The card has no way of detecting the chip type, so there is no way the stock clock speed and performance will differ between upgraded and stock version. But if you are going to customly clock (e.g. overclock) your card, the expansion chips might limit the maximal stable rate.

mkarcher wrote on 2020-10-09, 22:43:

The card has no way of detecting the chip type, so there is no way the stock clock speed and performance will differ between upgraded and stock version.

I'm not the original owner of this card so I don't know if the chips in the sockets are stock.

mkarcher wrote on 2020-10-09, 22:43:

But if you are going to customly clock (e.g. overclock) your card, the expansion chips might limit the maximal stable rate.

I wasn't looking to overclock the card. I was wondering if there would be a performance difference if all chips were the same type.

What mkarcher is saying is that mixing chips has no effect on performance except when using chips not capable of running at the card's rated speed (stock speed).
If the card runs fine swapping the slower chips for faster chips makes no difference in performance as long as the clocks aren't increased. The card does not increase its clockspeed in its own.
Using faster chips only makes a difference when overclocking. The slowest chip determines the fastest overclock.

See chip ratings as speed limits on a road. You're the one doing the driving, but if you exceed the limit for the road (chip), you might crash - although sometimes you can go quite a bit over the limit before that happens.

You say you want to run the card at stock, that's like saying you want to drive well below the limit. The difference between two different speed limits doesn't affect you there, equally, if this card doesn't exceed 60ns timings at stock, the chip mismatch doesn't matter. On a card that did require 50ns at stock, it might well matter.

dionb wrote on 2020-10-10, 21:40:

See chip ratings as speed limits on a road. You're the one doing the driving, but if you exceed the limit for the road (chip), you might crash - although sometimes you can go quite a bit over the limit before that happens.

You say you want to run the card at stock, that's like saying you want to drive well below the limit. The difference between two different speed limits doesn't affect you there, equally, if this card doesn't exceed 60ns timings at stock, the chip mismatch doesn't matter. On a card that did require 50ns at stock, it might well matter.

Hehee good way to explain it ! Have a S3 VLB with two diff speed sets of ram and it never gives me issues but am running it at 33Mhz bus, if I ran it at 40Mhz (it is rated for that) would probably have to add 1 wait state since the add-on ram is 10nS slower than the soldered...or maybe not.

Horun wrote on 2020-10-11, 02:13:

Hehee good way to explain it ! Have a S3 VLB with two diff speed sets of ram and it never gives me issues but am running it at 33Mhz bus, if I ran it at 40Mhz (it is rated for that) would probably have to add 1 wait state since the add-on ram is 10nS slower than the soldered...or maybe not.

Most likely not. Modern video cards (like ET4000, CL-GD542x, S3 stuff) have a dedicated memory or system clock that is used to generate memory timings. The memory speed requirements thus does not depend on the bus frequency. Older video cards (like CGA, EGA, classic VGA) have the memory clock derived from the dot clock, and fixed time slots the CPU may use to access the video memory. No matter what the bus speed is, the card wait states to the bus cycle until a memory access slot is available, and then runs the cycle at dot-clock derived speed. It uses the IOCHRDY line on the ISA bus to delay the bus until the memory is cycle is finished.

As a side note: The CGA in high-resolution text mode needs all memory access time slots to properly display the screen. But on the other hand, it may not add that many wait states that the processor can access the memory during horizontal retrace. The display period (from first to last character) is around 45 microseconds, but you may not have bus cycles exceeding 10 microseconds (that what some spec says) to not disturb the memory refresh happening every 15 microseconds. The IBM PC/XT is unable to refresh memory during an active bus cycle (and that includes wait states during a cycle), because refresh cycles are also sent via the same bus.

dionb wrote on 2020-10-10, 21:40:

See chip ratings as speed limits on a road. You're the one doing the driving, but if you exceed the limit for the road (chip), you might crash - although sometimes you can go quite a bit over the limit before that happens.

You say you want to run the card at stock, that's like saying you want to drive well below the limit. The difference between two different speed limits doesn't affect you there, equally, if this card doesn't exceed 60ns timings at stock, the chip mismatch doesn't matter. On a card that did require 50ns at stock, it might well matter.

Thanks to you and dornkaat for explaining what my post means. I want to add another practical perspective to it: As the card does not know there are 60ns chips in the sockets (that was the point of my post), it runs them as fast as the 50ns chips on board. I have no idea whether the 60ns chips are officially fast enough to handle the speed the card designers use (at the "full commercial temperature range", which usually means up to 70 degrees chip case temperature). Using 60ns chips thus does not decrease the speed of the card, but it might decrease the reliability of that card. If you start getting funky blinking miscolored pixels or incompletely filled window backgrounds (or any kind of 3D corruption) on hot summer days after the card warmed up, the 60ns RAM expansion chips might be the cause.