'Fusion' cards

[DGhost]

cost and complexity to implement would most likely outweigh any possible benefits. you still have to move the textures across to the card, and you may also increase the amount of data having to be pushed across to the card as DirectX and OpenGL both optimize their data *before* they send it to the card... and with *all* the data being moved across to the card for every screen refresh it could very easily wind up filling the AGP bus quickly.

i dunno tho, i'm not even an engineer, and while i will probably pursue it in the future, thats not gonna be for at least 6 months...

blue_helix: its called glide. it worked quite well until direct3d and opengl caught on, mostly due to technical reasons. glide was more complex to work with than either DX or OGL, but was *fast* because it didn't have alot of the overhead. UT on a V3 will still smoke just about any other card of equivalent cost, and even some more expensive.

[superfly]

Here's my 0.2 cents worth

I think that the best possible implementation for a video card would be that it would follow the same principle as a console.

Consoles are built using a unified memory bus in which the general purpose cpu is running on the same bus as the graphics and sound chip.

A unified bus architecture can never be implemented on standard pc since we'd lose the basic capability of upgrading them or adding new hardware,it's either one or the other.

Since pc's have inherent bandwith limitations that already limit the performance on current video cards and even more so on next generation cards,the logical solution would be to include a general purpose cpu with a graphics processor on the same card running on the same bus.

It's already possible to purchase a 700~800 mhz athlon or p3 for next to nothing,and the cost of memory it is at an all time low,so don't be surprised at all if we see 128 meg video cards very soon.

Agp bandwith issues would be rendered moot since the game code would only load once,and store all the important data locally on the video card and even if all the data coudn't fit on the video card the agp bus could work at peak efficiancy since there's no bus trafic going through the motherboard chipset(for use in the systems cpu,etc...).

This would ensure that the full performance of the video card is acheived,without developers having to compromise because of the wide ranging pc configurations on the market.

In short make the video card handle everything(AI,collision,clipping,gameplay,physics, T.l,raster functions,all Dx features,etc...),leaving only keyboard,mouse and sound processing duties the host system wich wouldn't require a fast cpu anyways.

There is only one little problem with the above scenario,it's called intel or AMD having a rather unpleasant conversation with the video card maker,if the company even thinks about doing a card like that and potentially not make gamers upgrade their systems as often.

[Evil_Donnyboy]

If the video card could do AI and all functions it wouldn't be called a video card. It'd be called a Co-Processor. To do all those funtions, it'd need an ALU, FPU and a whole buncha things CPU have, but video cards don't. But that would defeat the purpose of a video card. What allows the video card to do it's thing so much faster than a CPU can is because it's simplicity. It only has to be designed to do a small amount of funcitons, allowing manufacturers to design it do do those small variety of things very fast.

A CPU on the other hand must be designed to do a whloe lotta different things, from running browsers and OS's to FPU calculations for 3D apps. In fact, current CPUs are capable of doing the same effects that current vide cards, but it'd have to do software emulation, which is very very very slow.

In short, no, your idea of a video card does everything a video card shouldn't won't happen anything soon.

[xortam]

Go back and read my posts again. I've said that my proposal is that of a coprocessor card. I wouldn't think its necessary or desirable to have the game itself running on this card but moving some of the OS graphics kernel/driver onto the card to minimize memory transfers. The graphics APIs (OpenGL, Direct3D, etc.) would still execute out of system RAM via the CPU but data and control, at a common point just below these APIs, would then pass to the graphics coprocessor. DGhost ... isn't the texture data still passed through these APIs and therefore also pass though this common interface? The graphics coprocessor would contain a CPU (perhaps RISC based) and a graphics chip or GPU or perhaps a GPU can do it all. Once the data is on-board it would just need to move across a dedicated internal bus (or not all in the GPU only case).

[superfly]

But that's the beauty of it.

Not only will you get the full performance out the video card,but also make whatever system you may have to use it with irrelevant.

Have you seen the X-box demos and how a relatively slow cpu working on a unified bus with a very fast graphics chip can work wonders in terms of graphics.

You can have the fastest cpu out there running on the fastest bus system available,you'll never get the full performance out of your video card no matter what you do.

For instance i made a few calculations,and if you want to see games that have 100.000 polys per frame(a Geforce or radion can do this)at 30 fps requires about 500 megs per second of bandwith for that data to reach the card even before it can process it,and that's without including texture uploads at all.

The only way that it would be possible to upload that many polys,would be to implement some form the vertex compression to reduce the amount of bandwith needed to more manageable levels(similar to how s3tc works for textures).

Clearly,in a situation where data has to go through multiple buses before it gets to the video card itself(A.K.A pc's),you'll never see the full performance capabilities of recent video cards let alone next generation ones.

