So there's no way PS3 CPU is getting more than one PE~Cell?

Status
Not open for further replies.
nelg said:
Vysez said:
nelg said:
Then why would it not be better to have a R300 @ 2GHz instead of a R420@520MHz?
Maybe I don't understand your question, but I think ERP already answered that:
ERP said:
This ones relatively esay to answer....
It's cheaper to go wider than it is to increase clock. And as you mentioned graphics is so embarassingly parallel that it basically amounts to the same performance gain.
Can Intel and AMD clock so high due to better wafers or process tech. Intel seems to be able to produce >1GHz chips cost effectively. So what is it specifically about a 13nm GPU that prevents it from being clocked the same as a 13nm CPU?

Complexity, the fact that all transisters are not equal, and the layout of the chip. Are those valid reasons or am I mistaken. Why isn't a toyota corolla as fast as a viper?
 
basically think of it this way.

I have X dollars to spend to make the fastest graphics chip I can, I can go faster by adding more ALU's than I can by making "fast" ALU's. Which one do I pick.

The clock speed it a design decision, chips are designed to run at a particular clock speed (or rather in some range), you don't just make it and then work out how fast it runs.
 
nelg said:
Can Intel and AMD clock so high due to better wafers or process tech. Intel seems to be able to produce >1GHz chips cost effectively. So what is it specifically about a 13nm GPU that prevents it from being clocked the same as a 13nm CPU?
The main (and we might also say only) thing that prevent actual high end GPUs to go faster is their architectures.
Simply put, a "high-speed" architecture is created as such and asks for much more work "on each transistors" than an architecture that aims for reasonable clockspeed.
 
Besides design time, I would say power constraints and memory bandwidth are significant factors holding back GPU clocks.
 
nelg said:
(Serious question: why don't GPU ALUs clock much higher? I really don't know why we won't see 2GHz ALU in near future GPUs?)
Inquiring minds want to know. :)
Memory bandwidth is undoubtedly a factor as well. External DDR-type RAM scales poorly, maybe if they switched to XDR...
 
basically think of it this way.

I have X dollars to spend to make the fastest graphics chip I can, I can go faster by adding more ALU's than I can by making "fast" ALU's. Which one do I pick.

Which brings up a thought...

If Sony is going to 65NM with Cell for the PS3. That will allow for more geography of their Cell packageing; thus more performance. From this they could: 1. keep the 1PE model and add more SPU's possibly 10-16.

" Future versions of Cell chips could have more or fewer processing units depending on what device and software designers require, Kahle said. "There are a number of different ways to implement parallelism on the chip," he said. "

Source: cnet.com


Or 2. Keep the 1PE model increase clock and add fewer SPU's.

"The first version of the chip will run at speeds faster than 4GHz. Engineers were vague about how much faster, but reports from design partners say 4.6GHz is likely."

or 3. 2PE model.

This seems as the most likely scenerio, IMO.

Question: What would be the differnce IF you had a 1PE with 16 SPU's versus 2PE with 16 SPU's? Is there a point where the more SPU's you add you would need a PE ?
 
Mythos said:
Question: What would be the differnce IF you had a 1PE with 16 SPU's versus 2PE with 16 SPU's? Is there a point where the more SPU's you add you would need a PE ?

There sure is a point where those SPU`s do their work faster than 1 PE could feed all of them, but thats not the Prob.
The bigger issue IMHO is that the XBox2 CPU could (should) be more powerfull than 1 PE (assuming you dont use the SPUs). It will take time until a broad tool and library support for SPU`s is available, some tasks cant even be split upon the SPUs and then the PE also has to serve the SPUs. 2PEs would IHMO make sure there are no general-purpose bottlenecks comparing it to XBox2 (just think about Multiplattform Games).

I think 8SPUs are fine, but I still hope PS3 will have 2 PEs and 8-16 SPUs...
 
Mythos said:
basically think of it this way.

I have X dollars to spend to make the fastest graphics chip I can, I can go faster by adding more ALU's than I can by making "fast" ALU's. Which one do I pick.

Which brings up a thought...

If Sony is going to 65NM with Cell for the PS3. That will allow for more geography of their Cell packageing; thus more performance. From this they could: 1. keep the 1PE model and add more SPU's possibly 10-16.

