I got this X1950 XTX that is in physical "like new" condition. No missing components. It appears to be working perfectly in games and benchmarks BUT when the core reaches somewhere between 73c and 78c, it crashes. I ran R5MEMID and it passes perfectly.
The attachment 5.JPG is no longer available
The attachment 1.JPG is no longer available
The attachment 2.JPG is no longer available
Sometimes it will blue screen and sometimes it will go back to Windows with ATI's VPU recovery.
The attachment 3.JPG is no longer available
The attachment 4.JPG is no longer available
Last edited by zuldan on 2026-05-16, 06:12. Edited 1 time in total.
Interestingly your GPU-Z Screenshot shows the Fan not changing it's speed? It looks like it's fixed at 13%?
Btw. what a lovely card. Fond memories of me looking through magazines and imagining how awesome it would be to have this card. I had an X700 AGP at the time 😁
Can you monitor the VRM temperatures and VDDC externally?
Btw. what a lovely card. Fond memories of me looking through magazines and imagining how awesome it would be to have this card. I had an X700 AGP at the time 😁
Can you monitor the VRM temperatures and VDDC externally?
She’s beautiful!
I can use a thermal cam and possibly a multimeter. Will get back to you.
FYI, forgot to mention I replaced the core thermal paste and all the memory thermal pads but still got the same issue. I have some new cores on the way, just in case 😉
Interestingly your GPU-Z Screenshot shows the Fan not changing it's speed? It looks like it's fixed at 13%?
Btw. what a lovely card. Fond memories of me looking through magazines and imagining how awesome it would be to have this card. I had an X700 AGP at the time 😁
Can you monitor the VRM temperatures and VDDC externally?
Got some time to look at this again, I'm now certain it's a memory issue (or some issue with the power rail for the memory). I booted into Windows and immediately started getting artifacts on the desktop (without even running 3dMark and GPU cool). I pushed down on the top row of memory chips and I could get the artifacts to change.
The attachment Desktop.JPG is no longer available
I then ran R5MEMID again, and it passed.
The attachment R5MEMID_1.JPG is no longer available
I pressed down on the top row of memory cards and bingo I got artifacts on the screen
The attachment R5MEMID_2.JPG is no longer available
I ran R5MEMID again, finally got errors
1R5MEMID version 1.03, (c) Copyright ATI Technologies Inc, 2006 2Log file generation enabled to .\R5MEMID.LOG ... 3Extended logging enabled ... 4Reference data file (RDF) loading disabled ... 5No wait on error enabled (not effected by '-s') ... 6 [1] Mem production test FB001 7 8 Failing Bits 9 rk0 chH: 5 6 8 10 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 10 11 [2] Mem production test FB063 12 13 Failing Bits 14 rk0 chH: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 15 16 [3] Mem production test FB064 17 [4] Mem production test FB065 18 [5] Mem production test FB066 19 20 Failing Bits 21 rk0 chH: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 22 23 [6] Modified FB088 24 25 Failing Bits 26 rk0 chH: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 27 28 [7] Modified FB091 29 30 Failing Bits 31 rk0 chH: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33TEST RESULT SUMMARY: 34==================== 35R580/M59 (0x7240) detected. 36512M video memory. 37Test suite ran 1 of 1 times. 38Checking test status array ... 39[1 ] Mem production test FB001 : FAIL 40 Error ID G0FB096 41 1024 x 768 - argb8888 (60 Hz): TEST FAILURE 42 Error count: total[29295760] chH[29295760] 43[2 ] Mem production test FB063 : FAIL 44 Error ID G0FB096 45 1024 x 768 - argb8888 (60 Hz): TEST FAILURE 46 Error count: total[544048] chH[544048] 47[3 ] Mem production test FB064 : PASS 48[4 ] Mem production test FB065 : PASS 49[5 ] Mem production test FB066 : FAIL 50 Error ID G0FB096 51 1024 x 768 - argb8888 (60 Hz): TEST FAILURE 52 Error count: total[36076] chH[36076] 53[6 ] Modified FB088 : FAIL 54 Error ID G0FB088 55 1024 x 768 - argb8888 (60 Hz): TEST FAILURE 56 Error count: total[39245] chH[39245] 57[7 ] Modified FB091 : FAIL 58 Error ID G0FB091 59 1024 x 768 - argb8888 (60 Hz): TEST FAILURE 60 Error count: total[43036] chH[43036]
zuldanwrote on 2026-05-03, 00:18:Got some time to look at this again, I'm now certain it's a memory issue (or some issue with the power rail for the memory). I b […] Show full quote
Interestingly your GPU-Z Screenshot shows the Fan not changing it's speed? It looks like it's fixed at 13%?
