Have you read Donald’s books? MMIX is supposed is fully parallel, but only at the power of “2 or 3″ due to how Donald made MMIX-PIPE. Effectively Donald did make a few typos so that “3″ and “2+epsilon” gets mixed up, but I’m not perfect either. Are there anyone out there asking that we need a larger number than 3? Please tell me and I will investigate. A similar question: Does any one of you think you know what exponentation actually means? If that is the case I will recommend you to study von Neumann algebra.

“Language is the key to understanding the problem of language”

The spectacular increase in CPU speed in the past was not achieved only due to higher clock rate but also due to a better CPU design. Early Intel CPUs with CISC architecture, which spent a few clocks per instruction, has been replaced with the modern superscalar design, which can perform a few instructions per cycle. Even for modern processor we can see that high clock speed does not necessitate a superior performance. For instance, 1.83 GHz Core 2 Duo is significantly faster than the 3 GHz Pentium 4.

Do the hardware designers have run out of ideas? If so, it happened a long time ago. All modern desktop CPUs use a superscalar architecture, and the first superscalar processor was built by Seymour Cray in 1965, i.e. 6 years before Intel released its first microprocessor Intel 4004. Since then only EPIC was invented, and though EPIC was hailed as a way to achieve higher parallelism than it is possible with the superscalar architecture, EPIC has not lived up to its promise aside some specialized tasks.

Do we have an alternative? Seymour Cray had always resisted to massively parallel machines: “If you were plowing a field, which would you rather use: Two strong oxen or 1024 chickens?” but in the middle of 1990s, he started his own massively parallel design. Unfortunately, his sudden death made impossible for us to know what direction his work would take, but the start-up company that he led is now specialized in reconfigurable computing. It is early to say if that configuration is the future of computing, but whatever it will be, I think it will be much closer to the data-stream-based computing paradigm than to massive multiprocessing. Remember Knuth words: “Pipelines actually work for me, but threads don’t.”

As always, the answer may lie in vastly different applications. Perhaps games will rescue us from the idle cycles again, utilizing it for some AI opponent.

What’s interesting is what Intel and AMD so far does NOT do with the vastly increasing number of transistors when they ask the question if we want more cores or increased clock speed. Perhaps we want cheaper circuits, more energy efficient, or more functionality (all sorts of I/O and graphics spring to mind).

In theory, but I never get very usable results in practice. This may be an OS issue. A hosed winbox takes forever to respond to keyboard input, regardless of the priority of the hosing process.

It’s becoming more common for the “average” user to use photo editing, and video software, which can sink a fair bit of power. It’s also becoming more common for peripherals such as wireless cards to be software implemented. It’s obvious to even a fairly naive user that some of these apps are very wasteful of the computational power. From a financial and ecological point of view, buying more processor power so that software manufacturers can blatantly waste it is horrifying, but I will almost certainly break down and do it.

Well, you get the same kind of a effect with a single core CPU if time consuming tasks are run at a low priority. With a low priority you’re able to use the machine as usual and if you aren’t using much processing power elsewhere (e.g. just surfing around or reading emails), the process won’t take much longer.

I think we’ll probably move away from purely heterogeneous architectures over to multiple cores with a batch of vector processors. However, this of course raises the question if all that processing power is actually any useful for the average user. Rendering or encoding aren’t really everyday tasks for example.

NB this is for a really, really short scientific computing program. Shark is awesome on OS X for figuring out hotspots and free as in beer with OS X.

I’m breaking in my first quad processor at work, and trying to decide whether it makes sense to get one for home. I’m disappointed that the most power hungry engineering design app. doesn’t use the multiple processors for all jobs. The types of engineering design and analysis that make CPU a bottleneck in my design process are very different from the types the software company tried to optimize for multiple processes. But, I don’t see how they could have anticipated the intimate details of our design process; that information is often confidential.

However, the quad has a distinct advantage that is always a processor quickly available for user interface functions like email and documents, even when two or three other processors are 100% occupied with the pressing numerical problem of the day. This greatly improves my ability to work in parallel with the computer. Given that wetware is much more expensive than computers, this is key…