I wonder if the nVidia GPUs use their tensor cores for this. If they do it would be like comparing the M1 NPU to some standard GPU compute shaders.
I'd like to see an integrated GPU compared to the M1, for instance some intel Xe.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Apple M1 CPU & GPU speed is very disappointing
- Thread starter Appletoni
- Start date
- Sort by reaction score
- Status
I'd like to see an integrated GPU compared to the M1, for instance some intel Xe.
As far as I know M1 has not been oficially submitted by Apple and gotten a "real" spec rating for one of their systems yet. (Please correct me if I am wrong). We could only speculate why it has not been tested and submitted fully. I think Apple focuses more on questionable benches like geekbench and simple "one-threaded" type tests (native-only browser, h.256 playback etc.) to drive home their selling points for M1 systems.
I have never claimed M1 is "slow".. just that it is significantly slower than lower priced alternatives especially AMDs 7nm options for application workloads that really pushes the CPU to the limits )like chess apps).
Well, the fanbois over at Anandtech did post some SPEC results. But I guess they aren't serious because they were actually impressed with the performance. They don't run truly taxing workloads, like chess.
Please, post precise links to performance comparisons, with settings etc. These discussions are terrible. It's mostly posters trashing on Apple and its "fanboys". I'm not going to read through all this garbage to find something relevant.
That's some pretty low-quality content there, which you're bringing here yourself.
Last edited:
Then you think wrong. Apple never referred to geekbench.
Anyway, I've yet to see someone demonstrate how geekbench is "questionable".
As opposed to stockfish, Geekbench is designed to be a cross-platform benchmark tool and its results are consistent with SPEC tests.
I dont see the relevancy of this.. If you care about GPU perfromance and have $1500 to spend on a laptop you compare a m1 Mac against a $1000 5800H + RTX 3070 GPU similar priced..
For example take a look at..
Acer Nitro 5| 15.6-inch | AMD Ryzen 7 5800H| Nvidia RTX 3070 | 1080p | 144Hz | 16GB RAM | 1TB SSD |£1099.97 ....
5800H supports 16 threads on 8 full-speed cores..Much more porwerful even than the latest Max and Pro CPUs.. for less money.
If you don't care about GPU perfromance then both Xe and M1 is "fine".. and Xe-only alternatives will be much cheaper.of course.
My preference is to play people rather than software.
But a computer from 15 years ago would be more than enough of a challenge for me.
I don't make a living playing chess. It's simply a mental challenge to stem mental decline as I get older. I do make a living from my production stuff so it's very important to me that it runs well and it's really nice that it runs efficiently. Apple Silicon is not well known. I know a lot about x86 and vector instructions but I'm guessing that far fewer people know how to use Apple Silicon most effectively. Particularly the secret sauce. If chess were really important to Apple, they could add one or more chess engines in silicon. The areas where they are creating custom silicon indicates what they think their customers are looking for. It is rather difficult for any of us to disagree with their marketing department where they have vastly more resources and data to understand what their customers want.
Except it's missing ML cores, memory bandwidth, media encode engines, less GPU performance, etc.
And comparing total system prices is pointless : much of the cost of an Apple device is the nicer peripherals.
We could just compare the speakers on the 5800 of course.
The M1 does speculative/preemptive execution too, and while it doesn't have as bad an effect, it's not nothing and could be a security problem. I think just about all modern CPU's have it.
Well any benchmark that claims to be accurate for cross cpu architecture comparison, then rated for example an EPYC 7501 (32 core server processor) at a lower score than Apple A12X in multicore section, is flawed of course.. They have changed things but their whole cross-architecture benchmark doesn't really work to begin with.
So you're essentially saying "any benchmark claims by apple, that I don't even know what they are, are crap!"
Well... guess what I think of the chess benchmark?
Yeah, the M1 GPU is slower than an RTX 3070. News at 11. That's the case for all integrated GPUs that come in PC ultrabooks.
Comparing perf/$ between laptops (I suspect the other isn't an ultrabook) is a ridiculous criterion, because price covers the whole package, from chassis to display, speakers, camera, SSD, keyboard, battery life, thinness, weight and such.
Not at all the same. Intel actually has a separate branch target decoder, etc. much more convoluted than anything apple is likely doing.
Until someone publishes about a new side channel attack, I wouldn’t assume any new processor suffers from security leaks due to speculative execution.
A lot of things are easier to speed up when you have special designed engines on the Pro and Max chip.
I'm stunned that some threads take 21 pages or more trying to disprove the axiom of "The right tool for job." I went to read the links provided and stared at the not secure warning for the forum, wondering what decade have I entered. 8 pages of who owns TalkChess, never a good sign. I'm surprised Apple isn't all over this /s
It is called optimization of hardware. So now you are going to complain that Apple has some sort of unfair advantage.
Why don't you read the threads reffered to and the links in those all your questions about the settigns and compiles can be found..
Or better yet, Why dont you just try it yourself and postt some real results? if you think figures are somehow wrong or non representative?
Std banchmark with the benchmark option of stockfish after compiling the source from github is done by the -benchmark opttion https://github.com/official-stockfish/Stockfish/blob/master/src/benchmark.cpp
What results do you think is wrong?
This is a pretty good description of running the included std SF bench..
Similar procedures on all laptops according to chess forum.
Last edited:
Covert channel in Apple’s M1 is mostly harmless, but it sure is interesting
Technically, it’s a vulnerability, but there’s not much an attacker can do with it.
arstechnica.com
Why don't you read the threads reffered to and the links in those all your questions about the settigns and compiles can be found..
Or better yet, Why dont you just try it yourself and postt some rela results? if you think figures are somehow wrong or non representative?
Std banchmark with the benchmark option of stockfish after compiling the source from github is done by the -benchmark opttion https://github.com/official-stockfish/Stockfish/blob/master/src/benchmark.cpp
What results do you think is wrong?
This is a pretty good description of running the included std SF bench..
Similar procedures on all laptops according to chess forum.
What's your USCF?
Blender is still also very early days with Apple Silicon support.
You would maybe gain a few percent of performance by optimizing by hand compraed to the compiler that already optimizes for the CPU.
Apple probably have a dissadvantage of being a very closed, proprietary system, with poor or non-existing support and documentation to the open-source community trying to optimize code for Apple HW. This said Apple's sadly does not have some magic fairy dust that could be sprayed on their ARMv8.4 SoC to increase perfomance to even get closer to the lower end gen 4 8-core CPUs of AMD.
Can you be specific please.. What exactly optimizations do you think can be done on an M1 nott already done on stockfish and cFish c compiles with NEON. And how much would you think could gained exactly in theory and in practice..
This is a question everyone avoids here and just tries to sidestep, becuase they have no good answer!
- Status