Just tested the Daniel's tests against our 16 Core LX2160 that is still only clocked at 1.9Ghz (will be 2.2Ghz final production). These are our numbers
binarytrees: 37s mandelbrot: 13.6s fasta: 0.9s
I will give your benchmark a run once I have some time.
Okay I couldn't help myself, waiting for a build to finish.
./parse jsonexamples/twitter.json
Min: 0.00171284 bytes read: 631514 Gigabytes/second: 0.368695
Whoa, thanks. 368, huh? That's pretty impressive given the 1.9Ghz clock speed... By coincidence, the 'twitter.json' file was the one I was using as my slender data point.
Do note that at this stage the benchmark is so preliminary as to be largely meaningless - I haven't really done much more than eyeball the results. But as a preliminary data point, that seems to be a stronger showing clock-for-clock than the eMag one.
let me know if you want this data in github. Just testing our overclocking bring cpu clock up to 2.4Ghz and DDR to 3200
Also building with gcc 8.2 not I needed to change -march=native to -march=armv8-a+crypto
Min: 0.00111424 bytes read: 631514 Gigabytes/second: 0.566768
I'm definitely not at the stage where I can say anything definitive about performance. For a SIMD-intensive task, a preliminary run on a 3.3Ghz eMag core gets about 400 megabytes/second of JSON parsing relative to 2.2 gigabytes/second on a 4.0Ghz Skylake. This may come down to having a maximum of 2 x 128-bit NEON operations per cycle vs 3 x 256 AVX2 operations per cycle, as well as some clock speed differences. Will eventually post about this at branchfree.org when I don my Nomex long johns (required whenever doing any benchmark that might imply to anyone that any processor runs faster or slower than any other processor).