Archer57 wrote on 2025-07-09, 12:11:

So the question is - are they worth getting? Do they work reliably, or having the same AGP<=>PCI-e bridge - do they have similar issues to later HD cards? How is the compatibility with older games?

I remember there being someone who had serious issues with an AGP X700 card with Call of Duty and/or Medal of Honor: Allied Assault, while both worked just fine with my PCIe X700.

I don't think it was proven beyond doubt but the assumption was the AGP/PCIe bridge was causing issues.

Whats wrong with HD2xxx/HD3xxx? WinXP earlier driver that many recommend are catalyst 7.11 and its also first working table fog driver for ATi.

Up to x850 + maybe x1050 also win9x compatible.
x1950 is the first ATi h264 1080p card (powerdvd7)

Radeon 9xxx are native AGP cards for sure and maybe x7xx +x8xx also, but not x1xxx.

marxveix wrote on 2025-07-09, 12:44:

Whats wrong with HD2xxx/HD3xxx? WinXP earlier driver that many recommend are catalyst 7.11 and its also first working table fog driver for ATi.

Stability, regardless of drivers. And compatibility with older games, as well as performance in older games. I am not trying to use AGP system to run crysis or something and this cards just do not play nice with older stuff, even when/if they work.

marxveix wrote on 2025-07-09, 12:44:

Up to x850 + maybe x1050 also win9x compatible.

That's interesting, may be a fun thing to try.

marxveix wrote on 2025-07-09, 12:44:

x1950 is the first ATi h264 1080p card (powerdvd7)

I am not really interested in video decoding capabilities, but that's good to know.

marxveix wrote on 2025-07-09, 12:44:

Radeon 9xxx are native AGP cards for sure and maybe x7xx +x8xx also, but not x1xxx.

From what i've seen x7xx and x8xx tend to have bridges. I've seen something from this series mentioned as "the last ATI native AGP card", but i am yet to find such card...

ATi Radeon X1950 Pro and somewhat uncommon X1950XT AGP are the fastest cards that still have 16-bit dithering support. So that's already a big plus to compatibility. If you can live with Windows 2000/XP (no Win9x drivers) it should be mostly fine for early Direct3D titles. X1950 also can emulate Glide with Shader Model 3.0 .

I have personally used the following three cards:

Sapphire X700 AGP 128MB & 256MB (using Rialto bridge):

Both cards used DDR1. I had them in use as my daily driver for a couple of years. For what it's worth, I got them back in the day (2006) for around 50€, which was a decent price.
I used them on two boards back in the day: ASRock K7S8X and an MSI K8T-Neo2F.

They ran - fine. Sometimes the drivers where very fiddly to get going. Updating to something new wasn't always a good idea, but they worked relatively well. I used them with Windows XP and used to play the usual games back in the day: Counter-Strike, WoW, RTS games like settlers, Anno and Starcraft.

I still have one of them, it works just fine on my Epox EP-9NDA3, running happily whatever I throw at it. However on my MS-7075 it outright refuses to work correctly. I can't switch to high resolution or bit rate in Windows. No matter what driver I use. Other AGP cards work there just fine, which brings me to the next card:

Sapphire X1950GT 256MB AGP (using Rialto bridge:

I had this one in use in my retro machine a couple of years ago, mostly playing typical XP era games, didn't go anything below XP. It ran just fine, given that you provide a PSU with enough strength on the 12V rail (it pulls up to 80W from 12). It has the same issues on the MS-7075 and outright refuses to run there properly.

Sapphire X1300Pro AGP (using Rialto bridge:

Ran in my brothers PC back in the day, running with another MSI K8T-Neo2F. Worked just fine. It later moved to some Pentium 3 board from a friend, but I can't recall which one. It did work fine for her games like Sims etc. without issues. It ran just fine on both boards, but didn't like certain driver versions though. It would then just blackscreen when the driver is being loaded or go to 4bit color mode, similar to the X700.

