Download and install Rivatuner (I think it's version 2.21 you want).

In the settings, install the strap driver and then configure it to the 5950U.

If you want to permanently change it in the card's BIOS, see this thread:

The story of the QuForce FX 5800 Ultra...

My opinion is that it is foolish to overclock these cards, or rather, the RAM on these cards. The RAM speed was already pushing the limits of the lithography at the time very far. In fact, you would be wise to underclock the RAM if you want to get any longevity out of the card.

Meh, lithography limits aren't written in stone... oh wait 😁

FX5900U/FX5950U had 2.0ns GDDR, which is rated to work at 1 Ghz. Underclocking RAM without voltage tweaking is pointless.

Thank you for the responses,, this group never seems to disappoint on the technical side of things. From the old forum posts I read, I was of the impression that the FX5900 (non-ultra) and 5950 Ultra had the same Silicon, the difference was in the Bios settings, Vram capacity and the cooler which varied a lot between designs. Well, at least people were claiming that, maybe they did not know what they were saying, although the process node and transistor count are the same according to tech sites. I have always been a bit intimidated by Rivatuner, or modding my own bios. I will take a look at the links and see of that gives me what I need to decide how to proceed. I am not that concerned about the card itself it is nothing rare or valuable. Back then people were apparently just downloading the 5950U bios and flashing it onto the 5900 with good results at least in the short term.

Reflashing FX5900 to FX5950 is practically meaningless. You can just overclock it via driver or tweaker. FX5950 Ultra is an overclocked 5900 Ultra and was released as low effort solution to compete with Radeon 9800XT.

Got it, I thought there might be other differences in the bios like voltages or timings. I will try an OC 1st. One must admit that some of the real 5959U cards do look pretty cool though. Generally priced accordingly.

Here are some OC results I just got with 3DMark2001SE, 1600x1200x32 with 4xAA
CPU = Athlon64 3200+, OS = Win98SE

Score = 6272 3DMarks with PNY FX5900 non-Ultra, stock clocks 400/850
Score = 6959 3DMarks with Overclock at 475/950 (5950 Ultra clocks)

No crashes or visual glitches in the benchmark at either speed. GPU temp reached 65C with the OC, 63C at stock according to the driver temp monitor. So there seems to be some benefit from the OC, close to what is reported in the old reviews comparing 5900 and 5950U. This is as suggested, thanks for that.

I also ran the demo mode in 3DMark2001SE several times . At 1600x1200x32 with 4x AA and overclock, it tripped the Nvidia sentinel power monitor while initializing the GPU. Ran fine at same settings with no AA. Without OC, the Demo ran fine at all settings even with 4xAA. I'm not sure what the VGA sentinel is monitoring maybe voltage is dropping too much?

With this info I will keep it at stock and bump up clocks if I feel the need for a few more FPS in the right game.

The Serpent Rider wrote on 2021-08-28, 02:40:

FX5900U/FX5950U had 2.0ns GDDR, which is rated to work at 1 Ghz. Underclocking RAM without voltage tweaking is pointless.

Your contention is that running memory at lower speeds but at the same voltage doesn't stress the chips any less.

I reject that as empirically false.

mockingbird wrote on 2021-08-29, 21:20:

The Serpent Rider wrote on 2021-08-28, 02:40:

FX5900U/FX5950U had 2.0ns GDDR, which is rated to work at 1 Ghz. Underclocking RAM without voltage tweaking is pointless.

Your contention is that running memory at lower speeds but at the same voltage doesn't stress the chips any less.

I reject that as empirically false.

I reject your rejection as empirically false

maxtherabbit wrote on 2021-08-29, 21:53:

mockingbird wrote on 2021-08-29, 21:20:

The Serpent Rider wrote on 2021-08-28, 02:40:

FX5900U/FX5950U had 2.0ns GDDR, which is rated to work at 1 Ghz. Underclocking RAM without voltage tweaking is pointless.

Your contention is that running memory at lower speeds but at the same voltage doesn't stress the chips any less.

I reject that as empirically false.

I reject your rejection as empirically false

Heh. Down here on planet Earth it is easy to observe on such devices temperatures correlate with clock.