Why do you think that the increases in poly count in games have been relatively small over the last 3/4 years,it's because of those bus speed limitations,and only in recent games do we see games with at most 20/25 000 polys per frame which is still a joke for any T.l enabled card.

In fact the biggest reason why games look better today is mostly due to the fill rate increases on video cards,which allow for more complex textures.

And like i mentioned before,adding a general purpose cpu isn't that expensive,thanks to the insane speed race going on between intel and AMD,which is completely useless save for games or other very demanding software.

Don't come and tell me that you need ghz+ cpu's to run your o/s,browser or office application.

[DarkstarG]

How about a flashable ROM with the DirectX AND OpenGl built in that could be updated as both API's were updated.

Also to get the CORRECT answer up here again what the FUSION means is a video chip for UMA motherboards (you know, motherboards with built in sound, video, modem, etc...) Thats what the FUSION chip is.

[Ant]

was

[Gurm]

A thought on a card with "built-in" DirectX and OpenGL...

Do you really want to flash the card every time a new driver set comes out? Because what you're talking about is having really miniscule drivers on the PC and more robust drivers ON THE CARD.

Now, given how buggy some driver releases tend to be (especially the initial ones), do you really want that on the card where you have to flash the ROM to fix it?

I don't think so.

- Gurm

------------------
Listen up, you primitive screwheads! See this? This is my BOOMSTICK! Etc. etc.

[DGhost]

to superflys post: thats what SGI did with their visual workstations a while ago, if i understand correctly. remember the ones that had dual and quad procs?

with the x-box, it will never have to render above 640x480. the resolution of a tv is just way too limited. get a GF2, put it an a PIII 600, run it at 640x480 and see how fast it runs. it goes back to fillrate and lots o other things.

xortam: well... you either implement DX in hardware/software on the card, in which case you have to implement a full blown rasterizing engine on the card (for DX at least) and do all the work that directx does, but you don't see, on the card. Ie, polygon culling and clipping, z buffer processing and all the other stuff that DX does *before* it sends it to the card. a poorly written program that shoves lots of polygons at DX knowing that they will be eliminated will instantly max the bus out. or you do that in software which effectively makes it like it currently is.

also, if the AGP bus is such a limiting factor, why does AGP 2x and 4x make minimal difference. like .5 fps.

[superfly]

But you have to remember one thing,as far as cpu processing goes,there is absolutely no difference in the amount of work that the cpu has to perform from 640*480 to whatever resolution you want,in any game,as long as the same amount of polys are being used.

Changes in resolution or color depth only affect the video card itself and nothing more.

[Maggi]

<font face="Verdana, Arial, Helvetica" size="2">Originally posted by Ant:
was</font>

dry

[superfly]

just to prove that pc's are extremely bus limited and it applies even in the case of intel's 850/rambus solution,consider the following...

Over the last 7 years since the introduction of the first pentium systems,today's cpu's are at least 20~25 times faster than anything available back then,but bus speeds increases have been very slow in the same period of time,today's systems have bus speeds that are at most 5 times faster.

Still think you're getting the full performance out of your cpu???

think again.

The only time the full performance out of cpu's is being used is when they're processing data that is so intensive it might take the cpu 50 clock cycles(or more)to calculate it before it's ready to receive the next batch.

But games don't fall in that category,not even close.

Things like stress analisys,high end 3d rendering,servers,medical research,etc..,really do push today's systems to their limits.

[This message has been edited by superfly (edited 04 February 2001).]

[dZeus]

what fusion? never heard of it

[Bohrn]

Digital Viper X, ya forgot tile based rendering, back face culling, glide rendering and solar powered! Except on cloudy days ya can plug it into a sliced potatoe or citrus fruit. Laptoppies love it!


------------------
AsusP3B-F,P3 880 Slot1, 256mgsPC133,G40032megSH, Diamond MX300,13.5gigs of HD's,52X CDrom,WinME, PD 5.41 w/ 6.10ICD,Altec Lansing ATP3Subwoofer,
Envision 17",Terrayon Cable modem w/D-Link nic,1 grey cat,1 black cat & 1 calico

[Jorden]

But Gurm, we love to download a minimum of 12Mb zip files that when unzipped deliver us a minimum of 35Mb of drivers, which we then flash to our videocard

Just hope you don't get a power-interrupt.

Then again, with the EEPROM flasher that you can buy from MAtrox at an extra price, you can always flash it back to factory settings. Those drivers are a mere 2Mb

Jord

------------------
This cat, is cat, a cat, very cat, good cat, way cat, to cat, keep cat, a cat, fool cat, like cat, you cat, busy cat! Now read the previous sentence without the word cat...