I think it should be possible.. Why not..? With older graphics cards it will work the same way.. Also most VLB video board have an option to solder different type of memory on it..
I dont think there is something like a vendor lock on these voodoo 2 cards.. Like adding memory Dips to a graphics card.. It would detect the amount what is connected on the pcb.

It has been done but very very few people know that it is possible let alone what is needed to make it work properly. Cards that sold with 6mb, 8mb, and 10mb can be upgraded to the full 12mb however each TMU can support more than the normal 4MB each.

Voodoo 1 can be upgraded to 8MB by either finding higher density EDO ram (good luck) or by stacking and in that case one has to wire some of the pins. The Voodoo Rush is the same but one can not upgrade the memory for the 2D side.

The frame buffer on Voodoo 1 and 2 is limited to 4MB, TMU on the Voodoo 1 and Rush is limited to 8MB (stacking or high density) while the TMU on the Voodoo 2 can possible take up to 16mb each however it has never been done.

Before someone rages and throws a fit that what I said was BS http://www.falconfly.de/cgi-bin/yabb2/YaBB.pl?num=1366485713 The neck beards over there are modding VSA era cards like the 4500 and 5500 with double memory.

http://www.falconfly.de/cgi-bin/yabb2/YaBB.pl?num=1365950649

I don't know anything about Voodoo cards but I doubt it autodetects the memory. More likely you'd need to update the firmware to match the new memory size, as well as all the necessary timing settings that the RAM requires.

but I doubt it autodetects the memory.

Like poster above said... I upgraded many old video cards (ISA, VLB, PCI) with additional memory... except for the really old ones (1990/91) there are no jumpers or firmware upgrades needed to recognize new memory. They auto detect it. It's like upgrading main memory on a PC.

kixs wrote:

but I doubt it autodetects the memory.

Like poster above said... I upgraded many old video cards (ISA, VLB, PCI) with additional memory... except for the really old ones (1990/91) there are no jumpers or firmware upgrades needed to recognize new memory. They auto detect it. It's like upgrading main memory on a PC.

Do you mean cards that weren't intended to be upgraded, where you had to solder RAM chips to make the upgrade?

PC memory has configuration information included on the module. DIMMs have a serial EEPROM, SIMMs have soldered jumpers. That information is lacking when soldering new chips on a video card.

Interesting that video cards will autodetect the size with only raw memory chips. But come to think of it, I guess I've seen the same thing with common cards like the S3 Trio64s, whose upgrade consists of nothing but raw memory chips.
I guess the firmware just scans the address range to figure out what memory size is working without error. I think the timing details must be coded into the firmware, but as long as compatible chips are being used that wouldn't be an issue.

However, I've run into a Geforce MX card that reports half of it's correct memory size. I always figured it was a victim of somebody's firmware experiment. Maybe that's not it.

nforce4max wrote:

It has been done but very very few people know that it is possible let alone what is needed to make it work properly. Cards that sold with 6mb, 8mb, and 10mb can be upgraded to the full 12mb however each TMU can support more than the normal 4MB each.

Voodoo 1 can be upgraded to 8MB by either finding higher density EDO ram (good luck) or by stacking and in that case one has to wire some of the pins. The Voodoo Rush is the same but one can not upgrade the memory for the 2D side.

The frame buffer on Voodoo 1 and 2 is limited to 4MB, TMU on the Voodoo 1 and Rush is limited to 8MB (stacking or high density) while the TMU on the Voodoo 2 can possible take up to 16mb each however it has never been done.

You are saying a Voodoo 1 can be upgraded to 4 + 8 MB and a Voodoo 2 theoretically to 4 + 16 + 16 MB? I have heard of upgrades to these cards, but not more than some special editions. (eg. Voodoo1 with 4 + 4 MB, much more than most factory-made cards. More wouldn't be very useful anyway.) Do have any source for the 8MB-per-TMU limit, either for V1 or V2? I have never heard of that.

kixs wrote:

but I doubt it autodetects the memory.

Like poster above said... I upgraded many old video cards (ISA, VLB, PCI) with additional memory... except for the really old ones (1990/91) there are no jumpers or firmware upgrades needed to recognize new memory. They auto detect it. It's like upgrading main memory on a PC.

I don't think V1 or V2 cards have a firmware. The won't work as VGA cards so they don't need that, the drivers will do everything necessary. But IIRC there are some resistors (or something with a similar function) on the cards that have to be modded to indicate the amount of RAM and maybe the clocks. I suppose either the drivers reads those values and programs the chips, or the chips will reads those values on bootstrap.

Voodoo 1 and 2 are not like modern cards, they auto detect but there is no firmware that locks it all in. The memory configuration is all done in hardware like it was done in the days of old.