Archer57 wrote on 2025-07-09, 12:59:

marxveix wrote on 2025-07-09, 12:44:

Radeon 9xxx are native AGP cards for sure and maybe x7xx +x8xx also, but not x1xxx.

From what i've seen x7xx and x8xx tend to have bridges. I've seen something from this series mentioned as "the last ATI native AGP card", but i am yet to find such card...

Radeon X800XT for example is a native AGP chip, using the R420 under the covers. No rialto bridge involved.

Radeon X800/X850 cards are great for maxing out Win9x games, as in running them at 1600x1200 with all in-game settings at maximum, with 4xAA and 16xAF added on top via drivers.

There's no need to hunt down the rarer AGP versions, since PCIe does work under Win9x. You may need to install the drivers manually though. For games which use table fog, you can dual boot with WinXP and play them there, since it works perfectly when using Catalyst 7.11 or newer.

tehsiggi wrote on 2025-07-09, 13:06:

I have personally used the following three cards: […]
Show full quote

I have personally used the following three cards:

Sapphire X700 AGP 128MB & 256MB (using Rialto bridge):

Both cards used DDR1. I had them in use as my daily driver for a couple of years. For what it's worth, I got them back in the day (2006) for around 50€, which was a decent price.
I used them on two boards back in the day: ASRock K7S8X and an MSI K8T-Neo2F.

They ran - fine. Sometimes the drivers where very fiddly to get going. Updating to something new wasn't always a good idea, but they worked relatively well. I used them with Windows XP and used to play the usual games back in the day: Counter-Strike, WoW, RTS games like settlers, Anno and Starcraft.

I still have one of them, it works just fine on my Epox EP-9NDA3, running happily whatever I throw at it. However on my MS-7075 it outright refuses to work correctly. I can't switch to high resolution or bit rate in Windows. No matter what driver I use. Other AGP cards work there just fine, which brings me to the next card:

Sapphire X1950GT 256MB AGP (using Rialto bridge:

I had this one in use in my retro machine a couple of years ago, mostly playing typical XP era games, didn't go anything below XP. It ran just fine, given that you provide a PSU with enough strength on the 12V rail (it pulls up to 80W from 12). It has the same issues on the MS-7075 and outright refuses to run there properly.

Sapphire X1300Pro AGP (using Rialto bridge:

Ran in my brothers PC back in the day, running with another MSI K8T-Neo2F. Worked just fine. It later moved to some Pentium 3 board from a friend, but I can't recall which one. It did work fine for her games like Sims etc. without issues. It ran just fine on both boards, but didn't like certain driver versions though. It would then just blackscreen when the driver is being loaded or go to 4bit color mode, similar to the X700.

Archer57 wrote on 2025-07-09, 12:59:

marxveix wrote on 2025-07-09, 12:44:

Radeon 9xxx are native AGP cards for sure and maybe x7xx +x8xx also, but not x1xxx.

From what i've seen x7xx and x8xx tend to have bridges. I've seen something from this series mentioned as "the last ATI native AGP card", but i am yet to find such card...

Radeon X800XT for example is a native AGP chip, using the R420 under the covers. No rialto bridge involved.

Thank you for taking time to write this. It is useful. At least gives me some hope the cards may end up being useful.

It would be really nice to find one of those native AGP versions, but that'll probably require time and luck. Not sure why specifically this cards are so rare, i mean all the nvidia and later/earlier ATI/AMD stuff is there, just expensive, but this cards are impossible to find.

In your opinion - does the bridge need cooling? I've heard opinions that it does, i've also seen cards which worked for a while without. And given there is no easy/reasonable way to do it... kind of feels like it is best left alone. Or am i wrong?

Joseph_Joestar wrote on 2025-07-09, 13:11:

Radeon X800/X850 cards are great for maxing out Win9x games, as in running them at 1600x1200 with all in-game settings at maximum, with 4xAA and 16xAF added on top via drivers.

