Main CELL core is a MIPS design

[nondescript]

It would be very costly for Sony to put EE+GS@90nm on the PS3. Yes, its integrated and all that, but it's still 86mm2. That's bigger than a Pentium M at 90nm.

Besides, I don't think Sony has enough fab capacity as it is - Kutaragi is talking about moving PS2 production back up to the 150nm node, to free up 90nm capacity for PS3 production - so it hardly seems likely that he would add another 90nm chip unless absolutely necessary.

 

[Shifty Geezer]

Assuming EE emulation is straightforward on Cell, and eDRAM can be emulated through compression and total system BW, the only silicon needed for hardware emulation would be GS, which is only 15 million transistors right? And that could be squeezed onto RSX in place of say the video decoding hardware.

 

[PC-Engine]

It would be very costly for Sony to put EE+GS@90nm on the PS3. Yes, its integrated and all that, but it's still 86mm2. That's bigger than a Pentium M at 90nm.

Besides, I don't think Sony has enough fab capacity as it is - Kutaragi is talking about moving PS2 production back up to the 150nm node, to free up 90nm capacity for PS3 production - so it hardly seems likely that he would add another 90nm chip unless absolutely necessary.

Wouldn't that mean less PSTwos and more PS2s?

 

[Carl B]

Besides, I don't think Sony has enough fab capacity as it is - Kutaragi is talking about moving PS2 production back up to the 150nm node, to free up 90nm capacity for PS3 production - so it hardly seems likely that he would add another 90nm chip unless absolutely necessary.

Hey Nondescript, do you have a link or anything to where the move back to 150nm might be discussed? Hadn't heard about this and have to say it's pretty surprising/interesting. That would indicate a need for their fab build-outs beyond what was originally thought, if they're already feeling they will be supply constrained.

 

[one]

The M/B of the slim PS2 is designed to be able to use 90nm EE+GS or 110nm EE/GS. PSP is 90nm too, but the yield would be improved by now, and will be even better when they start the mass production of Cell and RSX.

 

[PC-Engine]

The M/B of the slim PS2 is designed to be able to use 90nm EE+GS or 110nm EE/GS.

If that's the case then why is KK talking about moving to 150nm? Why not just move it to 110nm?

PSP is 90nm too, but the yield would be improved by now, and will be even better when they start the mass production of Cell and RSX.

What does PSP yields have to do with using up fab capacity?

 

[Carl B]

If that's the case then why is KK talking about moving to 150nm? Why not just move it to 110nm?

Well, is he? If you've got the link to, go ahead and post it!

Everything else aside, I'm really really interested in this - it seems like such a large node to jump back to.

 

[PC-Engine]

If that's the case then why is KK talking about moving to 150nm? Why not just move it to 110nm?

Well, is he? If you've got the link to, go ahead and post it!

Everything else aside, I'm really really interested in this - it seems like such a large node to jump back to.

I'm just going off of what nondescript is saying.

 

[nondescript]

Hey Nondescript, do you have a link or anything to where the move back to 150nm might be discussed? Hadn't heard about this and have to say it's pretty surprising/interesting. That would indicate a need for their fab build-outs beyond what was originally thought, if they're already feeling they will be supply constrained.

It was mentioned in an interview with Kutaragi by PC Watch (One, I bet you've seen it, just don't remember it ). It was a while back. He didn't say that this was definitely going to happen - he only mentioned that this was a possibility, an option he would use if he needed it. (EDIT: He didn't mention 150nm specifically, he just said older process nodes. I assumed he meant 150nm, I forgot about 110nm.)

I'll get around to translating that article sometime (the relevant parts, anyways).

Wouldn't that mean less PSTwos and more PS2s?

Dunno. Probably. He didn't say.

 

[Carl B]

It was mentioned in an interview with Kutaragi by PC Watch (One, I bet you've seen it, just don't remember it ). It was a while back. He didn't say that this was definitely going to happen - he only mentioned that this was a possibility, an option he would use if he needed it.

I'll get around to translating that article sometime (the relevant parts, anyways).

Nondescript no big deal if you can't get around to it; I probably have read the interview, just blends in there with all the rest.

I have to say that regardless of what they do fab-side, I would be shocked if they would allow it to preclude the further production of the PSTwo - it's a superior money generator to the PS2, which just doesn't capture the consumers imagination in the same way any more with the PSTwo out there.

 

[nondescript]

Ok, here we go...it turns out my memory was off...

http://pc.watch.impress.co.jp/docs/2005/0617/kaigai191.htm[/url]]Currently, the PS2 chip is using a 90nm process, but if the PSP sells very well and we cannot produce sufficent volume, PS2 production can return to an older process. This (flexible allocation of manufacturing) is very important.

Whoops... Kutaragi was talking about PSP. But in theory, the same strategy could be used to meet PS3 demand, if the 90nm lines for PS2/PSP are the same as or similar to the 90nm lines for PS3. Since the interview was focused on PS3 production, I think this is a reasonable assumption to make.

 

[one]

PSP is 90nm too, but the yield would be improved by now, and will be even better when they start the mass production of Cell and RSX.

