Wow, this looks awesome!

DrAnthony wrote on 2025-06-02, 19:19:

Wow, this looks awesome!

Thank you!
One step further to de-mystify some properties of AGP era graphics cards!

Cool development, congratulations.

tehsiggi wrote on 2025-06-02, 18:36:

I am creating a reference value sheet in my repository as well.

Please specify the units of measurement.
I couldn't figure out for a long time whether it was Watts or Amps. 🙁

shevalier wrote on 2025-06-03, 05:10:

Cool development, congratulations. […]
Show full quote

Cool development, congratulations.

tehsiggi wrote on 2025-06-02, 18:36:

I am creating a reference value sheet in my repository as well.

Please specify the units of measurement.
I couldn't figure out for a long time whether it was Watts or Amps. 🙁

Will add that later today.. the header "Total Power" would imply watts, but I'll just add that in the lunch break 😁

Thanks!

Great job!
Very interesting device. And just in time when I am preparing another round of GPU tests. I am seriously considering building one. 😀
Perhaps modifying this design to measure PCI cards would be also possible. Different PCB to route PCI instead of AGP.

That's very nice!

AGP voltages at a glance and ability to mount the card sideways to work on it comfortably. For pci-e it is easy - there are plenty of raisers (thanks miners), but for AGP - not so much. Just recently had to work on a card and trying to measure things on the card itself sideways while not shorting something is quite annoying.

Measuring power is very interesting and useful too, may help troubleshoot issues with specific cards on specific systems...

havli wrote on 2025-06-03, 06:53:

Great job!
Very interesting device. And just in time when I am preparing another round of GPU tests. I am seriously considering building one. 😀
Perhaps modifying this design to measure PCI cards would be also possible. Different PCB to route PCI instead of AGP.

A PCI version would be even simpler from many aspects: Goldfinger design, PCB connector and of course, signal lines (there are fewer). I didn't have the need for a PCI variant, as my work is mostly around AGP right now. But the principle remains the same. Especially since most PCI cards I know don't use external power connectors anyway. That whole part could go away. If you don't care about the -12V rail, you can live with a single INA3221. If you don't need extensibility, instead of a full blown STM32F once might use a MCP2221 (USB to I2C converter) and skip the whole firmware part completely.

Sky is the limit I guess 😁

If you consider building one yourself, i might make small adjustments in the upcoming days to the AGP gold fingers. As mentioned in the repository, they are slightly offset on part 2 and 3. Nothing dramatic, but I can't sleep well 😁

Thanks for the kind words!

Archer57 wrote on 2025-06-03, 06:54:

That's very nice!

AGP voltages at a glance and ability to mount the card sideways to work on it comfortably. For pci-e it is easy - there are plenty of raisers (thanks miners), but for AGP - not so much. Just recently had to work on a card and trying to measure things on the card itself sideways while not shorting something is quite annoying.

Measuring power is very interesting and useful too, may help troubleshoot issues with specific cards on specific systems...

Thanks!

Yes, the "back up" thing of it is nice as well. You can easily access the rear side of the PCB. But it has its downsides as well: Large coolers (and collision with PCI cards). I have a PCI SATA controller because the Boards controller only supports SATA1. I needed to get 90° riser for that one, else it'll collide.

I thought about going with a flexible approach, however AGP is finicky as it is, so I left that out (for now). I even had cards that don't like to run in a riser at all. (tested with APM as well with a plain off-the-shelf riser)

Cheers!

tehsiggi wrote on 2025-06-03, 07:15:

Thanks! […]
Show full quote

Thanks!

Yes, the "back up" thing of it is nice as well. You can easily access the rear side of the PCB. But it has its downsides as well: Large coolers (and collision with PCI cards). I have a PCI SATA controller because the Boards controller only supports SATA1. I needed to get 90° riser for that one, else it'll collide.

I thought about going with a flexible approach, however AGP is finicky as it is, so I left that out (for now). I even had cards that don't like to run in a riser at all. (tested with APM as well with a plain off-the-shelf riser)

Cheers!

Most of the time i'll be poking around the card with multimeter something is wrong and it probably is not posting at all, so pci cards would not really be a concern. Though if you want to run stuff like 3dmark and measure current/power i see how it can be an issue.

By the way for those annoying issues with sata (via?) - IDE-SATA adapters work just fine most of the time, so that's an option and a way to avoid using PCI controller, which can often be even slower than using IDE...

tehsiggi wrote on 2025-06-03, 07:15:

A PCI version would be even simpler from many aspects: Goldfinger design, PCB connector and of course, signal lines (there are f […]
Show full quote

A PCI version would be even simpler from many aspects: Goldfinger design, PCB connector and of course, signal lines (there are fewer). I didn't have the need for a PCI variant, as my work is mostly around AGP right now. But the principle remains the same. Especially since most PCI cards I know don't use external power connectors anyway. That whole part could go away. If you don't care about the -12V rail, you can live with a single INA3221. If you don't need extensibility, instead of a full blown STM32F once might use a MCP2221 (USB to I2C converter) and skip the whole firmware part completely.

