Question about PS3s disabled SPE

[Xenus]

I have an interesting thought. Would it be possible for later when yields are better for the 8th SPE to be enable under Linux but not in games. It would seem like such a waste to not be able to use it for Linux apps if it was fully functional.

 

[Shifty Geezer]

I have an interesting thought. Would it be possible for later when yields are better for the 8th SPE to be enable under Linux but not in games. It would seem like such a waste to not be able to use it for Linux apps if it was fully functional.

That is a vague possibility, but it depends entirely on the way the SPE is disabled. If it's disabled in hardware, definitely not. If done in software, it would make sense for Sony to have it available in PS3 for tasks like Folding, in which case, depending how Linux goes, it may appear available there. Though I don't image they'd consider that until they're only using full 1:8 CBEs. You don't want customers paying the same amount for a console, and some getting a better machine than others when they run a Linux system evaluation app.

 

[Gubbi]

I think gholbine's point is that the physical difference may have an impact on code. The difference from one SPE to another when passing data will be at most one extra 'jump' in distance. Will that add latency or otherwise have a negative impact on code, when the dev writes for SPE0 to send data to SPE1 next door, but the data sent has to travel twice as far because SPE1 is knackered and SPE2 fills the role of SPE1?

That would add one EIB cycle of latency. If that breaks your code, your code is broken anyway.

Cheers

 

[PeterT]

Once (if that will ever happen) the yields/processes are good enough I don't expect them to produce cells with 8 SPUs and leave them all enabled. They'll probably just build a chip with only 7 on it in the first place.

 

[3dilettante]

Once (if that will ever happen) the yields/processes are good enough I don't expect them to produce cells with 8 SPUs and leave them all enabled. They'll probably just build a chip with only 7 on it in the first place.

That would just lead to CELLs with 6 active SPEs, or CELLs with 7 active SPEs and a big blank space in the chip.

It wouldn't save them any money or effort over keeping it as it is.

 

[Shifty Geezer]

That would just lead to CELLs with 6 active SPEs, or CELLs with 7 active SPEs and a big blank space in the chip.

Not if they redesigned the chip to fit 7 SPE's in. Of course where you get defects, you'd only have 6 SPEs, but the idea of the 1:7 Cell comes only once they get high yields, with the 1:6 defects making their way to other devices (if other devices ever get around to using them!)

 

[Npl]

Not if they redesigned the chip to fit 7 SPE's in. Of course where you get defects, you'd only have 6 SPEs, but the idea of the 1:7 Cell comes only once they get high yields, with the 1:6 defects making their way to other devices (if other devices ever get around to using them!)

A Cell with 7 SPEs would be complicated to cut. I havent seen a single chip with non-rectangular shape yet.

 

[Crossbar]

A Cell with 7 SPEs would be complicated to cut. I havent seen a single chip with non-rectangular shape yet.

Yeah, that would be an incredibly stupid thing to do.

I am sure they can reshuffle some PPE logic and cache memory if they want to make efficient use of the die. By saying that I am not assuming they will make an 7 SPE version, but if there is economy in doing it they probably will.

I am sure the EE+GS chip wasn't cheap to design but by the volumes the PS2 is still selling, it probably was a very good decision.

 

[Shifty Geezer]

A Cell with 7 SPEs would be complicated to cut. I havent seen a single chip with non-rectangular shape yet.

That's why I said 'if they redesigned the chip'. They'd need to shuffle things around to get a rectangular 1:7 Cell. Although thinking about the complexity of creating nanoscopic circuitry, you'd think a slightly irregular die wouldn't be the breaking deal...

 

[3dilettante]

That's why I said 'if they redesigned the chip'. They'd need to shuffle things around to get a rectangular 1:7 Cell. Although thinking about the complexity of creating nanoscopic circuitry, you'd think a slightly irregular die wouldn't be the breaking deal...

I see difficulty arranging a set of 7 identical rectangles into two rows with both rows being the same length, and it doesn't help to get too fancy with chip layouts.

Having irregular dies is an extreme no-no, given how manufacturing is done currently.
A diamond saw cuts straight across the wafer along the die boundaries.
Saws don't turn corners all that well, and irregular dies can leave wasted wafer space.

 

[Gubbi]

Not to mention the non-trivial work and cost associated with making a new layout, producing new masks etc. All to save 7% die area while losing a fair amount of fault protection.

Cheers

 

[Bohdy]

I'm just wondering, would it be at all possible/feasible for future games to detect and make use of an eighth SPE and/or extra redundant shaders on the RSX after the point where yields have increased to the point where most of the previously redundant units work?

 

[PeterT]

A very clear "No" on the SPU side, that would complicate development far too much for marginal benefits. The RSX quad is another story, that could perhaps be done transparently to the games (if they aren't too low-level optimized) but one has to wonder if the performance advantage would be all that without additional bandwidth. So, no again.

(This ignores the added factors of alienating early adopters, additional testing cost etc etc)

 

[3dilettante]

I'm just wondering, would it be at all possible/feasible for future games to detect and make use of an eighth SPE and/or extra redundant shaders on the RSX after the point where yields have increased to the point where most of the previously redundant units work?

Console programming is usually much closer to the hardware than it is in PC games. Sometimes even the smallest change can cause unexpected problems in software that isn't designed to handle them.

This might be more of an issue with the SPEs, since I think RSX has more abstracted from the programmer's view.

Activating extra shader units isn't guaranteed to help as much, since the rest of the system has already been balanced with the original number of shaders in mind.

Aside from all those concerns, Sony has an interest in not making the PS3 use 8 SPE CELLs.
They sell the fully capable chips for a lot more money to clients that want full performance, so making the commodity version equally as capable as their special edition chips will hurt their ability to create high-margin price segments.

Sony has a strong interest in making sure the extra units are never accessible to those who do not pay for the priviledge first, so it's likely those units are disabled in ways that are very difficult to change.