Is there an article similar to "Xbox 360 Graphics Demystified", but for the PS3?

Antenox

Newcomer
As per the topic.

I'm really curious as to how the PlayStation 3's architecture works, and how its design choices help or hinder its performance.

I know there is a thread on PS3Forums that details the PS3's hardware, but the writer is a CS junior in college. I'm wondering if there's a more professional article that covers a similar scope?
 
Nope. But fingers crossed now B3D is under new management, they'll start covering more varied hardware including the consoles, which are very interesting in the different approaches. Although ATi were happy to answer Dave's questions about Xenos, whereas nVidia are hush about RSX. I don't know how much information will ever be made available. It's as though we have lots of info on XB360's GPU but little on the CPU, and much info on PS3's CPU but little on it's GPU, and won't get much more than that!
 
Nope. But fingers crossed now B3D is under new management, they'll start covering more varied hardware including the consoles, which are very interesting in the different approaches. Although ATi were happy to answer Dave's questions about Xenos, whereas nVidia are hush about RSX. I don't know how much information will ever be made available. It's as though we have lots of info on XB360's GPU but little on the CPU, and much info on PS3's CPU but little on it's GPU, and won't get much more than that!

There has been quite a bit on the PS3 GPU actually, especially the related architecture in G70. You just have to be paying attention ;) Dave mentioned a long time ago when asked about RSX if he would do a similar article and his reply was when G70 was launched we would see why that wouldn't necessarily be necessary. The papers I have seen would re-emphasize this and a lot of people in the know have stated pretty much just that.

From a PR perspective there is absolutely no reason for Sony to give out any further information on RSX. They won a ton of bullet points at E3 2005 and with software coming it isn't very relevant any more. I imagine we will see a scenario very similar to the GCN where even 5 years later we have to fish for information if we want direct confirmation. I would expect more stealth and choice comments by Sony / Sony developers in place of any articles where Sony offers up insider information. The former is more effective and allows them to control PR and information in a positive light, whereas the later creates a situation where people may begin to compare the finer details. Not that it is relevant, but but keeping things NDA'd and using the lack of information as a positive PR tool is a powerful "weapon" in a "PR war".

Anyhow, read a couple G70 reviews, account for a closed box system, lower API overhead, take out 8 ROPs and reduce v-memory bandwidth to 128bit @ 700MHz, reduce dedicated memory size to 256MB, and then account for XDR access and FlexIO and some cache changes and we are getting pretty warm in regards to RSX.

Just that much tells us more about RSX in theory /and/ in practice than we know about Xenos. There are a lot of things about Xenos we may never know because it won't be coming in its current form to desktop PCs. And I would venture a guess we will learn more about Xenon through CELL's PPE than anywhere else (which is currently the case).
 
There has been quite a bit on the PS3 GPU actually, especially the related architecture in G70. You just have to be paying attention ;) Dave mentioned a long time ago when asked about RSX if he would do a similar article and his reply was when G70 was launched we would see why that wouldn't necessarily be necessary. The papers I have seen would re-emphasize this and a lot of people in the know have stated pretty much just that.

From a PR perspective there is absolutely no reason for Sony to give out any further information on RSX. They won a ton of bullet points at E3 2005 and with software coming it isn't very relevant any more. I imagine we will see a scenario very similar to the GCN where even 5 years later we have to fish for information if we want direct confirmation. I would expect more stealth and choice comments by Sony / Sony developers in place of any articles where Sony offers up insider information. The former is more effective and allows them to control PR and information in a positive light, whereas the later creates a situation where people may begin to compare the finer details. Not that it is relevant, but but keeping things NDA'd and using the lack of information as a positive PR tool is a powerful "weapon" in a "PR war".

Anyhow, read a couple G70 reviews, account for a closed box system, lower API overhead, take out 8 ROPs and reduce v-memory bandwidth to 128bit @ 700MHz, reduce dedicated memory size to 256MB, and then account for XDR access and FlexIO and some cache changes and we are getting pretty warm in regards to RSX.

Just that much tells us more about RSX in theory /and/ in practice than we know about Xenos. There are a lot of things about Xenos we may never know because it won't be coming in its current form to desktop PCs. And I would venture a guess we will learn more about Xenon through CELL's PPE than anywhere else (which is currently the case).
I thought the RSX was updated to the G71?

RSX has access to the XDR, but does the Cell have access to the GDDR3 in some way?
 
RSX is mainly just a G71, so there's really far less need for such an article.

Xenos on the other hand, contains a lot of new concepts.

Just imagine where we'd be in regards to Xenos knowledge without Dave's article though, shudder.
________
Miss_Snow cam
 
Last edited by a moderator:
There has been quite a bit on the PS3 GPU actually, especially the related architecture in G70. You just have to be paying attention ;)
Of course I've been catching all that talk, but it doesn't tell me what makes the RSX an RSX instead of a G70, unless the only difference is the ROPs and BW and different memory. ;)

Anyhow, read a couple G70 reviews, account for a closed box system, lower API overhead, take out 8 ROPs and reduce v-memory bandwidth to 128bit @ 700MHz, reduce dedicated memory size to 256MB, and then account for XDR access and FlexIO and some cache changes and we are getting pretty warm in regards to RSX.
Hrm, how do I account for cache changes when I don't know what they are? Are they bigger? Smaller? Faster? Slower? Intelligent cache? Dumbed-down cache? Also if those aspects take us to 'pretty warm' what do we need to know to get 'hot' in our understanding of RSX? Enquiring minds want to know!

Just that much tells us more about RSX in theory /and/ in practice than we know about Xenos.
As a general architecture, yes, and that may well answer the OPs question, except if there are difference to the RSX's cache or other components that affect performance over if it were a G70, that info would need to be known to answer the question. As to what makes RSX different, what makes it a different chip to G70 rather than just the same chip with a different bit of print on the top, we don't know much at all, which is what I was alluding to. Some people might be happy to just think of it as a G70 in PS3, but some of us want to know all the little details that differentiates it (if any). Likewise we may know XeCPU is pretty much a PPE, but it'd be nice to know what makes it different. eg. Is the cache and branch logic/performance the same? What differences if any are there in the multithreading mechanisms? Even if it wouldn't make for a very big article, just listing differences from a reference point, it'd be nice to have some report on the RSX and XeCPU that does this.
 
In fairness, getting down to a level of cache behaviour etc. is very very low level, and lower level than the type of information "Xbox 360 Graphics Demystified" generally presented too.

I agree in most regards with Acert, in that people need to just look at what's actually been said. When Sony says that it is G70 or NV47 based, that does imply quite an amount of information that is already out there. Does it need repeating? I know it would be nice to know where exactly RSX differs, but where it differs is likely to be at pretty low levels. People seem to complain about not knowing much about RSX, but TBH, the only notable thing that we don't know too much about, at least on a reasonably high level, is the number of vertex shaders. We know about its architecture, even if some of the parameters may have changed like cache or buffer sizes, we know about its bandwidths (measured bandwidth, even, to XDR), we know its clockspeed, we know the number of pixel shaders, the number of ROPs..
 
Back
Top