To "add" AltiVec should be no problem there. See GPul.

> To "add" AltiVec should be no problem there..

In case that's a response to my question above:
If minator is right and the devkits really have the PowerEN chip then they simply do not have VMX128 or standard AltiVec/VMX. How would it be added to the existing devkits? And assuming that was possible, how would the developers program for VMX128 as long as it has not been added to the devkits?

You'd probably not use it at this stage. I'd guess this is more about getting 16 core chips to developers to let them try using them. Handling all the cores is the difficult bit, vectorisation is relatively simple compared to that.

They had a similar situation before. The G5s they used for the 360 have Altivec but it's not the same as the version in the 360 itself. IIRC the version in the 360 has more registers and additional instructions.

> You'd probably not use it at this stage. I'd guess this is more about getting
> 16 core chips to developers to let them try using them. Handling all the cores
> is the difficult bit, vectorisation is relatively simple compared to that.

Agreed. That sounds reasonable. Let's wait and see if it's really the PowerEN chip in the devkits.

> They had a similar situation before. The G5s they used for the 360 have Altivec
> but it's not the same as the version in the 360 itself. IIRC the version in the 360
> has more registers and additional instructions.

Yes, VMX128 adds registers and certain instructions compared to standard AltiVec/VMX but also lacks certain standard instructions.

I suggest PowerEN because it's the only 16 core PPC I know of, but then I remembered the Blue Gene chips have 16 cores now as well. Both use the A2 core which sounds remarkably similar to the PPC core used in the 360 and Cell.

> I suggest PowerEN because it's the only 16 core PPC I know of, but
> then I remembered the Blue Gene chips have 16 cores now as well. Both
> use the A2 core which sounds remarkably similar to the PPC core used
> in the 360 and Cell.

Well, it could as well be a yet undisclosed chip different from PowerEN or Power BQC, but still be compound of 16 A2 cores. Or it could be a chip compound of 16 cores that are not A2 and which may be containing VMX128-compatible execution units, thus even be a sample chip of what's supposed to be used in the finalized console hardware.
If we take this current rumour serious, I guess the public information available (16 Power Architecture cores) is still much too sparse to hazard any educated guess of substance.

>Some people think that "16 cores" really means 16 threads, i.e. 4 cores with 4 threads each etc.

Or two main cores (SPEs) and 14 PPEs?

Andreas_Wolf,
Quote:

Could be, but those already exist and developing a new chip just for a dev kit makes no sense at all.

BTW here are some benchmarks:

Nvidia Tegra 3 Vs Qualcomm Snapdragon S4/

>> Some people think that "16 cores" really means 16 threads,
>> i.e. 4 cores with 4 threads each etc.

> Or two main cores (SPEs) and 14 PPEs?

Apart from the fact that you're confusing SPEs with PPEs, I very much doubt that Microsoft will use a chip that follows the Cell concept. Durango's (and its devkit's) CPU will most likely have symmetric cores, just like Xenon/XCGPU has.

> developing a new chip just for a dev kit makes no sense at all.

True, hence my suggestion of the devkit's chip being a "sample chip of what's supposed to be used in the finalized console hardware".

Andreas_Wolf,
Quote:

A PowerEN with AltiVec is impossible??
Remember Nintendos G3 with AltiVec?

> A PowerEN with AltiVec is impossible??

No, I said that the *existing* and *disclosed* PowerEN chips, which are compound of AltiVec-less A2 cores and are what minator explicitly referred to and what I replied to, do not have AltiVec. We were *not* talking about hypothetical future PowerEN chips, which may or may not be compound of AltiVec-enabled cores. The very reason that minator was suggesting the non-AltiVec PowerEN chip is that it already exists. It's just not possible to add AltiVec to already shipped devkits afterwards.
Apart from that, I wouldn't know why IBM should call a chip based on hypothetical AltiVec-enabled A2 cores and specially developed for Microsoft for use in a games console (and its devkit) "PowerEN" of all things, given that this name is already in use for a networking/server hybrid chip. It's not like any chip based on A2 cores is automatically a PowerEN chip (see the already mentioned Power BQC chip for instance).

> Remember Nintendos G3 with AltiVec?

No, I don't. Gekko doesn't have AltiVec.

https://morph.zone/modules/newbb_plus/viewtopic.php?forum=3&topic_id=7675&start=156
https://morph.zone/modules/newbb_plus/viewtopic.php?forum=3&topic_id=7794&start=1

16 symmetric cores?
No way, game code generally isn't well suited to parallel processing.

All this flailing around in the dark isn't terribly productive.

I believe its safe to assume that all the replacements for the leading game consoles will remain PPC based.
But so far, we've only received significant info on the Wii U.

More is known about the XBOX720's GPU then its CPU.

> All this flailing around in the dark isn't terribly productive.

Agreed, and it's everyone's own decision if he wishes to take part in it or just ignores it :-)

> I believe its safe to assume that all the replacements for the leading
> game consoles will remain PPC based.

Sure? The latest rumours on the PS4 (Orbis) say x86(-64).

Quote:

Sure?
- Characters can be updated separately.
- Physics can be done largely in parallel, collision detection might complicate things but even that can be divided up.
- Audio is separate.
- You can work on separate frames in parallel.


Desktop developers have been programming on serial machines for the last 30 years so apps including games reflect this. Now they have to be written in a different way.
This means looking at different ways to solve problems with different data structures and different algorithms.

All processors are heading this way so get used to it!

What'll be real fun is when people discover that cache coherence will hit a scaling limit so you'll have to do message passing between processors and this means relearning yet more.

Andreas_Wolf,
Quote:

I'd bet a kidney on it.
Sony has invested too heavily in the Cell to abandon it at this point.
Besides, what's the real advantage to 64bit X86 except memory addressing?
Does a game console need more then 4GB of memory?


