11F" instead of a "13F". That suggests to me that you might have the 110nm die shrink R351, not the 130nm R360. But I also see chips that end 12f and 11fl, so I don't really know.

The attachment Untitled.png is no longer available

Edit I took a look at a radeon 9550 and it was also 8.9mm on a side, so that looks the same

The attachment Radeon 9550.jpg is no longer available
Last edited by douglar on 2026-01-16, 20:10. Edited 2 times in total.