Btw. what a lovely card. Fond memories of me looking through magazines and imagining how awesome it would be to have this card. I had an X700 AGP at the time 😁
Can you monitor the VRM temperatures and VDDC externally?
Got some time to look at this again, I'm now certain it's a memory issue (or some issue with the power rail for the memory). I booted into Windows and immediately started getting artifacts on the desktop (without even running 3dMark and GPU cool). I pushed down on the top row of memory chips and I could get the artifacts to change.
The attachment Desktop.JPG is no longer available
I then ran R5MEMID again, and it passed.
The attachment R5MEMID_1.JPG is no longer available
I pressed down on the top row of memory cards and bingo I got artifacts on the screen
The attachment R5MEMID_2.JPG is no longer available
I ran R5MEMID again, finally got errors
1R5MEMID version 1.03, (c) Copyright ATI Technologies Inc, 2006 2Log file generation enabled to .\R5MEMID.LOG ... 3Extended logging enabled ... 4Reference data file (RDF) loading disabled ... 5No wait on error enabled (not effected by '-s') ... 6 [1] Mem production test FB001 7 8 Failing Bits 9 rk0 chH: 5 6 8 10 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 10 11 [2] Mem production test FB063 12 13 Failing Bits 14 rk0 chH: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 15 16 [3] Mem production test FB064 17 [4] Mem production test FB065 18 [5] Mem production test FB066 19 20 Failing Bits 21 rk0 chH: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 22 23 [6] Modified FB088 24 25 Failing Bits 26 rk0 chH: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 27 28 [7] Modified FB091 29 30 Failing Bits 31 rk0 chH: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33TEST RESULT SUMMARY: 34==================== 35R580/M59 (0x7240) detected. 36512M video memory. 37Test suite ran 1 of 1 times. 38Checking test status array ... 39[1 ] Mem production test FB001 : FAIL 40 Error ID G0FB096 41 1024 x 768 - argb8888 (60 Hz): TEST FAILURE 42 Error count: total[29295760] chH[29295760] 43[2 ] Mem production test FB063 : FAIL 44 Error ID G0FB096 45 1024 x 768 - argb8888 (60 Hz): TEST FAILURE 46 Error count: total[544048] chH[544048] 47[3 ] Mem production test FB064 : PASS 48[4 ] Mem production test FB065 : PASS 49[5 ] Mem production test FB066 : FAIL 50 Error ID G0FB096 51 1024 x 768 - argb8888 (60 Hz): TEST FAILURE 52 Error count: total[36076] chH[36076] 53[6 ] Modified FB088 : FAIL 54 Error ID G0FB088 55 1024 x 768 - argb8888 (60 Hz): TEST FAILURE 56 Error count: total[39245] chH[39245] 57[7 ] Modified FB091 : FAIL 58 Error ID G0FB091 59 1024 x 768 - argb8888 (60 Hz): TEST FAILURE 60 Error count: total[43036] chH[43036]
…Show last 2 lines
61Failure detected.
Just asking the obvious question: If the errors appear on boot with no pressure applied, do you have any measures to test the card for 20s with no cooler on memory so you can push down only on channel H?
Just asking the obvious question: If the errors appear on boot with no pressure applied, do you have any measures to test the card for 20s with no cooler on memory so you can push down only on channel H?
When I pressed down on individual memory chips I couldn’t pinpoint one that caused the most disruption on the screen.
