From the IO/AGP voltage selection, it does seem that it could predate the standardisation of AGP 2.0 and 4x/1.5V

Matth79 wrote on 2025-06-10, 12:33:

From the IO/AGP voltage selection, it does seem that it could predate the standardisation of AGP 2.0 and 4x/1.5V

Thank you for your reply. I greatly suspect that you are right. Now I have even more doubts. Maybe I should just put this Sapphire in the glass box and stick with my current FX5200.

Is there anything particular about the HD2600 that you wanted? Or just a "good quick buy"?

From my (year long painful) experience, I'd avoid any Rialto bridge powered card for a good experience. They can be very tricky to get happy with 🙁

tehsiggi wrote on 2025-06-10, 13:06:

Is there anything particular about the HD2600 that you wanted? Or just a "good quick buy"?
From my (year long painful) experience, I'd avoid any Rialto bridge powered card for a good experience. They can be very tricky to get happy with 🙁

Thank you for your reply. A good quick buy indeed) Had enough fun already - the Rialto was covered with some sort of cloth, and I have spent some time exploring whether this was just a dent protection, or part of a cooling system. Finally found a Youtube video that shows the full-scale aluminium cooling system mounted to the back. Not my board version(red), but same chip. So I have bough a copper radiator) Although the Rialto is proven not to burn immediately: the guy who sold me the card tested it in his machine as it was.

Meanwhile I have checked two things:
1)The Sapphire card fits to the TUSL AGP slot.
2)Everest software shows the current FX5200 card is running at 4x, which I understand means 1,5 volts. But the voltage figure is not specified in BIOS or in Everest.

That "cloth" is basically a thick thermal pad. They used it for protection of the exposed die. They usually had no cooling attached to it. With a relatively decent airflow, it's good enough.

If you search for pictures of other rialto based cards, like x700 AGP, X1950 AGP or HD3850 AGP you will see they never came with a cooler on rialto out of the box. (afaik)

tehsiggi wrote on 2025-06-10, 14:07:

If you search for pictures of other rialto based cards, like x700 AGP, X1950 AGP or HD3850 AGP you will see they never came with a cooler on rialto out of the box. (afaik)

Done that and see that indeed Rialto is typically not cooled, just padded around with this pink padding. We have to consider that these things were never meant to last 15-20 years though. Before this TUSL-2C I have had a Lucky Star 440BX motherboard that broke this winter. I tend to blame the 443BX chip - which had no heatsink and did okay for many years, but in the long run I suspect this turned out to be the weak spot.

Here is the video with the Rialto heatsink that I have found: https://www.youtube.com/watch?v=i1cU3xorwhI
The guy flips the card over at 0:10 to reveal a pretty massive heatsink.

Halofiber86 wrote on 2025-06-10, 21:30:

Here is the video with the Rialto heatsink that I have found: https://www.youtube.com/watch?v=i1cU3xorwhI
The guy flips the card over at 0:10 to reveal a pretty massive heatsink.

That's a good solution, even though it is not really "massive" - you have to remember that all this chips sink heat into PCB pretty effectively, which in large part is copper. Many SMD components are built/rated with this in mind. So that aluminum (presumably) plate is not really much better in terms of surface area than the PCB itself, and since it covers the PCB...

The issue i have with just gluing something to this chip is - it is a chip with bare die + some tiny SMD components on substrate, which are protected by those pad. Simply removing those pad and gluing something on top of bare die feels wrong. It will create mechanical load on the die and whatever compound is used to glue it to substrate which is not a good thing. Especially with how small the die is - the heatsink will either have to be inconsequentially small or you'll risk mechanically breaking the chip.

To properly cool it there has to be a way to mount the heatsink to the PCB itself, as it is done with, for example, chipsets which have bare die. And on this cards no such way exists.

As much as i'd like to attach a heatsink to the chip i can not think of a good way to do it and just gluing to bare die feels wrong and worse (more likely to damage the chip) than no cooling at all.

Also that guy poking at GPU with a screwdriver, then proceeding to scratch up copper heatsink with the same screwdriver and then use sandpaper on it... that's why i prefer to buy things nobody "serviced" before. That's just painful to watch. Use some alcohol and some cloth or paper to remove old thermal compound, no need to damage things.

Archer57 wrote on Yesterday, 02:02:

To properly cool it there has to be a way to mount the heatsink to the PCB itself, as it is done with, for example, chipsets which have bare die. And on this cards no such way exists.

Thank you, your thorough comment is much appreciated! I have already cut out a new thermal pad for the Rialto chip. I was planning to apply heatsink plaster between the Rialto and radiator, and the thermal pad was supposed to support the radiator on the perimeter. Actually there is a set of 2 smaller holes next to the 4 main chip large screw holes. These allow "click-on" plastic "screws" to be used in order to apply some kind of solution from the Rialto side. If only the matching sink could be found.

