Slashdot Mirror
Cell Workstations in 2005
yerdaddie writes "The cell processor will be introduced in graphics workstations before release in the Playstation 3, according to press releases by IBM and Sony. As previously discussed, IBM will be releasing more details in February 2005. However, apparently prototype workstations have already been "powered-on" and will be available in 2005. Since Windows on PPC was scrapped back in 1997, this leads to speculation that perhaps Linux, AIX, or BSD will be the operating system for cell workstations."
Actually it sound more like parallel processing to me, where many CPUs are connected together to form one larger CPU. Perhaps you can remover CPUs from the network while active?? Or maybe it is just easier to expand. Their page seems to be full of hype (in my opinion), but no description of concrete benefits from this technology. Also why is this in the games section ... seem more like hardware to me.
Philosophy.
The article says that each chip is running its own kernel. That seems like a lot of wasted energy to me. I agree that it could give a serious boost to performance. However what about the memory requirements (RAM specifically)? It sees to me that each micro-kernel is going to need some RAM of its own, and to get the promised performance you would need many of these micro-kernels. This technology may end up more limited by memory requirements than the speed of the chips.
Philosophy.
You do realize that the whole reason that Transmeta's processor works well at all is because it's hopelessly optimized for emulating x86 instructions. And their software took years to write and it still is not 100% correct. (they still have some bugs in the x86 emulation) It's not going to be easy to do such a thing and at the end of the day what would be the advantage of running emulation at that level when you can just run a user level process to emulate a PIC, or ultrasparc or whatever you want?
I don't see the point of being able to boot into a random chip because you also have to emulate the entire computer, not just the cpu.
Even if you could emulate an ultrasparc cpu, you can't just throw it into a PC case and boot solaris, you have to use an actual SUN computer that has the right video, network and ide cards in it otherwise you'll have a broken machine. There are lots of little things that will cause the machine to break. The cpu is the heart of a computer, but it's not the only piece. They all have to fit together or it won't work. Just like you can't go and install a copy of OSX on a motherboard for the MorphOS (you can, but it's through an emulation layer, Mac on Linux) It's not at the kernel level.
The same processor powering cell phone, PDA, gaming handhel device, gaming console and general porpose workstation can be a way out of porting-emulators hell which is handheld development is for now. However there will be different OS for handhelds still probably - for examle Nokia unlikely drop Symbian in favor of Linux...
I think IBM, Toshiba and Sony eventually will license and sell Cell technology to those who are interessted. One of the core ideas is that they want to spead this technology as far at they can since every Cell based machine tap on the computational power from all other Cell based appliances in its vicinity. The more the merrier!
Cell isn't one processor, it's a class of processors. The one that will go into the workstation is more powerful than one that will fit into a PDA, or a HDTV. I think that IBM will make one workstation, and Sony will make another. They will use different boxes and logos but they probably will use a common "Cell based" lable yet unseen, just like "Intel Inside".
- Henrik
- when the Shadows descend -
XBOX2 will be based on parallelized PowerPC G5 processors. It is not Cell tech. Game dev studios are using PowerPC G5 macs to develop next gen titles for the XBOX.
The POWER train seems to be in full motion. No more wondering why IBM is canning its x86 desktop crap.
I infer this means a full shift into Power based architecture from IBM, they will only retain x86 server products because customers may want them, but they will not play a large role in their roadmap.
And that could be a Very Good Thing. The Power architecture is superior to all x86 implementations, including AMD64, in every way. The sooner we can break out into full uncrippled 64 bit computing the better.
That's likely *the* key to success of this architecture. As far as I can tell, it isn't really new in a fundamental sense, parallel/distributed architectures have been around for some time. What IS new, is that this would be the first time that a) this new architecture and b) associated computing potential, hits the mass market, getting into ordinary people's hands.
You're absolutely right here. The real problem is not building such machines, but easy/effectively programming them. I suspect the success of this platform will depend on how easy IBM/Sony will be able to make that. If it's an incredible machine, but hopelessly difficult to program, I'd guess it would be a flop. But if programming it turns out relatively easy, it could be a huge success, and start a whole new era.
We'll see, time will tell...
From http://en.wikipedia.org/wiki/Transputer
If the API is written well enough the developers will have little to do to take advantage of the Cell processor network, they will however have to intentionally break their code up into relocatable threads that get all their data through the API and not through calls to the disk or memory or whatever. However, this is the kind of power that would make games like Fable actually work as they were envisioned. Other Cells can be precomputing graphics, meshes, et cetera. They could be used for car AI, to allow cars in racing games to precompute a line to follow based on what they saw/experienced in the last lap. Esoterically, simulation games could use them to compute weather effects. I'm sure I'm forgetting plenty of possibilities. If they pulled it off properly, and DID implement a compatible Cell-powered PC, you'd also gain benefits from having other Sony crap in your house when using that platform, which could be a big motivator. If Sony came up with a Cell-powered PC running Linux and ported OpenMosix to their kernel/architecture then I'd pretty much be sold :)
"You're right," Fisheye says. "I should have set it on 'whip' or 'chop.'"
They're comparable mediums because both are moving towards being two things at once: computer generated, and photorealistic. Neither genre has yet achieved completion in both at once, but both are sneaking up on it.
Games are going to reach the point where in terms of visual quality you will not be able to tell them apart from movies. In general some types of camera angle will not work in games, while others will not really work in movies. However, you do sometimes see movies with scenes in the first person, so while Half-Life 2 isn't the game, there may one day be a shooter that could be a movie.
