Thermal issue : Will EE3 be able to survive the heat?

[DeadmeatGA]

Another issue surounding the design of EE3 beside the previously discussed logic transistor density is its intense heat generation enough to make coffee and toast.

As you maybe aware, current top-end Athlons and Pentium4s are burning upto 70~80 watts of power at the peak while Power4 and Itanium2 have shot past 130 watts, requiring systems to have sufficient cooling in form of massive heat sink and fan. Now, these designs don't even have the sheer number of logic transistors that the dual-CELL cores EE3 is expected to have. Imagine all those FPUs and PPCs sucking up power to run at 3 Ghz and you are looking at one hot chip, probably hotter than any of server chips on the market. Unless SCEI puts the biggest heat sink and fan on ther market, the EE3 is not going to survive the massive heat generation and burn up.

So how will Sony be able to handle the heat generation? Slow down the EE3? Use some kind of liquid cooling technology?

 

[Vince]

Another issue surounding the design of EE3 beside the previously discussed logic transistor density is its intense heat generation enough to make coffee and toast.

Good topic and I'm anxious to see what some of the people here think. My responce would be that doing an extrapolation based on current x86 architectures isn't going to work and will (as in your previous arguments in which this base was utilized for prediction) be fallicious.

As for the thermal issue, IMHO it's more an issue of power usage (with the thermal aspects being a subset). To this end we have several known solutions that are being utilized in the Cell architecture:

• Designs that are inheriently effecient. For example, the IBM patent on a hybrid Scalar/Vector construct I posted talked significantly of their desire to maximize energy and area effeciencies. Perhaps you shoudl glance over it.
• Process Technologies like Silicon-on-Insulator, Low-K dielectric, Cu. These technologies have very significant effeciency aspects that can't be ignored (as you did) when trying to do what you're doing. SOI alone is a highly integral factor when a processor is designed around it's 1.7X to 3X power savings.
• To another large extent are the inherent thermal/power aspects of the lower lithography size.
  
• And then, finally, the physical methods of extracting and channeling any thermal dissipation.

It's a PPC core, not an entire Power4 die shrunk on there - big difference. You keep resurrecting these wrong ideas and throwing them at us like we're going to believe it the next time.

EDIT: Two Cores? You shall have a rude awakening.

And for God sake, can't you conform to the common nomenclature? It's not PSX3, it's not the EE3.

 

[Josiah]

PS3 will be just like PS2 was when it was introduced. the chips will be huge, expensive, and have terrible yields. they'll gradually improve production until they can make a profit off the hardware and eventually integrate all the chips together (which probably won't be possible until ~2008) I just hope they can make enough of them before launch so that there won't be shortages like there was with PS2. I'm not going to pay $1,000 on ebay for a PS3!

 

[Megadrive1988]

According to Sony, the PS3 was ment to have EE3 / Emotion Engine 3
by 2006. this comes from EETimes and was published in Next Generation
in late 1999. I think the December 1999 NG. As well as other sources, online and offline.

EE3 / Emotion Engine 3 is just a PS3-specific version Cell architecture, In all likelihood that is. It was said that EE2 / Emotion Engine 2, which never showed up, would be an enhanced EE architecture, but EE3 and GS3 would be totally new architectures. obviously Cell.

In short, EE3 / Emotion Engine 3 = PS3's version of Cell for a CPU. And likewise, GS3 / Graphics Synthesizer 3 = PS3's version of Cell for a GPU.

Panajev, would you not agree?

 

[Vince]

Megadrive1988, I know where it comes from. I also know where it's wrong. Do you think I didn't know or would just post something for the hell of it that doesn't contain blatent sexual innuendo?

EDIT: You just posted (like within 30min) in another thread how the pre-PS2 roadmap that projected a scaling of the EE into niches it never evolved into is incorrect and nolonger appliable. And then you post this?

 

[Fafalada]

Easy, they'll ship with liquid hydrogen tanks

Btw DM, to date, the only Sony product that is to be called PSX, is not due for release at least another half a year (or more?).
There never was a PSX yet, there never was a PSX2, and if there ever will be a PSX3 I suspect it's still close to a decade away.

