CELL Patents (J Kahle): APU, PU, DMAC, Cache interactions?

[j^aws]

You know that "PS3 architecture" image is fake, right?

, ...Well, it's never really been branded a fake since it was first posted on these forums by 'version' in this thread. And 'version' who posted it seemed a 'knowledgebale' PS2 developer, judging by the few posts he made. >>> before vanishing from the face of the Earth! Those Sony ninja's must've got him good! ...If you're still out there, please elaborate!

For any point that shows it's a fake, there are points to show it to be real! Anyway I could be completely wrong... it's quite easy to call it a fake but certain things just don't add up and point to the contrary!

E.g. from my earlier calc, were CPU=GPU=921.6 GFlops/Gops each by arbritarily choosing a 3600MHz & 800MHz CPU/GPU combination, a complete fluke :? ?

Or the 30 GPoly/Sec could well be 'micro polygon' figures for REYES. Or the 800 GB/sec eDRAM bandwidth being only around 20 times that of the PS2 48 GB/sec eDRAM bandwidth from 1999 etc. etc...

I was actually more interested in the two possible PixelEngine candidates from those two patents mentioned in that ppt presentation!

 

[3dcgi]

Whoever made that diagram you'd think they could add or draw better. 256 + 256 != 1024. I'm refering to the RDRAM. Plus, I'd be shocked to see 128 MB of RDRAM in a product priced near $300.

 

[one]

Whoever made that diagram you'd think they could add or draw better. 256 + 256 != 1024. I'm refering to the RDRAM. Plus, I'd be shocked to see 128 MB of RDRAM in a product priced near $300.

But it's the best bet done by experts that PS3 has 256MB (256 bit * 8 ) XDR DRAM. The launch price will be 39,800 yen ($380) when it debuts in Japan as it's the launch price of PS1 and PS2.

 

[j^aws]

Whoever made that diagram you'd think they could add or draw better. 256 + 256 != 1024. I'm refering to the RDRAM. Plus, I'd be shocked to see 128 MB of RDRAM in a product priced near $300.

Yeah...others have mentioned that the memory don't add up ! :? It's so blatently obvious that it doesn't even enter the realm of stupid mistake! Anyway others have mentioned that the 1024MB is a max figure and 512MB is the shipping figure (a memory upgrade via expansion port perhaps to 1024MB for the 'Home Server' version of the PS3?)

Also 128MB for a 2006 console doesn't sound much with PS2 shipping with 32MB of RDRAM disclosed in 1999...

...
But it's the best bet done by experts that PS3 has 256MB (256 bit * 8 ) XDR DRAM. The launch price will be 39,800 yen ($380) when it debuts in Japan as it's the launch price of PS1 and PS2.

Even 256MB in 2006 sounds pitifully low...

 

[one]

As for the clock speed, this old news in July 2003 claims that in 2005/2006 65nm Power 6 starts from 6Ghz It seems 3.6 Ghz is not aiming too high.

 

[j^aws]

As for the clock speed, this old news in July 2003 claims that in 2005/2006 65nm Power 6 starts from 6Ghz It seems 3.6 Ghz is not aiming too high.

Seems so... It's only aiming as high as the XeCPU rumoured 3.5 GHz+ ... In fact 800MHz for the GPU seems alot more ambitious than the CPUs against the rumoured 500Mhz+ XeGPU!

So all bases covered then!?

 

[Paul]

Well, it's never really been branded a fake since it was first posted on these forums by 'version' in this thread.

Actually someone made it and posted a thread about it way before version did, version simply found it and reposted it. The original thread is gone by now though.

 

[Vince]

People also don't pay attention to the fact that the team tasked with research into IBM's next generation ultra high frequency microarchitecture are people pulled from Cell, BOA, etc.

Nobody takes the time to do the background research on these projects, the people involved and what they attained. It's far easier to just say it's impossible and then go talk about how something as bland as Intrinsic's Fast14 is the best thing since Jesus and the invention of the phosphorescent condom.

Which is a shame since the - never discussed - IBM research is truely on an entirely different plane from the type of bullshit that get 5 page threads devoted to them in other forums here.

 

[MfA]

A 65 nm Cell processor with the same die size as the 250-nm EE running at 6 GHz? (Assuming most of it isnt eDRAM.) I would guestimate 250 Watt plus.

 

[Vince]

A 65 nm Cell processor with the same die size as the 250-nm EE running at 6 GHz? (Assuming most of it isnt eDRAM.) I would guestimate 250 Watt plus.

I do believe the 6GHz comments are in relation to Power6, which will be the second architecture designed with the 65nm node in mind and launch in 2006. As opposed to the Power5 base architecture (130nm).

