Kahenraz wrote on 2022-03-24, 00:51:

The BIOS chip is soldered and it's not worth the effort to remove and reattach it, with or without a socket.

Lol you do it through software haha (on the computer that has the card installed)

I think gpu-z can dump the bios and nbitor may be able to read the timings

Both Horun & Sphere478 called it that it could be a cap issue... I learned something today. Thank you.

Well the general principle is, if something works great at low clock or with less powerful part installed, but doesn't at high clock or with high powered part, then it's the power circuitry that's weak and prime suspect should be capacitors before anything else in it.

Kahenraz wrote on 2022-03-23, 21:25:

Bad caps are usually a little more obvious. Now I know that bad caps can be more subtle.

+1

It's not really related to this topic (-> DC smoothing caps), but capacitors really can be subtle as you say.
Because, one bad cap may still seem fine for DC applications, but then suddenly fails for high frequency (HF) or AC applications.
For example, when they're damaged, they will short (break down) when reaching a certain voltage level (as often used by HF).
Paper condensers of the past were notorious for such things.
Some say, these were defective by design, even. 😁
That's good to know when working with oscillators. 😉

Kahenraz wrote on 2022-03-23, 21:25:

Bad caps are usually a little more obvious. Now I know that bad caps can be more subtle.

Not based on my experience.
In fact, more often than not, (this is especially true for retro hardware), bad caps can look as good as new, even if they are dead or almost dead.
And worst of all, these guys are silent killers, they can permanently damage other parts of the circuit.

Se my post here, where I test two caps that look as good as new, but one of them is dead.

Where I had to downclock the memory to 290 Mhz before, it's now stable overclocked to 400 Mhz. That's quite an improvement.

This card seems to benefit the most from additional memory bandwidth. Increasing the core clock doesn't do much at all.

Kahenraz wrote on 2022-03-24, 17:18:

Where I had to downclock the memory to 290 Mhz before, it's now stable overclocked to 400 Mhz. That's quite an improvement.

This card seems to benefit the most from additional memory bandwidth. Increasing the core clock doesn't do much at all.

Add more caps in parallel on the back, see if it goes further

Is that to lower the ESR further?

1000 1uF capacitors in parallel for zero ohms ESR 🤣

Should further stabilize Ripple. 🤔

Sphere478 wrote on 2022-03-24, 00:44:

w00t! what did I win?

Street cred!😎

Edit: How could I misspell this?

There were four 1000uF capacitors that I removed. For reference, they measured: 940uF, 860uF, 840uF, and 540uF on my capacitance tester. I don't have any device for testing ESR.

Those cheapy component testers give a guesstimate... if you've got one that doesn't with the AVR chip in it, you can maybe get an updated firmware for it.

Taking a wild guess on the capacitance values, probably the 940 was okay... meaning still has a year or two in it, the 860 and 840 would have been marginal, and of course the 540 woulda been horrible. Though you can find caps still reading good capacitance, and not so good ESR.

ESR is most important capacitor characteristics of all.
Back in the day, I purchased the ESR meter in 2004 or so and was in tv repair shop, tv tech follow was not sure where to start on a toshiba tv. I spotted many ESR poor capacitors this way with this ESR meter tool and fixed that TV, This tool was real money maker from on now.

Cheers,

I already have a capacitance tester. I bought an LCR-TC1 you test for ESR. I'll report back on these same capacitors when it arrives. I'm very curious.