Pushing The 512MB Barrier On Video Cards

Hack Jandy writes "Remeber your ancient TNT graphics card that had 16MB of memory? ATI is pushing the texture barrier by incorporating 512MB in their newest X850 video card lineup. The catch? Even ATI acknowledges there will probably be no performance benefits to bumping the memory support from 256MB to 512MB as the cards are 'intended to demonstrate the next-generation capability to gamers." An anonymous reader points out that Gainward (which sells NVidia-based graphics cards), will shortly introduce its own 512MB card, according to Hexus.net.

Comments

512 is better

[mrtroy]

Because it is bigger than 256.

TWICE as big!!!

If my email tells me anything, size DOES matter.

Re:512 is better

[NanoGator]

"If my email tells me anything, size DOES matter."

Seeing as how most of the 'realism' of a 3D game comes from detailed textures, yes, size of texture ram does matter.

Re:512 is better

[maglor_83]

I find most of the realism comes from the physics engine. The texture just makes it look a bit prettier, but by no means makes the game any better.

Re:512 is better

[NanoGator]

"If my email tells me anything, size DOES matterI find most of the realism comes from the physics engine. The texture just makes it look a bit prettier, but by no means makes the game any better."

I didn't say it makes the game better. And yes, I should have defined 'realism' a little more clearly. I meant the rendered visuals of it, not the motion of it. You can do a lot more to make an image 'photo-real' with greater texture resolution than you can do with faster processing etc. Ask anybody who's played Doom 3. The normal mapping in that game, love it loathe it, did a great deal more to the visual detail of the game than adding a few more polygons to the scene.

Re:512 is better

[Ravenscall]

It is not the size of the RAM, but the motion of the engine.

Fast and Big mem

[Fox_1]

with 16 pixel and 6 vertex pipelines clocked at 540MHz. The graphics card's 512MB of DDR3 SDRAM operate at 1180MHz speed and have 256-bit memory interface.
Kinda sad but this card is more powerful then my PC on it's stats alone

translation

[British]

Even ATI acknowledges there will probably be no performance benefits to bumping the memory support from 256MB to 512MB as the cards are 'intended to demonstrate the next-generation capability to gamers."

Translation: Even though it's not practical, we'll sell it since gamers will buy it.

Well make it useful in a creative way

[FerretFrottage]

Why not create special drivers that allow you to use the unused vid ram as a ramdisk? If a game requires more than 256MB, then default the temp area back to file storage, but if you are only using 128-256MB for video, then let me do something useful with the remainder.

A use for this

[ZWheel]

I seem to remember someone writing a linux kernel module that lets you use extra video mem as a very fast virtual drive.

Now...

[inertia187]

Even ATI acknowledges there will probably be no performance benefits...

Now if we can just get those razor manufacturers to say the same about that 5th blade.

Scientific Applications

[ghoti]

This may not do much for games, but for scientific applications, especially visualization of large datasets, this is great. The visualization community has been using the advances made for gaming over the last years, and it's amazing what you can now do on the GPU: flow simulation, interactive visualization of large volumetric datasets with complex transfer functions, shading, etc.
For these applications, the more memory, the better.

General GPU Programming

[mjinman]

The move might not matter a whole lot to the normal gamer, but those of us who are researching/using video cards as fast vector coprocessors love this as it increases the matrix (texture) size we can do operations on. (I especially love it since some of my stuff runs 40x on my Radeon X800 than my Athlon 64 - its all linear algrebra, finite difference codes)

Almost Absurd

[ewhac]

"Would you like to mount unused graphics RAM as a swap device?"

Seriously, what's all that RAM used for when you're not playing games? It's still eating power; you may as well use it for something...

Schwab

Maybe Id's taking a year off dead for tax reasons.

[Thud457]

"It's the weird color-scheme that freaks me. Every time you try to operate one of these weird black controls, which are labeled in black on a black background, a small black light lights up black to let you know you've done it."

GL based window managers

[wowbagger]

Let's see:

Assume you were to use an OpenGL based window manager, wherein each window on your screen is little more than a polygon with a texture applied to it.

Assume you are working at 1600x1200 resolution, 24 bit color depth (padded to 32 bits for possible alpha channel).

Your frame buffer alone takes 7.3 MiBytes.

If you have a 32 bit Z buffer, add another 7.3 MiBytes.

Each 2D window in use will consume texture memory, so if we assume that the remaining 497.4 MiBytes of memory on the card as window memory, that lets us open roughly 68 full-screen windows before consuming all texture memory on the card.

If some of the windows are 3D windows themselves, you are going to want them to have their own Zbuffers - so double the memory usage for them.

While 68 windows may sound like a lot, given that most GL compositing schemes I've heard of want to keep ALL windows available, even if they are not mapped, to avoid expose events to the apps and to speed window open and close events, and I could see you getting to 30 windows pretty easily. Allowing double that for headroom doesn't seem like so bad an idea to me.

And I've ignored the XVideo overlay needs.

I can now die happy.

[raygundan]

The rare occurence of this sort of profoundly geeky post is why I still come to slashdot. God bless you, crazy GPU vector coprocessor finite difference code matrix guy!

Re:I can now die happy.

[sconeu]

I, for one, welcome our crazy GPU vector coprocessor finite difference code matrix guy overlords.

Re:This is why sound cards are no big deal!

[tsangc]

Ever wonder why GPUs are such a big deal and sound cards are such an after thought?

I think the reason why soundcards don't change very much because the fundamental methods of generating sound isn't compute intensive.

With 3D video, you're computing the display output, ray tracing, shading, whatever it is. Algorithms not samples define the visuals. Certainly there are "samples" (ie, texture maps) but these themselves need to be rendered through computation. At the same time, resolutions for display are increasing, requiring more computational horsepower. Hence a need for progressively faster CPUs to drive larger, more details and faster framerate visuals.

With audio, a lot of the audio world is still sample based--there usually aren't algorithms generating sounds from fundamental principals. If there are, it's in a highly specific use (ie, virtual instruments in something like Cubase, which uses the main CPU) or it's in some sort of environmental processing, like DSP effects, positioning etc which don't require that much performance past existing products today that have integrated DSPs. That and audio resolution in general isn't increasing--not at a rate compared to someone going from a 800x600 to 1920x1280 pixel display. Even adding extra channels doesn't seem to drive this requirement further.

As a result, I guess you just don't see the requirement to have "more powerful sound cards".