Some more R300 series fun in the lunch break.

So I build the jankiest temperature controlled setup possible to characterize the NTC on R300/R350/R360 GPUs. It consists of a styrofoam box, a 40mm fan, a 20W light bulb, Uni-T thermometer and a cheap module to control the lamps heating.

I used a voltage divider consisting of the GPUs NTC and a 100kOhm 0.1% low-temp-co resistor, powered by a lipo cell (gives nice, clean and stable voltage) to determine the resistance of the NTC.

The results look like this:

I know, it's no-where scientific, but for me it's honestly good enough.

So calculated my results look like this:

Beta Model Parameters
Beta (B): 4329.96 K
R25: 94652.81 Ω

Steinhart–Hart Coefficients
A = 0.0008714636826900755
B = 0.00020751383834629025
C = 0.00000006987140028267289

It's relatively safe to assume it's a beta 4250K part, which is pretty common.

This will now help me to read the temperature on R300 based cards, pretty close to the die. Unfortunately, no diode in the die. But hey, better than nothing.

Haha, love those setups.
Looks absolutely scientific 😁

Well, it at least got close enough and produced usable values 😁

And I only bought the thermostat module, as I needed that one anyways. The rest was laying around. Guess I'll make it a science fair project?

EDIT:

ChrisK wrote on Yesterday, 13:14:

Looks absolutely scientific 😁

Okay, I couldn't resist. MORE JANK! For science of course.

STM32 Bluepill, some I2C display + a quick code with the Steinhart-Hart values. I am impressed. The top temperature on the UniT is the thermal probe on the R350, the LCD shows the calculated temperature from the GPUs NTC.

@tehsiggi hope this photo helps, came from the guy I got the reference DELL 9800XT from. Currently can't unscrew the cooler as I need to source some 1mm or 0.8 Arctic pads. I do notice a transistor close to the core? Is that the specific part you're looking for?

PcBytes wrote on Yesterday, 15:37:

@tehsiggi hope this photo helps, came from the guy I got the reference DELL 9800XT from. Currently can't unscrew the cooler as I need to source some 1mm or 0.8 Arctic pads. I do notice a transistor close to the core? Is that the specific part you're looking for?

Yep, that picture is already good enough. Clearly using the transistor as the sensor for the core temperature. Thank you very much!

TechieDude wrote on Yesterday, 01:38:

Well, after my P3 Voodoo3 build started to throw RAM errors AGAIN, even after already replacing the DIMMs, it was probably time […]
Show full quote

Well, after my P3 Voodoo3 build started to throw RAM errors AGAIN, even after already replacing the DIMMs, it was probably time to retire it. That QDI Advance 10T served me well for a full decade, but failing DIMMs so soon after the previous RAM failure is VERY suspicious... I only had these in that board for 2 weeks...
But what could replace it anyway? The reason I had that system was to have the best possible compatibility with 90's games, while also having good performance. Not exactly period-correct, since I never really go for that, although its original specs were very much period-correct, with a Soyo SY-6VBA133, P2 400, SBLive! and Voodoo3, it quickly evolved to a PGA370 P3 (mostly because the original board died, and I ended up upgrading to the QDI), only keeping the Voodoo and SBLive! constantly. It even was a P4 build briefly, before settling on that QDI board (I really dislike Netburst, and for some reason that particular config caused problems with some games anyway.) That QDI in of itself went from P3 650MHz to Tualeron 1.2GHz, to P3-S 1266MHz OCed @ 1.3GHz. Surprisingly, the same Win98SE install took all those changes (even the motherboards) like a champ over the years without a hitch. I wouldn't really mind reinstalling the OS, it just didn't happen to be necessary.
So, what is the next step? Well, I have an MSI K7T266 Pro2 V2.0 (with the extra Promise FastTrak100 and NEC USB2 builtin), a few socket 462 CPUs, plenty of DDR DIMMs (although just one 512MB was more than enough while also being perfectly stable without patches), and more than enough time to rebuild everything. I settled on an Athlon XP 1800+ Palomino (~1.56GHz), 512MB DDR400 (allows for very tight timings on 266 speeds), and basically kept everything else the same. Build went nice, almost too nice...
Just for shits and giggles, I hooked up that same Win98SE HDD to see if that will work. And well... IT ACTUALLY DID! 🤯 No BSOD, no conflicts, no falling back to DOS Compatibility Mode, just automatically reinstalling the proper drivers for the new configuration, and after that, happily playing the startup sound, and loading my desktop. One reboot just for good measure, and everything is still fine. Now,I'm no stranger to getting Windows XP and later installs working on different motherboards even when SATA controllers were different (injecting drivers on existing installations is especially easy on Vista and later, and yes, I also did it just for fun), but that was from G41 chipset to Z77, and it still needed some finagling with drivers and reboots, even though when I was done, it was just fine. This? I expected it to be considerably worse, since everyone here knows how temperamental Win9x can be, but it was even smoother instead. It probably helped that they were similar enough VIA chipsets, and used the same drivers.
I also have another HDD on that PC with WinXP, mostly to see if I can get working Glide support in that, but the drivers have failed me there again and again. (Any recommendations for XP Voodoo3 drivers??). I also tried that, but ironically, that failed spectacularly with an UNMOUNTABLE_BOOT_VOLUME. Not good, and that's not due to the motherboard being different, it might be an actual HDD problem.

Dumbass me celebrated too damn early, I also got to experience the joy of a Sound Blaster Live! shitting the bed with a VIA chipset... For some reason, I had never encountered that bug, because it apparently only happens when the card shares IRQs with other devices, and through all my PCs, it somehow always had its own IRQ, but not here. No matter the slot, it just HAS to share with some stupid device now... Guess I might upgrade to an Audigy 2 ZS, though I REALLY like the Live Drive IR too much to get rid of it.

tehsiggi wrote on Yesterday, 15:39:

PcBytes wrote on Yesterday, 15:37:

@tehsiggi hope this photo helps, came from the guy I got the reference DELL 9800XT from. Currently can't unscrew the cooler as I need to source some 1mm or 0.8 Arctic pads. I do notice a transistor close to the core? Is that the specific part you're looking for?

Yep, that picture is already good enough. Clearly using the transistor as the sensor for the core temperature. Thank you very much!

You're welcome! For what I paid on thjis card (way too much for a R360, I know, but it's working and it passed the endurance tests I've gave it) it's nice to see it helps us both. Paid circa 93 euro on it, and it came with 24mW/K pads (holy overkill Batman!) from a brand called Upsiren + MX4 repaste on core, as well as the whole card cleaned up.

PcBytes wrote on Yesterday, 23:27:

You're welcome! For what I paid on thjis card (way too much for a R360, I know, but it's working and it passed the endurance tests I've gave it) it's nice to see it helps us both. Paid circa 93 euro on it, and it came with 24mW/K pads (holy overkill Batman!) from a brand called Upsiren + MX4 repaste on core, as well as the whole card cleaned up.

Well, good news is the card is using the reference 9800XT PCB, so you basically got a working 9800XT for 100€ - which I think is pretty decent. I'll think I'll never own one. The hunt is just to much for me to bear and most of my interest is more scientific than any real-world usage. Seeing how they are traded, it feels you'd not really want to use it in daily retro machines. Which I think is a pity, same as for the 9700Pros - which got so rare, people rather put them on shelves and collect them rather than using them.

I'll take a look today at the unloved child of the R300 Series, the 9500 Pro. Got one here that had a rough time, so I'll see if I can fix it up.

PS: The one cap near the slot bracket got a funky repair 😉