R700 Architectural Rumors (Inq :( )

E

Erasmus354

[H]F Junkie
Joined
Mar 12, 2004
Messages
9,450
http://www.theinquirer.net/default.aspx?article=35818

Ok first of all, yes it is the inq and nothing they post is to be believed. However I believe the rumor they started about the R700 is interesting. If ATI / Nvidia could come up with a simple and fast means of connecting smaller cores together it could be a boon (at least to their bottom lines if not to consumers as well). These swarms of GPU's on cards could make for simpler manufacturing, and no longer will we see the long wait times between release of a next gen high end part and the more affordable mid and low range parts.

Whether it is true or not for the R700 I think the architectural ideas have at least some merit, with readily apparent pros and cons. Personally I dont think they can get a fast enough interconnect between the chips, but they might not have a choice since next gen GPUs are reaching massive sizes already something needs to be done.
 
It would be interesting if they got to the point where discrete video cards turned into little miny multi-socket mobos that you bought GPUs and GDDR for. Me thinks it would be cheaper for us.

As for the Inq's R700 - it does sound interesting. kinda reminds me of Voltron. :rolleyes:
 
Actually, I was originally thinking G80/R600 would go this route to some degree. I had the idea that perhaps triangle/geometry/setup would be on one chip, while all pixel shading was on a second.

I do honestly see a future like is described in the article as a possibility. Think back to the Voodoo (separate texture chips etc...). We've already had this. I can see it again. Old things become new again.
 
What this means is R700 boards will be more modular, more scalable, more consistent top to bottom, and cheaper to fab.
Click to expand...

Hopefully $400 for top end? That would definitely be an awesome event.
 
Well I don't care if it's the Inq or not, this makes too much sense not to be true. I mean just look at where CPUs are going: dual core, quad core, then talk of even more.


 
yeah this is interesting and it does seem logical just like intel and the C2D's.
 
tisb0b said:
yeah this is interesting and it does seem logical just like intel and the C2D's.
Click to expand...

and unlike dual core/quad core and etc, with GPUs we could/should have the ability to see performance increases immediately :D
 
Go read "The singularity is near".
Talks about massive parallelism.
This is where all computing will be going.
 
Seems reasonable to me. We can't expect to continue to double transistor counts every generation. This modular, segmented system makes sense from a manufacturing complexity and yield standpoint, but how will each of these devices talk, and what sort of cooling solution will we need?

I'm not sure if this is the time to move to such arrangements, but I suppose it makes little difference.
 
The speed of transit between multiple cpus becomes less important the more units you have.
A thousand units with direct lines to each other running at low speed will be superior to two or four very fast units(250 or 500 times faster) working together, requiring just a couple very fast, very high bandwidth transit paths.
 
You must log in or register to reply here.
Back
Top