Video Games also have been getting higher and higher quality writing over time, though some argue that has already peaked in some of the more inventive and imaginative RPGs. As they gain more acceptance and thus greater exposure, more people have been able to become excited about video games specifically because they have become more polished.
I've played my way up through video game history (I'm just old enough to have been there for essentially all of it, so I guess there is something good about the time at which I was born) and I've definitely noticed video games and movies becoming more like one another. Most of the changes in movies (besides video games made from movies, only one of which have I ever bothered to watch) are subtle and I might be imagining most of them, but the changes in video games are pretty obvious and undeniable. Meanwhile, it is clear that video games are gaining wider acceptance and especially among groups into which they have traditionally had poor penetration.
"You're right," Fisheye says. "I should have set it on 'whip' or 'chop.'"
Well said. I think a lot of people watch ET and hear about the huge salaries and think, damn, all those actors are overpaid! What they don't realize is that about 1% of all actors make enough from acting to live on.
More to the point, it's not as if acting is the biggest expense on a movie. Most movies, the film stock alone costs more than most of the actors. When a film does have a huge actor salary, it's for a reason. The producers sign Julia Roberts for $20million because they know that her name alone will make them more than that at the box office. So bringing in a whole CGI team to replace the actors doesn't exactly sound like a cost-effective measure to me, since you'd need a few people plus a lot of equipment to do the work of one actor. I'm not even going to get into what acting actually involves and how you can't just program it, because if I have to argue that point with anyone, it's a waste of time.
These are console games, not PC. Players of console games have much bigger expectations than PC gamers as far as QC. If I play HalfLife 2, I might expect that there are some driver issues I would have to deal with, or problems if I don't have enough memory or hard disk space.
And you're completely wrong about level loading and the such. Console games run off disks. That means you cache the data in a preprocessed state in the exact order it will be read off the disk. When you game MUST load in 15 seconds (no exceptions or Sony refuses to publish your game), level loading is as completely optomized as it can be.
Where is the win with more than one processor? The bottleneck in level loading is not the processing time, its reading the data off the DVD, which on modern consoles is an asycnh process given to an IO processor with DMA. Generally the format read off the disc matches pretty closely whats in memory, because there isn't any type of LOD considerations, or considerations for how powerful the player's machine is. You can't increase texture levels because everyone displayes the same textures.
Usermods (mainly just maps), are fairly new in the cosole world. Right now I can only think of a few instances, and those are made with ingame editors and submitted to a common clearing house for approval by the publisher.
And like I stated, yes, you could offload AI, but what would be the poitn. Not graphics. The deadlines are too hard, and the bandwidth too intensive.
A problem with your optimism is this:
Let's say you use an 'outside CPU' to do AI computations. The game already had to do this as if you only had 'one' CPU (the PS3 itself), so it probably ended up using less than 20% of its CPU time (and probably a lot less than that) for AI. So you end up with an extra 20% of CPU time for better graphics - big deal. It's not worth the extra program complexity (and make no mistake - this capability will make games significantly more complicated).
All of this is ignoring whether the latency for network AI is even acceptable. It isn't for a lot of games. I have been playing a lot of the fighting game Dead or Alive 2 Ultimate lately. Many player actions in that game are measured in how many frames they take, with 60 total frames in each second. Many moves have periods of well less than 5 frames. The AI has to be able to respond at that kind of speed. The same is true for other popular genres like racing games.
I can't honestly think of any game task other than very non-responsive high-level AI (such as city traffic patterns or something) that would be suited to the latencies we are talking about. Maybe some types of physics calculations? That can be useful in some games, but you still go back to the problem that you aren't going to be able to realistically gain more than 30% or so of your CPU time back, without having game features in the first place that require more CPUs. That would be commercial suicide (at least in the forseeable future). Most of these high latency tasks just won't work using a 'low res' version.
There is no excellent beauty that hath not some strangeness in the proportion. -- Francis Bacon
I'm not being a luddite, I'm offering you a view from the marketing and technical standpoint of consoles. The console market is MUCH different from the PC market. While the PC is a very "open" platform, consoles are closed and done so by the manufactuers, who impose a very strict guideline on the content produced by publishers and developers for their systems. This means establishing a system of technical certifications every game must pass. These are things like loading times, stability, user interface, etc. All these guide games made for a platform towards a standard that the manufactuer wishes everyone to follow. And since they actually make the discs, you have no choice in the matter. So yes, I see that technically you COULD offload AI to enhance graphic performance, but like another poster replying to this message said, you have to do the game anyways at the same level without the extra help. Not doing so would risk fragmenting your market. Console players do not want to have to worry about having installed two extra CPU units, they just want to play the game on the back of the box. They aren't used to not getting a uniform experience. The dynamic LOD point was a case for "some platforms use lower poly models, some can perform with the full poly versions" Nothing to do with a console using dynamic LOD, and many games do. I was saying that because you have a standard platform, you can optomize all your data once for a specific set of hardware and be done with it. Even user created content can be preprocessed exactly as the memory and hardware would expect it to be by the creating console or publisher. The console is NOT a PC. The hardware remains consistant because that's one of its strong selling points - award winning and excellent games still come out 5 years after you've plopped down $300 for the system, unlike PCs where 3 years later you can't play any new games. Adding an upgrade curve back into the equation will only serve to fragment your market, and that means less dollars, and THAT my friend is why you won't see it happen....the bottom line, and that's it.