 

[dee]

Good topic and I'm anxious to see what some of the people here think. My responce would be that doing an extrapolation based on current x86 architectures isn't going to work and will (as in your previous arguments in which this base was utilized for prediction) be fallicious.

I'm not so sure about this, x86 seems at the moment to be leading the performance/power pack in laptop, desktop and workstation/server segments, via banais, p3-whatever (not sure about this actually) and opteron.

Regardless of this, would you accept that the IBM970 is reasonable power yardstick for the proposed PS3 core power spec - PowerPC core, powerful SIMD unit, top of the line manufacturing process. By all accounts I've read, this chip is burning a hefty amount of power, I've seen estimates of TDP on one aces/realworldtech/comp.arch (can't remember off hand which) being 90W.

 

[Grall]

I find it curious DMGA makes yet another "PSX3 (sic!) is D0000MED!" topic where he brings up heat as an issue after having seriously expected XB2 to feature a 6GHz tejas netburst-based CPU core just a few months ago. Maybe he still does, what do I know.

I mean, COME ON NOW! Are you really going to go there AGAIN? Like Vince said, current P4s aren't designed for, nor using either low-K or SOI. Nor do they use clock or power gating either, another set of techniques Cell might incorporate for all we know.

Cell in PS3 might feature as much as 4MB of SRAM and 64MB of eDRAM, and you're STILL going to maintain the core will have a high percentage of logic on it?

This is rediculous. Now I guess JVD's gonna keep complaining about all the PS3 topics swamping this forum while simultaneously NOT laying the blame for that squarely at your feet...


*G*

 

[Dural]

Btw DM, to date, the only Sony product that is to be called PSX, is not due for release at least another half a year (or more?).
There never was a PSX yet, there never was a PSX2, and if there ever will be a PSX3 I suspect it's still close to a decade away.

PSX was used by many way back when to abbreviate for PS, I believe it was an internal name Sony used before the offical Playstation name, not sure though. Many people also called PS2 PSY before it was given it's official name.

 

[BigGamer X]

^^^Yep. Some people have short memories. PSX was used in all of the game mags back in 1994 to describe Sony's upcoming console.

I wish people would stop this non sense about Cell vs. _____whatever CPU xbox2 will use. It's a non issue just like it is today with xbox vs. ps2. Unless Sony puts a REAL GPU in PS3, it'll most likely have similar performance to Nintendo's and MS's next machines, because they will have much better GPUs, you know, these little things that special in GRAPHICS.

 

[Paul]

I don't imagine heat to be a too big of a problem, you have it using SOI and a 0.65 micron process...

As far as the EE3/GS3 thing is concerned. PS3's chips could be named this, everything was part of a 3 part plan.

EE and GS - PS2

EE2 and GS2 - Which never really happened.

EE3 and GS3 - Sony claimed 1000X performance over the EE and GS with these chips, odd because that's what they claim with a Cell driven PS3. Could EE3 and GS3 simply be PS3's version of Cell? EE3 being the BE and GS3 being the Visualizers.

 

[Fafalada]

People always make up their own cute acronyms for things, doesn't mean they have any official basis though just because they are popular
For instance, the official acronyms for Nintendo consoles have always been GNC and AGB, but many people have trouble getting used to that.
The thing with DM though, is that he somehow manages to sound derogatory when using the said acronyms... :?

I wish people would stop this non sense about Cell vs. _____whatever CPU xbox2 will use. It's a non issue just like it is today with xbox vs. ps2. Unless Sony puts a REAL GPU in PS3, it'll most likely have similar performance to Nintendo's and MS's next machines, because they will have much better GPUs, you know, these little things that special in GRAPHICS.

General conzensus based on the Cell patent is that GPU will be cell based also.

 

[V3]

Slow down the EE3?

Yep, put those APUs to sleep, when they are finished doing their job.

Use some kind of liquid cooling technology?

I like that idea.

 

[Grall]

V3,

Sony used heatpipes in the first model(s?) of PS2. That's the kind of liquid cooling we'll see in a follow-up, I do not expect full-blown watercooling, that's too many parts that can fail. What if a seal breaks and the coolant gushes out, perhaps drowning an entire stack of hifi equipment for a customer? Would be ungood for Sony.

