It can sound silly but is possible (about the RSX)

version said:
g70 with 22GB/s membandwith a shit, i dont think sony do it
iam sure cell will be the vertexshader, and rsx only pixelshader

That certainly is a possibility... the Cell processor is quite capable as a geometry processor, but the question would then be is it more capable than the vertex pipelines in today's video cards. Removing the vertex pipelines from the NV50 would save about 10-20 million transistors.
 
"there would be an awful lot you would have to cut from the RSX to make room for that and maintain a 300 million transistor count"

:d) I always thought it was a 300 million + transistors...Never knew the exact number though...
 
rsx architecture ?
rey.JPG
 
scooby_dooby said:
They have not said they were working on RSX for 2 years. Their statement was vague for a reason.

Yeah, I appreciate that it was quite vague and I think that arronspink addressed its meaning quite well.

While I dont think that Nvidia have been designing RSX for (jeez where does the time go) nearly three years, I do believe that they were in effect a plan B from the outset. Not enough resources to dedicate a unique architecture (hence the G70 derivative) but enough time and prior thought to do better than an off the shelf 'quickie' if they were needed.
 
nVidia seemed to be pretty vague when they was talking about their products during the E3, and I was under the impression that the RSX was 300 million tranistors also. Then again... they was talking about their upcoming Geforce 7800GTX also. Who knows... we really won't know for sure until the RSX is publically known.
 
RSX was mentioned at the press confrence as having 300mil transistors, I dont have a link to prove that but I am sure you can find the press confrence vids somewhere around here but thats my $0.02 I got from the confrence when Nvidia spoke about the RSX.
 
expletive said:
Why? Maybe theyre just still working through kins in the 90mn process?

The same one that Sony has had in high volume production (and not ATI level production, but in all PlayStation2's, PSPs and EE powered CE devices) since fall of 2003? Perhaps.
 
Vince said:
The same one that Sony has had in high volume production (and not ATI level production, but in all PlayStation2's, PSPs and EE powered CE devices) since fall of 2003? Perhaps.

I doubt much of that would be relevant to rsx. You are talking about relatively miniscule parts.
 
AlphaWolf said:
I doubt much of that would be relevant to rsx. You are talking about relatively miniscule parts.

Well the point though is that Sony's 90nm process is mature, so they should have no issue bringing out these parts on the 90nm node. If they had aggresively gone for 65nm, well that might have heralded a problem in terms of yields. I think they learned from the GS disaster last time and are just going to a wait a little longer to go 65nm.
 
This may be a silly question, but would it be possible to change RSX's 128-bit memory interface to 256-bit, as it's in 7800GTX?
 
Last edited by a moderator:
london-boy said:
Why would they?

Double the memory bandwidth on the GDDR3 side. But I believe it was suggested before that a 256-bit interface could later present supply/cost issues. On the scale of potential "changes" from the announced E3 spec, I think it's a relatively reasonable one, but I still think really any change is very unlikely.
 
Titanio said:
Double the memory bandwidth on the GDDR3 side. But I believe it was suggested before that a 256-bit interface could later present supply/cost issues. On the scale of potential "changes" from the announced E3 spec, I think it's a relatively reasonable one, but I still think really any change is very unlikely.

I think it was agreed across the board that the RSX will be "something-else-limited" before it becomes bandwidth limited, therefore doubling the bandwidth to GDDR3 will not give you much in performance, except some cases. In the end it will cost too much for very little performance.

Or that's what people "in the know" said.
 
london-boy said:
I think it was agreed across the board that the RSX will be "something-else-limited" before it becomes bandwidth limited, therefore doubling the bandwidth to GDDR3 will not give you much in performance, except some cases. In the end it will cost too much for very little performance.

"something-else-limited" depends on the approach. I suppose you'd use a different approach of getting work done when you have double the bandwidth to memory to use.
 
Phil said:
"something-else-limited" depends on the approach. I suppose you'd use a different approach of getting work done when you have double the bandwidth to memory to use.

Of course. In the end you always work around what's there. But the 256bit bus is just expensive, so i guess devs will have to work around the fact that they'll have only so much bandwidth to work with.

I'd search the forum for posts about this but we can't search 3-letter words which makes this useless.
 
Back
Top