We sez: PlayStation 3's RSX is 240mm² and bigger than the PC equivalent

Farid

Artist formely known as Vysez
Veteran
Supporter
Read the whole article here
Some Belgian dude said:
Based on the RSX die shots released by websites following the PlayStation 3's launch in Japan yesterday, we have successfully estimated the chip's die size by comparing it to the nearby Samsung GDDR3 memory chips
We usually don't duplicate threads on the forum, but given that a lot of the Console Forum folks aren't visiting the PC forums ever, just as some PC folks don't browse the CF, I decided to start a thread about that short article in the Console Technology forum.
 
I would go to the PC forums, but talking about hardware that I will never be able to afford depresses me. I thought I was livin' large when someone gave me a GF4 MX to replace my GF2 MX...in 2003.
 
I would go to the PC forums, but talking about hardware that I will never be able to afford depresses me. I thought I was livin' large when someone gave me a GF4 MX to replace my GF2 MX...in 2003.
Heh heh. I know how you feel, but the PC forums, such as 3D Tech are great places if you want to hear about the latest techniques/technologies, and most importantly, you don't have to feel bad about being the only one unable to afford the latest piece of equipment because a lot of other folks on the PC forums are "broke enthusiasts". They have the enthusiasm, not the money. :LOL:

Actually, quite a few folks are running some antiquities, relics of ancient times when Shader Models were still limited in their instruction lenghts, these folks are commonly referred to as the braves. It's hard to talk about SM4.0, R600 memory scheduler and 40bits precision pipelines when your PC is rocking with a 9200, a FX5200 (with 512MB of RAM!!!) or a GF4MX, but they do it, nonetheless!
That's the true enthusiasm. ;)
So what are the extras?
Read the article, Uttar gives you quite some piece of info and add some interesting food for thought on the side.
Of course, I (or Uttar, but who cares about him?) could give you some more info, but where would be the fun with that?
 
So does this mean The RSX can use XRD just the same as GDDR for texturing, or close enough for it to be negligable.

Btw, is this little diagram right?

Cell --->20gb RSX
15gb<---

And would that mean that RSX has officially 40GB's of texturing bandwidth?
 
I thought they were the Penitentiary of B3D, where all those banned from the CF are sent for punishment and bettering their ways.

Me 2. It's the B3D gulag where people are being forced to read threads with other people droning on and on about..." There'll be 48 shaders in the upcoming nvRxxxxGSRTX ", "No, there'll be 64", "No, 48", "no, 64", "do you mean 48 dual (vec3/scalar) issue or 48 single issue", "64 dual yadda yadda" etc.... :)

Cheers
 
So does this mean The RSX can use XRD just the same as GDDR for texturing, or close enough for it to be negligable.

Btw, is this little diagram right?

Cell --->20gb RSX
15gb<---

And would that mean that RSX has officially 40GB's of texturing bandwidth?

Well it can be used for other things too, not just texturing. But yeah you could say that the 35GB/s link with Cell adds up to the VRAM bandwidth, which is nice.
 
Well it can be used for other things too, not just texturing. But yeah you could say that the 35GB/s link with Cell adds up to the VRAM bandwidth, which is nice.

Perhaps it should also be said that that RSX won't be getting 35GB/s over the Cell<->RSX link at any point unless it pulls a significant amount (or all but that's very unlikely) of bandwidth from XDR and then Cell generates a hefty amount of data procedurally. The XDR can only provide 25GB/s so Cell would then have to generate 10GB/s on top or rather Cell and RSX must communicate "something" to that amount of bandwidth (and more most likely as it is a bad thing that RSX regularly consumes all XDR bandwidth leaving Cell to starve.)

Personally Cell seems to sit best at the front and end of the graphic pipeline where it would do it's thing alone and send results of it's work at a lower bandwidth consumption rate. That is to say Cell can free up resources on the front end for RSX by handled geometry/particle systems/vertex work and then of course on the back end there is post-processing and element composition Cell could perform but it's difficult to identify opportunities where Cell and RSX can actively work together and would need to communicate such a large amount of data between one another for an extended period of time.
 
Actually, quite a few folks are running some antiquities, relics of ancient times when Shader Models were still limited in their instruction lenghts, these folks are commonly referred to as the braves. It's hard to talk about SM4.0, R600 memory scheduler and 40bits precision pipelines when your PC is rocking with a 9200, a FX5200 (with 512MB of RAM!!!) or a GF4MX, but they do it, nonetheless!
That's the true enthusiasm. ;)

that's me (but I have a ti4200, bettter than those crap cards :))
yay, true enthusiasm, or I'm a fool wasting his time. (more interesting to me than SM 4.0 though : G80's AA and texture filtering :cool: )

similarly..a hardcore gamer doesn't mean owning a new > $/€1000 PC :)
 
Perhaps it should also be said that that RSX won't be getting 35GB/s over the Cell<->RSX link at any point unless it pulls a significant amount (or all but that's very unlikely) of bandwidth from XDR and then Cell generates a hefty amount of data procedurally. The XDR can only provide 25GB/s so Cell would then have to generate 10GB/s on top or rather Cell and RSX must communicate "something" to that amount of bandwidth (and more most likely as it is a bad thing that RSX regularly consumes all XDR bandwidth leaving Cell to starve.)

The same thing can be said, but to a much greater degree of the XBox 360 - if Xenos uses up ram bandwith, then Xenon starves because of the unified architecture. The PS3 at least has separate simultaneously addressible XDR and GDDR ram. Also unlike Xenon or PC cpus, Cell's SPEs are designed and intended to run code and data from the local store, and well written SPE code will do exactly that. The need for Cell to use up external ram bandwidth is therefore much less than either Xenon or a PC cpu when running intensive computation.
 
The same thing can be said, but to a much greater degree of the XBox 360 - if Xenos uses up ram bandwith, then Xenon starves because of the unified architecture. The PS3 at least has separate simultaneously addressible XDR and GDDR ram. Also unlike Xenon or PC cpus, Cell's SPEs are designed and intended to run code and data from the local store, and well written SPE code will do exactly that. The need for Cell to use up external ram bandwidth is therefore much less than either Xenon or a PC cpu when running intensive computation.
I don't see what Cell or Xenos or PC CPU's have to do with this thread. Scifi was talking about BW for RSX.
 
I have a question about FlexIO between RSX and Cell... Can it operate 35 GB/s one way or the other, or just 20 one way and 15 the other?

And does the data from XDR travel through the EIB in Cell before going to RSX?
 
I have a question about FlexIO between RSX and Cell... Can it operate 35 GB/s one way or the other, or just 20 one way and 15 the other?

And does the data from XDR travel through the EIB in Cell before going to RSX?

bump
 
I have a question about FlexIO between RSX and Cell... Can it operate 35 GB/s one way or the other, or just 20 one way and 15 the other?

The latter.

And does the data from XDR travel through the EIB in Cell before going to RSX?

I think it depends on whether you are sending the data from Cell to RSX, or whether RSX is getting the data from XDR directly.
 
Back
Top