Besides, pumps wear out, and you still need a fan for the radiator... More moving parts, more cost, more everything, for little benefit.

*G*

 

[DeadmeatGA]

...

To Vince

My responce would be that doing an extrapolation based on current x86 architectures isn't going to work

Actually it should work; Pentium4s and Athlons aren't really X86 processors, afterall.

To another large extent are the inherent thermal/power aspects of the lower lithography size.

I am not so sure about this, because processors fabbed on smaller geomtry still burn up similar amount of power; the power consumption is proportional to the gate count and clockspeed than the transistor geometry.

EDIT: Two Cores? You shall have a rude awakening.

We will see.

Suzuoki Patent Filing

Server : 4 cores.
Workstation : 2 cores.
Visual Workstation : 1 core

To Faf

Btw DM, to date, the only Sony product that is to be called PSX

Officially no. Unofficially, the PSX existed ever since Nintendo dumped the PS(PlayStation) and Kutaragi swore his vengence.

To Grall

find it curious DMGA makes yet another "PSX3 (sic!) is D0000MED!" topic where he brings up heat as an issue after having seriously expected XB2 to feature a 6GHz tejas netburst-based CPU core just a few months ago.

Easy. P7(Pentium4) microarchitecture doesn't have as many logic transistors as EE3 to burn up power. Dual PPC cores and 18 VUs(16+2 spares according to patent filing) take hundreds of million logic transistors to build; that's more than 14 times the logic transistor count of Athlon.

Cell in PS3 might feature as much as 4MB of SRAM and 64MB of eDRAM, and you're STILL going to maintain the core will have a high percentage of logic on it?

You and I both know this isn't going to happen.

To V3

Yep, put those APUs to sleep, when they are finished doing their job.

What happens when games like GT6 would like to push PSX3 to its limit??? Run all VUs at full clock and then burn???

 

[Paul]

Server : 4 cores.
Workstation : 2 cores.
Visual Workstation : 1 core

You forgot about Figure 6, BE+4VS. It's never stated what Figure 6 was to be used for in the patent.

But looking at the pictures, a game console is the most likely use for it.

 

[Ozymandis]

People always make up their own cute acronyms for things, doesn't mean they have any official basis though just because they are popular
For instance, the official acronyms for Nintendo consoles have always been GNC and AGB, but many people have trouble getting used to that.

What does "GNC" stand for exactly? I assume AGB is "advanced GameBoy".

 

[Fox5]

Wasn't the Playstation called PSX by well....everywhere until the release of the redesigned PSOne?

BTW, Gamecube's abbreviation is GCN, Gamecube Nintendo, not GNC, Game Nintendo Cube, and also the name of some health supplement company.

 

[MfA]

I don't imagine heat to be a too big of a problem, you have it using SOI and a 0.65 micron process...

A .65 micron process doesnt help at all, smaller transitors ... but you have more of them, with more leakage (even with SOI) and a higher clock. On the whole if you try to get maximum performance out of a bit of silicon the smaller the process the more power you will be burning per area.

 

[Squeak]

So how will Sony be able to handle the heat generation? Slow down the EE3? Use some kind of liquid cooling technology?

Peltier pyramid.

In short, EE3 / Emotion Engine 3 = PS3's version of Cell for a CPU. And likewise, GS3 / Graphics Synthesizer 3 = PS3's version of Cell for a GPU.

Maybe it's time to stop thinking in such strict terms as "GPU" and "CPU"?
It looks like Cells ultimate goal, is to be a kind of "computational clay".

Obviously it would be best to have the whole system on one die, because bandwidth could be so much higher.
But of course it's impossible, with todays and the near futures technology, to get enough power for a system like PS3 on one die.
Therefore they have to use two dies.
One of these dies will have to have a rasterisiser and a framebuffer on it, but does that necessarily make it a GPU?
If the bus between the two chips were fast enough it wouldn’t matter how the tasks are being distributed between the two processors.
Of course this is purely a discussion of semantics, but nevertheless it is important to use the right words about the right things.