<center><img src=http://enterprise.watch.impress.co.jp/cda/static/image/2004/06/22/0622ibm4.jpg height=225 width=300></center>

Ohh, and while I'm posting. I don't claim like some to have seen it, but I hear that Sony (alone, pre-STI aggrement, circa 1999) was discussing a 130nm IC with two RISC cores with 256KB L1 caches and Global Storages of 2x2MB being ~250M transistors in an IC of 17 mm sq, ~1K I/O pins, and running at 1.2GHz. Their 90nm plans called for 8 RISC cores, etc.

I question how much of a difference there is once you had IBM come on board in 2000. From what I've heard, of the 10 principles on Cell, only 2 were from Sony. You be the judge.

 

[PC-Engine]

A 65 nm Cell processor with the same die size as the 250-nm EE running at 6 GHz? (Assuming most of it isnt eDRAM.) I would guestimate 250 Watt plus.

Uh..MfA, haven't you heard of the 81GHz diamond transistor? It'll be in PS3 for sure.

 

[ultimate_end]

I question how much of a difference there is once you had IBM come on board in 2000. From what I've heard, of the 10 principles on Cell, only 2 were from Sony. You be the judge.

This is quite a point of debate isn't it? Especially so given the difficulty to track down hard info.
I always thought that Sony was the driving force behind cell, until I read an IBM patent for a multi-processor that pre-dates the Masakazu Suzuoki cell patent. It was more or less the same except that it only had only 5 APUs per PE. That was some time ago, so I can't remember the filing date, it may have even been 1999. Of course it doesn't mean that it wasn't produced out of some sort of collaboration between Sony and IBM.
It's also important to note that the Suzuoki patent fleshed out the idea quite significantly (software cells, sandboxes etc).
But then, IBM is attributed with most of the real design work (the real amount of input may never be known of course) and had to figure out how to actually implement it in silicon.

I have a feeling however that Sony's cell may not necessarily end up being the exact same as IBM's cell. Or not. But this is based on another IBM cell patent that I read (I believe it was to do with clock distribution) where it mentioned that the bus that connects the APUs does in fact run at Double Data Rate (as in DDR RAM), but that it also ran at only 1GHz when most of the rest of the chip runs at indeed 4GHz. It was quite an intriguing patent.
Now from what I have read (Ask Paul, he will point you in the right direction), Sony intends to use XDR and Redwood interface technology not only with the broadband engine, but also in the Broadband Engine! Now you can draw all of your own conclusions from that.

Sorry if you have heard all of this before. I'm new here.

 

[Vince]

I always thought that Sony was the driving force behind cell, until I read an IBM patent for a multi-processor that pre-dates the Masakazu Suzuoki cell patent. It was more or less the same except that it only had only 5 APUs per PE.

You're refering to a post-STI agreement patent that was filed in December of 2000 and granted in August of 2004. The actual patent, entitled, "Symmetric multi-processing system with attached processing units being able to access a shared memory without being structurally configured with an address translation mechanism."

It's filed by the following:

• Altman, Erik R.; Danbury, CT
• Capek, Peter G.; Ossining, NY
• Gschwind, Michael; Chappaqua, NY
• Hofstee, Harm Peter; Austin, TX
• Kahle, James Allan; Austin, TX
• Nair, Ravi; Briarcliff Manor, NY
• Sathaye, Sumedh Wasudeo; Lagrangeville, NY
• Wellman, John-David; Hopewell Junction, NY
• Suzuoki, Masakazu; Austin, TX
• Yamazaki, Takeshi; Austin, TX

The Patent references the Suzuoki patent as well as one by SCEs Yasuyuki Yamamoto on Data transfer method and device. The Suzuoki Patent then cross-refernces several IBM patents on DMA control, transfers and RPCs. As well as the second patent you commented on that concerns the Ring-Bus IIRC.

 

[one]

It was more or less the same except that it only had only 5 APUs per PE. That was some time ago, so I can't remember the filing date, it may have even been 1999. Of course it doesn't mean that it wasn't produced out of some sort of collaboration between Sony and IBM.

Isn't that just another multi-core CPU or a prototype of BlueGene? IBM released the dual-core Power4 in 2001, so they had plans before 1999. Also note that the Cell is almost natural evolution of Emotion Engine which was designed by Toshiba.

 

[ultimate_end]

I always thought that Sony was the driving force behind cell, until I read an IBM patent for a multi-processor that pre-dates the Masakazu Suzuoki cell patent. It was more or less the same except that it only had only 5 APUs per PE.

You're refering to a post-STI agreement patent that was filed in December of 2000 and granted in August of 2004. The actual patent, entitled, "Symmetric multi-processing system with attached processing units being able to access a shared memory without being structurally configured with an address translation mechanism."

