PDA

View Full Version : The Ideal Graphics Card


khellandros66
07-30-07, 11:56 AM
Hmmm I wonder why no one is working on this yet, but here is my idea.

Core 1 = Geometry = 500Mhz 256bit mem interface w/ 256MB GDDR3
Core 2 = Physics = 500Mhz 256bit mem interface w/ 256MB GDDR3
Shader Core = 2Ghz 384bit mem interface w/ 512MB GDDR3

Each core could handle all its taskes, once done Core 1 & 2 pipes through the faster shader core for a complete mix. Instead of pumping cost with useless faster ram why not put it to better use? And if say the geometry add/or physics core needs more ram, there should be a hard coded way to que up into the system ram... This system would also allow a CPU to be free to do the AI, OS Background Processes, etc

~Bobby

BronzeGod
07-30-07, 12:43 PM
:headexplode:

DRen72
07-30-07, 12:52 PM
http://www.nvnews.net/vbulletin/showthread.php?t=95241

SLippe
07-30-07, 01:13 PM
Hmmm I wonder why no one is working on this yet, but here is my idea.

Core 1 = Geometry = 500Mhz 256bit mem interface w/ 256MB GDDR3
Core 2 = Physics = 500Mhz 256bit mem interface w/ 256MB GDDR3
Shader Core = 2Ghz 384bit mem interface w/ 512MB GDDR3

Each core could handle all its taskes, once done Core 1 & 2 pipes through the faster shader core for a complete mix. Instead of pumping cost with useless faster ram why not put it to better use? And if say the geometry add/or physics core needs more ram, there should be a hard coded way to que up into the system ram... This system would also allow a CPU to be free to do the AI, OS Background Processes, etc

~Bobby
I can only imagine the cost of something like that! :thumbdwn:

khellandros66
07-31-07, 12:13 PM
Well cost isn't gonna be that much more, especially if they were to use 65nm, and just imagine two of these in SLI with proper memory management. Granted we see only one source of ram now its because the cards stock up the same geometry and textures on both cards' ram, if they were to operate independantly, then well...you can grasp the significance of its improvemnets.

~Bobby

Madpistol
07-31-07, 12:56 PM
3 cores @ 65nm a piece on a single card???

:headexplode:

Good idea, but it's not going to happen unless some of those cores run on the same chip. Also, If the shader core memory uses 512MB, by your definition, it has to be either 256-bit or 512-bit. 384-bit only works for 768MB or 384MB of memory.

jolle
07-31-07, 04:48 PM
reason why that wont happen is pretty much the reason they went with unified shader core in the first place.
You will get alot of deadtime on one or 2 of the cores while one is fully loaded.
The physics core is gonna be idle alot since all games wont support it.
Better to assign a couple of the shader units to physics when implemented in the software, then to have a dedicated part often sitting idle.

Altho if you can make a arbiter that can balance out tasks to a number of GPUs, you could perhaps have say 3 GPUs on a board, have the arbiter hand out tasks to them and keep them balanced and as loaded as possible.
and allow all GPUs to share the RAM over dedicated busses, if that is even possible.
If all GPUs could work on the same frame, that would offer alot of aggregate bandwith I guess.

john19055
08-04-07, 02:20 PM
It looks like a multiple CPU inside it right now.






27063