tehsiggi wrote on 2025-10-22, 18:46:Hmm.. strange.. Can you try to wipe the complete log file and run it again? I can recall this happening to me as well once.
So, it passed test 1? I'm now going to try the * variable to run them all.
Okay, I finally got the r3memid results. Here's what I got:
R3MEMID version 1.07, (c) Copyright ATI Technologies Inc, 2003Log file generation enabled to .\R3MEMID.LOG ...Reference data file (RDF) loading disabled ...No wait on error enabled (not effected by '-s') ...TEST RESULT SUMMARY:====================Rv350 (0x4152) detected.128M video memory.Test suite ran 1 of 1 times.Checking test status array ...[1 ] Fill : PASS[2 ] Data line toggle (GUI) : PASS[3 ] RW page (GUI) : PASS[4 ] RW channel (GUI) : PASS[5 ] Data mask : PASS[6 ] Macro + Micro8x2 tiling read : PASS[7 ] Byte swap macro+micro8x2 read : FAILError ID 0VC0181024 x 768 - 32 bpp ( 60 Hz): TEST FAILUREfailing bit : MDA1 0 1 2 3 4 5 6 7 ... MDB1 0 1 2 3 4 5 6 7 ...[8 ] HOST_DATA/<f,b>/M2L : PASS[9 ] Line slopes/dir/P/solid : PASSFailure detected.
So only one test failed. I do not know what this means.
Played a bunch of GLQuake and HL1 on OpenGL at 1024x768 without incident after this. So it's not busted, but something is off.
not rereading it all:
is the problem the same on VGA and DVI?
the card draws power through the mb only?
Retro PC warning: The things you own end up owning you.
ux-3 wrote on 2025-10-23, 16:45:not rereading it all:
is the problem the same on VGA and DVI?the card draws power through the mb only?
Yup, just uses motherboard power. I haven't tried DVI but because the video output appears perfectly clean most of the time I doubt that's the problem.
Just played the rest of the shareware episode of Quake in GLQuake at 1024x768 without a problem. On the other hand Return to Castle Wolfenstein ran for about 5 seconds at 640x480 before crashing out.
No guesses form anyone? It feels like the card isn't totally broken but has something busted but I lack the wherewithal to understand the log file.
Could you provide a picture from the back of the card? I'd like to take a look at it, if possible.
tehsiggi wrote on 2025-10-24, 18:13:Could you provide a picture from the back of the card? I'd like to take a look at it, if possible.
Here you go.
Whelp, I guess there's not an obvious answer. I'll take the system back to a Radeon 9000 or Quadro4 380 XGL and see if someone figures this out. Both will be adequate for 90s games but it would be nice to have a working higher end AGP card.
I took a look at it this afternoon. I can't see anything obvious. furthermore, having MDB1 and MDA1 with the same bits failing is odd, from a technical point of view. It does not seem like this is directly related to something on the memory. It could be, that those two ICs are faulty, but that would be very astonishing, I rarely saw a 3.3ns chip fail. The Chips that would be affected according to your R3MEMID run are U76 (channel A) and U78 (channel B), both located at the back of the card. You could check if application of pressure changes the pattern / behavior.
Update of sort - I tried another card and getting similar results. Memtest86+ gave a ton of errors, so it's either system memory or CPU I think.
Did you get this sorted?
I have a system that won't boot with R9550/9600, also drawing all power from the AGP slot.
I noticed the PSU whines when I try - so I'm wondering if R9550/9600 actually cause the motherboard to pull more than the PSU can deliver causing voltage ripple and instability? I have no problem with the molex-powered Radeons, which I suspect might be stressing the AGP power lines less?
MattRocks wrote on 2025-11-24, 01:18:Did you get this sorted?
I have a system that won't boot with R9550/9600, also drawing all power from the AGP slot.
I noticed the PSU whines when I try - so I'm wondering if R9550/9600 actually cause the motherboard to pull more than the PSU can deliver causing voltage ripple and instability? I have no problem with the molex-powered Radeons, which I suspect might be stressing the AGP power lines less?
In idle the 9600 series is consuming very little power. So i doubt this issue will be due to power constraints.
tehsiggi wrote on 2025-11-24, 04:24:In idle the 9600 series is consuming very little power. So i doubt this issue will be due to power constraints.
It doesn't show if the cards are pulling from the AGP slot, or from the Molex (separate rail).
Also, during POST they are not idle - they are charging their capacitors.
MattRocks wrote on 2025-11-24, 20:05:tehsiggi wrote on 2025-11-24, 04:24:In idle the 9600 series is consuming very little power. So i doubt this issue will be due to power constraints.
It doesn't show if the cards are pulling from the AGP slot, or from the Molex (separate rail).
Also, during POST they are not idle - they are charging their capacitors.
It shows. I measure both. Just click on the card name.
Those laughable couple 1000uF at max are not an issue for any PSU of the ATX era, unless it is broken.
ubertrout wrote on 2025-10-21, 03:36:I got a lot of cards as untested, and by far the most interesting was a Radeon 9600 XT. Put it in a system and I can't figure o […]
I got a lot of cards as untested, and by far the most interesting was a Radeon 9600 XT. Put it in a system and I can't figure out if it's broken or it's just troublesome.
The good - regular 2d performance on the VGA port in Windows 98 is just fine. No artifacting, nothing weird. Windows sees it and drivers install fine. HWinfo32 sees 128 MB of SDRAM on the card and that the card is running at 499.5 Mhz and Memory is running at 297 Mhz.
The bad - I pretty much can't get it to do anything 3D. Quake 2 and Unreal Tournament both pretty much refuse to work. Even the DirectX 3D test fails. Weirdly GLQuake seems to work great. Not sure what's going on there.
