Does Cell Have Any Other Advantages Over XCPU Other Than FLOPS?

[Panajev2001a]

I think he had enough presence of mind not to actually say. Unless maybe he doesn't *like* working in the games industry any more

Wibble, he did not say... but you know somehow I have a hard time believing that SCE provided GCC beats out MS's compiler in code generation/optimization .

Take that as you will .

 

[Shifty Geezer]

That's actually a point I'm not clear on. The PS3 compiler is supplied by SCE right, and not IBM? Does it benefit from IBMs developments at all? How does it relate to the recently released Cell SDK? Seem very inefficient to have two companies working on seperate compilers for the same hardware, especially when Sony hasn't as much experience in that field (AFAIK).

 

[pc999]

Put a Kernel on the SPUs wouldnt use a lot of local memory?

 

[Titanio]

That's actually a point I'm not clear on. The PS3 compiler is supplied by SCE right, and not IBM? Does it benefit from IBMs developments at all? How does it relate to the recently released Cell SDK?

I believe the Cell port of GCC is solely in SCE's hands. The one that comes with the SDK is from SCE at least. XLC is IBM's stuff.

 

[Fafalada]

How does it relate to the recently released Cell SDK?

Well licensing issues aside, technically we COULD use the XLC at the moment - at least until Sony decides to put some clause into TRC that we're not allowed to use it.

Unless maybe he doesn't *like* working in the games industry any more :smile:

Well who knows - I have this impression that ever since they took him away from tinkering with lowlevel stuff he's been kinda grumpy.

 

[[maven]]

have you recently benchmarked the supernoisy thing?

Yep

That's not fear...

Yes, it is...

 

[ADEX]

Was ERP's implication that Xenon's cores are clock-for-clock faster than the PPE, or vice versa?

No idea but an XCPU should be faster than a PPE simply due to the larger cache.
It'll be especially faster on vector workloads due to the additional 96 registers in the VMX++ unit. Vector workloads are lot more common than people think e.g. memory copies / moves, simple look up tables etc.

I would guess memory latency on the XCPU will be higher but I don't know for sure, a lot of stuff is highly latency sensitive, bandwidth is lower of course.

There's different OSs and different compilers involved as well.

 

[mckmas8808]

One: Yes, like any other processor out there you need some kind of kick to start (amiga anyone? ), after that SPEs can mostly run without any PPE help.

Thankyou so we can finally put that one to rest.

 

[[maven]]

[XCPU]'ll be especially faster on vector workloads due to the additional 96 registers in the VMX++ unit. Vector workloads are lot more common than people think e.g. memory copies / moves, simple look up tables etc.

Huh? Are you talking about generating them or looking stuff up? I think the time of (simple) look up tables is over (see memory barrier). And table lookups (where you'd want to resolve each entry in your vector independently) are not suited at all for SIMD instructions...

 

[Fafalada]

And table lookups (where you'd want to resolve each entry in your vector independently) are not suited at all for SIMD instructions...

They could be suited perfectly fine if IBM didn't make us use their stoneage SIMDs.

 

[London Geezer]

They could be suited perfectly fine if IBM didn't make us use their stoneage SIMDs.

There were SIMDs in the stoneage?!!

 

[[maven]]

They could be suited perfectly fine if IBM didn't make us use their stoneage SIMDs.

Touché.

 

[Asher]

I think he had enough presence of mind not to actually say. Unless maybe he doesn't *like* working in the games industry any more

He did say if you can read between the lines.

The DD2 core is "closer" to the Xenon cores than the DD1 core.

As I've said before, Xenon and the PPE were based on the same processor but they've both had considerable work done on them, in different directions. Kind of like a Pentium 3 vs a Pentium M.

 

[version]

intel and amd will use cell architecture in about 2010, why cell, why not xenon???
ahhh waste my time.

 

[Ragemare]

If you say so.
I agree and never said otherise. SynpaticSignal was saying 'These devs say Cell is hard, so are are they crap programmers or is Cell hard?' My response was 'it's hard for people to adapt.' Once they know Cell, it won't be impossibly hard, but it will of course be harder than a hefty x86 OOE core. But it will of course be hard for a hefty OOE x86 to decompress 12 HD streams on the fly or process a 16 million element FFT quickly. The cost of greater performance is grater responsibility and effort needed by the devs.

Well I'm glad we got that sorted out. The problem is though, that it sounds like your saying once you get used to using Cell, it will be just as easy to code for, despite being more complex to code for. There are instances where a more complex architecture can be easier to use once mastered than a simpler one, but that doesn't apply when the programmer has to do more work in the complex architecture. There are of course exceptions to this, but I'm speaking in a general fasion

 

[Squeak]

I think DeanoC impllied in another thread that the Cell core had better multithreading, or am I mistaken?

 

[nAo]

I think DeanoC impllied in another thread that the Cell core had better multithreading, or am I mistaken?

I think you misunderstood Dean, what Dean meant is that...

 

[Shifty Geezer]

I think you misunderstood Dean, what Dean meant is that...

When you were a kid, did you used to fry ants with a magnifying glass, and pull the legs off spiders?

 

[nAo]

When you were a kid, did you used to fry ants with a magnifying glass, and pull the legs off spiders?

LOL, I used to put ants underwater (they can surprisely surive for a lot of time!) and pull legs off crickets!

 

[Shifty Geezer]

Well now I know what a drowning ant feels like