Cell details (Nikkei Electronics)

[Mikage]

http://techon.nikkeibp.co.jp/article/HONSHI_LEAF/20050223/101976/
http://techon.nikkeibp.co.jp/article/TOPCOL_LEAF/20050224/102031/
http://techon.nikkeibp.co.jp/article/TOPCOL_LEAF/20050225/102098/

Nikkei Electronics magazine is the most famous technical magazine in Japan.
The magazine is doing a big feature on Cell at this issue.

I try to translate some sentence. (But I am poor at translation...)

The chip size was 185mm2@90nm by the original plan. But if Cell has
8*SPE, the chip size turned out to be too big. So the engineers considered
whether they reduced SPE to 6. But Ken Kutaragi definitely rejected
this proposal. He said "In the computer world, the involution of 2
is most beautiful. This is my aesthetics."

The first Cell chip completed in April 2004. Logical motion
was no problem and booting OS was successful.But there was
some problem at electricity. This problem at last settled
in the second half of 2004(by Mr. Suzuoki). The first chip
already motioned @5.2GHz.

If Cell has more SPEs(e.g. 24 or 36) on ring bus ,there can
be one solution that Cell increase ring bus plurally and connect
each ring bus with bridge circuit.(by Kahle)

Is 256KB(LS) small? By Suzuoki, "LS being divided 2 and using by
the composition of dubble buffer, I think there is practically
nothing problem. if one buffer is SIMD type and calculate 16 byte
per one cycle, calculation can end at about 4000 cycle. Date transfer
delay from main memory is several thousands. So during the time,
another buffer can retrieve.

whether each OS can have What percent of oeration resouces of PPE, it can be specified
individually. Each SPE is under cotrol of any OS that work on PPE. This virtual function
is not new. "When we planed Cell at the time, we thought this function very new. But now
this technology is not as new as we expected."

SPE plays a part that speed up specific task in a OS of several OS working on PPE.
So we planed so that many SPE can work separately and each SPE can concentrate on
one's own work.

Q:What percent is the achievement rate of Cell?
Kutaragi:It's a miracle. Miracle! we achieved more wonderfully than we expected.

Q:To realize a dream, real technolog also is necesary.
Kutaragi:I think that we are particular about 90nm technology as semiconductor
technology. I know that some people expect advanced technology as 45nm or 65nm.
But we can't make product only in a dream. I as a manager is particular about
a mass production @90nm.

 

[version]

yes 6 is bad, 8 16 or 32 good
any machine transate?

 

[nAo]

4?

 

[Panajev2001a]

4?

.

 

[Megadrive1988]

64 SPEs and 8 PPEs are good. I want my 'MNE' 8)

 

[j^aws]

...
The chip size was 185mm2@90nm by the original plan. But if Cell has
8*SPE, the chip size turned out to be too big. So the engineers considered
whether they reduced SPE to 6. But Ken Kutaragi definitely rejected
this proposal. He said "In the computer world, the involution of 2
is most beautiful. This is my aesthetics."
...

Welcome and thank you for the translation!

Can you determine what context is "2" that KK mentions, i.e. 2 CELLs?

 

[Mikage]

...
The chip size was 185mm2@90nm by the original plan. But if Cell has
8*SPE, the chip size turned out to be too big. So the engineers considered
whether they reduced SPE to 6. But Ken Kutaragi definitely rejected
this proposal. He said "In the computer world, the involution of 2
is most beautiful. This is my aesthetics."
...

Welcome and thank you for the translation!

Can you determine what context is "2" that KK mentions, i.e. 2 CELLs?

Sorry, my English is bad... 2 is the number 2. So the involution becomes
2, 4, 8, 16 etc....

 

[j^aws]

...
The chip size was 185mm2@90nm by the original plan. But if Cell has
8*SPE, the chip size turned out to be too big. So the engineers considered
whether they reduced SPE to 6. But Ken Kutaragi definitely rejected
this proposal. He said "In the computer world, the involution of 2
is most beautiful. This is my aesthetics."
...

