Danger Manfred wrote on 2025-06-29, 19:29:

Unfortunately, I do not have a device for measuring this.

The only 'device' you'd need to measure maximum TDP is a kill-o-watt type meter. You get a baseline of the system running with a PCI videocard (that can't consume more than 1-2 watts), then you run some graphics intensive benchmark and you subtract the difference and voila, you got your TDP.

mockingbird wrote on 2025-06-29, 19:32:

The only 'device' you'd need to measure maximum TDP is a kill-o-watt type meter.

Sure, but as stated previously, I do not own one.

Have you seen this thread? AGP Power monitor - diagnostic hardware tool

weedeewee wrote on 2025-06-29, 19:48:

Have you seen this thread?

I have not, and it is exactly what I am looking for!

And to top it off, the OP is located in the same country as me, so I might be able to help out, since I have a *lot* of AGP cards that could be measured, I could suggest lending them to him or maybe buy such an AGP power meter off him and do it myself!

Thank you so much, this was exactly what I'm looking for!

One thing worth remembering is that TDP is not power consumption, it is a measure of cooling system requirements. And worse - the way it is calculated is not universal and may vary per manufacturer, so, for example, Intel TDP and AMD TDP are not necessarily directly comparable.

Typically peak power consumption will be significantly higher than TDP, average power consumption - lower.

So numbers you get by measuring current are not TDP, they are power consumption. Probably average power consumption, but that depends on methodology.

And you also have to consider efficiency losses in various power supplies along the way before power gets to GPU. For example efficiency of old ATX power supplies can easily be as low as 60-70% and if you measure with something like "kill-o-watt" you have to take it into account.

Also board power consumption is not the same as GPU power consumption - GPU VRM is not 100% efficient either and there are other things on the board, like memory, fan(s), etc. So different boards with the same GPU can consume different amount of power.

TDP is a lie, none of them are even close to being accurate tho AMD does tend to be a bit better at it than Intel or novideo, take it as a guide for what level of cooling will be required and not as gospel. (They also all calculate it differently since there is no industry standard for for the calculation or what figures are used for it)

As others have mentioned before, TDP itself is a pretty useless metric. It could server as a ballpark number what to expect, however its definition is not same among vendors, generations etc. So basically it's a fabricated number that says value X.

While in the era you are interested in (AGP) people where not as interested in it, it (and its flaws) become more popular with modern PCIe graphics cards and their increasing power demand.

As you already seen in my power monitor thread, I had the same concern / issue with the lack of data for those old graphics cards.
That is why I decided to measure the total board power (TBP) of AGP cards. It will not take into account what the actual GPU consumes, but what the complete graphics card uses.

As the card is not converting the electrical energy into anything else than heat (apart from the kinetic energy of the FAN), it'd be safe to assume that the amount of total power the graphics card consumes will eventually have to be transported out of the case in form of heat.

Since that is your concern in that specific use-case, the total board power measurement should help you out to determine which cards suits your needs best.

Techpowerup lists TDP for many AGP cards, but not all of them. I'm also unsure how accurate that info is, but it's better than nothing, I guess.

Sometimes, you can also find a card's TDP in old reviews on Anandtech and similar websites.

You don't need to know anything about TDP, just slap on the biggest heatsink/fan combo that fits the card and add some thermal glued heatsinks on those memory chips. If the case ventilation is really that bad then I'd choose a different case as well with better cooling options.

Try a Radeon 9500/9600, GeForce FX 5700 LE, or GeForce 4 MX.

Lutsoad wrote on 2025-06-30, 08:04:

You don't need to know anything about TDP, just slap on the biggest heatsink/fan combo that fits the card and add some thermal glued heatsinks on those memory chips. If the case ventilation is really that bad then I'd choose a different case as well with better cooling options.

Sometimes people want to use the best off the shelf part that fits their build. And in this case the goal appears to have a very small case. Therefore an evidence based choice of a graphics cards seems logical.
And there are key differences.. Especially in non-demanding titles / situations, you can have vast differences. A Radeon 9200 (non-SE) draws at minimum around ~6W, wheres a 9550/9600 can go as low as 3W, which doesn't sound much, but in confined space is a lot of difference.

IMO from engineering point of view "just use the biggest cooler" is terrible idea and making informed decision based on actual numbers is a good way to do things.

Also one more quite interesting thing to consider is performance/power. The fact a card can use certain amount of power does not mean it has to. So it would be interesting to see power draw measured, for example, in some game running at the same settings and FPS. Here usually newer cards will do better. Perhaps a card with higher maximum power draw may be still good if actual power draw in intended applications is lower.

Just as an example - think something like comparing 8600gt and gtx750...

tehsiggi wrote on 2025-06-30, 10:07:

Therefore an evidence based choice of a graphics cards seems logical.

Logical, yes. Practical, no. Cooler is better and lets the card run longer with better OC options.

Lutsoad wrote on 2025-07-01, 06:55:

tehsiggi wrote on 2025-06-30, 10:07:

Therefore an evidence based choice of a graphics cards seems logical.

Logical, yes. Practical, no. Cooler is better and lets the card run longer with better OC options.

Practical? As well.
A radeon 9800 for example consumes around 30W in idle. Even if I slap a big cooler on it that keeps it cool, that energy goes somewhere and contributes to the overall temperature of the system. Knowing that, I might rather go for an 9600 for example, which consumes around 7W in idle, which keeps my overall system cooler. Especially given that I might not even need the horsepower of the 9800.

Again, we are talking about a size constraint here, so if you can optimize the choice of components, why wouldn't you.

And overclocking those 20 year old cards is something you can do, but the golden age of squeezing every .1 fps out of them is basically over. Give them their well earned rest in a nice and cozy system.

There are also edge cases like using a modern PSU with just 20A on the 5V rail in an AthlonXP system.

Low power cards like a GeForce 2 MX400 may work fine there, but high power ones like the GeForce 3 Ti 500 might not. I think Phil showed that in one of his videos.

P.S.

There's a power consumption chart for various graphics cards here: https://www.vintage3d.org/rgraph/single/cons2.php

Joseph_Joestar wrote on 2025-07-01, 07:18:

There are also edge cases like using a modern PSU with just 20A on the 5V rail in an AthlonXP system. […]
Show full quote

There are also edge cases like using a modern PSU with just 20A on the 5V rail in an AthlonXP system.

Low power cards like a GeForce 2 MX400 may work fine there, but high power ones like the GeForce 3 Ti 500 might not. I think Phil showed that in one of his videos.

P.S.

There's a power consumption chart for retro graphics cards here: https://www.vintage3d.org/rgraph/single/cons2.php

Yep, there are some cards that rely heavily on 3.3V and 5V, which is specifically why I went through the hassle of measuring every power rail. Cards like the Radeon 9700Pro will just start artifacting if the external 5V power input has a voltage too low.

tehsiggi wrote on 2025-07-01, 07:10:

Again, we are talking about a size constraint here

That's when you go for a better case that immediately solves this problem.

Joseph_Joestar wrote on 2025-07-01, 07:18:

There's a power consumption chart for various graphics cards here: https://www.vintage3d.org/rgraph/single/cons2.php

Awesome chart, all my cards are listed! Thank you very much for posting this here!

Some of the values are lower than my measurements as well as the reference values from ATI, looking at the 9800 Pro.