VOGONS


First post, by Ozzuneoj

User metadata
Rank l33t
Rank
l33t

I'm sure I've posted about this before, but I hope I at least haven't made this exact thread already and forgotten about it... because this is a problem that's been bugging me for years.

Is there any program or benchmark that can accurately identify the specs of a video card? I'm mainly concerned with the parts from the late 90s and early 2000s, when cards often had 20 different variants and performance ranged from good to abysmal.

Being able to identify physical characteristics is a nice idea, but it doesn't always work. For example, the number of memory chips normally will tell you when a card definitely does NOT have a 128bit bus (one or two chips), but you can't use it to determine if a card definitely DOES have a 128bit bus (some cards with 4 chips are either 64bit or 128bit, even within the same model).

I use Everest to figure this stuff out sometimes but it often reports nothing for certain cards (mainly Rage 128 cards). Rivatuner can sometimes shed some light on these specs but it seems only for later cards.

Any suggestions?

Also, if we can find a 100% reliable way to do this, I will post pictures of cards I have and the results.

Now for some blitting from the back buffer.

Reply 1 of 12, by Putas

User metadata
Rank Oldbie
Rank
Oldbie

There is no simple way. But bus width is the easiest of them all. Experienced eye and few tests for confirmation and you are set.

Reply 3 of 12, by candle_86

User metadata
Rank l33t
Rank
l33t
Ozzuneoj wrote on 2020-05-08, 04:54:
Putas wrote on 2020-05-08, 04:48:

There is no simple way. But bus width is the easiest of them all. Experienced eye and few tests for confirmation and you are set.

Care to elaborate? That's the entire point of this thread. 😀

Read the memory chips and look if they are 8bit, 16bit or 32bit modules then multiply by number of chips.

Reply 4 of 12, by chrismeyer6

User metadata
Rank l33t
Rank
l33t

Have you tried using GPU-z I've had luck with it and older cards.

Reply 5 of 12, by Ozzuneoj

User metadata
Rank l33t
Rank
l33t
candle_86 wrote on 2020-05-08, 12:07:

Read the memory chips and look if they are 8bit, 16bit or 32bit modules then multiply by number of chips.

That's a good idea. Not very fast when dealing with a ton of cards, but if that is an accurate way to tell, then at least it is something.

chrismeyer6 wrote on 2020-05-08, 13:49:

Have you tried using GPU-z I've had luck with it and older cards.

I tried it a few years ago and it didn't seem to provide info for cards that old. Has it been updated for "retro" stuff?

Now for some blitting from the back buffer.

Reply 6 of 12, by chrismeyer6

User metadata
Rank l33t
Rank
l33t

It seems to of gotten better it was able to correctly identify my GeForce 4 ti4200 and my GeForce 4 MX460

Reply 7 of 12, by Ozzuneoj

User metadata
Rank l33t
Rank
l33t
candle_86 wrote on 2020-05-08, 12:07:

Read the memory chips and look if they are 8bit, 16bit or 32bit modules then multiply by number of chips.

Can you help me interpret this?

As a quick test I just checked a Winfast 3D S320 II (TNT 2) with 32MB of SGRAM (4 chips on each side). Here is the datasheet for the chips on the card:

https://www.digchip.com/datasheets/parts/data … G112Q-6-pdf.php

512K x 32bit x 2 banks

If these are 32mbit (4MB) chips, and the card has 8 of them, that makes 32MB, which is correct. But if I try to find the bus width the same way that would be 32bit x 8 which is 256bit, which isn't correct, so there must be more to it than simply multiplying the number of chips by the bits as shown in the datasheet... at least in the case of this card. If you only multiply the ones on each side, how do you know which cards are designed this way and which aren't?

chrismeyer6 wrote on 2020-05-08, 15:38:

It seems to of gotten better it was able to correctly identify my GeForce 4 ti4200 and my GeForce 4 MX460

Those are quite a bit newer than the cards I need to test. Of course, it is super handy to be able to test Geforce 4MX cards because the specs are an absolute mess... but I believe Everest works for those as well.

