Edit: Ah, I thought it looked low! That number is just for the iPhone CPU. I can't find an official number for the GPU but one Linpack run got 1,023 Gflops. That means it would rank in 2004.
The iPhone 6 CPU is significantly faster than 1.3 gigaflops. The 5s is capable of 10.4 gigaflops on LINPACK, and the 6 is faster still. Would have made Top500 in the late 90s.
My experience is that iPhone 6 GPU can be up to 50-70 times faster than the CPU - for single precision floating point (i.e. Swift running on the CPU, and Metal on the GPU). See http://memkite.com/blog/2014/12/18/gpgpu-performance-of-swif... for an example (comparison with Accelerate framework)
That's almost a "what not to do" for using Accelerate. You're introducing multiple passes over the data, and introducing extra dummy arrays that need to be passed over as well, which blows up the load-store traffic further. You're also using vvpowf to compute a simple reciprocal, which is wildly inefficient.
I don't mean to pick on you, but it's a misleading comparison. A basic transform that only gets rid of the extra working arrays and does no other optimizations is ~5-10x faster in my quick timings:
vvexpf(&result, &negx, &localcount)
let one = Float(1)
vDSP_vsadd(&result, 1, &one, &result, 1, localcount)
vDSP_svdiv(&one, &result, 1, &result, 1, localcount)
Titan Z can do 8 teraflops. From the Top500 List - June 2001, the #1 system at Lawrence Livermore could do R(max) 7 teraflops and R(peak) 12 teraflops.
I now see that I missed the gist of this, which was about mobile performance. The Tegra X1 can do a teraflop, which would put it at #2 on the Top500 List - June 1998.
"LINPACK" (without qualification) is a double-precision benchmark, and doesn't measure raw compute flops, but what is actually achieved solving a linear system. K1 does double-precision at a rate of 1/24, and good libraries typically achieve ~90% of rpeak on similar hardware, so something around 13.7 gflops is a reasonable estimate.