Your input and @tehsiggy's good advice earlier make me think I should just lay the Sapphire aside for a while as it is. It is just not good for my system. I have done extra research and stumbled upon a ca. 2009 thread of the ACTUAL OWNERS of this Sapphire 2600 AGP. The thread is 40-50 pages long, which gives you some perspective. Many, many people came and complained of blue screens, games not loading, systems damaged. Just does not feel right. For example, several people report that system ran fine for 2-3 days and then was damaged. This is what solutions I have sifted out:

- there should be at least 400W main power supply. The 300W can work for several days, but then the motherboard is at great risk to be damaged (northern bridge especially) because of the lack of power (something about power fluctuations that northern bridge is able to compensate up to a point)
- the Rialto should be definitely cooled
- all games should be run in DX9, NOT DX 10 mode (-dx9 modifier in the game's label)
- all AGP speeding options (fast write, etc. should be OFF)
- the Sapphire is not likely to be damaged in the older motherboard
- 8.442-071204a1-057556E-ATI drivers should be used
- the general system should be something like a 2GHz-processor based

I think this Sapphire is just not right here. Even though I use a laptop hard drive, I indeed only have a 300W power supply and a Celeron 1200. Now my system is very stable and ticking fine. Yes, there is no nice girl picture on my FX5200, but it gives me everything I need, I have accustomed to its few limitations and I even have an extra FX5200 in reserve. I think that I just have to acknowledge that I'm not the first man to be tricked into trouble by the picture of a nice girl that is just not for me)))

I have pretty much the same card - 2600XT. So far i am unable to even determine conclusively if the card is faulty, is just... garbage or if i am having some compatibility issues.

It does not work right, 3dmark crashes, games crash, BSODs happen. I've tried all the drivers (on XP, so no DX10 here), fooled around with all the AGP options from fast writes to frequency and voltage, different motherboards (though all nforce2 socketA), PSUs, everything.

It, however, suddenly works fine if i use CPU with 133Mhz FSB. No idea of how or why since it does not affect any other frequencies or anything related to the card, but it is what it is. This may be points towards compatibility issues with chipset?

Also found and replaced at least one leaky/dead cap.

If i plug 7600GT or 7300GT (with AGP-PCIe bridges too, but from nvidia) into the same system it is perfectly stable with any settings. And nvidia cards do have cooling for the bridge... though no pretty girls here either, just simple black heatsink which works rather well (this cards are from palit).

I also think some of the things you've picked up are... exaggerated. I do not think there is hard cutoff for PSU and unless the PSU is poorly designed or faulty it should not damage anything regardless. The bridge... it is better to cool it, but evidently the cards can work as is for a long time too.

Now i am waiting for an opportunity to grab some completely different motherboard with AGP, perhaps S478, to test this card....

And ultimately yeah, IMO this card is not a good acquisition... it theoretically has good performance, but i am not sure it is worth all the trouble...

Well, I could tell you stories.

I've been using an Sapphire X700 AGP 128MB as my main card back in the days. First on an ASRock K7S8X Rev. 3.0 (SiS Chipset) and then later on an MSI K8T Neo2-F 2.0 (Via K8T800 Chipset). I have the exact same card still and tried using it on a MS7075 (K8T800) and having only problems. Got it to working by setting the AGP Aperture Size to 1G. Same issue with a separate X700 with 256MB as well. Only driver that works is catalyst 5.3 for some reason. Overall, it's just really really odd.

The same card runs without any issues on an ASUS A7N8X (nForce2)

I think most of it stems from signal quality of the AGP slot. I would not state it's the chipset, since I have different issues with the card on different boards with various chipsets.
The X1950GT I'm having is basically behaving the same (most of the time).

I have a 7600GS AGP and that one just works happy anywhere. In general the HSI bridge from Nvidia appeared to be *way* more stable than their ATI pendant - rialto bridge.

So that's why I said: For a good AGP experience without doctoring around what the actual heck is going on, avoid rialto chips.

Btw. my X700s Rialto reaches around 50°C tops during 3DMark03, so that's not too bad. (Free air, no additional fan or anything).

The HD2000 AGP Series were known to behave even worse.. there was this notion of an AGP hotfix driver, but only ATI knows what the heck it's about.

Hi all...
I will tell you about my own experience.
The video card from Sapphire has its own tricks in the frequency correction of the PWM controller of GPU and VRAM (I would (and have) done it closer to the datasheet), which affects overclocking.
In principle, it reaches the frequencies of the HD3650.
11000 in Dmark2003
Works great on Aopen MX3S (i815).
To lubricate the turbine, you need to pull the impeller up forcefully.
Rialto Bridge does not require an additional radiator.
As ATI themselves wrote, the chip is simple, overheating is not a problem.
On a working motherboard this video card does not cause any problems.

But.
This is the HD2x00 generation, i.e. DirectX 10
This has nothing to do with Tualatin.
Compatibility with 98/XP games is poor.

shevalier wrote on Yesterday, 15:01:

Compatibility with 98/XP games is poor.

Thank you very much for taking your time to write about your experience! In fact I'm running Windows 2000 here. I'm now totally convinced that this Sapphire is not good for me anyway, maybe should look for some NVIDIA option instead.

I have checked the Retroweb.com and they state that your Aopen MX3S is i815E (Solano-2), with a universal AGP port. This is very close to my TUSL-2C (i815EP, Solano-3) . However as far as I catch the drift in this whole Sapphire-Rialto business, these slightest differences define success from failure. If I'm allowed to pry a little here, maybe you could tell me how much power supply do you have on this Aopen system?

shevalier wrote on Yesterday, 15:01:

On a working motherboard this video card does not cause any problems.

My opinion is different. If multiple motherboards and cards work fine in any combinations, but then one card has issues - the card is the part which is not working properly.

Judging by how common this issues are - my guess would be that something, be it signalling or power, is out of spec on this cards, which happens to be tolerated by some motherboards.

The fact that software part is very poor too does not help things either - when everything is half-functional, from software to hardware, it becomes quite hard to figure out specific issue in specific situation.

Halofiber86 wrote on Today, 00:02:

maybe you could tell me how much power supply do you have on this Aopen system?

Enlight hpc-250
atx 1.0, 250w 😀

but I don't recommend it, it has a bad standby power supply circuitry
if the control electrolytic capacitor dries out, then everything burns out, including the transformer