We sez: PlayStation 3's RSX is 240mm² and bigger than the PC equivalent
- Thread starter Farid
- Start date
RSX reaback from SPE's LS?
Afaik RSX is connected to both RAM types, though XDR creates higher latency, which is covered by bigger caches according to latest news though...
For RSX to write data to XDR.
Nope, RSX does not have a direct connection to XDR, only to GDDR and Cell. To get to XDR, RSX needs to go through Cell.
I know from this example and some previous discussions that SPUs can initiate DMA write transfers to the RSX.
I would expect that it is possible for data to go in the other direction as well, from RSX to the SPUs local store.
After the ps3 craze, I began visiting this "place"...
So they say.
I thought Nvidia had a pretty good yields for the G71 when it was released at 90 nm as both the GT and GTX board had 24 pixel shaders, which shipped in volumes at launch, compared to the G70 where the GT board had one quad disabled. It almost gave the impression they introducxed some technology to improve the yield in the G71.
But I guess one possibility does not exclude another.
Shogmaster
Regular
I know I'm gonna get reamed for this (but I'm used to it 
) but couldn't it be possible that some parts of RSX is for aiding so called "software BC" that the later iterations of PS3 is going to employ without the EE+GS on the mobo and that's part of the reason RSX is so big to the PC equivilent in comparasin? It seems very unlikely that a pure software BC method (without going the painful 360 route for each game) is going to give you anything remotely close to 100% like Sony's been promising (especially since even the EE+GS on the mobo is not 100%).
In my crazy theory, I'm thinking that Sony already built in parts (or whole) of GS into the RSX, but knew that it won't quite get the BC software to utilize it ready in time for launch, so they designed rev 1 mobo with EE+GS plus the DRAM to do pure hardware BC in the mean time.
Commence with the reaming.
) but couldn't it be possible that some parts of RSX is for aiding so called "software BC" that the later iterations of PS3 is going to employ without the EE+GS on the mobo and that's part of the reason RSX is so big to the PC equivilent in comparasin? It seems very unlikely that a pure software BC method (without going the painful 360 route for each game) is going to give you anything remotely close to 100% like Sony's been promising (especially since even the EE+GS on the mobo is not 100%). In my crazy theory, I'm thinking that Sony already built in parts (or whole) of GS into the RSX, but knew that it won't quite get the BC software to utilize it ready in time for launch, so they designed rev 1 mobo with EE+GS plus the DRAM to do pure hardware BC in the mean time.
Commence with the reaming.
I guess that's still a possibilty.
You don't read B3D forums enough, if you call that theory "crazy".
It sounds like a perfectly valid possibility to me. Actually, just the other day, I was discussing about the possible integration of some GS parts in RSX with someone who might had knew something about that. He didn't seem to know, but he didn't rule out the idea.
In any case, as of now, I find it hard to imagine how they could get a 100% software BC solution up and rolling that has a high level a compatibility. I'm not calling it impossible, but it really strikes me an extremely complex task. That's why the integration of some GS elements in RSX, in the actual version or the next -in which case, the first gen PS3s wouldn't be able to support any of the probable image quality ameliorations, results of the RSX rendering-, does sound like an eventuality.
In any cases, it's a scenario that we (Well, I) can't rule out for now.
Shogmaster
Regular
I just remember getting tag teamed by faffy and some of the other Sony devs when I made this prediction months back so....
The EE + GS is a seperate chip on the motherboard. I doubt very much they would throw 2 EE+GS's for the hey of it.
Let's be fair though, for many people the main reason that theory got started was the notion that Sony would embed 4MB of extra DRam to RSX, and make it available for more then BC.Vysez said:
I think some kind of dynamic recompilation of GS display lists is more likely to be at work to get BC going though - ie. something that reschedules frequent state changes that used to be free on GS into order that is more RSX friendly (for some states like rendertarget swaps, this is "easy" because order on PS2 was a speed optimization that doesn't affect rendering results, for some others, it would be more tricky).
Whether this could work reliable without per-game domain knowledge I wouldn't know though.
Last edited by a moderator:
PCSX2 the major PS2 emulator on the scene has been working on something like that for the last few releases it still is buggy as hell and who knows what the compatibility of it will be but it gives a hell of a speed boost when it does work. So I guess it is a plausible theory whether they go that route or not.
We're not talking a whole PS2 chipset on RSX. We're talking some hardware aspects, such as...GS bytewise compatibility on the RSX graphics pipes or such.
The presence of the full EE+GS is because the emu isn't running yet, but when it is, the EE+GS comes out, and the hardware support in RSX (if it exists) kicks in.
TechOn measured the die themselves and put it at 258mm2 (16.2mm × 15.9mm).
http://techon.nikkeibp.co.jp/article/NEWS/20061124/124359/
P.s. Since it's mentioned in the B3D article... TechOn also measured Cell to be 228mm2 (19mm x 12mm), slightly smaller than the oft quoted value.
http://techon.nikkeibp.co.jp/article/NEWS/20061124/124359/
P.s. Since it's mentioned in the B3D article... TechOn also measured Cell to be 228mm2 (19mm x 12mm), slightly smaller than the oft quoted value.