There's no need to hunt down the rarer AGP versions, since PCIe does work under Win9x. You may need to install the drivers manually though. For games which use table fog, you can dual boot with WinXP and play them there, since it works perfectly when using Catalyst 7.11 or newer.

Hmm, that's good to know. Definitely something to try.

I've seen a few people mention better AF quality compared to GF6 cards, how true is that? I kind of always skipped those image quality arguments as they always felt similar to "amd vs intel" holy wars, so no idea if any of that stuff actually is true...

Archer57 wrote on 2025-07-09, 14:50:

I've seen a few people mention better AF quality compared to GF6 cards, how true is that?

Not sure, I never bothered with testing it myself. There's some info in this Anandtech article but the differences seem to be very minor.

On a possibly related note, Phil did a comparison between GeForce FX and GeForce 6 cards on his channel recently. From his tests, the FX cards handled AF better in games where you can lean left/right.

Archer57 wrote on 2025-07-09, 14:50:

Thank you for taking time to write this. It is useful. At least gives me some hope the cards may end up being useful.

It would be really nice to find one of those native AGP versions, but that'll probably require time and luck. Not sure why specifically this cards are so rare, i mean all the nvidia and later/earlier ATI/AMD stuff is there, just expensive, but this cards are impossible to find.

In your opinion - does the bridge need cooling? I've heard opinions that it does, i've also seen cards which worked for a while without. And given there is no easy/reasonable way to do it... kind of feels like it is best left alone. Or am i wrong?

Does the bridge benefit from cooling? Like any integrated circuit I say yes.
Does the bridge need cooling? From my experience: No.

My X700s never had any extra cooling on the rialto bridge, just the thermal pad that is there for protection of the exposed die, that's it. They ran happy. My case had a decent airflow, but even the X1300 I ran survived years in a very bad ventilated OEM case.

On my testbench (convection only) the bridge rarely exceeds 55°C on the die. Imho totally fine.

There were a lot of posts in the web about "cards die due to the bridge overheating". I have not seen any real proof for that, nor anyone really going that deep into the debugging to figure out 100% that this is true. Perhaps I'll buy a dead one of ebay and will convert it to a PCIe Frankenstein. If that works, we'll know rialto is dead.

R420 and R481 are native AGP chips. So with some luck, you can find card based on them. For Example Radeon X850 Pro AGP shouldn't be too hard to find. Like this one https://hw-museum.cz/vga/133/ati-radeon-x850-pro-agp

Regarding AF quality - GF6/7 are rather bad, that is a fact. Their AF is very angle dependent. GF FX (especially in high quality mode) is much better at this... but of course for the price of much bigger performance hit. GF3/4Ti are aslo perfect when it comes to AF quality, however it is too slow to actually use it.

Radeon R300 and R400 based are quite similar to GF6/7. R500 (X1300 and later) are much better once you enable the high quality mode. I found some pictures I made 10 years ago... ouch, really that long ago 😁

tehsiggi wrote on 2025-07-09, 17:05:

Does the bridge benefit from cooling? Like any integrated circuit I say yes. Does the bridge need cooling? From my experience: […]
Show full quote

Archer57 wrote on 2025-07-09, 14:50:

Thank you for taking time to write this. It is useful. At least gives me some hope the cards may end up being useful.

It would be really nice to find one of those native AGP versions, but that'll probably require time and luck. Not sure why specifically this cards are so rare, i mean all the nvidia and later/earlier ATI/AMD stuff is there, just expensive, but this cards are impossible to find.

In your opinion - does the bridge need cooling? I've heard opinions that it does, i've also seen cards which worked for a while without. And given there is no easy/reasonable way to do it... kind of feels like it is best left alone. Or am i wrong?

Does the bridge benefit from cooling? Like any integrated circuit I say yes.
Does the bridge need cooling? From my experience: No.

