A little standard known as Thunderbolt has made its way around the industry for its high bandwidths and promise of transporting data optically rather than electrically. Intel, the creator, eventually needed to drop optical communication from the spec with a return loosely planned but firmly believed. For the last year, Apple was the only source for Thunderbolt-capable computers; starting in April, several PC manufacturers are expected to participate in adopting Intel’s technology.
Intel weighed in on the adoption of the standard in a statement to their partners.
To speed up the standardization of Thunderbolt, Intel is cooperating with Apple and Apple is the sole vendor currently to have PC products featuring Thunderbolt technology. As demand for the technology has seen obvious growth, Intel is ready to release the technology for public use.
While I am not too fond of the more proprietary platforms, several technologies await better external busses: high-speed storage and dockable processing accelerators such as external video cards are two very good examples. We will also at some point need to break free from electron transistor-based computing methods; optical integrated circuits based on photonic crystals appear to be a logical albeit distant next step. Advancements in optical bus technology for Thunderbolt, now, would be applicable for the advancement of that technology when it becomes ready, much later.
What would you do with a faster external bus? The crazier the prediction, the better.
If an external GPU will run
If an external GPU will run on this – why not external CPUs? Dual-socket workstation motherboards may go extinct if this is fast enough for them. I can see homemade beowulf clusters, or possibly even a distorted realization of what the Cell architecture originally advertised – seamless integration of multiple processors without end user hassle.
You hook your laptop up to the Thunderbolt bus, your desktop automatically sees it as an additional X number of processor cores and Y amount of RAM and your laptop sees the desktop in the same way.
QPI is around 25 GBPS, thunderbolt is advertising 10 GBPS. It’s not quite an insane notion, but it’s not down to earth either. 😉
I’d bet this happens around 2015-2017.
While I don’t think that external busses will get efficient enough to eclipse the need for internal ones… a great prediction.
the amount of latency would
the amount of latency would increase exponentially for every ‘cpu’ you add to the chain because of the amount of ‘layers’ the data would have to travel through before it was even sent off from the host machine to the ‘add-on’ one, then back to the host machine. It honestly wouldnt make it viable, regardless of bandwidth. … lets not get it confused with distributed computing.