XENON to have two PPC Cores?

[Panajev2001a]

The PPC cores that will be in Xbox are modified G5's, I thought that was well known. They have been modifed quite extensively apparently.

Uhm...

Name: David Wang (dwang@realworldtech.com) 2/7/05

Singh, S.R. (swaranNOrajSPAMsinghPLEASE@hotmail.com) on 2/7/05 wrote:
---------------------------
>David Wang (dwang@realworldtech.com) on 2/7/05 wrote:
>---------------------------
>>Today is the day that the CELL processor family gets announced.
>
>One noteworthy point from the question session today was that the
>streaming vector unit in CELL is about 20% dynamic circuits by area (not
>including the RAM arrays, I assume). The register file and forwarding macros heavily
>employ dynamic logic for performance, while the ILB uses dynamic
>circuits for the purpose of increased density.
>
>If anyone has any questions for the presenter of the first CELL paper (7.4),
>either David or myself would be glad to ask them during the author
>discussions this evening.

I just spent about 20 minutes with the gentleman who designed the dynamic logic gates for CELL, and he confirmed my suspecion that the PPE (powerPC processing Element) is based on a "new" 64b PPC core, and the lineage of the processor came out of the research papers presented in ISSCC in 2000. He also gave me his analysis/opinion of the scalability of the dynamic logic down to 45 and 65nm. (short summary: It's just more work.... If dynamic stops working, we'll go static, but keep dynamic for the speed/power/area as long as the process can do it.)

Interesting stuff.

This new processor core seems to have VERY similar requirements/features to the cores used by the Xbox 2/Xenon's CPU: 2-way in-order cores with hardware Multi-threading (SMT), very high clock speed, etc...

 

[Pugger]

This is a bit to technical for me Panajev2001a but could a modified G5 be described as "new" 64b PPC" ?

 

[Panajev2001a]

This is a bit to technical for me Panajev2001a but could a modified G5 be described as "new" 64b PPC" ?

Yes, but it would hardly come close to the frequency target they have barred major re-working of the chip which would present more troubles and more headaches than trying a more clean-sheet design (instead of taking the G5 and modifying it until it runs at those kinds of frequencies... yes, they could take the core... save some ALUs and stuff and redo everything else ).

The G5 is a very wide and fairly complex beast and this is not what they looked for to be the basis for this new PowerPC core IMHO.

 

[one]

A PPE in Cell, which is not derived from an older core, and a G5 (PowerPC 970) are only compatible by Power ISA + AltiVec.

confirmed my suspecion that the PPE (powerPC processing Element) is based on a "new" 64b PPC core, and the lineage of the processor came out of the research papers presented in ISSCC in 2000.

So, indeed, it matches well with what I guessed in the last year (here and here) about Sang Dhong, one of the core Cell architects, and his world-first 1GHz microprocessor

http://www.isscc.org/isscc/2000/ap/ap/Ssn_5.html

5.4 A 1GHz Single-Issue 64b PowerPC Processor 3:15 PM
H. Hofstee, N. Aoki, D. Boerstler, P. Coulman 1, S. Dhong, B. Flachs, N. Kojima, O. Kwon, K. Lee, D. Meltzer 2, K. Nowka, J. Park, J. Peter, S. Posluszny, M. Shapiro 3, J. Silberman 2, O. Takahashi,
B. Weinberger
IBM Austin Research Lab, Austin, TX
1IBM Server Development, Austin, TX
2IBM T. J. Watson Research Ctr., Yorktown Heights, NY
3IBM Microelectronics, Austin, TX

A single-issue short-pipe 1GHz PowerPC processor uses 0.22 um (0.12 um L eff), 6-level copper interconnect CMOS. The design uses nearly 100% dynamic circuits and includes 64kB caches, address translation hardware, and dual precision low-latency floating point unit.

This new processor core seems to have VERY similar requirements/features to the cores used by the Xbox 2/Xenon's CPU: 2-way in-order cores with hardware Multi-threading (SMT), very high clock speed, etc...

Though I don't think PPE in Cell and Xenon CPU core are the same thing because the former is pretty much unproven thing, is it in the leaked document that the Xenon CPU has in-order cores?

 

[one]

http://www.theinquirer.net/?article=20920

Xbox, Apple are wins for the Hypertransport Consortium

Where's that angel Gabriel when you need him?

WHILE STUMBLING through the thickets of the World Wide Wibble, we came across a Word document that contained something that we'd never seen before.

The Word document is a copy of Digital Video and Multimedia News, and it contains a summary, as you might expect, of news in the digital video sector. This particular issue covers February.

In its writeup of the Electronics Summit 2004, a conference for Electronics editors that we've never heard of, it mentions some of the design wins of different technologies. Number 7 on the list is Hypertransport, that well known technology used in AMD 64 chips and Power Macs. As well as IBM and AMD, who we know use the protocol in those systems, the Xbox is listed as a design win. No big shakes, of course, since we already know that it will use IBM for its processing.

But perhaps the most interesting confirmation is that Hypertransport has a design win for the Apple PowerBook. This would appear to be confirmation that a G5 PowerBook is definitely turning up and that despite thermal problems, it is at least in the works. We dug and dug, but couldn't find any other document pertaining to this win on the Wibblesome Wobble, so consider yourselves informed. µ

 

[archie4oz]

A PPE in Cell, which is not derived from an older core, and a G5 (PowerPC 970) are only compatible by Power ISA + AltiVec.

You mean PowerPC ISA..

 

[one]

A PPE in Cell, which is not derived from an older core, and a G5 (PowerPC 970) are only compatible by Power ISA + AltiVec.

You mean PowerPC ISA..

Anyway, "not derived from an older core" is a bit misleading, I should have written "not derived from an older core available on the market".

So a "leaked info" at an Apple news site about a newly developed embedded Power processor (65nm PPC3xx) is apparently a fake or not related to Cell (and the Xenon CPU).

 

[Guden Oden]

The PPC cores that will be in Xbox are modified G5's, I thought that was well known.

It's more like well-HOPED, amongst the internet fanbois... Dude, G5 is a HUGE chip. To hope for three of those, even modified ones, is completely unrealistic IMO. The G5 works in a very curious fashion with lots of oddball exceptions for how instructions may be issued together with what other instructions etc (read the Ars article on the subject), using THAT particular chip as a basis for an extremely cost-sensitive product like a console sounds really bad to me. Especially as MS voiced quite vocal dissatisfaction with the price of the original box in the past!

 

[DeanoC]

The PPC cores that will be in Xbox are modified G5's, I thought that was well known. They have been modifed quite extensively apparently.

People who follow Beyond3D forums have known for some time the Xenon CPU isn't based on G5's....

Do people listen to me, NO <grumble, grumble>

 

[pc999]

The PPC cores that will be in Xbox are modified G5's, I thought that was well known. They have been modifed quite extensively apparently.

People who follow Beyond3D forums have known for some time the Xenon CPU isn't based on G5's....

Do people listen to me, NO <grumble, grumble>

We listen, in fact we would like to listen much more , to bad that it is impossible , anyway dont worry because we listen very,very,very carefull

 

[Entropy]

The PPC cores that will be in Xbox are modified G5's, I thought that was well known.

It's more like well-HOPED, amongst the internet fanbois... Dude, G5 is a HUGE chip. To hope for three of those, even modified ones, is completely unrealistic IMO.

Uhm, not to disagree with the general gist of your post, but the G5 isn't huge by any stretch. It's 60 mm2, and that's with 512kB cache. Three of those cores, and with 1 MB of shared cache, would come in at roughly 150 mm2 or so. This first Cell type processor that has been shown off is still 50% bigger.

That said, it definitely seems a possibility that the xBox2 will use cores similar to the PE, in which case three of them is no problem in terms of die space, nor would that 3.5 GHz target be a problem even at modest power draw.

 

[Nite_Hawk]

The PPC cores that will be in Xbox are modified G5's, I thought that was well known.

