CELL Programming

[Frank]

For general office work, the main reason to buy a fast computer is because you want to run XP and .NET and such, because applications require that. But, for example, a large multinational around here still uses NT. So that can be done. (Although they are going to upgrade everything this year.)

Windows XP requires at least 256 MB and a reasonably fast processor. But Word, Excel, Powerpoint, SAP and all the other business applications don't. So, for office work, a cheap computer would do. Dual core and a fast videocard might become interesting when Longhorn introduces the new GUI, that needs a lot of processing power.

So, the main reason to buy a fast computer would be games, artwork and video encoding. And that's about it, and only if they support multiple threads. Which might just be the things Cell could do better than a generic multicore processor.

The power of a single PPU is more than sufficient to run everything else.

 

[jvd]

Diguru even people in the office will need faster pcs . As time goes on people will become better at multi tasking and it wont be uncommen for them to have a spread sheet up , word , a net browser and something else going and them wanting to be able to switch seemlessly between those programs. Right now you simply can't esp after a few hours on the pc

 

[Shifty Geezer]

Can't you?! :? I can. AthlonXP 2500, 1 GB RAM. I can open several browsers, have a paint app running, Word, small programming environment, and MP3 (or .ogg in my case) player, and still have plenty of resources spare. The only reason I upgraded from my PIII 800 was for raytracing - it would have sufficed very well (with a fast HD though).

I think the things that really slow 'development' down are image manipulation, music encoding, video decoding, and stuff that Cell ought to be good at. Oh, and gaming. My machine isn't the best gaming rig though lays GW well and i play consoles for games. Same next-gen.

I don't think any amount of technology is going to make wordprocessing or spreadsheets faster (databases, yes), nor web surfing. My only interest in upgrading my PC at a future date is for faster raytracing (can't EVER get enough power!!!) but as a small aspect of my use it dosen't warrant big shell-outs.

 

[scooby_dooby]

Diguru even people in the office will need faster pcs . As time goes on people will become better at multi tasking and it wont be uncommen for them to have a spread sheet up , word , a net browser and something else going and them wanting to be able to switch seemlessly between those programs. Right now you simply can't esp after a few hours on the pc

I dunno about that, for $400 CAD I can buila Sempron 2400+ system, with a GB of ram. That wil llet you run photoshop, dreamweaver, word, outlock, ws_ftp, msn and anything else, all open at the same time, all switching quickly.

DigiGuru is right on, Games, Artwork and Video, that's it.

Howver I don't think that will matter, I don't see CELL going anywhere in the PC marketplace, AMD and Intel are too established, there's no room for a 3rd guy.

 

[Carl B]

Yeah I really think we have to narrow the range we're talking about.

Cell PC's to me are nothing more than PS3's with hard drives; any additional functionality will be added by Sony through those means. Don't get me wrong, that could take off and become a huge vehicle for Linux, GUI depending, but you know - I'm just not seeing a Cell PC for sale at Best Buy anytime soon.

Other applications for Cell I think are in workstations and supercomputing and it's sub-sets (imaging for example, as recently mentioned by IBM).

Maybe home servers (and the whole tv and consumer electronics scene obviously), depending on what Sony and Toshiba do there.

 

[Shifty Geezer]

For Home Servers I can't see what advantage Cell would offer though. I'd have thought serving is one of those areas Cell is bad at. But really isn't that just hard-disk thrashing sending data out, so not too demanding?

I don't imagine Cell PC's are planned by anyone, and eyes will be on the PS3 HDD for performance and uptake. I'm extremely curious as to what appears softwarewise on the platform and whether it gains any strong applications, or if KK's suggestions are just dreaming.

 

[Carl B]

Well that home server idea was Kutaragi's recent suggestion in an interview; I wouldn't have otherwise have thought it. 2 6-SPE Cell chips or something. Yeah I wouldn't think it 'killer,' but since it seemed aimed at DVR functionality, maybe if it did that 'auto-upconvert' HD thing, people would find it worthwhile.

 

[Frank]

You would have to switch to Linux to use a Cell on your desktop, though. That will be the main problem, not a lack of power.

So, unless Linux offers a decent desktop replacement and the likes of IBM and Novell get the market for business software going, it probably isn't going to happen.

 

[Carl B]

Well right, which is why I definitely think that a *Cell PC/desktop* for all intents and purposes will always be a PS3. Now how robust a computer on the software side that ends up becoming will be interesting to see, btu besides PS3 and consumer electronics, I don't think the average person will ever have much interaction with Cell. Some companies might though.

