Bit noise: qualcomm desched, apple confused, fast moths, risc-v chips

Bit noise: Qualcomm desched, Apple confused, fast moths, RISC-V chips

Qualcomm has called the smartphone processor snapdragon 888, the first chip with the strong arm core cortex-x1. The snapdragon 888 is expected to expect about 25 percent faster than his transaction snapdragon 865 – that’s already, but far too little to keep up with apple’s m1. Because the snapdragon 888 – its type designation points to the chinese chuck number eight – containing four strong cores, but only one of them is a cortex-x1, the others are about 18 percent weaker cortex-a78. Unlike 2018 and 2019 qualcomm has announced no new snapdragon for windows notebooks this year.

Bit noise: Qualcomm desched, Apple confused, fast moths, RISC-V chips

The giant processor cerebras cs-1 is almost as coarse as a vinyl record and not only with ki algorithms.

Apple developed with the m1 the previously impressive arm chip for notebooks, but confuses the faithful follower but with nebuloses promises to its unified memory architecture (uma). All internal calculators of the m1 cpu cores, gpu and neural engine – can therefore access ram addresses without first copying data into separate storage areas (zero copy). This is actually a very good idea because it avoids intermediate steps and thereby increase the performance, she’s neither new nor unique. Amd has been the concept since 2011 as heterogeneous system architecture (hsa) "accelerated processing units" (apu) advanced, combining cpu and gpu cores. I love amd coarse under the table falling that intel generally combined cpu and gpu cores. And in intel documents from 2014 to use the built-in "hd graphics" as opencl calculation accelerator one can read how there the common "zero copy"-access from cpu cores and gpu to the same memory areas works. Remains to be hoped that apple uses the beautiful uma advantages under macos better for programmers, because under windows they are rarely used for use.

Smart wrinkle

Friends of distributed computing for science via folding @ home (f @ h) must now be very strong: google has overholding right, with artificial intelligence (ki). Google’s research division deepmind has won the competition casp souveran with an alphafal’s offered ki system. This could dramatically accelerate the search for protein compounds for new medications. Because instead of with naked arithmetic power possible, many molecular fittings cancel, at short alphafold off the way through ki predictions – so to speak with opposites instead of force. When this creates practically usable application packages for scientists, it is still open.

Also for ki algorithms art swift has called by esperanto technologies a pci express compute accelerator with over 1000 risc-v cores. However, esperanto – there also works the former transmeta-land dave ditzel – his et-maxion and et-minion cores already announced in 2017 and called neither concrete delivery dates nor prices. Anyway, et-graphics cores for a risc-v gpu are planned at any rate.

The company micro magic claims that its in-house 64-bit risc v core is the most efficient and easily creates 5 ghz. However, the envision lacks many details in order to be able to classify the success – for example, it is not clear which variant of the 64-bit risc-v technology is used at all.

How the market develops for the many new ki acceleries seems more uncertain than ever. On the one hand, the need for ki computing power continues to grow, on the other hand, cloud giants like google, amazon, alibaba and baidu develop their own ki chips instead of shopping for suppliers. Amazon set on the house conference re: invent the own "trains"-accelerator for exercising ki algorithms before; the "inferentia" for ki applications is already available in the aws instances ec2 inf1. Against this background, cerebras reports that their huge cs-1 processor, which is almost as coarse as a 30-centimeter wafer, is not only at ki quickly, but also in the solution of certain linear equation systems. The boundaries between ki and high performance computing (hpc) blur.

Of course, the greatest new chips use little if you can not buy them. Currently, the it industry is plagued by delivery funds, which may be in the year 2021. Who can deliver in this situation, secures market shares – and so intel will seem to get rid of two-nanometer processors. So it’s not just about having to have the fastest, but also enough of it

Like this post? Please share to your friends:
Leave a Reply

;-) :| :x :twisted: :smile: :shock: :sad: :roll: :razz: :oops: :o :mrgreen: :lol: :idea: :grin: :evil: :cry: :cool: :arrow: :???: :?: :!: