Fri Dec 28, 2018 4:24 am
Thanks. Sorry. But we can now with the latest gpu's, which are much better than a ARM at various things.
I've tracked the arm since 1987 or earlier, since before the original Archimedes computer came out, but maybe before as I was thinking of an Electron computer with Arm, and the BBC Arm CPU add on module was out, and heard about them doing a CPU around 1983. Was a large stepping stone in the CPU community which I'm part of. But it has become bloated with nearly as many instructions supported as Intel (well over 1000) where my own design proposal has less than 16 instructions aiming at 5 nanometer to fit the CPU core, and a design that's meant to replace GPU and DSP, FPGA, and asic in various circuit, hopefully 5Ghz low powered in today's processes (though if only 2Ghz is achievable, that is a significant improvement. But these are going be much bigger than 5nm), and much much more in tomorrow's technology, with I don't know how far past 10Ghz in today's technology on a high energy circuit. Many years of thinking to get get around the many constrictions of today's designs and constrictions applied from today's process technology, and shifting from normal CPU chip process technology to another.
At the moment they are much better off looking at next generation ARM technology aimed to compete with Risc-V, the current open source challenger, and scrap the existing instruction set support. They literally would he better investing in bringing RISC V up to scratch over the current ARM.
I'm not doing much on this lately and not keeping up with the whole processor advancements and scene, but arm is just a step in the right direction decades overdue, but not a Quantum Leap, but not like the Quantum Leap (reference to a computer I had, promised much, but was just a bit better, but nice machine).
Back before or after at the announcement of the Archimedes computer, they were talking about the arm being a main business machine CPU, but that went quiet. They probably found the market was tied up and people locked into Intel/Motorola were not willing to play ball, plus the collaborate power PC processor and Intel's x86 replacement, would have focussed attention on more feature mature technologies. If it wasn't for an outbreak of viruses, it might not have mattered so much, as the Arm was the most powerful desktop CPU by far (though the mob I was supporting had a more mature innovative powerful CPU, in a gate array format mind you) and things like strong arm kept that tradition alive for a while. The integrated arm chip technology was a leader, making cheap business machines possible, which would give them market for years, and time and money to mature the technology for unix like systems abd features to move higher. Simply, the simpler ARM design allowed lower power/higher performance in a power envelope but needed the investment to design that. The Risc V misstep also promises this over the ARM using a lot of architectural CPU design lessons over the decades. In the end, that might not matter so much as the CPU core becomes less and less of a greater design where processing is more and more concentrated outside the CPU. So, on a 64 core chip, great over Intel, still better than arm, but on an integrated chip where the CPU is only a little of the processing ability, not so great. My design is meant to be most of the design, so it matters more. At this scale we come down to task perform per unit of energy, ultimate performance, and to a lesser extent performance and energy per unit of area. Existing cpu level benchmarks don't matter as much, everything changes so much, you are looking at benchmarks more like application benchmarking. So, you would look between 10-100x the performance density and up to 1000x on today's machines. Even of that only translates into 10x the application performance (or 100x giving a heap of applications in parallel). That's still great. You notice the figures look a bit funny, that is because a lot of the performance per unit of energy comes from lower power, not greater throughput. The processing is also light weight instructions, but because you have lower power, you can pack processors more densely, but because you still get heat build up you can only go so dense and have to reduce speed. However, what I'm working on now is next generation process technology that goes a lot faster. But, I seriously need big money right now, and a team of scientist engineers. Jim and Grant are wasting money. The low cost custom silicon manufacturing scheme I mentioned elsewhere, which could deliver a $100 8k or 16k camera with storage could use this technology, but simply it can be as small as a cinema sensor (storage allowing, as that might require 3D array which is something not so simple).
aIf you are not truthfully progressive, maybe you shouldn't say anything
bTruthful side topics in-line with or related to, the discussion accepted
cOften people deceive themselves so much they do not understand, even when the truth is explained to them