My X700s never had any extra cooling on the rialto bridge, just the thermal pad that is there for protection of the exposed die, that's it. They ran happy. My case had a decent airflow, but even the X1300 I ran survived years in a very bad ventilated OEM case.

On my testbench (convection only) the bridge rarely exceeds 55°C on the die. Imho totally fine.

There were a lot of posts in the web about "cards die due to the bridge overheating". I have not seen any real proof for that, nor anyone really going that deep into the debugging to figure out 100% that this is true. Perhaps I'll buy a dead one of ebay and will convert it to a PCIe Frankenstein. If that works, we'll know rialto is dead.

Thanks. That sounds similar to what i was thinking - will probably stick to just trying to have decent airflow so that PCB itself does not get too hot. Should probably help the bridge chip stay a bit cooler too.

I understand that any electronics can generally benefit from better cooling, but achieving lower temperatures becomes exponentially harder as it becomes closer to ambient and it generally makes sense to stop at some reasonable values...

havli wrote on 2025-07-09, 17:13:

R420 and R481 are native AGP chips. So with some luck, you can find card based on them. For Example Radeon X850 Pro AGP shouldn' […]
Show full quote

R420 and R481 are native AGP chips. So with some luck, you can find card based on them. For Example Radeon X850 Pro AGP shouldn't be too hard to find. Like this one https://hw-museum.cz/vga/133/ati-radeon-x850-pro-agp

Regarding AF quality - GF6/7 are rather bad, that is a fact. Their AF is very angle dependent. GF FX (especially in high quality mode) is much better at this... but of course for the price of much bigger performance hit. GF3/4Ti are aslo perfect when it comes to AF quality, however it is too slow to actually use it.

Radeon R300 and R400 based are quite similar to GF6/7. R500 (X1300 and later) are much better once you enable the high quality mode. I found some pictures I made 10 years ago... ouch, really that long ago 😁

The attachment FX_5900_8xAF_def.PNG is no longer available

The attachment FX_5900_8xAF_HQ.PNG is no longer available

The attachment GF_6800GT_16xAF_HQ.PNG is no longer available

The attachment X850_Pro_16xAF.PNG is no longer available

The attachment X1650_Pro_16xAF_HQ.PNG is no longer available

Thanks for the info. Interesting, it seems like this is less related to GPU manufacturer and more to the fact there apparently was a period of time when both manufacturers implemented some AF optimizations affecting quality...

It's a shame because it is yet another disadvantage of late/faster AGP cards... perhaps i'll try to see any difference between the cards i have - FX5900XT and 7600GT...

And yeah, as much as i'd like to get something like that X850 Pro i think i'll just have to wait and see if something reasonable comes up - there are no offers at all on local ebay equivalent...

I've spent whole bunch of time digging up even those X800GTO, simply going through all the "cheap old PC" offers and trying to see if there is something with AGP and if it has some interesting cards.

Also got cheap X700pro sold as "dead" with no description what's wrong, but visually with no signs of attempted repair and warranty seals on the heatsink still intact, so we'll see how that goes...

Archer57 wrote on 2025-07-10, 01:56:

Thanks. That sounds similar to what i was thinking - will probably stick to just trying to have decent airflow so that PCB itself does not get too hot. Should probably help the bridge chip stay a bit cooler too.

I understand that any electronics can generally benefit from better cooling, but achieving lower temperatures becomes exponentially harder as it becomes closer to ambient and it generally makes sense to stop at some reasonable values...

Some people went to extra ordinary lengths to cool the rialto chip.. with more or less success. Some people insisted it needed cooling, since it had a thermal pad. It was around that time that I pulled myself out of online PC tech forums because no matter how hard you tried to argue with logic, people were just stuck on the "my opinion is a fact" bandwagon. Well, that didn't change much I guess 😁

Due to the nature of AGP platforms, there is usually the northbridge right next to the upper side of the AGP slot. This means a potential secondary heat source is pretty close nearby. Especially in bad ventilated cases I can see how there will be a "hotspot" there, including the warm air coming from the CPU cooler.

