Cell Technology

[MBrown]

I've been reading up on it. It seems cool. I just want to talk about it.

This technology seems so awesome and powerful. It makes complete sense to me. I seems so much more efficient than just dual core. Reading more on it makes me wonder why is their even an effort to make dual core. The industry should just go strait into Cell technology. I think it may even be cheaper than Dual Core. Reading up on it is make me so anxious to get a PS3. But there is a problem I see with Cell. Some one who just uses a computer for word processing, school work and email will have no need for it. It would be over kill I think. Only the people who do heavy stuff like gaming and video stuff will see a benifit from it. But still I cant wait.

Just my little thought of it. You might think it sounds random or that it doesnt make any sense at all, but oh well What do you all think of it? What do you think will happen to it?

 

[imported_michaelpatrick33]

By the time it catches on there will be quad/eightway cores coming from AMD/Intel. Also, until I see hardware/applications taking full advantage of it I will keep the wait and see attitude about it. The PS3 has great specs on paper (but is rumored to be a monster to code for) but we will see two years after the PS3 release and developers master (or get close to mastering) the intricacies of the cell what it is capable of doing.

 

[coomar]

cell was planned as the next step after dual/quad/eight cores, it will be slow until all software is multi-threaded in the PC environment

 

[irwincur]

Hold it.

There is a reason IBM dropped the cell from their long term development plans. It is good at simple single-integer calculations (well that is all they have released at least so this is assumed) but terrible at the more complex calculations found in say a PC or server. In short, it is just not an efficient option. Hence, IBMs decision to drop it from a wider release.

As much as I would like to see it work, seems as if it will have a decent home in set top boxes and console video game systems. Glorious, no, but these computers are the bread and butter of the world today.

 

[MBrown]

Originally posted by: irwincur
Hold it.

There is a reason IBM dropped the cell from their long term development plans. It is good at simple single-integer calculations (well that is all they have released at least so this is assumed) but terrible at the more complex calculations found in say a PC or server. In short, it is just not an efficient option. Hence, IBMs decision to drop it from a wider release.

I didn't know that. That sucks because cell seems so awesome.

 

[Lonyo]

Originally posted by: MBrown

Originally posted by: irwincur
Hold it.

There is a reason IBM dropped the cell from their long term development plans. It is good at simple single-integer calculations (well that is all they have released at least so this is assumed) but terrible at the more complex calculations found in say a PC or server. In short, it is just not an efficient option. Hence, IBMs decision to drop it from a wider release.

I didn't know that. That sucks because cell seems so awesome.

Key word is *seems*, because it's not.
I guess you missed the Anandtech article on the XBox 360/PS3 CPU's and how they aren't so great for numerous reasons (it's been pulled,possibly to protect sources), but it was a rather scathing commentary on how useless the Cell and even the cores on the XBox 360 are.

 

[Gatt]

Cell seems awesome in theory, but just it's lack of OOE alone makes it impractical for anything but highly specialized, highly predictable use.

Without OOE code has to be ordered in such a way to minimize that hit, versus OOE processors which don't have to have that kind of care taken. Sure, once everything's in order Cell can probably kick butt. But when code isn't specifically ordered, it's going to be very slow. Not good, because it adds to the already monumental programming task now that we're talking Multi-threading. I'd at least like to trust that I don't have to order everything as well as think in several different threads at the same time. Alot more headache for a small gain.

Especially when you consider there's only so much you can share out to the secondary processors. 4 cores should be sufficient for any gaming task, and personally I'd prefer to have 4 full fledged cores than several cores with limited features.

I think Cell's going to turn out to be a Cro-Magnun-Multi-Core solution, something that almost was great, but just kind of missed the key to evolution.

Plus, I also think Sony's going to choke on it because it's got too many negatives from a development standpoint to make it a worthwhile system to develop for. Too complicated, Too limiting to a single platform, Too expensive to develop for. Develop for Cell, you only get Cell. Develop for X-box 360 or PC, you get something you can more readily port and reach a new large market for little cost.

 

[Furen]

Originally posted by: Lonyo
Key word is *seems*, because it's not.
I guess you missed the Anandtech article on the XBox 360/PS3 CPU's and how they aren't so great for numerous reasons (it's been pulled,possibly to protect sources), but it was a rather scathing commentary on how useless the Cell and even the cores on the XBox 360 are.

Ya, ya. The only saving grace on the Xbox 360 is that microsoft owns the patent on Procedural synthesis, so dont expect this to happen on a PS3 near you.

 

[dunno99]

Originally posted by: Furen
Ya, ya. The only saving grace on the Xbox 360 is that microsoft owns the patent on Procedural synthesis, so dont expect this to happen on a PS3 near you.

Actually, I don't know how much the patent can hold up in the courts. It's not the first time that the concept of procedural synthesis has come up. PS is pretty much the result of the proposed "geometry shader" back in WGF 2004 (or maybe earlier). I'm pretty sure this idea rose out of Academia first and percolated its way to MS. If Sony is to challenge MS on the patent, I think the patent will be the one thrown out.

And to respond to other posts, no, the Cell isn't all that great for general computing. Sure, it's a superb supercomputer on a chip type of thing, but think of it as a GPGPU (General Purpose GPU) with the "scatter" functionality built in, and some looser constraints. And except for applications that demand more flexibility/precision than a GPU can offer, the SPEs are outshine by the essentially n vector SPEs (for, say, the 360, that'd be a "48") and ridiculous memory bandwidth of a GPU. In general, think of the SPEs as something between a CPU and a GPU...yeah, that's about right.