It's filed by the following:

• Altman, Erik R.; Danbury, CT
• Capek, Peter G.; Ossining, NY
• Gschwind, Michael; Chappaqua, NY
• Hofstee, Harm Peter; Austin, TX
• Kahle, James Allan; Austin, TX
• Nair, Ravi; Briarcliff Manor, NY
• Sathaye, Sumedh Wasudeo; Lagrangeville, NY
• Wellman, John-David; Hopewell Junction, NY
• Suzuoki, Masakazu; Austin, TX
• Yamazaki, Takeshi; Austin, TX

The Patent references the Suzuoki patent as well as one by SCEs Yasuyuki Yamamoto on Data transfer method and device. The Suzuoki Patent then cross-refernces several IBM patents on DMA control, transfers and RPCs. As well as the second patent you commented on that concerns the Ring-Bus IIRC.

I think you may be right. Maybe. I read it quite a while ago.
I guess I should've said "IBM patent application". The only one I could find was patent application number 20020078308, which is I believe is the same as your issued patent, except that is doesn't mention Suzuoki etc , and doesn't reference any patents other than two that are blanked out, so I wasn't to know at the time.

If this is infact the one I read, then yes it was filed December 14 2000. Strangely, I could have sworn it was earlier than that. But at the time I was furiously reading through many many patents, so I'm probably very much mistaken .

As for the other patent: the ring-bus patent and the clock distribution patent aren't the same. Again, I am referring to patent applications here, so there may be an issued patent that is different. I guess I will have to do a lot more research on these patents.

But it certainly lends some credibility to the 4GHz cell claims doesn't it? Especially when you consider that Rambus technology or DDR may be integrated into the Broadband Engine, which will help the chip to run at high speeds by having the buses run slower (yet supporting excellent bandwidth).

-edit-
I found the 4GHZ/DDR patent application (or applications as it turns out) and and there actually is a resemblence to the ring-topology bus patent. These are the co-pending patents 20040078613 and 20040076189. The first goes into a lot of detail, but the second's diagram is the one that looks most like the ring-topology/APU diagram. Note that these patents only show 4 APUs.

 

[ultimate_end]

It was more or less the same except that it only had only 5 APUs per PE. That was some time ago, so I can't remember the filing date, it may have even been 1999. Of course it doesn't mean that it wasn't produced out of some sort of collaboration between Sony and IBM.

Isn't that just another multi-core CPU or a prototype of BlueGene?

Now Blue Gene/L is quite another beast entirely. Blue Gene/C on the other hand...

Now we all know that it was originally called cyclops, but could the C now possibly stand for "Cell"?
It makes a lot of sense when you are trying to build a PetaFLOPS supercomputer and you happen to have a bunch of "TeraFLOPS" chips lying around, doesn't it?

Also note that the Cell is almost a natural evolution of Emotion Engine which was designed by Toshiba.

Oh I don't doubt that Sony would have gone ahead with a massively parallel architecture, even without IBM's help. Kutaragi and Co absolutely love the idea. And there are some similarities in philosphy between cell and the EE.
I wouldn't say so much that Cell is the natural evolution of the EE, but rather PS3 will be a natural evolution of the EE. The EE's VU0 was originally intended to be a physics/animation co-processor. That was very important. Judging from Masakazu Suzuoki's Broadband engine/Visualizer combination, I would be willing to wager that Physics/animation processing capability bas been very much at the front of people's minds at Sony.

 

[nAo]

This is the last iteration of the main CELL patent AFAIK:
Resource dedication system and method for a computer architecture for broadband networks

Inventors:
KUNIMATSU ATSUSHI (JP); SUZUOKI MASAKAZU (JP); YAMAZAKI TAKESHI (JP); JOHNS CHARLES RAY (US); ASANO SHIGEHIRO (US); WATANABE YUKIO (US)

Assigned to:
SONY COMP ENTERTAINMENT INC (US); IBM (US); TOKYO SHIBAURA ELECTRIC CO (US)

That's the first time we see a CELL-related patent assigned to Sony, IBM and Toshiba at the same time..

ciao,
Marco

 

[cthellis42]

It's far easier to just say it's impossible and then go talk about how something as bland as Intrinsic's Fast14 is the best thing since Jesus and the invention of the phosphorescent condom.

I must protest! You are by FAR diminishing the impact and societal importance of the phosphorescent condom by bundling it with the likes of Jesus!

 

[qwerty2000]

Hey Jaws

It seems you know what you're talking about can you give me your estimate on ps3 specs and try be realistic as you can be please. Thanks in advance.

 

[Deepak]

Some old PS3 related articles...

PS3 Chip nears completion

Fast, Faster and IBM's PlayStation 3 Processor