PS3 cell

[madcatmk3]

Any completly know about this thing and how powerful it really is?

[Áedán]

Define "Powerful". On data manipulation that requires multiple passes, it's very effective. If you mean how well does it suit general purpose computing, then the answer is not amazingly well.

[fantomfreq]

Powerful enough to decode 48 mpeg-2 streams simultaneously.

http://www.engadget.com/entry/1234000213041232/

[madcatmk3]

thats a muscle flex tesy I have no doubt they permade the cell to do that for a flx.
But by pwerful I mean wha type of prrocessor and is it more powerful then say an fx-57 or intel EE. Any chance the design will later be used for a general perpose PC?

[Gizmo]

You can't compare the Cell to an FX-57 or a P4EE. That's rather like asking which car is better, a VW Beetle or a Ferrari. It all depends on what your end goal is. The Cell is designed to fullfill a very different capability than a general purpose CPU like the FX-57 or P4EE.

[Áedán]

Originally Posted by madcatmk3

thats a muscle flex tesy I have no doubt they permade the cell to do that for a flx.
But by pwerful I mean wha type of prrocessor and is it more powerful then say an fx-57 or intel EE. Any chance the design will later be used for a general perpose PC?

Actually, the Cell was designed to do a whole bunch of things. It certainly wasn't specifically designed to decode MPEG2.

In terms of processor, think of it more like a G5 (Of Apple fame, but actually IBM's POWER architecture) core hooked up with 8 sets of digital signal processors around it. It isn't like any type of processor that's used in a PC. It is rather transputer-like in certain ways.

It doesn't suit general purpose computing very well. However, also remember that a high end GPU has significantly more processing power than an FX57 or an Intel EE, because a high end GPU *IS* a form of CPU. It just happens that the GPU doesn't suit general purpose computing very well either.

For certain types of operation, a Cell processor will run rings around the FX57/ EE. For other types of operation, the FX57/EE will run rings around the Cell.

[madcatmk3]

I know about the gpu my X-fi has the same horsepower as a p4 3.4 Ghz. And what I say with the mpegs was they perposely mad the one or the cell able to do that high number for bragging rights. Can you (to the best of your knowlage) explain the cell it's advantage and disadveantages compared to a PC CPU. Mabey that a bit much but I'm into concole gamming so I want to understand both types.

[Áedán]

Originally Posted by madcatmk3

I know about the gpu my X-fi has the same horsepower as a p4 3.4 Ghz.

~75% of the "horsepower" of the X-Fi is dedicated for converting audio at one sample rate to audio at another sample rate, and hence isn't generally available to do anything else. The requirement that the X-Fi does all that resampling is down to a design decision made by EMU many years ago. Since then Creative has been stuck with it. Other sound card manufacturers used a different design, and hence didn't need to do all the resampling. The Creative take on that is to take a disadvantage, and turn it into a marketing feature.

If you don't have a good grounding in processor architecture, it's not so easy to explain the difference between the Cell and a IA32/x86-64 processor. Do not consider the cell processor to be the domain of consoles! For certain types of operation (that many supercomputers are used for), the Cell architecture is far more ideal than a general purpose computer.

Originally Posted by madcatmk3

And what I say with the mpegs was they perposely mad the one or the cell able to do that high number for bragging rights.

No. As I stated earlier, the Cell was not specifically designed to decode MPEG data. It was specifically designed to handle certain types of task that involve performing the same basic instructions across a large amount of data. It so happens that MPEG (and JPEG, 3D, audio, F@H and SETI et al) are highly reliant on performing the same basic instructions on large amounts of data. Thus, the Cell processor is rather good at it. It's something known as SIMD (Single Instruction Multiple Data). Thus, the Cell's PowerPC core is freed up to do things like AI and general logic tasks, whilst those specalised data processing cores get on with the rest of the work such as generating the triangle setup needed for the 3D card and generating the audio.

[madcatmk3]

can you explain the cell anyway. Please and Never the less the X-fi still has that horse power even if most of it ins't useable(my 2 gb of RAM malkes up for that).
Thanks

[Áedán]

The best article for understanding the Cell processor is here.

My comments about the X-Fi were made simply because it's like having a 6L engine in a car, but 4 of those liters are reserved solely for winding down the windows.

