If it follows the layout of the voodoo 1 then the lower 4 should be the framebuffer I think. That’s just a guess though. You’ll probably just have to replace 4, see if it works and replace the other 4 if not. It’s entirely possible only 1 chip is bad though so after finding which 4 are bad, you might try to narrow it down to just 1, if you care to. Personally I’d probably just replace 4 and if that didn’t work replace the other 4.

Actually the banshee might have combined ram rather than separate framebuffer and texture memory. So you might just replace all 8.

Thank you! Any ideas which other memory chips could be used as a replacement?

Banshee uses unified memory, so it can be any of the eight chips indeed.

To replace all eight, you should get chips with similar layout and voltage, doesn't have to be as fast as 6ns.

Putas wrote on 2023-10-22, 08:06:

Banshee uses unified memory, so it can be any of the eight chips indeed.
To replace all eight, you should get chips with similar layout and voltage, doesn't have to be as fast as 6ns.

Thank you, it's a bit more clear to me now.
So basically any chip marked "3617161D" should work, do i get that right?
Does it have a difference if it ends with DT, not D?

Edit:
This Winbond chip W9816G6CH-6 looks like exact replica, can someone give an expert opinion?

Edit2:
Datasheet for W9816G6JH shows it should work, and it's still produced. Would love a confirmation on this.

By looking at readable photos of Banshee i've found this types of memory in TSOP50 package was used:
NEC D4516161AG
IBM 0316169CT3D-10
SAMSUNG KM416S1020CT-G10
And Samsung still has a datasheet available!

1M x 16 SDRAM
512K x 16bit x 2 Banks
Synchronous DRAM
LVTTL

Ok, now to find what could be locally sourced

Finding video cards using SDRAM in 16MB size is kinda hard to find. Loose chips are easier to get.
Actually yours is TSOP 54 (54 pins not 50 pins).

Same SDRAM as voodoo3 2000.

Ebay: search: 1M x 16 bit SDRAM anything faster than 7.5ns.

TNT and TNT2, and Radeon 16MB uses these.

This one uses 7.5ns which is about 150Mhz. Yours is 10ns which is about 66 to 83MHz.
https://www.ebay.ca/itm/134535882023?hash=ite … ABk9SR-Tc7JnrYg

This in loose chips:

https://www.ebay.ca/itm/114131492965?epid=111 … ABk9SR6Dni5rrYg

Cheers,

pentiumspeed wrote on 2023-10-22, 22:50:

Actually yours is TSOP 54 (54 pins not 50 pins).

Thanks for the links! Are you sure for TSOP54? I've counted the legs and that's 2x25
--
Anyway, to test my luck ordered HY57V161610DTC-7 from AliExpress.

My (sadly fried) PCI Voodoo Banshee has the KM416S1020CT-G7 chips and those have the same pinout / voltage as the Hynix chips you're getting 😀 The Banshee cards I've seen seem to have memory around 8ns to 6ns range so 7ns should be fine.

It'd be nice if diagnostic data existed for the 3DFX cards on which chip / location ends up causing which graphical glitches - TAGG XOC on youtube has done something similar for Geforce 3 > 9800 series cards, I think by selectively disabling (either shorting or disabling VREF on) individual chips on working cards: https://www.youtube.com/watch?v=JCnmt4Dt_74
I have a partly working AGP Voodoo Banshee with a similar memory problem and will be just randomly swapping out SGRAM chips to see what helps.

Thermalwrong wrote on 2023-10-23, 14:37:

My (sadly fried) PCI Voodoo Banshee has the KM416S1020CT-G7 chips and those have the same pinout / voltage as the Hynix chips you're getting 😀 The Banshee cards I've seen seem to have memory around 8ns to 6ns range so 7ns should be fine.