Sky is the limit I guess 😁

If you consider building one yourself, i might make small adjustments in the upcoming days to the AGP gold fingers. As mentioned in the repository, they are slightly offset on part 2 and 3. Nothing dramatic, but I can't sleep well 😁

Thanks for the kind words!

I am in no rush. I still need to finish some other projects that will take few more months at least. After that I can start building your power measuring device 😀
Yeah, PCI version could be somewhat cut-down. I'll take a look at that in the future. My goal is to benchmark cards from the beggining of the 3D era and many of these are in PCI form.

Btw - how hard would it be to add frequency measuring to the board - similar to this? Original Hardware: PCI / ISA / AT Clock measuring device.

And one more question - I assume this power monitor is compatible with both 1.5V and 3.3V AGP, right? The universal connector must be there for a reason 😀

Archer57 wrote on 2025-06-03, 07:27:

By the way for those annoying issues with sata (via?) - IDE-SATA adapters work just fine most of the time, so that's an option and a way to avoid using PCI controller, which can often be even slower than using IDE...

My board has.. issues. Let's put it this way. I'm currently not in the mood to take care on it. Not my wish platform for my AGP testing, but that'll be a future project. I've got one of those SATA -> IDE adaptors here and it works just fine. The setup using the SI3114 card is just so convenient. I did not miss the old IDE cables, that's for sure!

havli wrote on 2025-06-03, 07:34:

I am in no rush. I still need to finish some other projects that will take few more months at least. After that I can start buil […]
Show full quote

I am in no rush. I still need to finish some other projects that will take few more months at least. After that I can start building your power measuring device 😀
Yeah, PCI version could be somewhat cut-down. I'll take a look at that in the future. My goal is to benchmark cards from the beggining of the 3D era and many of these are in PCI form.

Btw - how hard would it be to add frequency measuring to the board - similar to this? Original Hardware: PCI / ISA / AT Clock measuring device.

And one more question - I assume this power monitor is compatible with both 1.5V and 3.3V AGP, right? The universal connector must be there for a reason 😀

Regarding frequency measurement: Iirc, you can measure around 30MHz with the STM32F103. So for the AGP clock either some prescaling or co-processing would be needed. I considered that for a minute, but then it slipped off my mind, since I was more interested in the power management side of things.

It should work just fine with any AGP voltage. There is no limitation from the implementation side of things. However I just don't have a 3.3V platform laying around right now, so I haven't tested it with that.

shevalier wrote on 2025-06-03, 05:10:

Cool development, congratulations. […]
Show full quote

Cool development, congratulations.

tehsiggi wrote on 2025-06-02, 18:36:

I am creating a reference value sheet in my repository as well.

Please specify the units of measurement.
I couldn't figure out for a long time whether it was Watts or Amps. 🙁

I've update the document with mention of all values on the page being in Watts and added the GeForce 7600GS and 6800XT to it as well.

Cheers.

tehsiggi wrote on 2025-06-02, 18:36:

[*]Radeon X700 256MB AGP
[*]Radeon X1950GT AGP

Do these cards have an AGP to PСI-е bridge?

RIALTO theoretically should be powered by VDDq, its consumption and compliance with the AGP3.0 standard are interesting.
https://www.playtool.com/pages/agpcompat/agp30.pdf
p.95

PS. Half of the owners of ATI/AMD cards with a bridge are happy, another half - do not work anything.
Although, perhaps, this is something karmic. 😀

shevalier wrote on 2025-06-16, 07:28:

Do these cards have an AGP to PСI-е bridge? […]
Show full quote

tehsiggi wrote on 2025-06-02, 18:36:

[*]Radeon X700 256MB AGP
[*]Radeon X1950GT AGP

Do these cards have an AGP to PСI-е bridge?

RIALTO theoretically should be powered by VDDq, its consumption and compliance with the AGP3.0 standard are interesting.
https://www.playtool.com/pages/agpcompat/agp30.pdf
p.95

x1950 does have the bridge, x700 should be native AGP.

And yeah, it would be extremely interesting to see how this cards behave in terms of power requirements and signalling. Something must be off somewhere - way too many people are reporting issues. They can be ignored saying "you hardware is faulty" or "you are doing something wrong", but IMO what this means is - something is different about the cards, either something is out of spec or very close to the limits and some motherboards are slightly below spec.

shevalier wrote on 2025-06-16, 07:28:

PS. Half of the owners of ATI/AMD cards with a bridge are happy, another half - do not work anything.
Although, perhaps, this is something karmic. 😀

😁

I do not know what i did to anger AMD gods or something, but in the last few years i've had really bad luck with their cards. They just do not want to work correctly on my systems. Do not know what it is, may be it is something about my motherboard, PSU or whatever choice, but i am not making this up - it is a real problem.

I consciously try to not be a fan of some company and always check all options, but when something does not work multiple times on multiple systems - buying more of it just feels wrong.

By the way i've plugged that 2600XT into dell workstation and it does work flawlessly, but that has really slow 1.8Ghz willamette so not sure if that proves anything.

Hey there.

Both of the cards are AGP to PCIe via the Rialto bridge.
The X700 is using the RV410 GPU which is natively PCIe.