I just fetched a broken X1600 from ebay to see what's the deal with it. I hope it is a "not working at all broken" and not just some "artifacts due to bad memory". I'll then go ahead and remove the rialto and convert it to PCIe, perhaps just 1x for the beginning. We'll see how that goes. I love those kind of experiments. They don't serve a specific purpose other than my curiosity.

tehsiggi wrote on 2025-07-10, 04:33:

Due to the nature of AGP platforms, there is usually the northbridge right next to the upper side of the AGP slot. This means a potential secondary heat source is pretty close nearby. Especially in bad ventilated cases I can see how there will be a "hotspot" there, including the warm air coming from the CPU cooler.

Actually airflow from CPU cooler can be pretty effective at cooling things. Including northbridge, CPU VRM and probably back side of the videocard. As long as the cooler is not tower with fan sideways. In fact i have a suspicion that many motherboards were designed with this in mind and later, when tower coolers became common before northbridges went extinct, this may have caused some failures by itself...

Yes the air is warm, but it is not warm enough to be useless or harmful.

tehsiggi wrote on 2025-07-10, 04:33:

I just fetched a broken X1600 from ebay to see what's the deal with it. I hope it is a "not working at all broken" and not just some "artifacts due to bad memory". I'll then go ahead and remove the rialto and convert it to PCIe, perhaps just 1x for the beginning. We'll see how that goes. I love those kind of experiments. They don't serve a specific purpose other than my curiosity.

This would definitely be a curious experiment. Would not necessarily give any definitive answers, but still interesting.

One thing which can cause confusion here is if simply heating the board up while desoldering the chip makes it work. Perhaps heating up the GPU to see if it is the typical bumps failure before messing with the bridge would be a good idea...

So, X800GTO has arrived. The card is not in amazing condition - it is definitely well used and disassembled a bunch of times, despite warranty sticker being intact.

GPU was covered with silly amount of thermal compound and someone apparently tried to drive a screw through one of heatsink mounting holes doing damage to the board and separating layers:

Fan connector is missing its plastic part and also does not work:

And i absolutely hate sapphire coolers:

Why did they need to place screws holding plastic cover on the underside of the heatsink? Have to take whole cooler apart and replace thermal compound just to clean the dust out...

Good thing - the card works. After removing the dust, replacing thermal compound and wiring the fan to 12v directly...

And unlike AMD HD stuff it does not crash all the time. The drivers are still garbage, i ended up with 8.5 + CCC from 7.11 or something because CCC from 8.5 did not work and 7.11 driver did not work either (crashes), but after wasting some time on this - everything works. Which kind of implies AGP-PCIe bridge has nothing to do with later card issues.

Seemingly a very decent card. At a glance performance is slightly above 6600GT/7300GT (around 16500 in 3dmark01se, AthlonXP 3200+) and i did not notice FPS drops/stutters i've seen on HD2600XT.

Lack of temperature monitoring is annoying, but measured with finger - it does not seem very hot. The hottest part is those heatsink on the bridge (which was already there when i got the card).

It is R430 too, so may be i'll try playing around with unlocking into X800XL, should be fun.

Archer57 wrote on Yesterday, 02:46:

And unlike AMD HD stuff it does not crash all the time. The drivers are still garbage, i ended up with 8.5 + CCC from 7.11 or something because CCC from 8.5 did not work and 7.11 driver did not work either (crashes), but after wasting some time on this - everything works. Which kind of implies AGP-PCIe bridge has nothing to do with later card issues.

Catalyst 7.11 should work fine with that card. It does require .NET Framework 2.0 so be sure to have that installed. Also, if you had previous versions of ATi/AMD drivers installed on that system before using 7.11 you might get some weird errors. Some of those driver cleaner utilities may be able to fix that, and it certainly doesn't happen on a fresh WinXP+SP3 installation.

