This thread has some good info in it.

Geforce2 Ultra

I have GeForce 2 Ultra from DELL. Connector for TV Out is not soldered.

Hah, that is EXACTLY the card I just posted about here. I googled the part numbers and found roughly equal numbers of people claiming it to be a GF2 GTS, GF3 and Quadro2 Pro.

^^ my card, for the record.

I JUST threw together a test rig and found out Linux reports it as a Quadro2 Pro with 16MB RAM.

Note that this isn't always 100% accurate, but it seems likely in this case.

xjas wrote:

I JUST threw together a test rig and found out Linux reports it as a Quadro2 Pro with 16MB RAM

16MB seems really odd. Don't those cards usually have 64MB? Even the lower tier GF2 cards have 32MB. I wonder about whether that part of it is accurate.
Unfortunately the heatsinks make it difficult to verify directly.

Yeah, the Quadro 2 Pro has 64MB.

I usually drop mystery cards into a 98SE system with 45.23 installed and just go with what the drivers say it is.

It's only the Dell variants made by Elsa that cause all the confusion. If you really want to be 100% sure that you are getting a genuine GF2 Ultra and not a Quadro 2 Pro, don't buy an Elsa make from Dell. And don't buy one with DVI. Some GF2 Ultras have DVI, but no Quadro 2 Pro doesn't have DVI.

Linux will go with whatever the PCI ID is, so maybe a Frankenstein board?

shamino wrote:

16MB seems really odd. Don't those cards usually have 64MB? Even the lower tier GF2 cards have 32MB. I wonder about whether […]
Show full quote

xjas wrote:

I JUST threw together a test rig and found out Linux reports it as a Quadro2 Pro with 16MB RAM

16MB seems really odd. Don't those cards usually have 64MB? Even the lower tier GF2 cards have 32MB. I wonder about whether that part of it is accurate.
Unfortunately the heatsinks make it difficult to verify directly.

Based on further research, I'm inclined to believe it's a 64MB card rather than trust lspci. I have found a couple cases where lspci isn't completely accurate; it also reports my Rage II+ as having 1MB which doesn't seem right (pretty sure it's a 4MB or 8MB card.)

It's most likely a Quadro 2 Pro. I bought mine just a couple years ago that claimed to be a GeForce 2 Ultra for only $15 (!!!). But yeah, not actually the case.

Granted, even Quadro 2 Pros have gotten expensive for some reason, and I see one for $45 listed right now that isn't a bad price. I've discussed it with Phil, and although I have experience hard-modding these to be a GeForce 2, he's successfully modded one by software using a combination of BIOS flashing and RivaTuner. So that gives you some options.

I'm not familiar with the program the seller used, but maybe it's only reading the GPU type and making an assumption off that?

OK, I confirmed it's a GeForce 2 Ultra.

The seller provided a high resolution image of the back, enough to judge the position of the resistors on the card. In order to be a GeForce 2, resistor R122 cannot be populated, while resistor R121 should be. That is indeed the case for the card in that listing.

eBay card (click to zoom)

My card after modding into a GF2

So this confirms the card is a GeForce 2, and not a Quadro.

That appears the be the case for this card as well, for only $30.

Thanks for the clarification. It's great to have a definitive marker for this line of cards.

EDIT: Just looked at my Quadro 2 Pro: R121 unpopulated, R122 populated.

So... is this thread implying that there are no differences between mentioned cards, just what the card thinks it is and thus what it reports itself as? Is there anything unlocked by modding a card to make it think its something else?

shamino wrote:

16MB seems really odd. Don't those cards usually have 64MB? Even the lower tier GF2 cards have 32MB. I wonder about whether […]
Show full quote

xjas wrote:

I JUST threw together a test rig and found out Linux reports it as a Quadro2 Pro with 16MB RAM

16MB seems really odd. Don't those cards usually have 64MB? Even the lower tier GF2 cards have 32MB. I wonder about whether that part of it is accurate.
Unfortunately the heatsinks make it difficult to verify directly.

Linux only supports OpenGL, in OpenGL there is NO way to directly query the size of the devices memory. So maybe it just lies o.0.

