Daily Pathtracer, Part 5. Let's make a naïve, suboptimal, port to GPU via Apple Metal instead. http://aras-p.info/blog/2018/04/03/Daily-Pathtracer-Part-5-Metal-GPU/
Next up: I dunno, maybe a DX11 GPU implementation?
@aras Funny i've also noticed that the MacBook pro integrated intel gfx is running OpenCL code faster than the dedicated Nvidia card. ram / bus / contention ?
@jakob that's my takeaway from other benchmarks too; the Intel one is actually *better* at computing power, it only is slower if bottlenecked by memory bandwidth
@aras also - now make 'Burst' target the GPU :^)
@jakob meh, that sounds like work. I don't like work :)
@aras How about a Vulkan one ? :)
@groud too hard for me
@aras Wait. Why is the intel on board faster than the 750?
@ScreennameTaken in my experience it just is (on some other workloads too). According to the internets, it can do more "math" indeed (Intel Iris Pro: 832 Gflops, GeForce GT 750M: 722 Gflops). It doesn't have that much memory bandwidth, so YMMV.