Lack of temperature monitoring is annoying, but measured with finger - it does not seem very hot. The hottest part is those heatsink on the bridge (which was already there when i got the card).

Both GPU-Z and ATi Tray Tools should be able to show you the temperature of an X800 card under WinXP.

Joseph_Joestar wrote on Yesterday, 03:02:

Catalyst 7.11 should work fine with that card. It does require .NET Framework 2.0 so be sure to have that installed. Also, if you had previous versions of ATi/AMD drivers installed on that system before using 7.11 you might get some weird errors. Some of those driver cleaner utilities may be able to fix that, and it certainly doesn't happen on a fresh WinXP+SP3 installation.

For some reason with 7.11 some games (including 3dmark) refuse to run with an error that they cannot initialize d3d device or something like that. Those which work have no issues. 8.5 fixed that, but CCC crashes on start. So i installed 8.5 with no CCC and then installed CCC from 7.11. I needed CCC mostly because this is the second card i've seen which detects max monitor resolution incorrectly and i had to fix that in CCC.

I probably do need to use a fresh copy of XP because messing with a few of this drivers may have indeed screwed something up, but at least this configuration seems to work fine.

Joseph_Joestar wrote on Yesterday, 03:02:

Both GPU-Z and ATi Tray Tools should be able to show you the temperature of an X800 card under WinXP.

I have not tried ATi Tray Tools, but all the usual suspects like GPU-Z and everest/aida do not show GPU temperature.

Archer57 wrote on Yesterday, 02:46:

Lack of temperature monitoring is annoying, but measured with finger - it does not seem very hot. The hottest part is those heatsink on the bridge (which was already there when i got the card).

https://www.techpowerup.com/articles/94

Archer57 wrote on Yesterday, 02:46:

And i absolutely hate sapphire coolers:

Well, this is a great cooler, what are you grumbling about?
Dust is just dust, it needs to be cleaned periodically.
It is in this implementation from Sapphire that the fan does not blow through it properly - due to the small diameter of the fan and too low speed.
Only for such longitudinal heatsink need to use a turbine, not a fan
The turbine has higher flow and pressure.
Like this

shevalier wrote on Yesterday, 06:08:

https://www.techpowerup.com/articles/94

Does not work though. May be that's software and i'll try with fresh copy of windows later, but i doubt it.

The card has a bunch of components not soldered (right where the chip is shown in those article), so perhaps manufacturer decided to save a few cents at some point:

shevalier wrote on Yesterday, 06:08:

Well, this is a great cooler, what are you grumbling about?
Dust is just dust, it needs to be cleaned periodically.

Ease of maintenance. It is simply a bad design to force full disassembly each time instead of doing this:

Kind of like those laptops where you need to tear whole thing down to clean the heatsink. Not related to specific card though, sapphire did a lot of this...

shevalier wrote on Yesterday, 06:08:

It is in this implementation from Sapphire that the fan does not blow through it properly - due to the small diameter of the fan […]
Show full quote

It is in this implementation from Sapphire that the fan does not blow through it properly - due to the small diameter of the fan and too low speed.
Only for such longitudinal heatsink need to use a turbine, not a fan
The turbine has higher flow and pressure.
Like this

Yeah, the fan surprised me a little, does not make much sense using regular fan blowing into solid surface.

Perhaps i'll replace it, though seems to work decently well and is not too noisy, even at 100% RPM (not sure if this card had fan control).

Archer57 wrote on Yesterday, 06:44:

The card has a bunch of components not soldered (right where the chip is shown in those article), so perhaps manufacturer decided to save a few cents at some point:

not sure if this card had fan control).

The thermal diode inside the core appeared in the Radeon 9500, as well as in the nV FX5700. Until the monitoring IC was built into the core, everyone continued to save a few cents.
Nope, LM63 is a "Remote Diode Digital Temperature Sensor with Integrated Fan Control"