Read more.And an Affinity Photo dev has tested the new M1 SoC against an Intel Core i5-9600.
Read more.And an Affinity Photo dev has tested the new M1 SoC against an Intel Core i5-9600.
That ties in with TH report that it outpaces an RX560: https://www.tomshardware.com/uk/news...cs-performance
The RX560 is a 3B transistor part, though I would expect Apple would have to add more execution units so they can effectively underclock/undervolt to hit this tdp.
Possibly fast enough for Apple to create a headset and claim they invented VR?
Cinebench seems ok, assuming it's platform agnostic, nothing special in all honesty for someone who does 3D though.
Affinity... honestly their benchmarks make me question how well optimised the programs were before the transition... could it be that the 'arm' code is better optimised (the iOS affinity apps are 'fully featured') than the x86 or maybe it's the fact that the m1 kind of has 24 'processing' cores....
Now I'd actually test affinity myself to see how it performs on windows but I can't for the life of me find the benchmark lol.
Last edited by LSG501; 16-11-2020 at 02:42 PM.
I find it sadly rare that people optimise that closely for performance at all let alone on a given platform.
The fact that the memory is on package might help it though. That partly helps with power needed to drive the ram, but could also mean with no PCB traces and connectors to drive the memory speed could be quite fast. Detailed benchmarks will be very interesting there.
Unified memory could mean so many things though. An AMD APU could be said to have unified memory, as could an Intel IGP. M1 unifies having an AI inference processor with access to the same memory. Memory in a PC stops being unified when, for very good reason, you slap in a GPU with its own memory. Even then, with the variable BAR stuff that seem to be so in the news that would from the software side make memory *appear* unified and that is enough to give a performance boost.
But clocking the t**ts off a memory chip because it is right over the CPU? Yeah, I could see the benefit in that
TBH I'm not fully sure what the deal is with the unified memory but from what I can tell it's just one pool of memory for gpu and cpu...just a 'bit closer' than using intel etc
The m1 could also basically be acting like say a x64/x86 cpu and a cuda (or opencl) graphics card where the system is using everything and apple can set up the OS to use the m1 how they like after all. I'm sure Apple will do everything they can to maximise performance and show their decision was the right thing to do'. The performance gain I've seen in some instances when using cuda (which has a version of unified memory in it's specs) to speed things up is huge, so if apple leverages the gpu in a similar way it will show big gains because intel (and amd, due to opencl, to be fair) is pretty lacking in that area.
edit: supposedly those cinebench r23 numbers for the m1 are from it being run in emulation (or maybe the air), macrumors has some different numbers from the macbook pro showing 7508 multi core and 1498 single core
Last edited by LSG501; 16-11-2020 at 11:52 PM.
This story has been updated. For some reason Bits and Chips seems to have shared an Apple A12Z Cinebench R23 result and claimed it was from an Apple M1. I don't know why someone would do this... Anyway the article has been updated with the much better newer results, said to be from an Apple MacBook Pro M1 8GB (1498 single core, 7508 multi core). Thanks for everyone who contacted me via the form to update on this.
Last edited by mtyson; 17-11-2020 at 12:03 AM.
Tiger-lake holding up with a 12w TDP config.
There are currently 1 users browsing this thread. (0 members and 1 guests)