IBM bets POWER8 processor farm on hardware acceleration For 50 years, Moore’s Law held good. The axiom, formulated by Intel founder Gordon Moore in 1965, stated that overall processor power would double every year. In 1975 this was revised to every two years. But in July this year, Intel CEO Brian Krzanich revealed in an earnings call that in the wake of the introduction of 14nm …
Closer to 5 to 10, or non-existent for real world performance? Unless you can use more than 2 cores without I/O bottle neck. Per core performance for believable laptops compared with single CPU 2.2GHz P4 Laptop in 2002?
Anyway it WAS NOT CPU power but number of Transistors?
It was 12, then 18 then 24 months ...
It was never a law but an observation cum target?
A recent peer reviewed Applied Math paper describes a unified theory of optimization. Several fields of statistical/big data problem solving have been unified. Google has used its expertise in optimization coding to supply TensorFlow code to all those who might need code to derive benefits from the unified theory of optimization. Even proprietary unicorns will eventually realize that the idea of seeking AI help with big-data research is to big not to share.
It's called a high-end graphics card, and many programs can take advantage of them - notably in gaming.
Now, if you are not gaming, or editing video, where exactly can accelerated computing help ? With an Excel spreadsheet ? Doubtful.
The types of workload determine the computing needs. Apart from the two domains I already outlined, most, if not all, other domains Joe User can be dabbling in hardly need any accelerated grunt to get through.
....all of which feeds into the idea that computers have been 'fast enough' for the past 3+ years. The thing slowing down your Facebook browsing is now local and network IO. SSD's are fixing the local problem and the prevalence of broadband is (much more slowly) solving the rest of it.
My 3 year old laptop's CPU is within 10% of the performance of the latest and greatest 6th-gen offerings. My motivation for upgrade at the moment is more about having a nicer screen and a faster GPU - something that thankfully *is* still following a Moore curve to some degree.
Any push for faster computation is coming from governments who do love to play with virtual nukes, or the odd researcher trying to simulate lumps of organics or the business end of black holes. Bit of a niche market.
16:9 is worse than 4:3 for laptops.
Unless it's a "retina" screen you get 1080 pixels high.
Non-Shiny is hard to get.
Cheap models have poorer viewing angle than 14 years ago.
2002 there were 15" totally matt, ultrasharp, wide view angle 1600 x 1200 screens on some laptops. Good luck finding one even as good as that in PC World.
Some high end Lenovos and Apples have better screens, but you need 17" due to the 16:9 form factor
HD Video curse, reducing resolution of PCs! Maybe we'll see some decently priced UHD/4K laptop screens, still be cursed by video factor .
Encryption, Compression, Data Aggregation, Search - all these could be hardware accelerated as hardware acceleration has greatest benefit where there are very simple, and well-defined algorithms that can be parallelised across a dataset.
None of these are particularly useful on a mobile or desktop device, but on the server-side they can accelerate many data-centric tasks ten-fold or more.
At the moment, all those thousand-node x86 bigdata clusters out there take multiple clock cycles to apply an operation to a single data item - SIMD allows multiple data per clock cycle, but hardware or silicon based acceleration provide the opportunity to apply an operation to an entire dataset with the CPU hardly being involved at all.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