There are ways of estimating how much you are using (as an application), and there are various nV and ATi extensions for getting more dedicated information about the GPU you are running on, but nothing standard.

^^ AFAIK lspci doesn't have anything to do with OpenGL, it just probes all the hardware on the PCI bus (which includes AGP & PCIe) directly. It's an OLD command from way back before OpenGL was even a thing.

I believe the "memory at e000..." line 'should' be the physical memory on the graphics card and "memory at d800..." is the AGP aperture, but obviously it's misreporting those somehow.

I don't think it's meant to report the full amount of memory on the card. I just tried that command on a system with a 1GB GT430 in it, and it reported the same 16MB "32-bit non-prefetchable" block that your Quadro2 reported. It also had a couple 64-bit ranges, but none of it added up to 1GB.
I think it's a reflection of how the card is mapped into system memory space, but the card's whole RAM isn't all mapped there. This would also be why using a 1GB card under WinXP32 doesn't chop off 1GB of addressable system memory.
The 3D programmers probably know more about how the memory mapping works.

I mean, from a fab perspective, they are probably an identical die.

Packaging wise, if you take off the heatsink/fan, a GF2U will be printed on the chip where as a Quadro2 Pro will have Quadro2 Pro printed on the chip.

So to sum this all up what are the things to look for in these cards to determine what it may be?

fsmith2003 wrote:

So to sum this all up what are the things to look for in these cards to determine what it may be?

You don't feel like you've got an answer yet?

xjas wrote:

^^ AFAIK lspci doesn't have anything to do with OpenGL, it just probes all the hardware on the PCI bus (which includes AGP & PCIe) directly

Ah yes, missed that vital piece in the puzzle o.0...

xjas wrote:

It's an OLD command from way back before OpenGL was even a thing.

OpenGL isn't that new 😉 92/93-ish, which I think predates lspci somewhat.

shamino wrote:

I don't think it's meant to report the full amount of memory on the card. I just tried that command on a system with a 1GB GT430 in it, and it reported the same 16MB "32-bit non-prefetchable" block that your Quadro2 reported. It also had a couple 64-bit ranges, but none of it added up to 1GB.
I think it's a reflection of how the card is mapped into system memory space, but the card's whole RAM isn't all mapped there. This would also be why using a 1GB card under WinXP32 doesn't chop off 1GB of addressable system memory.
The 3D programmers probably know more about how the memory mapping works.

Yes nVidia uses UMA so there is no explicit address for the frame buffer, it's one big black box, and rightfully so. Other than how you populate your vertex buffers (interleaving etc), no knoweldge of mem mapping required who knows what buffer is stored next to what texture etc.. who cares, thats the drivers job to do (hence why bad drivers can have such an ipmact on an actual gfx performance).

dexvx wrote:

I mean, from a fab perspective, they are probably an identical die.

This was my understanding, Quadros also shipped with more mem? and perhaps better mem? So bit like Xeons to non-Xeons back int day, hand-picked better examples of tha same die 😀... or marketing BS?. The firmware was different and thus the drivers used at runtime and allowed stuff like multiple clipping planes, logic-operations, various anti-aliasing and overlays to be done in hardware rather than in software by the driver.

Tbh I have always pretty much worked with Quadros (because I don't have to buy them o.0), but I have always wondered what the differences were and if these could be justified by the ever increasing price-hike nV puts on this line... so I looked it up. Here is an nVidia document from '03 describing these differences, in case anyone was wondering what feature differences Quadros actually give over GeForces.
http://nvidia.custhelp.com/app/answers/detail … and-quadro-gpus
These days GPGPU takes prescendence (increase of that is very much welcomed, and FP precision can be important), but no doubt there is still merit in Quadros IQ for just standard geometry display, and the ability to have shite loads of textures (labels etc) and raw vertex buffers in mem. CAD doesn't tend to need 'effects'...Worth it or not? I leave to the reader o.0

They vbios's are interchangeable, since the actual chip and memory are the same in most cases.

Thus, you have to look at the actual chip package by prying off the heatsink/fan.

https://en.wikipedia.org/wiki/GeForce_2_serie … 2_Ultra_GPU.jpg

So Reputator's R121/R122 distinction isn't definitive enough for folks?