Welcome and thank you for the translation!

Can you determine what context is "2" that KK mentions, i.e. 2 CELLs?

Sorry, my English is bad... 2 is the number 2. So the involution becomes
2, 4, 8, 16 etc....

That's okay...I understand now...Thanks.

 

[one]

Well, some more details from the magazine (paper edition) flied in... (don't quote me I don't have a copy)

4Ghz is the target clock from the start of the Cell project, calculated by a logarithmic graph as the clockspeed of 2005.

The mass production of the Cell processor is done in the 90nm process that could be matured through the PSP production rather than in the 65nm process. It's because you can't manufacture things only with dreams, according to Kutaragi

UMA is preferable when a Cell is coupled with a 3D graphics chip because of the cost reason.

For actual Cell products, hardware and software are developed in the facilities in Japan and the U.S. In the sanctuary, there is a development board the size of a chair cushion, with 2 Cell processors + air-cooled heatsinks. 3D graphics libraries for games and HDTV decoder software are being developed on this board.

 

[j^aws]

...
UMA is preferable when a Cell is coupled with a 3D graphics chip because of the cost reason.
...

UMA is a strong case for eDRAM, IMO!

 

[ERP]

...
UMA is preferable when a Cell is coupled with a 3D graphics chip because of the cost reason.
...

UMA is a strong case for eDRAM, IMO!

Not necessarilly, XBox was UMA and didn't have any

 

[j^aws]

...
UMA is preferable when a Cell is coupled with a 3D graphics chip because of the cost reason.
...

UMA is a strong case for eDRAM, IMO!

Not necessarilly, XBox was UMA and didn't have any

Yes, but I betchya wanted it too!

 

[psurge]

The disclosed incarnation of Cell (90nm) is a full custom design using dynamic logic. From what I understand, this means a significant amount of work is required for a process shrink. I wouldn't be at all surprised with a 90nm launch process for Cell.

 

[gamepower]

UMA is preferable when a Cell is coupled with a 3D graphics chip because of the cost reason.

quit with the blantent bashing of sony please

 

[Trawler]

Edited by moderator:



No more trolling from you Trawler, this is your last warning. Keep it up and you will be banned.

 

[akira888]

Uh huh... yeah... right...

nota bene: What does Kutaragi mean by an "involution" of 2? Obviously he's using the word in a different context than the meaning of which I know.

 

[one]

Uh huh... yeah... right...

nota bene: What does Kutaragi mean by an "involution" of 2? Obviously he's using the word in a different context than the meaning of which I know.

Kutaragi as a computer engineer is obsessed with 2^n or pow(2, n) or powers of two
8 SPE rathar than 6 SPE. Simple as that.

 

[StefanS]

Q:What percent is the achievement rate of Cell?
Kutaragi:It's a miracle. Miracle! we achieved more wonderfully than we expected.

Q:To realize a dream, real technolog also is necesary.
Kutaragi:I think that we are particular about 90nm technology as semiconductor
technology. I know that some people expect advanced technology as 45nm or 65nm.
But we can't make product only in a dream. I as a manager is particular about
a mass production @90nm.

So they better yields than expected? That's a big surprise! Plus initially Cell is on 90 nm.

 

[Acert93]

The disclosed incarnation of Cell (90nm) is a full custom design using dynamic logic. From what I understand, this means a significant amount of work is required for a process shrink. I wouldn't be at all surprised with a 90nm launch process for Cell.

Put your flame suit on A few here are pretty adament that the PS3 CELL is coming in at 65nm as Sony planned a long while back. Next thing you know people are going to be suggesting CELL wont have any eDRAM

 

[pahcman]

what happen to sony highly advanced 65nm tech? 90nm isnt what intel already have today? what happen to the ps3 cell was from scratch a 65nm design? why did kk say a dream??

i want edram!

is it true cell first born may have only 4 spe?