It'd be nice if diagnostic data existed for the 3DFX cards on which chip / location ends up causing which graphical glitches - TAGG XOC on youtube has done something similar for Geforce 3 > 9800 series cards, I think by selectively disabling (either shorting or disabling VREF on) individual chips on working cards: https://www.youtube.com/watch?v=JCnmt4Dt_74
I have a partly working AGP Voodoo Banshee with a similar memory problem and will be just randomly swapping out SGRAM chips to see what helps.

Thanks!
Today i had some spare time, so i decided to do a solder pass on all memory chips on the card.
And my shaky hand did a short on one of the chips. As a result - picture become exactly like on that video. 😀
Fixed that and it's back into previous state.
Interesting part - i had a minor problem (instead of those short lines) on cold start - some letters were like scrambled for a 10-30 secs. It's now gone. Now it looks like this (some pixels are jittering) and i'm even more sure that one chip has faulty cells.

n0p wrote on 2023-10-23, 07:02:

Thanks for the links! Are you sure for TSOP54? I've counted the legs and that's 2x25 -- Anyway, to test my luck ordered HY57V161 […]
Show full quote

pentiumspeed wrote on 2023-10-22, 22:50:

Actually yours is TSOP 54 (54 pins not 50 pins).

Thanks for the links! Are you sure for TSOP54? I've counted the legs and that's 2x25
--
Anyway, to test my luck ordered HY57V161610DTC-7 from AliExpress.

My mistake, 50 pins.

Cheers,

Bought this chips on Aliexpress, advertised as Hynix.
Didn't work, tried two of them,
Card somewhat works with them, but with that chips black lines appear, so they are either faulty or timings are off.
Soldered back original RAM, now i'm sure it's not first or last chip 😀
I'll keep looking.

Did seller show other photos after this? The last few photos is the correct one that seller is selling. I ran into this before with a CPAP blower, no not for me, a special project now ruined.

This IC is not Hynix. And throw them out.

Instead: This time this one is -6ns
Buy qty of two to get 8 chips.
https://www.ebay.ca/itm/225848042688?hash=ite … ABk9SR9Ld_ryAYw

Cheers,

@pentiumspeed, thanks for link!
Seller advertized those as Hynix, but due to some problems and very low cost of memory itself i didn't open a dispute in time.
Anyway i decided to try those 😀 - as i was sure replacing one chip at time would work.
Now i need some time to embrace myself for second try - my soldering skills indeed improved and i was very careful, but anyway, some traces have moved - didn't break, just moved a bit when i was removing excess tin, so second attempt might be last for this Banshee 😀
I wonder though, have anyone seen projects for remaking PCB for these?
Looking at the board - it's not that complicated to make a replica.
And completely offtopic - can some kind of PiVoodoo be possible?

So, i decided to look what is possible to obtain here of the closest specs. Found two chips, AOTOM (6ns) and ESMT(5ns).
Guess what? Both worked. But that was not places of failing chip 😁
So i assume it safe to replace chips by one if access time is same or lower.

Success!
--
Time to clean the card 😀
--
If i'd pay more attention from the start - framebuffer organized in eight vertical blocks, line by line. So you can count pixels to fail region. It was the chip number three in sequence.
Horisontal line is: 12345678...12345678

Well done. It's always satisfying to hear about a 3DFX card being brought back to life.

Congratulations. Have you tried overclocking with the cool new memory?

Thanks guys! I'm very happy myself - and i'm no more afraid of microsoldering.
Modern fluxes are miracle.
As for overclocking - i didn't, but i'll try 😀 This card replaced my fully dead (killed by own power convertor) Rush in P200MMX PC, so i think bottleneck will be CPU, not the Banshee.
Carmageddon Splat Pack plays great 😀

So, to logically close my work one the card -
I've installed 40x40x10mm Sunon medium-speed maglev fan.
Connector on card is PH2.0, cut and replaced original 2-pin on fan, voltage is 5V.
Plastic pins are from chassis fan, used a round file to make fan holes a bit bigger.
Not exactly silent, but not loud either.
As for long wires - i hate short ones so i tend to leave a bit more than needed.
And it will make next PCI slot unusable, but i don't have anything there anyway.