Tim Tyler: On parallelism

A video discussing the state of modern computer architectures - in particular their degree of parallelism.

Transcript: http://alife.co.uk/essays/on_parallelism/

Category:

Science & Technology

Tags:

• parallel
• serial
• computer
• architecture
• fpga
• machine

[jjsajd]

Computers are not as serial as made out in this video ie multi-core processors, super computers, stream processors in GPUs. The argument that computing is stuck in serial land is not really true computing has many serial elements but is rapidly shifting to parallel designs. The issue is there are difficult problems in parallelism that need to be solved. This video is really an oversimplification of what is going on.

[wonderist]

Regarding high-level languages and the difficulty of compilers: Google "language oriented programming". Also see onboard(.)jetbrains(.)com(/)ar­ticles(/)04(/)10(/)lop

The idea is to have specialized languages (DSLs) to aid in creating new languages. This bootstrapping would allow for programmers to easily and rapidly develop specialized languages for any domain.

Consider that such a technique could readily be applied to the domain of parallel computing.

[selfhaytred]

Interesting, I've always wondered about this very thing. Although, true parallelism in computing may only be useful for certain problems, like AI for instance. An entire neural network could theoretically be represented in a large parallel circuit. Coupled with a learning/training unit, such a network could make the human brain obsolete.

[hord]

Multi Cores, Raid, SLI + Crossfire, Triple/dual Channel are all ways which components are going more paralle. However changing computers in one big sweep then nothing will work.

Also Nvidias new Tesla cards are many multi core processors of which programmers can use for super computing in a small machine in which Cuda can be used.