Coprocessors all over!
Back a few months ago, I predicted that coprocessors were about to make a come back. They have in the corporate world, with Cray's introduction of the XD1 supercomputer, using Xilinx's Virtex FPGA alongside AMD's Opteron. I went a few steps further, and predicted that coprocessors would start to make some waves on the desktop side, just like the 3d accelerators that we take for granted these days.
So today I'm reading some news and come across this new company called Aiseek. They make a product called Intia which is a hardware/software combo that aims to give a whole new level of realism in computer games. This product has yet to really get picked up yet, and honestly I hope it never does.
Even though I like the idea of better AI in games, I don't think that we need another $300 card to make it happen. Just like the Ageia Physx card, which has yet to show off any improvements, I think this Intia card is just a bunch of hogwash.
What we need is a field-programmable gate array (FPGA) that can be programmed on the fly to do whatever acceleration is needed. Say you're playing a game that has very high physics demands, turn that FPGA into a physics accelerator and you're off. Shutdown that game and start up one that needs more AI, not a problem. Doing some serious Photoshop? We have that covered too.
ATI is starting to make some moves in that direction. They want to use a second or third graphics card for it's floating-point abilities to help speed up Havok's new physics engine. There's also been talk of using the GPU's power to research complex protein folding. If we can do this without another card, I'm all for it, but don't make me buy a card that I only need for one game.
One add-on board I didn't talk about yet is the Killer NIC from Bigfoot networks. Now this is actually a good product I think, as network processing is something that everyone can use. As broadband speeds near 10Mips, our aging machines are having trouble sorting out all that data. Purchasing a card that takes over that load could potentially help everyone out.
So the coprocessor is back, and I'm sure that Physx, Killer NIC and Intia are just the tip of the iceberg. Let's just hope we don't get stuck swapping cards depending on the game we want to play.
So today I'm reading some news and come across this new company called Aiseek. They make a product called Intia which is a hardware/software combo that aims to give a whole new level of realism in computer games. This product has yet to really get picked up yet, and honestly I hope it never does.
Even though I like the idea of better AI in games, I don't think that we need another $300 card to make it happen. Just like the Ageia Physx card, which has yet to show off any improvements, I think this Intia card is just a bunch of hogwash.
What we need is a field-programmable gate array (FPGA) that can be programmed on the fly to do whatever acceleration is needed. Say you're playing a game that has very high physics demands, turn that FPGA into a physics accelerator and you're off. Shutdown that game and start up one that needs more AI, not a problem. Doing some serious Photoshop? We have that covered too.
ATI is starting to make some moves in that direction. They want to use a second or third graphics card for it's floating-point abilities to help speed up Havok's new physics engine. There's also been talk of using the GPU's power to research complex protein folding. If we can do this without another card, I'm all for it, but don't make me buy a card that I only need for one game.
One add-on board I didn't talk about yet is the Killer NIC from Bigfoot networks. Now this is actually a good product I think, as network processing is something that everyone can use. As broadband speeds near 10Mips, our aging machines are having trouble sorting out all that data. Purchasing a card that takes over that load could potentially help everyone out.
So the coprocessor is back, and I'm sure that Physx, Killer NIC and Intia are just the tip of the iceberg. Let's just hope we don't get stuck swapping cards depending on the game we want to play.
posted by Nate Solberg at
10:22 AM
0 Comments:
Post a Comment
<< Home