Not sure which ones you'd consider weird 😉
The mobility radeon 9600 from Danger Manfred is weird in my world.

I don't have any 3dfx cards in my collection, so that'd be on someone else.

I only have a V3 3000 AGP - would have to test that in a different rig from the others, because my normal benchtable is incompatible with the oldest AGP cards (ASRock ALiveDual-eSATA 2, AMD Phenom II X2 555 Black Edition @ 4GHz, 2x2 GB DDR2 1066 CL5-5-5-15).

I do have the Radeon 9000 series mostly complete (no 9700 Pro) and most of Geforce 2 to 7 cards, too. Even some unicorns like the Mobility Radeon 7000 and 7500 cards by Vampower which carry additional power for professional displays. Or the Geforce 6600 LE, I guess that one's for masochists.
The Medion Radeon 9000 cards might be considered weird, too: Radeon 9600 TX (sounds like a 9600 XT but is actually a 9500 Pro with cheaper TSOP RAM with more loose timings but making up for it with higher clockspeed), or the 9800 XL (XT core but below Pro clockspeeds and cheapo RAM) and XXL (also XT core, but almost XT clockspeeds and cheapo RAM).

Added a plain Radeon 9700 to the list.
It's consumption is a bit higher than a 256bit 9500, which is to be expected.

I took a minute and added some charts to it, making it easier to identify which rail is used most by a quick look.

I added three more cards:
- Radeon 9500 Pro - slightly lower consumption than a 9500 non Pro. While it still has the same amount of memory chips, the halved memory bus compensates for the power consumption of the more pixel pipes.
- Radeon 9600XT - just because I have it - glad to have it on the list
- Radeon HD3450 512MB - Curiosity just got me here. And now that I've measured it, I'm not sure what drugs the engineers of that card have been on. It could easily be powered by AGP slot only.

I'd be really interested in seeing some numbers for early Nvidia cards. Riva 128, TNT, TNT 2, 256 etc

I think I have a TNT2 laying around somewhere.. might get to test it next week. Apart from that I don't have many cards from "that era" if that makes sense, my stash starts with DX8 mostly.

tehsiggi wrote on 2026-03-16, 13:25:

Radeon HD3450 512MB - Curiosity just got me here. And now that I've measured it, I'm not sure what drugs the engineers of that card have been on. It could easily be powered by AGP slot only.

Since it's predominantly 12v line, I think they just played it safe, because 20-pin ATX connector has only one wire with 12v.

The Serpent Rider wrote on Yesterday, 18:01:

tehsiggi wrote on 2026-03-16, 13:25:

Radeon HD3450 512MB - Curiosity just got me here. And now that I've measured it, I'm not sure what drugs the engineers of that card have been on. It could easily be powered by AGP slot only.

Since it's predominantly 12v line, I think they just played it safe, because 20-pin ATX connector has only one wire with 12v.

I suspect as well they just went "this is the same BOM parts as for the regular PCIe card, we'll power it from AGP, AGP has bad 12V power delivery so we use an external input".

I wonder however, if they couldn't just use the 5V rail for it instead. But whatever, they've done it the way they did. Now we at least know for sure where this card draws its power from.

tehsiggi wrote on Yesterday, 15:13:

I think I have a TNT2 laying around somewhere.. might get to test it next week. Apart from that I don't have many cards from "that era" if that makes sense, my stash starts with DX8 mostly.

I'm interested in early AGP for Super Socket 7 use.

A lot of SS7 have very poor voltage provision for the AGP slot and the vrms cook themselves when cards draw too much power.

I'm keen to dial in on what a safe power draw is for these boards alongside working out which cards fall into that category.

My Voodoo 3 for example draws too much and the VRMs on the AGP hit 100 degrees C.

Lostdotfish wrote on Today, 11:02:

I'm interested in early AGP for Super Socket 7 use. […]
Show full quote

tehsiggi wrote on Yesterday, 15:13:

I think I have a TNT2 laying around somewhere.. might get to test it next week. Apart from that I don't have many cards from "that era" if that makes sense, my stash starts with DX8 mostly.

I'm interested in early AGP for Super Socket 7 use.

A lot of SS7 have very poor voltage provision for the AGP slot and the vrms cook themselves when cards draw too much power.

I'm keen to dial in on what a safe power draw is for these boards alongside working out which cards fall into that category.

My Voodoo 3 for example draws too much and the VRMs on the AGP hit 100 degrees C.

My Riva TNT2 works, I'll have some complete measurements for you tomorrow. First test showed 5W idle on desktop and 9W peak in Half-Life 1 - coming from the 3.3V rail.

Lostdotfish wrote on Yesterday, 14:33:

I'd be really interested in seeing some numbers for early Nvidia cards. Riva 128, TNT, TNT 2, 256 etc

I have all these cards measured. I will post the results in few days when I get back to my workshop 😀

havli wrote on Today, 13:25:

Lostdotfish wrote on Yesterday, 14:33:

I'd be really interested in seeing some numbers for early Nvidia cards. Riva 128, TNT, TNT 2, 256 etc

I have all these cards measured. I will post the results in few days when I get back to my workshop 😀

I take from that that the APM is doing it's job for you? Great stuff!

Oh yes, works perfectly 😀
I have measured dozens of AGP cards - all the old parts since the beggining of AGP bus till GeForce2 / Radeon DDR... Including Voodoo3/4/5, Matrox, Savage4 and some less common GPUs. There are still some left unfinished. For example Rage Fury MAXX - it refuses to work with my test system... I must find more compatible board. 😀 When I'm done, I will post all the results here and also add them to my articles on hw-museum. For some cards I even have the logs saved, if you are interested I can send them to you, so you can add them to your database.

havli wrote on Today, 15:54:

Oh yes, works perfectly 😀
I have measured dozens of AGP cards - all the old parts since the beggining of AGP bus till GeForce2 / Radeon DDR... Including Voodoo3/4/5, Matrox, Savage4 and some less common GPUs. There are still some left unfinished. For example Rage Fury MAXX - it refuses to work with my test system... I must find more compatible board. 😀 When I'm done, I will post all the results here and also add them to my articles on hw-museum. For some cards I even have the logs saved, if you are interested I can send them to you, so you can add them to your database.

That would be much appreciated! Glad to hear it proves to be a useful tool. Did we ever talk about calibration? I'm not sure anymore.
I love to see those results.. finally we have real validated data and no more guesswork.. and my collection just doesn't go back that far 😁

Yes, you told me how to perform calibration. In the end, I didn't do it and left default readings uncalibrated. I like the clean design of the PCB without jumpers 😁 And I figured it would be better to calibrate each rail and thus have several correction factors. This would require modification of the software and I was too lazy to do so 😁
Even with slight inaccuracy, the device is still great.

Interesting enough, the deviation for the rails was basically the same for all of them in my case, since all ina3221s came out of the same batch.