Any thoughts as to what's going on? Nice AGP cards are getting harder to find and I'd rather not give up on this card quickly.
what a beaut! I miss the red boards, it may not work as a graphics but i bet you can pump some computations through it
MattRocks wrote on 2025-11-24, 01:18:Did you get this sorted?
I have a system that won't boot with R9550/9600, also drawing all power from the AGP slot.
I noticed the PSU whines when I try - so I'm wondering if R9550/9600 actually cause the motherboard to pull more than the PSU can deliver causing voltage ripple and instability? I have no problem with the molex-powered Radeons, which I suspect might be stressing the AGP power lines less?
Yeah, I pulled one memory stick and the system got much more stable, and most games were working with _some_ stability. Whether lingering instability is Windows 9x nonsense or a problem with the card isn't clear. I should re-run the memtest with that stick out...
ubertrout wrote on 2025-11-25, 21:22:Yeah, I pulled one memory stick and the system got much more stable, and most games were working with _some_ stability. Whether lingering instability is Windows 9x nonsense or a problem with the card isn't clear. I should re-run the memtest with that stick out...
Are all the sticks the same? If you have a mix of timings and voltages then you might need to manually tweak the BIOS to find a stable setting - might involve specifying a voltage.
But sounds like your Radeon is working fine. Congrats.
tehsiggi wrote on 2025-11-24, 20:09:It shows. I measure both. Just click on the card name.
Those laughable couple 1000uF at max are not an issue for any PSU of the ATX era, unless it is broken.
I tentatively suggest re-evaluating the meaning of the data. If I read your data correctly, the R9550 draws ~5W from AGP and R9700 draws ~7W from AGP.
Today I swapped the PSU and the bad Radeon 9550 jumped to life. This result shows the "no video out" symptom must have been caused by insufficient/unstable power on AGP slot.
However, the Radeon 9700 Pro always consistently worked fine with the bad PSU. This shows the ~7W R9750 must experience less ripple from AGP than the ~5W R9550. If my understanding of the data is right, then the power from the molex must be mixed with and stabilising ripples in power from the AGP slot. A ripple of 1000uF-2000uF on the AGP slot might be something the graphics card capacitors can iron out, which might matter to a GPU memory controller?
Side-note: The only other thing that changed in my brief test is a small consequence of my sloppy testing - the HDD was unplugged (not normally the case), and leaving the HDD unplugged might have influenced power supply during POST. However, I strongly doubt that changes anything because the HDD is not on the same rail as the AGP slot - and any draw from the HDD would have impacted the R9750 more than R9550.
In either case there is a lesson learned: Low-powered GPUs without molex are (perhaps obviously) more dependent on the AGP slot (and motherboard) than high-powered GPUs with molex.
MattRocks wrote on 2025-11-28, 17:59:I tentatively suggest re-evaluating the meaning of the data. If I read your data correctly, the R9550 draws ~5W from AGP and R97 […]
tehsiggi wrote on 2025-11-24, 20:09:It shows. I measure both. Just click on the card name.
Those laughable couple 1000uF at max are not an issue for any PSU of the ATX era, unless it is broken.
I tentatively suggest re-evaluating the meaning of the data. If I read your data correctly, the R9550 draws ~5W from AGP and R9700 draws ~7W from AGP.
Today I swapped the PSU and the bad Radeon 9550 jumped to life. This result shows the "no video out" symptom must have been caused by insufficient/unstable power on AGP slot.
However, the Radeon 9700 Pro always consistently worked fine with the bad PSU. This shows the ~7W R9750 must experience less ripple from AGP than the ~5W R9550. If my understanding of the data is right, then the power from the molex must be mixed with and stabilising ripples in power from the AGP slot. A ripple of 1000uF-2000uF on the AGP slot might be something the graphics card capacitors can iron out, which might matter to a GPU memory controller?
Side-note: The only other thing that changed in my brief test is a small consequence of my sloppy testing - the HDD was unplugged (not normally the case), and leaving the HDD unplugged might have influenced power supply during POST. However, I strongly doubt that changes anything because the HDD is not on the same rail as the AGP slot - and any draw from the HDD would have impacted the R9750 more than R9550.
In either case there is a lesson learned: Low-powered GPUs without molex are (perhaps obviously) more dependent on the AGP slot (and motherboard) than high-powered GPUs with molex.
There is not much to re-evaluate on the data. The rails of the molex connector and agp slot are not connected on the cards PCB for a Radeon 9700, the schematic clearly shows it. Some parts use the molex for powering, some use the Slot.
As for your card, i wouldn't blame it on ripple (which is a strange thing to just suspect in the first place) but perhaps the voltage being too low in the first place for proper regulation. If you want to know for sure, anything else but measuring is guesswork.
So my statement is still, if a PSU isn't able to supply a card properly through the AGP slot alone, it's unsuitable or broken.
Unless your PSU has true real split rails, everything connected puts load on the same rail your PSU. The only thing you can rule out by having the HDD on a different connector is the voltage drop over the wire. But the AGP slot is still powered over the same voltage rail. If the PSU regulation is not proper, an additional load on the let's say 5V rail will still impact the AGP slot.
If there is a tutorial on where to put multimeter probes I'd report the figures.
Using what I know, I can only ask: With PSU#1, why does R9700Pro work fine and R9550 fail to initialise? We can agree the PSU is at fault, but the difference in the effects on R9700Pro and R9550 still need to be coherent.
Also, the definition of faulty PSU is unclear. PSUs gradually decline from day one and eventually pass a threshold that constitutes failure, but where that threshold is mysterious.