What does PSP yields have to do with using up fab capacity?

It's about the maturity of their 90nm lines. Though the die size of the PSP chip is not disclosed, I assume many potential problems were fixed in manufacturing it.
http://techon.nikkeibp.co.jp/english/NEWS_EN/20050426/104211/

I intend to stick to 90nm technology. I know some people who suggested going into mass production using such outrageous technologies as 65nm or 45nm, but you can't make something only by dreaming. In my management role, I'm determined to start mass production with the 90nm. Unlike in the past, it has become extremely difficult to establish a single process technology. What used to take six months after the delivery of manufacturing equipment now takes two or even three years. In other words, once technology uses fine scale processing, you now have to face ten or twenty new problems that bubble up along the way, where in the past, one action was enough. It takes strong conviction and confidence to move on. That is why a stable process technology is so important for the mass production of Cell. Whether we like it or not, we cannot slow down the adoption of fine scale processes in the future, and that is a fact. In this sense, the significance of "PSP" is enormous. Through mass production of PSP chips, Sony succeeded in stabilizing 90nm manufacturing technology. And with that, we were fully prepared-including the injection of new material - to take on new challenges for Cell.

 

[PC-Engine]

PSP is 90nm too, but the yield would be improved by now, and will be even better when they start the mass production of Cell and RSX.

What does PSP yields have to do with using up fab capacity?

It's about the maturity of their 90nm lines. Though the die size of the PSP chip is not disclosed, I assume many potential problems were fixed in manufacturing it.
http://techon.nikkeibp.co.jp/english/NEWS_EN/20050426/104211/

I intend to stick to 90nm technology. I know some people who suggested going into mass production using such outrageous technologies as 65nm or 45nm, but you can't make something only by dreaming. In my management role, I'm determined to start mass production with the 90nm. Unlike in the past, it has become extremely difficult to establish a single process technology. What used to take six months after the delivery of manufacturing equipment now takes two or even three years. In other words, once technology uses fine scale processing, you now have to face ten or twenty new problems that bubble up along the way, where in the past, one action was enough. It takes strong conviction and confidence to move on. That is why a stable process technology is so important for the mass production of Cell. Whether we like it or not, we cannot slow down the adoption of fine scale processes in the future, and that is a fact. In this sense, the significance of "PSP" is enormous. Through mass production of PSP chips, Sony succeeded in stabilizing 90nm manufacturing technology. And with that, we were fully prepared-including the injection of new material - to take on new challenges for Cell.

All that says is PSP helped for a more stable 90nm process for CELL. What does that have to do with 90nm fab capacity? Seems to me they're going to have PS3 shortages unless IBM helps out with fabbing CELL.

 

[one]

All that says is PSP helped for a more stable 90nm process for CELL. What does that have to do with 90nm fab capacity? Seems to me they're going to have PS3 shortages unless IBM helps out with fabbing CELL.

Line A and B can produce 100 X chips in each line a day. If the yield gets 100% better, Line A can produce 200 X chips. If you don't want more X chips, you can use Line B to produce Y chip. Simple as that.

Also, you seem to assume the sales/demand of PS2 will continue to grow in 2006 at the same pace as now, which I think is unlikely.

 

[Carl B]

Thanks for the link Nondescript! Ok well in that context everything makes a little more sense; and since One pointed out the PSTwo's motherboard is set up to accept a die size for the EE+GS up to 110nm, that's probably the node Kutaragi would drop back to if they needed further capacity for 90nm Cell chips.

Seems like their efficiencies have greatly improved in that regard, however - so that may not be necessary.

 

[Barbarian]

Assuming EE emulation is straightforward on Cell, and eDRAM can be emulated through compression and total system BW, the only silicon needed for hardware emulation would be GS, which is only 15 million transistors right? And that could be squeezed onto RSX in place of say the video decoding hardware.

I wouldn't qualify EE emulation as being straightforward on Cell. We are talking IOP+EE core+VU0+VU1 and a DMA engine. PS2 games build giant dma chains, how would an emulator patch those, realtime? Not to mention little endian, dual issue, multimidia instructions, mpeg decode logic, scratch pad. Even if we assume VU0 and VU1 can be emulated on 2 SPUs, Cell at 3.2 Ghz has only 10x advantage over EE. Can an emulator sustain this kind of speed?
Emulators are known to be branch heavy, and we know the PPU is not exactly strong in that department.

 

[one]

I wouldn't qualify EE emulation as being straightforward on Cell. We are talking IOP+EE core+VU0+VU1 and a DMA engine. PS2 games build giant dma chains, how would an emulator patch those, realtime? Not to mention little endian, dual issue, multimidia instructions, mpeg decode logic, scratch pad. Even if we assume VU0 and VU1 can be emulated on 2 SPUs, Cell at 3.2 Ghz has only 10x advantage over EE. Can an emulator sustain this kind of speed?
Emulators are known to be branch heavy, and we know the PPU is not exactly strong in that department.

Is static recompilation that difficult for EE to Cell? Cell is designed to work in deterministic, real-time environment.