A New Feature of the RSX has been REVEALED!

Status
Not open for further replies.

babcat

Regular
A NEW FEATURE OF THE RSX HAS BEEN REVEALED.

Although, it does not seem overly significant, it would be interesting to hear what the experts have to say.

Lets hope Sony gives us a FULL report of the RSX's features sometime very soon (at least before launch).

I love it when the PEOPLE get the TRUTH! Power to the CONSUMERS!


http://www.ps3land.com/article-839.php

In addition, when Floodgate is used in conjunction with the Gamebryo for PS3 renderer, it can take advantage of a cool feature of the RSX to verify completion of graphics data transfers. That avoids having the Cell wait for such transfers to complete before issuing draw calls. Gamebryo for PS3 already does this automatically when it renders particle systems, but it can be used all kinds of of dynamic or procedural data, whether it’s geometry or textures.
 
Seriously, nice catch, babcat. I don't know how much Nvidia (or ATI) chips use DMA on the PC platform to initiate data transfers (rather than tying up the CPU, and thus necessitating asynchronous notification of this kind), but it makes sense that the system would have some means for notifying processor elements in the Cell when a transfer has completed.

It'd be interesting to know whether this is using the PPU/SPU 'mailbox' notification that is a feature of the Cell, or whether this is something more in the way of a conventional interrupt. It'd also be interesting to know whether support for this notification is essentially a necessary component of implementing FlexIO support in RSX, or whether they went above and beyond a little bit to reduce any impact on Cell.

Good eye, in any case.
 
Sounds like it is able to perform DMA transfers and then signal an IRQ once the transfers are done. Such capabilities at the hardware level are not anything new and really quite mundane these days; what is new/interesting here is more that the software layers/drivers for the RSX actually seem to be exposing it.
 
Interestingly, it seems that this part

That avoids having the Cell wait for such transfers to complete before issuing draw calls.

could be interpreted two ways. The first is that they mean that the Cell would not have to 'busy wait', and can do other things while waiting for the DMA to complete (standard DMA/interrupt scenario).

This could also be interpreted literally, suggesting that the command protocol to RSX allows draw calls to be queued up someplace (XDR? GDDR3? internal caches?) and then automatically drained when the transfer is complete.

The Cell's XDR controller has some ability to enqueue transfers, I believe, but if this is really an RSX feature, something slightly more interesting may be going on.
 
It's going to be sad that everyone who could tell us more won't be able to answer due to NDAs.

Barbarian did tell us previously that the RSX has extra texture lookup logic, and maybe this is something somewhat related. Although the developer said that this function can be used for any kind of data and not just textures.

The whole key it seems to the PS3's power (even though the RSX is turning out to be significantly more than the PC NV47) is the CELL/RSX interaction and the ability for them to communicate as efficently as possible. I would guess that this is probably just one of a set of features to enhance the functionality of the CELL and RSX working together.
 
Yes, I would say so.

First of all, we have the extra large texture caches which are large enough to hold DXT1 compressed textures.

Secondly, we have the larger post lighting and transform vertex cache.

Third, we have extra texture lookup logic which this bit of information is probably associated with.

Fourth, we have *SEVERAL* developers and others bragging on the shading ability of the RSX. We have presentations about OPENGL ES and PSGL saying that the graphics of the PS3 are, "all about shaders!" Also, in one of those documents it is stated that in some of the software the NVIDIA shader paths are replaced with the RSX shader pathways or something similiar to that. The developer of "Untold Legends Dark Kingdom" said in his blog that the RSX has a lot of special features that he cannot talk about due to NDAs. If these same exact features were in the G70 then he could have talked about them. Additionally, we have the recent rumors here and there about the RSX *maybe* having an extra mini-ALU or "extra shading instructions." Basically, everything points to the RSX having at least some degree of extra shading clout.

Fifth, according to the human eye I see the PS3 pulling off some pretty stunning graphics.

I think according to all this and what we probably do NOT already know that all these little things add up to an ammount of extra performance that is AT LEAST not "insignificant" and could be said to be "significant." It's obviously not a huge giant leap and the RSX is *not* some mega super miracle chip. But it's something significant over the PC part.
 
Well, it's all and good if the RSX has extra aces up its sleeves, but I wouldn't trumpet stuff like this too loudly. These kind of features may well have been present in the NV4x architecture all along and simply suffering from poorly or non-existant exposure in the PC driverset due to PC API limitations or other such reasons. I'm sure at least some here remember ATi's supersampling AA or F-buffer and such that pretty much never actually became enabled in windows drivers for example.

This HAPPENS in the PC biz. It doesn't mean it's a new feature that is exclusive to RSX. Also, old features (such as interrupt signalling upon task completion) are often repackaged under a flashy marketroid-invented name or simply portrayed as a new invention simply to trick gullible (fan)people and make the product look more impressive. This has happened many many times in the 3D biz. Just look at any marketing sheet of a deltachrome GPU for many examples of that! :LOL:

So, let's not get carried away mmkay? Well, at least not without some kind of official confirmation...
 
Guden,

That's the problem. There is no official confirmation. Also, those working for the big gaming media websites and magazines are not even asking about the RSX. Sony is not commenting, NVIDIA is not talking, and on the rare occasion the three letters of RSX are spoken it's only in passing.

