Dual-Core CPU Opens Door To 1080p On Smartphones

An anonymous reader writes "Following Qualcomm, Samsung is also close to launching a new smartphone processor with two cores. Based on ARM architecture, the new Orion processor promises five times the graphics performance of current chips and to enable 1080p video recording and playback. Next year, it seems, dual-core smart phones will be all the rage. Apple, which is generally believed to have the most capable processor in the market today, may be under pressure to roll out a dual-core iPhone next year as well."

Comments

Okay then

[Jbcarpen]

My parallel programming professor likes to harp on the fact that nearly all new computers in the future will be multicore. Apparently he's right.

Re:Okay then

[ledow]

The GP2X is a dual 200Mhz processor handheld console that runs off 2 AA batteries and was released in 2005. It was not that revolutionary, it was not that expensive, it was not that difficult to program. You can program on both chips so long as you don't mind temporarily ditching the in-built MPEG-decoding firmware that runs on one of them (it's replaced on each bootup) or being very careful where you tread. The Nintendo DS was multiprocessor. And not in the unconventional sense of "one CPU and a handful of support chips", two real, full, general purpose, programmable chips. Most modern games consoles have more than one real chip, many of them using multiple specialised processors (e.g. IBM's Cell in the PS3 which is a "9-chip" component).

Massive parallelism has been around forever and it's in consumer electronics already and has been for quite a while. It might not be *designed* like that but you have always had home computers with multiple processors that can be programmed to operate in parallel. There were people misusing floppy drive controller chips and sound processors to do all sorts, and the GPU is "another" processor now, and one that's extremely good at running lots of things in parallel. Do you think that your CPU could ever keep up with software-rendering on a modern game? Or that a single-core "general purpose" GPU could?

Your professor is right, except for his timescales - parallelism is already here, right down to tiny embedded systems, and has been for years. Just that hardly anybody uses it without some seriously specialised programming design beforehand. That tells you quite a lot about how expensive it is to use effectively. Hell, to a new programmer, even threading and re-entrancy can be a huge nightmare for them and our solution at the moment is blocking locks and things like that (so that we can *think* about them being seperate linear instances). If you can become an expert in parallelism you'll probably have a good career ahead - but a specialist one. Everyone else is just waiting until they can pass everything off to a dedicated chip and get told when it's finished - they don't *WANT* to properly program parallel programs (tongue twister!), they just want everything to happen serially and faster, or the closest approximation to that that they can get.

Seriously, when you program any graphics library, you just throw a huge amount of 3D object data into another processor and let it get on with it. You don't care if it runs at 6GHz, or whether it's got a separate internal processor for every pixel of the final image running at 1MHz, so long as it damn-well draws the screen as quickly as possible and tells you when it's done. Parallelism is a hidden facet of modern architecture precisely because it's so necessary and so damn complicated to do it right. Programmers, on the whole, are much happier with linear processes. It's taken games this long to pass off physics, AI, etc. to a separate thread, and we've had threading for DECADES.

Parallel things are just a placeholder at the moment because we can't up the GHz much and if we could, it wouldn't help that much with the tasks we want to do. So even operating systems are handing off everything they can to other chips - Ethernet TCP offloading, GPU, sound-processors, RAID, you name it. It's all about making the general-purpose CPU's do as little as possible so the main use of the computer can continue uninterrupted. And parallelism is only used to increase the things we can do, not break tasks into much more efficient subtasks. Most people who have dual-core or above in the early days wanted it so that other things (e.g. antivirus) got out of the way of the real process (e.g. games!) so it could have as much time running in it's single thread as possible.

Parallelism can see fantastic gains but it needs to be done from the design stage each time. Mainstream multi-core products are too much of a moving target to do anything more fancy than run a number of huge, single-threaded programs at the same time. That's *not* the same thing, and never will be. Parallelism is specialised. Games programmers would give their right arm for everyone to have a single 100GHz processor instead.

Re:Killer feature.

[sznupi]

Not necessarily; some tests by ARM/Symbian/Nokia strongly suggest that a n-core chip of x frequency is a good way to get considerable energy savings over a singlecore chip of n*x frequency. Of course whether or not it would be used that way is another thing...