It's more like well-HOPED, amongst the internet fanbois... Dude, G5 is a HUGE chip. To hope for three of those, even modified ones, is completely unrealistic IMO.

Uhm, not to disagree with the general gist of your post, but the G5 isn't huge by any stretch. It's 60 mm2, and that's with 512kB cache. Three of those cores, and with 1 MB of shared cache, would come in at roughly 150 mm2 or so. This first Cell type processor that has been shown off is still 50% bigger.

Not to pick nits, but the G5 on 90nm is a bit bigger (65.1mm2).

Nite_Hawk

 

[Entropy]

The PPC cores that will be in Xbox are modified G5's, I thought that was well known.

It's more like well-HOPED, amongst the internet fanbois... Dude, G5 is a HUGE chip. To hope for three of those, even modified ones, is completely unrealistic IMO.

Uhm, not to disagree with the general gist of your post, but the G5 isn't huge by any stretch. It's 60 mm2, and that's with 512kB cache. Three of those cores, and with 1 MB of shared cache, would come in at roughly 150 mm2 or so. This first Cell type processor that has been shown off is still 50% bigger.

Not to pick nits, but the G5 on 90nm is a bit bigger (65.1mm2).

Nite_Hawk

Not to pick nits?!?
I don't know where you got that number (but it seems to have more significant digits than mine, although mine is from IBM). Still, the 970fx is not just core+cache, there is chip area devoted to driving pins and whatnot that wouldn't be triplicated, so my seat of the pants 150 mm2 is probably still in the ballpark. In fact, I wouldn't be surprised if they could squeeze it a little bit tighter still. So three cores of roughly 970 complexity would be a possibility in terms of chip size. In terms of power draw at 3.5 GHz however, forget it, even taking into account that the XBox2 diagram where from more lithographically optimistic days.

Actually, given how tiny the PE is, Microsoft should be able to put more into the XBox2 CPU than just three of them + 1 MB of shared L2 cache. Particularly given Gates' comments about the power draw of the thing. But perhaps they opted to prioritize for manufacturability/cost.

The XBox2 CPU is still a bit of a mystery, but three cores of PE type would seem to best fit the rumours floating about.

 

[Nite_Hawk]

The PPC cores that will be in Xbox are modified G5's, I thought that was well known.

It's more like well-HOPED, amongst the internet fanbois... Dude, G5 is a HUGE chip. To hope for three of those, even modified ones, is completely unrealistic IMO.

Uhm, not to disagree with the general gist of your post, but the G5 isn't huge by any stretch. It's 60 mm2, and that's with 512kB cache. Three of those cores, and with 1 MB of shared cache, would come in at roughly 150 mm2 or so. This first Cell type processor that has been shown off is still 50% bigger.

Not to pick nits, but the G5 on 90nm is a bit bigger (65.1mm2).

Nite_Hawk

Not to pick nits?!?
I don't know where you got that number (but it seems to have more significant digits than mine, although mine is from IBM). Still, the 970fx is not just core+cache, there is chip area devoted to driving pins and whatnot that wouldn't be triplicated, so my seat of the pants 150 mm2 is probably still in the ballpark. In fact, I wouldn't be surprised if they could squeeze it a little bit tighter still. So three cores of roughly 970 complexity would be a possibility in terms of chip size. In terms of power draw at 3.5 GHz however, forget it, even taking into account that the XBox2 diagram where from more lithographically optimistic days.

Actually, given how tiny the PE is, Microsoft should be able to put more into the XBox2 CPU than just three of them + 1 MB of shared L2 cache. Particularly given Gates' comments about the power draw of the thing. But perhaps they opted to prioritize for manufacturability/cost.

The XBox2 CPU is still a bit of a mystery, but three cores of PE type would seem to best fit the rumours floating about.

That was a very long reply to a very simple post.

Btw, here's a link to where I got the size:

http://www-03.ibm.com/technology/power/newsletter/august2004/files/powerpc_device_quick_look.pdf

Nite_Hawk

 

[bbot]

Maybe it's time to reconsider an old idea about xbox2. Maybe it will have three dual-core processors.