Apple Destroyed my Expectations.

[Emily Young]

We finally got our hands on Apple Silicon and put it through its paces to answer the question on everyone’s mind: Can it really replace Intel and provide a smooth desktop experience?

 

 

Buy Apple Mac Mini M1
On Amazon (PAID LINK): https://geni.us/JEzh
On Best Buy (PAID LINK): https://geni.us/mmRBV
On B&H (PAID LINK): https://geni.us/b8lWK1

[givingtnt]

6 minutes ago, GabenJr said:

We finally got our hands on Apple Silicon and put it through its paces to answer the question on everyone’s mind: Can it really replace Intel and provide a smooth desktop experience?

 

 

Buy Apple Mac Mini M1
On Amazon (PAID LINK): https://geni.us/JEzh
On Best Buy (PAID LINK): https://geni.us/mmRBV
On B&H (PAID LINK): https://geni.us/b8lWK1

Now that's automatic beyond belief! 

[eeeee1]

and my wallet.

[Kaleb Janzen]

I don't know much about apple

[SlashedM]

Anyone have this wallpaper?

[Emily Young]

1 minute ago, SlashedM said:

Anyone have this wallpaper?

 

[SlashedM]

1 minute ago, GabenJr said:

 

Thanks Anthony!

[NeuesTestament]

It would be very interesting to see how good Apple can scale this up. Because otherwise x86 will rule supreme.

[SupaKomputa]

On the chart written 5600 (without x), a typo or a clue....??

[Nystemy]

I can't help but look at this image and wonder.
"Why such a specific number?"

Like nothing wrong with expecting roughly 2x performance from having 2x as many cores in a highly multithreaded application.

Though there can be other bottlenecks and limitations when moving to a higher core count solution that limits that peak performance.
Partly due to Amdahl's law, but more realistically due to bandwidth, power stability, or just cooling.

I am personally of the opinion that it would have been better to put a range for the theoretical score, like 12000-14400. Since it is an estimate, and not real world performance at current. Though, maybe Apple has an 8 core chip as well and you are just "discretely" putting out an actual measured value, but I somehow doubt that, for multiple reasons....

[Commodus]

1 hour ago, NeuesTestament said:

It would be very interesting to see how good Apple can scale this up. Because otherwise x86 will rule supreme.

Probably fairly well. Like Nystemy suggested, doubling the cores wouldn't necessarily lead to double the performance. But Apple has a knack for wringing a lot of performance out of its chips, and there's a real chance we'll be wondering how higher-end MacBook Pros and lower-end iMacs are competing well against higher-end PCs.

[Mnky313]

from my very limited testing I couldn't get more than 4 displays (on an RTX 2080S) with MST hubs, so I would assume if it's limited to 2 displays you probably won't be able to get more with an MST hub (though maybe Apple has a different limitation).

[Lord Vile]

2 hours ago, GabenJr said:

We finally got our hands on Apple Silicon and put it through its paces to answer the question on everyone’s mind: Can it really replace Intel and provide a smooth desktop experience?

 

 

Buy Apple Mac Mini M1
On Amazon (PAID LINK): https://geni.us/JEzh
On Best Buy (PAID LINK): https://geni.us/mmRBV
On B&H (PAID LINK): https://geni.us/b8lWK1

Still doesn't know how to read a graph I see

[Jeppes]

2 hours ago, NeuesTestament said:

It would be very interesting to see how good Apple can scale this up. Because otherwise x86 will rule supreme.

Scale up and put a real gpu in. They were nice enough on the video not to use 2070S in Blender with optix or not to do a gpu enabled Davinci Resolve benchmark.

[Senzelian]

How does this thing breathe? Looks like the entire underside is covered with that giant black plastic foot.

 

 

And underneath there are ventelation holes? I'm confused.

 

[Spindel]

Great work @GabenJr and Linus.

 

As some have seen I have a 16 GB M1 Mini and I, as a long time Mac user, am very happy with it. 

 

Only thing I thing you should have emphasized more on is the responsiveness of the system (I just don't know how you could do that), because that is a thing I feel is hard to get across to people that haven't tried an M1 Mac. 

 

Also someone at LMG should get Linus to daily drive a M1 mac for a month, I would like to see what Linus would think of a macOS once he has used it long enough to get out of the "it should be this way because it is like this in windows" mentality. 

[Anstarmus]

Am I misreading this chart? It looks like the i7 Mac Mini was beating the M1 by 23 seconds (12:32 on YouTube). Other than that I loved the video.

[captain_to_fire]

14 hours ago, NeuesTestament said:

It would be very interesting to see how good Apple can scale this up. Because otherwise x86 will rule supreme.

There’s already large scale use of ARM chips in servers. I think Apple has already had that figured out especially for the likes of the future Mac Pro refresh, heck they can even bring back Xserve should they wish if only macOS Server is as good as Fedora, RHEL, or even Windows Server. 

[porina]

Wrote a short comment on floatplane about Prime95, but want to expand here now I tried to do some more digging.

 

Prime95 makes good use of FP64 resources, mainly through AVX instructions of modern x86 CPUs although older SIMD instructions are also supported. At least since Sandy Bridge and Zen. It is also the case these instructions consume relatively higher power than average. So x86 CPUs running hot on Prime95 relies on the CPU having a good FP64 execution rate. FMA instructions were added with Haswell which saw a good jump in performance over Sandy Bridge. Zen always supported FMA, but Zen and Zen+ only had two 128-bit execution units, and didn't match Intel's two 256-bit execution units until Zen 2. 