Now for some blitting from the back buffer.

Reply 8 of 12, by candle_86

User metadata
Rank l33t
Rank
l33t
Ozzuneoj wrote on 2020-05-08, 15:51:
Can you help me interpret this? […]
Show full quote
candle_86 wrote on 2020-05-08, 12:07:

Read the memory chips and look if they are 8bit, 16bit or 32bit modules then multiply by number of chips.

Can you help me interpret this?

As a quick test I just checked a Winfast 3D S320 II (TNT 2) with 32MB of SGRAM (4 chips on each side). Here is the datasheet for the chips on the card:

https://www.digchip.com/datasheets/parts/data … G112Q-6-pdf.php

512K x 32bit x 2 banks

If these are 32mbit (4MB) chips, and the card has 8 of them, that makes 32MB, which is correct. But if I try to find the bus width the same way that would be 32bit x 8 which is 256bit, which isn't correct, so there must be more to it than simply multiplying the number of chips by the bits as shown in the datasheet... at least in the case of this card. If you only multiply the ones on each side, how do you know which cards are designed this way and which aren't?

chrismeyer6 wrote on 2020-05-08, 15:38:

It seems to of gotten better it was able to correctly identify my GeForce 4 ti4200 and my GeForce 4 MX460

Those are quite a bit newer than the cards I need to test. Of course, it is super handy to be able to test Geforce 4MX cards because the specs are an absolute mess... but I believe Everest works for those as well.

2 banks is your giveaway

Reply 9 of 12, by Ozzuneoj

User metadata
Rank l33t
Rank
l33t
candle_86 wrote on 2020-05-09, 14:37:
Ozzuneoj wrote on 2020-05-08, 15:51:
Can you help me interpret this? […]
Show full quote
candle_86 wrote on 2020-05-08, 12:07:

Read the memory chips and look if they are 8bit, 16bit or 32bit modules then multiply by number of chips.

Can you help me interpret this?

As a quick test I just checked a Winfast 3D S320 II (TNT 2) with 32MB of SGRAM (4 chips on each side). Here is the datasheet for the chips on the card:

https://www.digchip.com/datasheets/parts/data … G112Q-6-pdf.php

512K x 32bit x 2 banks

If these are 32mbit (4MB) chips, and the card has 8 of them, that makes 32MB, which is correct. But if I try to find the bus width the same way that would be 32bit x 8 which is 256bit, which isn't correct, so there must be more to it than simply multiplying the number of chips by the bits as shown in the datasheet... at least in the case of this card. If you only multiply the ones on each side, how do you know which cards are designed this way and which aren't?

chrismeyer6 wrote on 2020-05-08, 15:38:

It seems to of gotten better it was able to correctly identify my GeForce 4 ti4200 and my GeForce 4 MX460

Those are quite a bit newer than the cards I need to test. Of course, it is super handy to be able to test Geforce 4MX cards because the specs are an absolute mess... but I believe Everest works for those as well.

2 banks is your giveaway

Not trying to be obtuse, but... what does that mean?

If one chip says it is 512K x 32bit x2 banks, that makes it sound like it should be 32bit x2 per chip (64bit), which is even more wrong when there are 8 chips. If the x2 is referring to how many chips are in a bank, then that also can't be right because the card is 32MB, so each chip has to be 4MB (which would be 512x32x2 per chip). Do you actually DIVIDE by the number of banks to determine the bus width of each chip?

I guess the reason I'm struggling with this is that it seems to be the opposite of how how the term "bank" was used back when I first heard the term, when consumer boards first started having dual channel memory. On an nForce 2 Ultra board, for example, it was described as having two banks, each with their own 64bit width channel. By utilizing both banks (usually up to two RAM slots for one channel, one for the other) you had dual channel which gave you a 128bit memory bus. This is 64bit x 2 banks = 128bit. If 32bit x 2 banks on a memory chip on a video card somehow equals 16bit, this is why I'm confused.