You can go looking around on that forum and you will find almost everything you would need to perform such a mod yourself but for the voodoo 2 it is a bit harder but it can be done. The easiest method is stacking but you will still have to wire a few pins for the timings to work correctly. 4MB limit on the frame buffer and 16mb each on the TMU according to the documentation from 3DFX. The reason why it was likely never done as a product was cost and at the time the need for that much memory wasn't apparent.

shamino wrote:

PC memory has configuration information included on the module. DIMMs have a serial EEPROM, SIMMs have soldered jumpers. That information is lacking when soldering new chips on a video card.
.

Well, then show me where it is stored and how this stick is different from raw chips:

For DDR2 and DDR3 it's true. They have a special chip for it. But EDO RAM is always autodetected.

RacoonRider wrote:

Well, then show me where it is stored and how this stick is different from raw chips: […]
Show full quote

shamino wrote:

PC memory has configuration information included on the module. DIMMs have a serial EEPROM, SIMMs have soldered jumpers. That information is lacking when soldering new chips on a video card.
.

Well, then show me where it is stored and how this stick is different from raw chips:

I think that SIMM is flipped the wrong way to see them.

Pins 67-70 (Presence Detect 1/2/3/4):
http://pinouts.ru/Memory/Simm72_pinout.shtml
Those pins are at the left in your picture, but the jumpers are probably connected on the other side of the PCB.

A couple examples from eBay, not my pictures. Configuration jumpers are at upper right:

I found a couple SIMMs and checked the resistance across the closed jumpers, and they were 0 ohms. I thought they'd use a weak pulldown resistor but apparently not, the signals are dead shorted to ground.

The SIMM I showed is one-sided, there are no elements on the other side.

If you are correct, then where are those jumpers on all the 30-pin modules? Why are there sticks with no jumpers at all?

I wasn't originally thinking of 30pin SIMMs. I looked up their pinout:
http://pinouts.ru/Memory/Simm30_pinout.shtml
They don't have any Presence Detect pins. I tried to find out how their capacity is detected but didn't find any relevant links.
I assume the BIOS uses some software algorithm to test various addresses until the functioning capacity can be deduced. So this would indeed be a blind/jumperless situation on the 30pin modules.
The pinout info for 30pin says the unused high address lines are left floating, but I don't know if the host system is able to electrically sense that. If it can, that would be another way to get the capacity.

For 72pin:
I dug through my 72pin SIMMs but only found one that didn't have any jumpers on it. When I tested pins 67-70 vs Ground, I found that the Presence Detect signals were still intact and indicated it's capacity. Apparently the PCB was hardwired for that configuration.

I saw some references to IBM having a slightly different mapping for the Presence Detect. This suggests that 72pin memory wasn't tightly standardized across the industry, at least early on. I will leave open the possibility that 72pins might exist without any Presence Detect signal connected, but I wasn't able to find an example in mine. Those that don't have jumpers can still have the signals wired.

If 72pin without Presence Detect exist, I think it would be earlier modules, probably not EDO. I also wonder how well such memory would be supported by various systems. Since 30pin didn't have those signals, I can imagine early 72pin systems could inherit the older behavior and ignore the Presence Detect pins.

It appears JEDEC standards require the Presence Detect signals on compliant 72pin memory:
http://www.ele.uri.edu/iced/protosys/hardware … arpoint-8MB.pdf
However, that's not a complete document, so I'm inferring a bit about the context. JEDEC's web site wouldn't let me download without an account.

shamino,
Thank you! You learn something new every day 😀

So there must be a way of finding out SIMM capacity without checking datasheets or trying to install it?

4 jumpers give 16 variations of total module size. I will try to find consistent data for 2, 4, 8, 16, 32 Mb modules from what I have.

Just checked 20 72-pin EDO and FPM modules I have in spare parts.

1 of them uses jumpers. It's 32MB EDO module, pins 67,68,69 are connected to pin 72. On other 19 modules of all common sizes pins 67,68,69,70 are not connected.

Interesting. It makes me wonder whether PC motherboards typically ignore those signals or actually use them. If the signals aren't always present, then what's the point?

Only pins 67-68 are used to identify the size, so the codes get recycled some. 1MB and 16MB have the same coding. It seems like SIMM motherboards typically only claim to support 4 different sizes of SIMMs and I think this is the reason, but that only makes sense if they're actually looking at these signals.

Pins 69-70 are supposed to identify the access speed.

Pin 48 is supposed to indicate if it's a Parity module. I don't know if GND means parity or non-parity. I didn't think to check when I had my SIMMs out, and the info on web sites seems to be inconsistent.

I found 4 modules that had pins 67-70 all open. They were all 8MB 60ns and the PCB says they are from a Packard Bell. The pinouts.ru link from earlier implies that this would be a valid coding for 8MB 60ns. However, this page:
http://www.mcamafia.de/mcapage0/memorypd.htm
disagrees, and says that 8MB 60ns should have all 4 pins connected to Ground. It says having all 4 pins open would mean "free or invalid". Unless all those modules you looked at are 8MB 60ns, then I assume "free or invalid" is the way it gets interpreted.