Brand new cores for this GPU have arrived. I’m tempted to replace the core and if that doesn’t work, reball the memory chips (new chip replacements are almost impossible to get, Samsung GDDR4 K4U52324QE - B609 0.9ns). You know I love replacing parts 😂
On a side note, the small rectangular heat sink covering the VRM section wasn’t screwed to the board (I must have unscrewed it in my last repair session and forgot about it), so when I lifted the card up the heat sink “slid” out and damaged a VT1115SF chip (https://www.aliexpress.com/item/1005010604836806.html), totally my fault. There is an error LED that light that shows up now where you plug in the PCIE connection. Hopefully the replacement part should be here in 2 weeks.
zuldanwrote on 2026-05-06, 08:59:When I pressed down on individual memory chips I couldn’t pinpoint one that caused the most disruption on the screen. […] Show full quote
Just asking the obvious question: If the errors appear on boot with no pressure applied, do you have any measures to test the card for 20s with no cooler on memory so you can push down only on channel H?
When I pressed down on individual memory chips I couldn’t pinpoint one that caused the most disruption on the screen.
Brand new cores for this GPU have arrived. I’m tempted to replace the core and if that doesn’t work, reball the memory chips (new chip replacements are almost impossible to get, Samsung GDDR4 K4U52324QE - B609 0.9ns). You know I love replacing parts 😂
On a side note, the small rectangular heat sink covering the VRM section wasn’t screwed to the board (I must have unscrewed it in my last repair session and forgot about it), so when I lifted the card up the heat sink “slid” out and damaged a VT1115SF chip (https://www.aliexpress.com/item/1005010604836806.html), totally my fault. There is an error LED that light that shows up now where you plug in the PCIE connection. Hopefully the replacement part should be here in 2 weeks.
Ok not impossible if you happy to pay 🤣
I found that too 😝 but $10 USD a chip, that’s $80 USD to replace all the chips, ouch!
In theory you could remove the assumed chip. If it's really channel H, it'll probably work as bad as it does now, since it appears all 32 bits of it are affected. That would tell if your assumption is right. Not sure what path I'd go here.. having to swap the ASIC in hope it fixes it is also a bold move 😁
Ok not impossible if you happy to pay 🤣
I found that too 😝 but $10 USD a chip, that’s $80 USD to replace all the chips, ouch!
In theory you could remove the assumed chip. If it's really channel H, it'll probably work as bad as it does now, since it appears all 32 bits of it are affected. That would tell if your assumption is right. Not sure what path I'd go here.. having to swap the ASIC in hope it fixes it is also a bold move 😁
That is a good idea. I will try it.
I’ve swapped ASICS a few times with success. It is scary but the card is unusable at the moment. I also have another fully functional X1950 XTX so not too worried if this card becomes For Parts card.
I got this X1950 XTX that is in physical "like new" condition. No missing components. It appears to be working perfectly in games and benchmarks BUT when the core reaches somewhere between 73c and 78c, it crashes.
Yeah, that last sentence might be part of the problem as to why.
Anything upwards of 70C is a little too hot for these cards, IMO.
While ATI didn't officially admit to a "bumpgate" issue, it's from the same fab and made on the same process as nVidia GPUs at the time, so I'm pretty sure some similar issues existed. The Xbox 360, whose GPU is also from the same years and fab is a proof of this: the Xbox 360 ran hot and consequently the GPU was the most failure-prone part... and by far.
Start with a better cooler and see if you can keep the GPU under 65C under load. Under 60C could be even better, but that may prove difficult unless you do something "wild", like attaching a large(r) CPU heatsinks (something along the size of a stock Core 2 Quad heatsink). I've done that to an artifacting HD4850 that I reflowed, and so far it's still holding up. One thing I noticed from the HD4850 (and also some HD3870 cards) is that when the GPU core is at some XX temperature, the "MEMIO" probe could be up to 10C higher, as the MEMIO area is burried deeper into the GPU core (IIRC) and thus running hotter. So when you see 70-75C temperatures, parts of the GPU silicon could be even higher than that.
momakawrote on 2026-05-14, 16:56:Yeah, that last sentence might be part of the problem as to why.
Anything upwards of 70C is a little too hot for these cards, IM […] Show full quote
I got this X1950 XTX that is in physical "like new" condition. No missing components. It appears to be working perfectly in games and benchmarks BUT when the core reaches somewhere between 73c and 78c, it crashes.
Yeah, that last sentence might be part of the problem as to why.
Anything upwards of 70C is a little too hot for these cards, IMO.