" Future versions of Cell chips could have more or fewer processing units depending on what device and software designers require, Kahle said. "There are a number of different ways to implement parallelism on the chip," he said. "

Source: cnet.com


Or 2. Keep the 1PE model increase clock and add fewer SPU's.

"The first version of the chip will run at speeds faster than 4GHz. Engineers were vague about how much faster, but reports from design partners say 4.6GHz is likely."

or 3. 2PE model.

This seems as the most likely scenerio, IMO.

Question: What would be the differnce IF you had a 1PE with 16 SPU's versus 2PE with 16 SPU's? Is there a point where the more SPU's you add you would need a PE ?

you mean a point where the additional SPUs/SPEs would need a PU/PPE to give them orders, right?
 
Right.

Let me also clarify point 2. Keep the 1PE model increase clock and add fewer SPU's. [1PE model= 1:8, + higher clock, plus 2-4 SPU's.]

"The first version of the chip will run at speeds faster than 4GHz. Engineers were vague about how much faster, but reports from design partners say 4.6GHz is likely."

Source: cnet.com
 
ERP said:
basically think of it this way.

I have X dollars to spend to make the fastest graphics chip I can, I can go faster by adding more ALU's than I can by making "fast" ALU's. Which one do I pick.

The clock speed it a design decision, chips are designed to run at a particular clock speed (or rather in some range), you don't just make it and then work out how fast it runs.
Do you mean in engineering or fabrication?
 
Does it strike anyone else that this is probably all still up in the air....I'm sure the version of Cell they introduced at ISSCC 2005 is being used in early dev kits right now, and what makes it into the retail PS3 machine will be dependant upon their manufacturing capability.
 
Re: So there's no way PS3 CPU is getting more than one PE~Ce

Megadrive1988 said:
Is there no way that Playstation3 CPU or CPU-system is getting more than one of these 234 Million Transistor Processors ?

If Sony/Toshiba are far along on the "floating body cell" edram, they could use that instead of the 6-T SRAM. Under this scenario, the "CELL" based PS3 CPU may clock lower, but the saving of space in die area would make it more feasible to more than one PE. With all the SRAM a PE has, the advantage of using eDRAM without a capacitor would be huge.

I'm very optomistic that when Sony unveils their PS3 CPU, it will be very aggressive. They've got to much to loose by being timid at this point.
 
A 6 cycle load-use latency, at 4GHz, is <2ns access time into 256kbyte of SRAM. From what I heard, access time for a single bit of Z-RAM is already ~3ns. I don't think this is going to replace L1/L2 SRAM anytime soon, but for large L3 or embedded memory it sounds great.
 
What are the chances that the GPUs going into Xenon and Revolution will be ~500MHz but have double/triple pumped ALUs?
 
PC-Engine said:
What are the chances that the GPUs going into Xenon and Revolution will be ~500MHz but have double/triple pumped ALUs?

They could , but i think they will be higher than 500 mhz . I'm thinking most likely around 700mhz , mabye a 12x1 or 16x1 set up
 
jvd said:
They could , but i think they will be higher than 500 mhz . I'm thinking most likely around 700mhz , mabye a 12x1 or 16x1 set up

How do you come to this conclusion?
 
He thinks bigger numbers = always better like with cylinders in a car engine; he's american after all. They think like that.

I don't expect there to be an amazing amount of pixel pipes in console GPUs, it's not plain textured fillrate they need lots of when running at TV resolutions. Instead there'll be more emphasis on shading.
 
Guden Oden said:
it's not plain textured fillrate they need lots of when running at TV resolutions. Instead there'll be more emphasis on shading.
TV resolutions next gen means resolutions up to 1920x1080 and at the very least 1280x720.
 
cybamerc said:
Guden Oden said:
it's not plain textured fillrate they need lots of when running at TV resolutions. Instead there'll be more emphasis on shading.
TV resolutions next gen means resolutions up to 1920x1080 and at the very least 1280x720.

..Or 640x480 with industrial quantities of AA, like most people will play at.
 
If you're still using a 480p SDTV/EDTV in 2006-2011, you're a loser. :) 720p sets will exist at the sub-$1000 price point by then, and 1080p sets will be sub-$2000

IMHO, with HD-DVD and Blu-Ray supporting 1080p, and all major manufacturers moving to 1080p, designing a next-gen system to support anything less than 1080p framebuffers as standard is a big mistake.
 
Status
Not open for further replies.
Back
Top