 

My concern is that we don't know how well the M1 runs Prime95 x86 code. If it runs it poorly, this will show up as lower power consumption. So throughput of Prime95 needs to be checked and compared also. There is a built in benchmark which will run various FFT sizes and give a throughput score. Different size FFTs can't be directly compared with each other. A rule of thumb guide is that FFT size multiplied by 8 gives the data set size. Multiply by number of workers for effective total data set size. A worker may have multiple threads working on the same data. The total data set size may be compared to L3 cache size to see if it can run "on CPU" or if it then also becomes subject to ram bandwidth considerations at bigger sizes. For time efficiency it may be worth manually running a small and large FFT size for comparison, as opposed to the whole range.

 

Quote

The four 128-bit NEON pipelines thus on paper match the current throughput capabilities of desktop cores from AMD and Intel, albeit with smaller vectors. Floating-point operations throughput here is 1:1 with the pipeline count, meaning Firestorm can do 4 FADDs and 4 FMULs per cycle with respectively 3 and 4 cycles latency. That’s quadruple the per-cycle throughput of Intel CPUs and previous AMD CPUs, and still double that of the recent Zen3, of course, still running at lower frequency.

From: https://www.anandtech.com/show/16226/apple-silicon-m1-a14-deep-dive/2

 

Above is the only write up on the FP execution potential of the M1 I've managed to find so far, and I find it confusingly written. 4x128-bit pipelines does match Intel since at least Sandy Bridge, and Zen 2, both of which essentially have two 256-bit execution units. It is unclear to me if the M1 can do FMULs across its entire width, or does it count separate FADDs to make up the numbers? I believe Intel and AMD can do FMAs across the width. Intel FMA latency is 4 cycles from memory (in Skylake), without looking up to double check, so that's comparable to the M1's FMUL. I don't see how the M1 can have higher cycle throughput as claimed in the article based on what's presented. In the best case I see parity. Also as far as I'm aware there was no change to the FP execution units in Zen 3 vs Zen 2 so I wouldn't expect a factor of two difference there.

[Radium_Angel]

As someone who doesn't follow Apple's business plan, the switch the non-x86 (ARM I assume?) is somewhat surprising to me.

Anyone wanna enlighten me as to why Apple has made this move?

[LAwLz]

6 minutes ago, Radium_Angel said:

Anyone wanna enlighten me as to why Apple has made this move?

1) Dramatically lower power consumption.

2) Potentially higher performance.

3) Lower heat output.

4) Bigger control over their own release schedule. A delay from Intel will no longer delay upgrades of Apple's own products.

5) Easier to maintain compatibility and share resources with their iPhone and iPad development and designs.

6) Probably lower cost for the manufacturing of the chips. If they are buying chips from Intel, Intel wants to make a profit as well. If Apple makes their own chips, they do not have to share the profits with Intel.

 

There are also a lot of drawbacks, such as a massive risk, and needing a ton of investment, but a lot of that is something they are already doing because they develop their own iPhone and iPad chips.

[Radium_Angel]

5 minutes ago, LAwLz said:

Bigger control over their own release schedule

I thought nVidia was trying to buy ARM, is this a case of Apple licensing the ability to produce their own ARM chips so they control the workflow (so to speak) and what happens to ARM (the company) is immaterial?

 

[A51UK]

8 minutes ago, Radium_Angel said:

As someone who doesn't follow Apple's business plan, the switch the non-x86 (ARM I assume?) is somewhat surprising to me.

Anyone wanna enlighten me as to why Apple has made this move?

I think this move make total sane for apple. There three reason:

 

1: is apple love total control over everything from  software, hardware to reaper, by going ARM with they own CPU now there have total control over that part of hardware. I did think there may more back to Power (PowerPC) or something different CPU. I think there will try to block Linux and Windows from running maybe on M2. 

 

2: To make them more different from PC, so there can say there not a PC it also make easy to hard problems or performance.  

 

3. There was make ARM CPU anyway for the Iphone so by doing this could have to cut costs. 

 

They did a really good job with the M1, I think for reason two they will not every sell this or any CPU for other company or PC. 

 

[A51UK]

1 minute ago, Radium_Angel said:

I thought nVidia was trying to buy ARM, is this a case of Apple licensing the ability to produce their own ARM chips so they control the workflow (so to speak) and what happens to ARM (the company) is immaterial?

 

It most likely nvidia will keep licensing ARM instruction set out, but apple can added new instructions to make them self different from others.  Who know how long the licensing term are for, it could be 10+ year. 

[LAwLz]

4 minutes ago, Radium_Angel said:

I thought nVidia was trying to buy ARM, is this a case of Apple licensing the ability to produce their own ARM chips so they control the workflow (so to speak) and what happens to ARM (the company) is immaterial?

Yes exactly.

Nvidia might buy ARM, but ARM in and of itself is just an instruction set, just like x86 is an instruction set. Just because a CPU is an ARM CPU does not mean it is just like all other ARM CPUs, in the same way an Intel CPU is not the same as an AMD CPU.

Both accept the same instructions (x86 in the case of Intel and AMD, ARM in the case of Apple, Qualcomm, etc).

 

So even if Nvidia buys ARM, the cores in Apple products are designed in-house. Apple (ARM CPU maker) won't be at the mercy of Nvidia (potential ARM owner) anymore than AMD (x86 CPU maker) is at the mercy of Intel (x86 owner).