How much memory on video card?

I bought a new used PCI-ex16 video card that says on it

"256MB onboard supporting 512MB DDR2".

Where are these other 256MB supposed to be?

Plugged into the card?

In Pittsburgh?

I suppose when it's installed, software can tell me how much is in
place now, but lets say I'm at a hamfest and I want to tell by
looking. Even with a magnifying glass, I can't actually see any
identifiable memory, let alone a socket where the second 256M could
have been plugged in, or removed, or could be plugged in in the
future.

Micky wrote:
I bought a new used PCI-ex16 video card that says on it

"256MB onboard supporting 512MB DDR2".

Where are these other 256MB supposed to be?

Plugged into the card?

In Pittsburgh?

I suppose when it's installed, software can tell me how much is in
place now, but lets say I'm at a hamfest and I want to tell by
looking. Even with a magnifying glass, I can't actually see any
identifiable memory, let alone a socket where the second 256M could
have been plugged in, or removed, or could be plugged in in the
future.

That's Hypermemory. Just stolen from system memory :-(

https://en.wikipedia.org/wiki/HyperMemory

https://en.wikipedia.org/wiki/TurboCache

"The Name goes on, before the Quality goes In".

At the marketing meeting, DozingRAM and SnoreMemory
did not make the cut as descriptive terms for the adverts.
The two above terms were the only ones left on the list.

*******

An equally silly idea, were some of the motherboard chipsets
with built-in GPU. Now, that configuration normally relies
entirely on system RAM. And so some genius decided to put a
32 bit interface on the Northbridge and connect a single 32 bit
wide DRAM chip, as additional frame buffer. The chip capacity
was tiny, and the impact of the chip, just about zero. The
motherboard makers had probably concluded the engineers
who designed it were on LSD when they thought it up. And
so each motherboard had this "hood ornament" soldered next to
the Northbridge. But, it was all in the name of marketing
and feature differentiation. Naturally, it disappeared
as a concept, never to be seen again. "Now With Hood Ornament".

ATI already has a solution for that sort of thing, where they
connect RAM chips (four of them) to the top of the GPU. And
that's popular in things like laptops. Numerically you need
a few chips, to give a decent texture RAM capacity. Four chips
is better than one chip.

*******

The cards with plugin RAM were circa 2000 or so.
A 4MB card with room for a 4MB plugin module,
that sort of thing. More marketing shenanigans.
Back then, we really needed that RAM, as there
was barely enough for a frame buffer. They should
have just soldered it all down, and dispensed with
a socket.

Before you go to these hamfests, you should make
yourself a list of "do buy" or "don't buy" video
cards. And then you'll have some idea what to stay
away from. On modern cards, you can get a worthwhile
card for $30 to $50, with RAM soldered to it. Enough
RAM to hold textures. Even if the buffer transfer
rate hasn't improved on those cards, since about
the year 2003. They're perfectly fine if you're not
playing Crysis, and your tastes are more SimCity.
And with a modern card, they're "buzzword compliant".
The card has shaders, can accelerate Firefox or Flash,
has a built-in video decoder (less CPU usage for DVDs).

(Note - site owners no longer add new video cards to list...)

http://www.gpureview.com/videocards.php

To give a comparison, my Test Machine has an HD6450,
which was around $50.

Radeon 9800Pro 128MB 21.76GB/sec 2003-04-02
Radeon HD6450 GDDR3 12.8 GB/sec 2011-04-18

So you can see how technology is really "taking off" :-)
Whoosh. I think my 9800Pro is faster than more than
50% of the video cards in the house. Some have 6.4GB/sec
video memory on them. Zazoom.

I don't think I've played any games on the HD6450 at
all. So I have no idea how fast the Sims race
around the screen.

If you look in that list, cards like this are $30 to $50
and give you buzzword compliance and a chance of driver
support for about 5 years (to the end of Win10 maybe).
With that ole familiar smokin frame buffer. At least
they don't skimp on the amount of RAM, because it's
hard to make smaller chips :-) I don't think
NVidia is particularly getting rich off these.

GeForce GTX 720 DDR3 14.4GB/sec 2014-03-27

Paul

Micky wrote:

I bought a new used PCI-ex16 video card that says on it

"256MB onboard supporting 512MB DDR2".

Where are these other 256MB supposed to be?

Plugged into the card?

In Pittsburgh?

I suppose when it's installed, software can tell me how much is in
place now, but lets say I'm at a hamfest and I want to tell by
looking. Even with a magnifying glass, I can't actually see any
identifiable memory, let alone a socket where the second 256M could
have been plugged in, or removed, or could be plugged in in the
future.

It would be a damn old video card to have an sockets on it. That a PCB
could support more memory modules does not necessitate that there
actually be more sockets or even more foil paths and socket patterns on
the PCB to which more memory modules could be soldered. If the memory
was not already installed (meaning the foil patterns on the PCB exist to
which those modules can be soldered), don't expect "support" of greater
memory to mean there is a physical means of adding more memory.

Laptops are the same. The same mobo may be used with several models.
On some [later or more expensive] models, the foil patterns exist to
which a socket is soldered so the user could add more memory. However,
in those more expensive models, they already come with more memory. Due
to cost efficiency, the PCB foil patterns may exist on the mobo PCB but
nothing was soldered to them to make use of those foils. So the same
mobo may support 8MB of system RAM but a lower less capable model can
only support 4MB because there is no socket in which the user could add
another 4MB memory module.

Did the used video card you bought actually have sockets into which you
can insert more modules? Socket cause signal attenuation and
degradation and why, for a long time now, VRAM has been directly
soldered to the PCB. If there are no sockets, it may be that you are
required to remove and discard the old memory modules and buy and insert
new higher capacity memory modules. That is, the controller and foils
exist that can support higher addressing so you can replace old memory
with higher capacity addressing. You lose the old low-capacity memory
modules to replace them with new high-capacity modules, assuming those
foils and controller have the address lines available to support the
higher addressing.

Memory modules on video cards is pretty easy to see. They are the next
biggest chips on the PCB from the processor chip. If you cannot see any
memory on the video card's PCB then it has some small amount of memory
(needed to boot time since the video card's BIOS has to get loaded
before the system BIOS) then it steals its memory from the system RAM.
System RAM used to be much slower than VRAM (video RAM) but I haven't
kept up with DDRAM and VRAM to know how much difference in performance
there is nowadays. Any video cards that steal system RAM to usurp for
their own are considered low-end video cards. This is where onboard
video gets its memory and why onboard video is always slower benchmarked
than the separate daughtercard video cards with their own VRAM or
synchronous DRAM.