Tested a whole bunch of early-2000s video cards. Mostly FX 5200 and MX440 cards. Then updated some missing information on the nvidia GPU chart wiki to save my sanity. It's kind of crazy how the charts have such detailed information, and yet some things just seem to be missing entirely. Like the fact that MX440 (NV17 or NV18 AGP 8x cards) cards with 64bit 166(333)Mhz memory exist. I think I have one single Geforce 4MX card which has a 128bit memory bus... and probably a dozen or more that are 64bit, and those are almost all labeled "MX440", and nearly all of those have 166Mhz memory.
I had to oil a fan on a PNY Verto Ti 4200 64MB (a beast of a value card at the time!) and one of the screws had clearly been mangled by a previous owner at some point. All of my cheapo mini screwdriver\bit sets were doing a terrible job of getting the stubborn screw out because the Phillips head had been almost entirely ground out. The metal on my screwdriver bits was just too soft. Thankfully, my almost new Ifixit repair kit had some much higher quality bits that were hard enough to get the job done. I am SOOOO glad I picked up that kit with my MicroCenter "New Customer" discount when I traveled past one a few months ago. I was disappointed that there were really NO good deals on anything useful in the store (highly overrated place outside of the crazy deals that pop up from time to time), so I just used the coupon on something that I otherwise would never be able to get cheaply. Got what I thought was an overpriced ~$30 kit for like $5, and I see now that it was absolutely worth it. I've only ever had good quality large tools... never good quality tiny tools.
Also, as my final retro activity worth mentioning... After many many runs of 3dMark2001SE testing the hardware mentioned above, I threw in a Quadro FX3000 to see how much better some of the tests ran. Then it hit me... I haven't run 3dmark2001SE on my main PC in probably 15 years... at least! So I installed it on my main PC (Ryzen 5800X3D, RTX 3060 Ti) and wow... it's amazing how bad it makes the textures and models look when they are pegged at 999fps at 1080P with 8xAA. I didn't run the full benchmark suite, but just as a quick comparison, the FX3000 (basically a Geforce FX5900 non-ultra) on an Athlon XP@2Ghz got like ~87fps average in Nature at 1024x768, default settings. My modern system managed like 1150fps at 1920x1080 with 8xAA, and just under 1300fps at default settings. The high poly count 8-light test was like 27Mvertices\sec on the FX3000 and around 2300Mvertices\sec on the new system. And the difference in noise level between the GPUs is... non-trivial. 🤣
And my 3060Ti is barely even worth mentioning compared to the latest high end GPUs. Truly amazing how far real time computer graphics has come in the past 20 years, in spite of the thermal, power and process limitations.
EDIT: I would just like to add... I don't want to hear anyone complaining about whether the latest GPUs are named "properly" for their performance\price. I just tested an XFX GeforceFX 5600XT. This card has the exact same configuration as a 5200 (non-ultra) except it has 64bit 166Mhz DDR and a 235MHz core clock, compared to the average 5200's 64bit 200mhz DDR and 250Mhz core clock. So, despite the (awful) active cooler, it is measurably slower than even the cheapest OEM FX5200 card with passive cooling. Wow... What a joke. At least the caps are okay, despite being "RIFEKING" brand. I'm glad XFX upped their game after this era.