It beats my 2010 MacPro 2.94Ghz 12 core on single thread...ah that's progress for you. 
CMP still >2x the score on multi-thread, but nowhere near the efficiency per core or Watt of the A12z. Quite impressive so far for a development system.
The A12x benchmarks about the same as a GTX 950 for Metal gfx performance (~9100), just below an nVidia 105o (~10,100), so the A12z should be about on par for Metal as it has 8 gpu cores instead of 7 in the A12x. Still well behind an RX 580 at 52,800. No idea about OpenGL, though according to Apple that is also slowly being deprecated for their OS releases.
CMP still >2x the score on multi-thread, but nowhere near the efficiency per core or Watt of the A12z. Quite impressive so far for a development system. The A12x benchmarks about the same as a GTX 950 for Metal gfx performance (~9100), just below an nVidia 105o (~10,100), so the A12z should be about on par for Metal as it has 8 gpu cores instead of 7 in the A12x. Still well behind an RX 580 at 52,800. No idea about OpenGL, though according to Apple that is also slowly being deprecated for their OS releases.
Karl wrote:
Jacques wrote:We can agree to disagree.... my comment was not precise, but was in terms of ARCHICAD and Twin Motion, neither of which will perform usably or at all with any on-chip GPU known today, but which require powerful, heat-generating discrete GPUs at the moment. With today's (and next year's) technology, computers without discrete GPUs are not usable in a production ARCHICAD workflow.
Sorry, wrong.
I don’t think this is the case. There is nothing limiting the power of an integrated GPU, and it has the benefit of using shared memory, meaning it can potentially perform faster. One of the largest bottlenecks for the GPU is the speed at which data can be transferred between the CPU (which packages the data for each frame) and the GPU (which calculates it, and then sends it back). If that turnaround time can be reduced, it has a huge benefit overall.
The OpenGL deprecation in favour of Metal is a big cost for developers, but a huge win for consumers. This means developers will be required to rewrite their graphics pipelines on a MUCH faster framework. My (albeit limited) understanding is that openGL provides a number of APIs to program the GPU, but because its cross platform and works on multiple chips, it disguises the actually instructions it’s using on the GPU itself. So you tell the program to do an instruction, but beneath that, openGL has to determine how to run those instructions on its given chipset and then translate those instructions to be run. Metal provides an API that more closely matches the actual hardware instruction sets that are on the chips, so it doesn’t have to abstract itself. You can run similar graphics on much less computing power
So now we potentially get an integrated chip that has faster data transfer between the CPU and GPU, an instruction set that runs with FAR less overhead on the GPU, and the added benefit of unified memory between the cpu and GPU. I think there’s a lot of power that will come from that increased efficiency. I mean consider how incredible their graphics performance is now on mobile, and that’s with a decent amount of throttling for energy efficiency.
With that, while it’s technically possible for apple to include discreet GPUs in their future products, im not sure it’s going to be such a necessity. They already provide the option for eGPUs, and even support