[Áedán]

Creative claim that the X-Fi is capable of 1,200 MFLOPS (1200 million floating point operations per second). A 3.2GHz P4 is capable of around 6,000 MFLOPS.

One of the eight SPEs in the Cell processor is capable of around 32 GFLOPS (32,000 MFLOPS). That's a total of 256,000 MFLOPS not including the PowerPC core.

[madcatmk3]

thanks for the link aedan

[madcatmk3]

doe sthe cell support OOOE
http://en.wikipedia.org/wiki/Out-of-order_execution (incease like it was possible someone didn't know what I meant)

[Gizmo]

I'm sure the PowerPC core does. I'm also sure the vector units do not. Out-of-order execution doesn't even make any sense for the vector units, as the result of the next calculation depends on the result of the current calculation, so such considerations don't even have a place in the architecture.

[Phat Pat]

The downfall of the cell processor, as I understand it, is that all those cores share one tiny cashe.

[dsio]

Originally Posted by madcatmk3

doe sthe cell support OOOE
http://en.wikipedia.org/wiki/Out-of-order_execution (incease like it was possible someone didn't know what I meant)

To answer that:

RISC Strikes Back



The trend in CPUs over the last 15 years has been to increase performance by not only increasing the clock rate but also by increasing the Instructions Per Cycle. The designers have used the additional transistors each process shrink has brought to create increasingly sophisticated machines which can execute more and more instructions in one go by executing them Out-Of-Order (OOO), many modern desktop CPUs do this, exceptions are Transmeta, VIA and Intel’s Itaniums. OOO is something of a rarity in the embedded world as power consumption considerations don’t allow it.



The PPE is completely different however. It uses a very simple design and contains no OOO hardware, this is the complete opposite approach to IBM's last PowerPC core, the 970FX (aka G5).

No part of Cell supports OOO, neither the PPE nor the SPEs.

The whole point of the design is that its cheap to make, easy to design, and uncomplicated in nature. That doesnt mean programming for it is easy, but cheap, fast and easy is the whole point of a RISC architecture.

[Áedán]

OOO is both unnecessary and undesirable for the Cell processor. OOO allows the processor to reorder the instructions to allow it to fill more execution units at the same time. It's only really necessary in the PC world to overcome the fact that the instruction set used is badly suited to modern processors. It also helps to cover up the fact that code written to run on modern PCs is not well optimised.

As the Cell is a new processor, there is no backwards compatibility to worry about. The compiler determines the instruction order, and will re-order as necessary. This is far better than using OOO, as OOO has a very limited view of instructions in code. The compiler has a far better view of what instructions it needs to generate, and hence can optimise them far better. This is even the case on processors that do have OOO - Having the compiler doing the instruction optimisation results in far better performance than OOO itself can manage.

Originally Posted by Phat Pat

The downfall of the cell processor, as I understand it, is that all those cores share one tiny cashe.

That's not really true. The SPE units cannot execute code from main memory - they're not designed to. If they were, the design would never be able to fetch instructions from main memory or from the cache fast enough.

Instead, each SPE is entirely self contained. That means there is no need for cache, as the SPE has it's own local RAM to run from. The SPE is set up to run by the PowerPC core, and is then left to do it's own thing. If the SPE were in the PC world, it would be a single chip that contained a processor, RAM and chipset.

[madcatmk3]

The downfall of the cell is there melting. I heard one of the proposed souldtiosn wat Water cooling which adds more ot the price. I think if they do go with water cooling it because the xbox has an extremly loud cooling fan like a hiar dryer.

[Áedán]

I'm afraid that's just vicious rumour - along with the rumour that Sony fired all their PS3 engineers. Neither of which make any sense.

As far as the Cell processor goes - they already have Linux up and running on it, which rather squashes any rumours of the cell processors "melting". IBM will be shipping Cell based workstations (not consoles!) in volume in 2006. See developerWorks for the SDKs for developing linux software to run on Cell. Again, this is completely seperate to the PS3.

Toshiba will be shipping their developer systems for Cell based workstations in April 2006.

[Áedán]

I note that some of the climate modeller scientists are already looking very closely at the Cell processor. They're looking for an increase in computing power to the tune of several orders of magnitude to enable them to perform some of the modelling that they've not been able to. Why haven't they been able to? Lack of computing power - current supercomputers don't scale as well as they'd like them too!