Graphics cards *can* use VDDQ, but that is usually only for their I/O buffers from what I've seen. The current used up on average is low, which kind of matches my assumption.

Sidenote:
The X700 and X1950 straight up refuse to work correctly (with or without a riser) on my K8T800 based MSI mainboard. Driver install works, but the image is just 4bit, low resolution. No setting change or anything helps. I tried *everything*. Any other card, native AGP from ATI or Nvidia, even Nvidia ones using bridge chips, work without any issues on that board.

So to test these cards I switch to my A7N8X, where they just work fine.
This gives an idea on how picky these can be. I used a X700 AGP for a couple of years in my main rig and I can tell you: It's been a diva!

Back to the question:

Rialto is *not* only powered from VDDQ (which would have surprised me a bit). I have the schematic of a gigabyte X700 AGP and that one uses 3.3V from the AGP slot via a linear regulator (MOSTFET + OPAMP) to generate the 1.2V, 1.5V and 1.8V it uses. It also uses the 3.3V from the AGP slot directly for some I/O voltages.
It also uses VDDQ for the AGP part, which again, will probably be for the I/O buffers of the port.

I quickly tested the X700:

IDLE:

• 12V: 1.876 W (1.030 W / 3.153 W)
• 5V: 0.121 W (0.099 W / 0.142 W)
• 3.3V: 4.907 W (4.810 W / 4.929 W)
• VDDQ: 0.042 W (0.037 W / 0.043 W)
• 12V 6pin: 0.000 W (0.000 W / 0.000 W)
• 12V 4pin: 3.433 W (3.312 W / 3.795 W)
• 5V 4pin: 9.947 W (9.842 W / 10.028 W)
• Total Power: 20.327 W (19.131 W / 22.090 W)

3DMark03:

• 12V: 1.879 W (0.414 W / 3.215 W)
• 5V: 0.120 W (0.099 W / 0.141 W)
• 3.3V: 7.325 W (2.803 W / 9.266 W)
• VDDQ: 0.047 W (0.012 W / 0.439 W)
• 12V 6pin: 0.000 W (0.000 W / 0.000 W)
• 12V 4pin: 11.476 W (3.302 W / 22.225 W)
• 5V 4pin: 10.395 W (1.535 W / 13.374 W)
• Total Power: 31.241 W (8.165 W / 48.660 W)

The current on VDDQ never exceeded more than 30mA (which is a mean value, high frequency peaks might be higher.

tehsiggi wrote on 2025-06-16, 10:26:

The current on VDDQ never exceeded more than 30mA (which is a mean value, high frequency peaks might be higher.

https://schematic-x.blogspot.com/2018/04/giga … atics-pack.html
GIGABYTE 7VM400M-RZ - REV1.0.pdf

2*APM3055L = 2*12Amp (0.1Ohm/2)

When you pick up the hardware - "Holy gifts of the Omnissiah, a fine example of the Dark Age of Technology"
When you look at the schematics/reverse engineering - "Heresy! I wish I could burn this..."

tehsiggi wrote on 2025-06-16, 10:26:

The X700 and X1950 straight up refuse to work correctly (with or without a riser) on my K8T800 based MSI mainboard.

offtopic
JetWay K8T8AS - K8T800 too
With HD2600 Pro AGP there were constant glitches.

shevalier wrote on 2025-06-16, 14:40:

https://schematic-x.blogspot.com/2018/04/giga … atics-pack.html GIGABYTE 7VM400M-RZ - REV1.0.pdf […]
Show full quote

tehsiggi wrote on 2025-06-16, 10:26:

The current on VDDQ never exceeded more than 30mA (which is a mean value, high frequency peaks might be higher.

https://schematic-x.blogspot.com/2018/04/giga … atics-pack.html
GIGABYTE 7VM400M-RZ - REV1.0.pdf

2*APM3055L = 2*12Amp (0.1Ohm/2)

When you pick up the hardware - "Holy gifts of the Omnissiah, a fine example of the Dark Age of Technology"
When you look at the schematics/reverse engineering - "Heresy! I wish I could burn this..."

On my MS-7075 VDDQ is directly generated through a PHD45N03LTA from the 3.3V rail. Not shared with anything else. Directly driven through the MS-6 ACPI controller.
I suspect this board causes most issues because its - imho - not well made.
I had K8T Neo2-F (MS-7094) back in the days which worked - waaay - better with Rialto bridge cards. And it has the same power delivery schematic as my current board, just using the P3055LD instead.

I think it's fair to state that those Rialto Bridge based cards were and are not your first choice for a stable platform.

Cheers

Very curious to try a Voodoo3 3500TV & Geforce3 Ti500 & Geforce4 Ti4600/4800.

All the more "pricey" cards for a collection, none of which lay around here 😉
I had a Ti4200 laying around years ago, but that was way before I got invested in actually repairing retro hardware. It's long gone.

Perhaps I'll snack one Ti 4600 of ebay at some point that has obvious memory errors and repair it.

For what it's worth:
GeForce 6200 AGP will follow shortly for the reference data.
And I plan on having a proper page for my reference data that is a bit more readable than a markdown table in github.