While ATI didn't officially admit to a "bumpgate" issue, it's from the same fab and made on the same process as nVidia GPUs at the time, so I'm pretty sure some similar issues existed. The Xbox 360, whose GPU is also from the same years and fab is a proof of this: the Xbox 360 ran hot and consequently the GPU was the most failure-prone part... and by far.
Start with a better cooler and see if you can keep the GPU under 65C under load. Under 60C could be even better, but that may prove difficult unless you do something "wild", like attaching a large(r) CPU heatsinks (something along the size of a stock Core 2 Quad heatsink). I've done that to an artifacting HD4850 that I reflowed, and so far it's still holding up. One thing I noticed from the HD4850 (and also some HD3870 cards) is that when the GPU core is at some XX temperature, the "MEMIO" probe could be up to 10C higher, as the MEMIO area is burried deeper into the GPU core (IIRC) and thus running hotter. So when you see 70-75C temperatures, parts of the GPU silicon could be even higher than that.
So IMO deal with the temperature first.
Under load it's normal for these cards to run 70c to 80c (this is the temp range I've seen on three X1950 XTX so far). Under extreme stress it can even reach 90c (during demanding benchmarks). Unfortunately, ATI didn't put great coolers on these cards. Yes, bumpgate is possible. Appreciate your insights 😀
During testing I cracked a voltage regulator (VT1115SF). I didn't screw in the VRM heatsink on the side and it fell out while lifting the card, hitting the VT1115SF. If anything is wrong with the VRM the EXT_PWR light turns on when trying to power the card. Unlike Nvidia cards (VRM still turns on if there is an issue), the entire VRM is powered off. The power usage of the card was at 0.3 amps.
The attachment Broken voltage regulator.JPG is no longer available
This has now been replaced with a new VT1115SF and the card came back to life. EXT_PWR light stayed off and the card now used 1.8 amps according to my custom PCI-E power meter. A noob version of tehsiggi's AGP power monitor 😉
The attachment Power Meter.jpg is no longer available
Anyway, proceeding with the repair, before replacing or reflowing chips on a card I always take some resistance measurements to make sure I haven't shorted something after the procedure. I've left a copy here to help someone else in the future.
The attachment X1950XTX Front - Resistances.jpg is no longer available
I ran ATI memory tool a final time before the repair, this time I got errors with chip H and chip G (a few errors). I decided to reflow memory chip H and reflow the core. After reflowing, all the resistances checked out fine but this time when testing the power usage of the card only went up to 1.68 amps instead of 1.8 amps. Not having much hope (I was getting prepared to replace the core), the card passed perfectly with the ATI tool. I pressed on the memory chips, tried my best to get the memory test to fail and I couldn't. WTF
1R5MEMID version 1.03, (c) Copyright ATI Technologies Inc, 2006 2Log file generation enabled to .\R5MEMID.LOG ... 3Extended logging enabled ... 4Reference data file (RDF) loading disabled ... 5No wait on error enabled (not effected by '-s') ... 6 [1] Mem production test FB001 7 [2] Mem production test FB063 8 [3] Mem production test FB064 9 [4] Mem production test FB065 10 [5] Mem production test FB066 11 [6] Modified FB088 12 [7] Modified FB091 13TEST RESULT SUMMARY: 14==================== 15R580/M59 (0x7240) detected. 16512M video memory. 17Test suite ran 1 of 1 times. 18Checking test status array ... 19[1 ] Mem production test FB001 : PASS 20[2 ] Mem production test FB063 : PASS 21[3 ] Mem production test FB064 : PASS 22[4 ] Mem production test FB065 : PASS 23[5 ] Mem production test FB066 : PASS 24[6 ] Modified FB088 : PASS 25[7 ] Modified FB091 : PASS 26No failures detected.
This is my card that has always worked (in a different machine).
The attachment X1950 XTX Original Working.png is no longer available
This is my card that was repaired.
The attachment X1950 XTX Broken Working.png is no longer available
I've done my best to try get the card to fail but I can't. It's passed everything I've thrown at it with flying colors. Either the issue was bumpgate OR the original owner bumped the card at some stage causing some solder joints under the CPU to crack. If I get the same issue in the future I'll replace the core but at the moment the card is 100% WORKING. WOOHOO!!!