So at that point the motherboard/device would either fail to see the memory, or it would need to figure out the capacity by some other means. Maybe the Presence Detect bits are required by some devices/boards but not by others. If they're never required then I'm not sure what would be the point of having them.

Seems like they had a big problem with standardization. SIMMs are weird. 😀

The SIMMS I tested were of various sizes from 2MB to 32MB both EDO and FPM. I added some, so for now of 23 SIMMs tested 3 use pins 67-70. They are top right modules, 32MB EDO (67,68,69 connected) and 2x16MB FPM (67,70 connected).

Notice that most SIMMS have the soldering pads for jumpers. the 2MB FPM stick branded PNY/IBM even has lead balls on top of them.
As far as I know, when EDO first appeared in the news, they claimed that new memory would be 100% compatible with FPM-only boards. However, it was soon discovered untrue. That was due to the fact that chipset makers partially ignored JEDEC specifications, and while the new EDO RAM was compatible on paper, real life was a different thing. I might be not exactly precise here, but as far as I remember, for the sake of better performance, in FPM era "read" and "write" commands were not terminated in a propper way, but rather superseded by the next comand. It was OK in fast page mode, but in EDO mode this was the key point of burst read/write operation.

I think that was the case with detection pins as well. While standartised by JEDEC, these pins were never or almost never used. As soon as RAM manufacturers found that out, they started to reduce cost of their modules by removing now unnesessary components without redesigning PCB.

My guess is that on average my modules are produced later and were initially cheap (as most PC components from that era found in Russia). Cheap modules would have been the first to be affected by cost-cutting.

Could you look up the production dates of your modules/chips?
Mine are as follows:

Pin not connected (20):
1993 - 1
1995 - 2
1996 - 1
1997 - 7
1998 - 3
Unknown - 6

Pins connected (3):
1996 - 2
1995 -1

Sorry for the slow reply. I went through my SIMMs and this is what I came up with. Found some more that were taped together into a brick several years ago. I freed them from their bonds and will store them in a more sensible container.

Each line is a unique module design.
The number on left is the number of identical pieces for that module, I suppose duplicates shouldn't really count.
Next to that is the apparent year of production, followed by some specs, then manufacturer if marked.
"np" means it's non-Parity
I noted "tin" or "gold" as another symptom of cheapness, but really only 3 of my oldest ones are gold, the rest are all tin.
I noted the country of origin when it was marked, but most of them weren't marked. For some I noted the origin of the RAM chips if they were marked.

1Presence Detect connected:
21	1997	16MB FPM  "LG Semicon"	np	tin		KR
31	Unknown	32MB FPM (bought around 96-97)  "MGV"	np	tin	nocoo
42	1996	4MB FPM		"PNY-CA"		np		tin		chips:US
51	1993?				chips "IBM 93 14"	par		Gold	nocoo
61	1994	4MB FPM		Kingston			par		Gold	nocoo
72	1997	16MB EDO	"Lite On" Micron chips		np	tin		chips US, PCB unmarked
82	1997	16MB EDO	SMART Modular PCB, Fujitsu chips, HP branded pn 1818-6430	np, parity PCB layout	tin	PR
92	2000	32MB EDO	Viking Components	np, parity PCB layout	tin	nocoo
101	Unknown	1MB FPM	 Toshiba module "8806AAA" chips say "8753HCK" - I don't think 1987-88 is possible.  Parity  Gold  JP
112      1998  32MB EDO   Centon  nocoo   chips are 50ns but PD is set to 60ns
12
13Presence Detect NOT connected but might be valid coding for 8MB 60ns:
142	1996	8MB EDO 60ns	"Packard Bell"		np	tin	chips:SG    No jumper pads
152	1996	8MB EDO 60ns						np	tin	chips:KR   Jumper pads present/empty
16These have all pins 67-70 open, but some info implies it could be a valid coding for 8MB 60ns.
17
18Presence Detect NOT connected and definitely invalid:
192	1998	16MB EDO 60ns	"Lite On" PCB, Hyundai chips	np	tin	TW
20<several>	1999	32MB EDO 60ns	np	tin
21These have all pins 67-70 open and this is not valid for the capacity.

====
Ignoring duplicates, the total of distinct different modules by year work out to:

Presence Detect connected (10):
Unknown - 1 <old 1MB>
1993 - 1
1994 - 1
1996 - 1
Unknown 1 <~1996-97 32MB>
1997 - 3
1998 - 1
2000 - 1

PD open but unclear if invalid for specs (2):
1996 - 2

Presence Detect disconnected and definitely invalid (2):
1998 - 1
1999 - 1

So my 2 module types with clearly invalid and disconnected PD are from 1998 and 1999, making them very late 72pin SIMMs.