Now for some blitting from the back buffer.

Reply 10 of 12, by Tetrium

User metadata
Rank l33t++
Rank
l33t++
candle_86 wrote on 2020-05-09, 14:37:
Ozzuneoj wrote on 2020-05-08, 15:51:
Can you help me interpret this? […]
Show full quote
candle_86 wrote on 2020-05-08, 12:07:

Read the memory chips and look if they are 8bit, 16bit or 32bit modules then multiply by number of chips.

Can you help me interpret this?

As a quick test I just checked a Winfast 3D S320 II (TNT 2) with 32MB of SGRAM (4 chips on each side). Here is the datasheet for the chips on the card:

https://www.digchip.com/datasheets/parts/data … G112Q-6-pdf.php

512K x 32bit x 2 banks

If these are 32mbit (4MB) chips, and the card has 8 of them, that makes 32MB, which is correct. But if I try to find the bus width the same way that would be 32bit x 8 which is 256bit, which isn't correct, so there must be more to it than simply multiplying the number of chips by the bits as shown in the datasheet... at least in the case of this card. If you only multiply the ones on each side, how do you know which cards are designed this way and which aren't?

chrismeyer6 wrote on 2020-05-08, 15:38:

It seems to of gotten better it was able to correctly identify my GeForce 4 ti4200 and my GeForce 4 MX460

Those are quite a bit newer than the cards I need to test. Of course, it is super handy to be able to test Geforce 4MX cards because the specs are an absolute mess... but I believe Everest works for those as well.

2 banks is your giveaway

So is there a way to tell how many banks a graphics card has?

Whats missing in your collections?
My retro rigs (old topic)
Interesting Vogons threads (links to Vogonswiki)
Report spammers here!

Reply 11 of 12, by Putas

User metadata
Rank Oldbie
Rank
Oldbie
Ozzuneoj wrote on 2020-05-08, 15:51:

If these are 32mbit (4MB) chips, and the card has 8 of them, that makes 32MB, which is correct. But if I try to find the bus width the same way that would be 32bit x 8 which is 256bit, which isn't correct, so there must be more to it than simply multiplying the number of chips by the bits as shown in the datasheet... at least in the case of this card. If you only multiply the ones on each side, how do you know which cards are designed this way and which aren't?

The chips on each side will share positions, therefore don't multiply 2x. They are in so called clamshell mode, sharing access, simplifying board design and saving some power.

Reply 12 of 12, by Tetrium

User metadata
Rank l33t++
Rank
l33t++
Putas wrote on 2020-06-03, 04:08:
Ozzuneoj wrote on 2020-05-08, 15:51:

If these are 32mbit (4MB) chips, and the card has 8 of them, that makes 32MB, which is correct. But if I try to find the bus width the same way that would be 32bit x 8 which is 256bit, which isn't correct, so there must be more to it than simply multiplying the number of chips by the bits as shown in the datasheet... at least in the case of this card. If you only multiply the ones on each side, how do you know which cards are designed this way and which aren't?

The chips on each side will share positions, therefore don't multiply 2x. They are in so called clamshell mode, sharing access, simplifying board design and saving some power.

Niiiice. I learned something new today 😀
I had never heard of clamshell mode before (so I had no words to google with), but it makes sense and as a bonus this is usually easy to identify visually.
From what I gather, it's a new thing (it mentions specifically GDDR5) but the principle is probably much older.

This way of manufacturing GPU cores would also make it more convenient to disable part of the GPU so the remaining part of the GPU can still be used with half (or other divider) part of its memory bandwidth.

Whats missing in your collections?
My retro rigs (old topic)
Interesting Vogons threads (links to Vogonswiki)
Report spammers here!