We have nothing to grasp but straws here. Some of what we already though we knew has even been removed from Sony's websites. Developers usually (except in rare cases in which heroic developers dare to give a *few* details) just say they can't speak about it due to NDAs which is of course understandable.

In the rare case someone who might know something about the RSX speaks about it's features we all need to jump on the information before it's put into the memory hole or deleted from the web. Heck, I saw one official next-gen game forum claim their policy was to delete any leaks of information about any console or game. Delete any information that could potentially be a leak of such information. And to also delete any information that was claimed to be a leak regardless if the information was already commonly available. They had to do this due to NDA and legal reasons.

Basically, those of us who are facinated with the RSX and are not game developers programming for the thing are very frustrated right now. The PS3 seems to be coming together very nicely in our minds and we know a good bit about the console and it's components. We all know about the memory types, the FlexIO is pretty strait forward, the CELL is well documented, and Blu-ray is obviously well understood. But when it comes to the GPU it's like there is a black stony silence.

I'm at the point every time I hear about the CELL processor which is talked about again, and again, and again in interviews about the PS3 or it's games I just want to vomit. Yes, it's interesting. Yes, the CELL is very powerful. But we keep on hearing the same thing five thousand times over. The CELL is used for physics, animation, sound, vertex work, and it's SPEs are very powerful. We know that by now. However, what we don't know is the full and complete nature of the RSX. Everytime I hear a developer go on and on about the CELL while just brushing past the RSX like it does not exist I'm appauled. I'm not angry at them of course, but the fact the secrecy is so tight they are terrified to talk about it due to the fact they could violate an NDA. So when someone is interviewed and gives even a tiny nugget of information about the RSX I am thrilled, because it's so rare.

This one feature of the RSX is probably nothing too amazing. It's probably a nice feature to have. And it's certainly possible that the G70/NV47 has the same feature. But how are we to know if Sony does not reveal the specifics about the RSX?

Quite frankly, I wish Sony would stop trying to hide this strong piece of hardware and act like they have a little bit of pride in the GPU of their latest console. They need to show it off and tell us all about it instead of trying to stick 100 layers of NDA's on top of it.
 
Can anyone translate to english for us mortals?

I agree, all I have understood its that rsx can have a 'verify' function on some data sent to mem/cell and that a guy is saying that rsx have additional ALU units because he think the games of ps3 are awesome anche because the devs for the sony platform are talking of 'shader power' (maybe because the ps2 have no shaders inside? think it over)
 
I agree, all I have understood its that rsx can have a 'verify' function on some data sent to mem/cell and that a guy is saying that rsx have additional ALU units because he think the games of ps3 are awesome anche because the devs for the sony platform are talking of 'shader power' (maybe because the ps2 have no shaders inside? think it over)

Also, doesn't the PS/2 have some sort of problem where streaming data from the VU to the GS can be a right pain to get right due to it not having exactly the sort of mechanism described above (or having one but it being buggy or badly designed)? In which case, PS/3 devs could just be jumping up and down singing Sony's praises for not hopelessly screwing up interrupt driven CPU -> GPU DMA transfers this time...
 
Last edited by a moderator:
First of all, we have the extra large texture caches which are large enough to hold DXT1 compressed textures.
So you are telling me that if I have a 2048x2048 DXT1 texture with all its mipmaps (2.66 MBytes worth of data) RSX can fit it all in its texture cache!! WOW! almost 3 megabytes of cache!
[note for the reader: I'm being sarcastic ;)]

Secondly, we have the larger post lighting and transform vertex cache.
where did you get this?

Fourth, we have *SEVERAL* developers and others bragging on the shading ability of the RSX. We have presentations about OPENGL ES and PSGL saying that the graphics of the PS3 are, "all about shaders!"
What would you expect them to say? welcome to the world of fixed function GPUs? :)

Additionally, we have the recent rumors here and there about the RSX *maybe* having an extra mini-ALU or "extra shading instructions.
Yeah.. anyone could come here, claim he has signed an NDA, breaking it trying to tell you something about RSX..and getting it wrong at the same time :)

Fifth, according to the human eye I see the PS3 pulling off some pretty stunning graphics.
This is what you should care about in the end. Would you buy a machine with amazing specs that produces shitty graphics? I don't think so.

BTW..basicly any GPU out there can transfer some data or perform some operation and then notify the CPU, welcome to 199x ;)
 
do you hear voices in your head telling you to poll the GPU? :)

Not in my head :LOL:

For instance, the reason the Folding guys use for not being able to run a CPU folding core at the same time as the GPU one is because it needs to keep polling the GPU. And I've read the PS2 emulator's (PCSX2) blog stating that a second core helps free up the cpu time that isn't used for polling.
 
For instance, the reason the Folding guys use for not being able to run a CPU folding core at the same time as the GPU one is because it needs to keep polling the GPU. And I've read the PS2 emulator's (PCSX2) blog stating that a second core helps free up the cpu time that isn't used for polling.
The problem is ..they are not writing drivers but applications.. ;)
 
Status
Not open for further replies.
Back
Top