[ Edited by Jim 12.04.2012 - 17:54 ]

>>> I believe its safe to assume that all the replacements for the leading
>>> game consoles will remain PPC based.

>> Sure? The latest rumours on the PS4 (Orbis) say x86(-64).

> I'd bet a kidney on it.

From "flailing around in the dark" to betting a kidney, that went pretty fast ;-)

> what's the real advantage to 64bit X86 except memory addressing?
> Does a game console need more then 4GB of memory?

I don't know, but the reason for 64-bit in hypothetical x86 consoles could be the same as for 64-bit in existing PPC consoles like the PS3 and the XBox360, don't you think?

Andreas_Wolf,
Quote:

Call it SIMD ... You name it ...

Remember GPul ...

> Remember GPul ...

I purposely ignored this reference the first time you gave it, but as you keep coming back with it, I'll react:
It's actually a good example, as the cut-down POWER4 with added AltiVec was not the server chip POWER4 anymore, but the desktop/workstation chip PPC970. So why do you think a cut-down PowerEN with added AltiVec or VMX128 would still be a PowerEN (remember, "EN" means "Edge of Network") and not something else with a new name, especially since what makes the PowerEN a PowerEN and differentiates it from other chips with the very same cores is what's on the chip *besides* the PPC cores, i.e. the on-chip peripherals that make the PowerEN a hybrid networking/server chip? And that's not even taking into account that adding AltiVec or VMX128 to what's currently the A2 core would make it something else than the A2 core.
To summarize: If you remove most of the on-chip peripherals from the PowerEN chip, you have no PowerEN chip anymore, but a new chip that happens to be based on the same A2 cores (refer to Power BQC for another chip that is based on A2 cores but is no PowerEN). And if you went even further adding AltiVec or VMX128 to what's known as the A2 core, it wouldn't be the A2 core anymore.
And all that removing and adding things from/to chips and cores wouldn't change anything in the devkits that had already been shipped before with, say, a PowerEN chip with AltiVec-less A2 cores, same way existing POWER4 machines didn't magically turn into PPC970 machines.
It may be, as minator suggests, that the Durango devkits have the existing PowerEN chip with AltiVec-less A2 cores (I don't believe that though). And it may very well be that IBM is about to take the A2 core, add an AltiVec or VMX128 unit to it, put several of those new cores together with some glue logic and on-chip peripherals suited for gaming consoles and voila, have the chip for Microsoft's new console. However, calling such new chip a 'PowerEN chip with added AltiVec/VMX128' is pure nonsense.

Andreas_Wolf,
Quote:

I like risk. ;)

Andreas_Wolf,
Quote:

Andreas, when did I said it is a PowerEN chip and when did I said it will be still called PowerEN?????

Yes, it could be a "variant of PowerEN" or a PowerPC chip with A2 core.

Mit dir zu diskutieren ist echt verschwendete Zeit ... Welcher Idiot wird auf die Idee kommen zwei verschiedene CPUs den selben Namen zu geben ....

> Andreas, when did I said it is a PowerEN chip and when did I said it will be
> still called PowerEN?????

Okay, you were not talking about calling it "PowerEN". Sorry for that, my bad. But as you just confirmed, you'd still call it a "variant of PowerEN", which I consider nonsense (see below).

> Yes, it could be a "variant of PowerEN"

No, if you strip off the on-chip peripherals that make the PowerEN the PowerEN (ignoring for once that this is not how chip development works anyway), i.e. a chip that is targeted at hybrid networking/server applications (that's what "EN" stands for after all), you do have neither a PowerEN nor a "variant of PowerEN" but just a bunch of A2 cores (which is what they really would start with btw). And if you then add AltiVec to those cores and some on-chip peripherals which make sense in a gaming console, you end up with a new chip that is as much a "variant of PowerEN" as it is a variant of Power BQC, i.e. not at all. Or do you really believe Microsoft would have any use for most of PowerEN's on-chip peripherals in a gaming console?

> or a PowerPC chip with A2 core.

Exactly, that's what it would be, as I've been saying all along. Whether A2 with AltiVec included would still be A2 is debatable, though, and would largely depend on how it was classified by IBM.

> Welcher Idiot wird auf die Idee kommen zwei verschiedene CPUs den selben
> Namen zu geben ....

Theoretisch wäre es doch kein Problem, einen neuen Chip zum Beispiel "PowerEN 720" (für XBox720) o.ä. zu nennen. Warum es aber Blödsinn ist, für einen Chip, dem mangels entsprechender On-Chip-Peripherie inhaltlich das "EN" fehlt, "PowerEN" auch nur als Namens-*Bestandteil* zu verwenden, oder ihn eine "Variante des PowerEN" zu nennen, nur weil er identische (A2) oder verwandte (A2+AltiVec) CPU-Kerne verwendet, habe ich schon mehrfach zu erklären versucht. Es käme doch auch keiner ernsthaft auf die Idee, einen solchen Konsolen-Chip nur wegen der Kernverwandtschaft als Variante des Supercomputer-Chips Power BQC zu bezeichnen, oder?

Interesting comments from the CTO of Sony's console division:

“We are looking at an architecture where the bulk of processing will still sit on the main board, with CPU and graphics added to by more digital signal processing and some configurable logic.”

Looks like it's going to be even more exotic than the PS3.

Sounds like a CPU + GPU combo surrounded by SPEs and an FPGA thrown in for good measure!

Full interview here.

For those of you thinking that SPUs are the answer for PS4 or whatever it's called, Apple's patent on Interleaved memory should be expired in late November 2016 thus making faster memories available for cheap. That's what I'm betting on: Unlimited burst lengths.