 

[hey69]

if its something like AMIGA2 with CELL , count me in

 

[Laa-Yosh]

wouldnt Cell be very fast in 3dmax, maya and lightwave rendering?
i believe these are FPU hungry?

Indeed they are, but they also require at least 64 bit floating point precision for the geometry calculations. And Cell isn't much faster than Intel/AMD chips at that, it's strenght lies in its 32bit FPUs.
This is a compromise that suggests - at least to me - that Cell was never intended to run most of the scientific and highend professional applications, but rather it was always meant to be the CPU for a gaming console.

 

[one]

wouldnt Cell be very fast in 3dmax, maya and lightwave rendering?
i believe these are FPU hungry?

Indeed they are, but they also require at least 64 bit floating point precision for the geometry calculations. And Cell isn't much faster than Intel/AMD chips at that, it's strenght lies in its 32bit FPUs.
This is a compromise that suggests - at least to me - that Cell was never intended to run most of the scientific and highend professional applications, but rather it was always meant to be the CPU for a gaming console.

It has some uses in the field though
http://gear.ign.com/articles/615/615521p1.html?

The next demo was based on a new cloth simulation algorithm being worked into Maya. Again using two Cell processors, the demo was able to run 16 separate simulations simultaneously. Each piece of cloth was defined by 300 vertices, but the real kicker with this demo is that the algorithm incorporated self-intersecting physics, keeping the cloth from flowing through itself. This sort of simulation is much more computationally-intensive than simulating a cloth against another object.

 

[Laa-Yosh]

You can always fall back to lower precision, the question is, will it be good enough? Requirements - or maybe expectations - are usually very different between a console game and a high budget movie VFX shot.
Also, just how much RAM can a Cell CPU address? I've heard that it's memory controller is limited to 256MB, which is far from enough.

And Cell has to fight a very slow market here. It took ILM many years to replace their animators' SGI O2s with some Pentium4 machines, which turned out to be 5-10 times faster. Cell does not seem to offer such a big jump with FP64, so it may not be enough to replace their 1000-CPU Intel rendering farm.
The workstations might be a good market though, but then there's compatibility, software support etc. We'll see - I would certainly be happy to get a faster machine.

 

[Carl B]

Also, just how much RAM can a Cell CPU address? I've heard that it's memory controller is limited to 256MB, which is far from enough.

Common misconception, which I myself was under for quite some time.

http://www.realworldtech.com/page.cfm?ArticleID=RWT022805234129

Bottom half o' the page.

 

[Shifty Geezer]

It took ILM many years to replace their animators' SGI O2s with some Pentium4 machines, which turned out to be 5-10 times faster. Cell does not seem to offer such a big jump with FP64, so it may

What's the double precision performance of a P4/Athlon 64? Cell is looking at up to 30 GFlops which AFAIK is a significant gain, plus aspects of the pipeline could probably be run in single precision, such as procedural textures wich I imagine Cell to stomp all over.

 

[Laa-Yosh]

Athlon/Intel performance will probably be at least 25-50% per CPU at the time a Cell workstation can enter the market. Hop together two of them and you're at the same level, so price and software related issues would most likely be more important than the hardware itself.

Mixing precisions might be possible in a game engine pipeline, but I'd belive that a rather large and complex 3D application's developers would rather keep their code clean and upgradeable. There's some killing competion in this market with Lightwave getting pushed out by agressive pricing from Maya/XSI, and Max has never been too big in the movie VFX front. So they need an open, easy to expand architecture, which a tightly optimized solution probably couldn't be.

 

[Laa-Yosh]

xbdestroya: hm, so it can be redesigned to support more RAM, but it can't do it right now, correct?

 

[Npl]

xbdestroya: hm, so it can be redesigned to support more RAM, but it can't do it right now, correct?

If you need large Amounts of Ram you could at a Memory-Controller on the FlexIO-Side and add whatevr type of RAM and in whatever amount you like.

 

[Carl B]

xbdestroya: hm, so it can be redesigned to support more RAM, but it can't do it right now, correct?

Well, I think it can - just a matter of how the FlexIO is divied up bit-wise; so I think it can be done right now - but I think you're right that it's not being done right now. They'll probably give the memory controller some more work as well in future revisions, I would imagine.

 

[version]

"The Vector Floating-Point Unit in a Synergistic Processor Element of a CELL Processor":

http://web.axelero.hu/varga1973/synerg.pdf