Kahenraz wrote on 2021-08-09, 22:31:

All other things being equal, is there any difference to be had in performance between these this kinds of RAM chips on a video card?

IMG_20210809_182729.jpg

IMG_20210809_182744.jpg

It would help if markings were readable on second photo.

It would also help to know what GPU is connected to the RAM on each card .

First photo has 2 HY57V161610D chips with 2 16-bit banks each at 143MHz . So, I would guess two chips afford either 32-bits or maybe 64-bits (if both banks can be accessed simultaneously, which I doubt, but I could be wrong ).

How can they be the same card if the pcb is obviously different?

In this context, the TC-7 is likely 70nS RAM, while the -10 is more likely 100nS
Now on some cards, it doesn't matter, the chipset has a minimum spec requirement and anything faster has no advantage, while some do have the potential to run faster RAM for more performance.

Looking back, 143MHz rings a bell, sounds like the RAM speed for a Geforce 2MX model - The Creative one with slow DDR, and they DID have choices of RAM speed and type

One looks like SDRAM and one SGRAM.

Matth79 wrote on 2021-08-10, 18:41:

In this context, the TC-7 is likely 70nS RAM, while the -10 is more likely 100nS

This is synchronous RAM. The timing no longer signifies the time from RAS to data availability (that's how access time is defined on classic, i.e. non-page-mode, FPM and EDO) RAM, but instead the minimum clock period. It is no longer specified in tens of nanoseconds, but in nanoseconds. So TC-7 is 7ns minimum clock period (143 MHz), and -10 is 10ns minimum clock period (100 MHz). The access time is longer than a single clock cycle, because it takes multiple clocks (RAS-to-CAS delay + CAS latency) from RAS to data availability. IIRC, in PC100-222 SDRAM, the first "2" is the CAS latency and one of the other "2"s is RAS-to-CAS-delay, so the access time is 4 clocks of 10ns, i.e. 40ns.

So -7 RAM allows faster clocks than -10 RAM. This means faster bursts at least, but without looking at the full datasheet, I cant tell wheter access time (from inputting the address to getting the first data) is faster too. If the -7 chip needs 3 cycles latency, and the -10 chip needs just two cycles, access time is about the same.

SGRAM was also optimised to work faster than regular SDRAM. Kinda like old VRAM vs regular DRAM on more old video cards.

https://www.ardent-tool.com/tech/memory/vrmsgrm.pdf

1SGRAM Features
28-column Block Write and Write-per-Bit modes.
3Block write permits you to write the data stored in
4the Color Register to eight consecutive memory
5locations or columns in a single cycle. Mask data
6stored in the Mask Register coupled with Write-perBit (WPB)
7mask data present at the inputs (DQs) are used to mask
8specific bits and prevent them from being written.

good for planar graphic modes (old/obsolete) and 1bit per pixel color depth (stencil buffers/shadows).