GPU battle with Tensorflow and Apple Silicon

ML with Tensorflow battle on M1 MacBook Air, M1 MacBook Pro, and M1 Max MacBook Pro.

#m1 #m1max #m1pro #ml #tensorflow #macbook

Benchmark code:

(Take 15% off any premium NativeScript course by using the coupon code YT2020)

— — — — — — — — —

❤️ SUBSCRIBE TO MY YOUTUBE CHANNEL 📺

— — — — — — — — —

🏫 FREE COURSES

— — — — — — — — —

📱LET'S CONNECT ON SOCIAL MEDIA

[fwajed]

Since we don't know how to compare CUDA cores and M1 GPU cores, it'd be nice to see a TensorFlow comparison between CUDA and M1 GPU cores.

[somaking]

Summary:
M1 Max - 21m 48s (1x)
M1 Pro - 40m 25s (1.85x, expected 2x)
M1 - 116m 40s (5.35x, expected 4x)

[ricosrealm]

Definitely would like to see the Nvidia vs M1 Tensorflow tests

[briancase]

Ah, so here we see a dramatic illustration of the true effect of thermal throttling. Four times ~20 minutes should have been about one hour and twenty minutes. But the Air took an additional 36 minutes. That's all from the M1 clock speed being reduced in an effort to keep the chip cool enough to stay within its operating limits. So you see a very good illustration of the fact that the Air is very good at "hurry up and wait" type tasks such as rendering a Web page and then idling while you look at it, but the Air is not designed to excel at long-running tasks like NN training or even long builds, though it can do better there because when compiling code, there's significant wait time when accessing the SSD. Hurry up and wait is why your cell phone can run all day on a battery....

[bujin]

8:35. Not sure why the Max was left out of the summery shot. Would have been nice to see all three sets of numbers up on screen at the same time.

[frankelierrosalesgonzalez]

I’d love to see Nvidia vs M1 GPU to compare execution time and GPU usage

Really like this developer-oriented channel ! Kudos

[ev]

I would have liked to see the m1 Max in High Performance mode. It can give you a not-insignificant performance bump when running GPU heavy tasks.

[mscottsummers]

Nvidia vs M1 Max showdown. Fight!
I’m gonna check this out on my 14” M1 Max Pro.

[laminbarrow]

I would certainly like to see a metal vs CUDA test for machine learning. If CUDA wins, then one can definitely cheap out and buy a cheaper windows laptop or pc with an Nvidia RTX GPU for machine learning and data science.

[daveh]

Does the TensorFlow Metal benchmarking allow the combined use of GPU+CPU/AMX+ANE? It would be good to see the M1s firing on all cylinders then comparing that with PCs' CPU+GPU performance rather than ignoring the M1's full capability.

[tty]

Please do a test comparison between the Nvidia laptop and m1! I am wondering which one should I buy.

[NTeach]

I'd be interested in seeing a setup with pipenv rather than conda

[corvinyt]

Umm, why use GPU instead of Neural Engine? Can you benchmark with ANE optimized models? It would be good to test against GPU. :)

[andrewgremlich]

Yeaaaahhh!!! I was hoping you'd pull out that contraption!!! :D

[makingtheweb]

A smart man aways uses a walking freezer. I get better performance.🤣

[lucasecci]

The name can’t be anything but “the starter”

[PedroTeixeira]

Would be great to see this for the new Mac Studio!

[praffulpanwar]

Planning to buy m1 Pro 14th inch. Hope its good for my development. i am full stack web developer and work on at least 4-5 project same time so 4-5 phpstorm windows and navicate, apache, mysql, node, docker, kubernetes, teams, ms word, postman, filezilla, textastic all at same time.

currently i am using macbook pro 2019 and it keep on saying memory full quit some application.

should i go for air or m1 pro is only best for me. 16th inch is out of budget. Also if any other window laptop that are good as m1 pro.

[jithingeorge]

nice reviews my friend. i would like to see that nvidia vs m1 comparison though. thanks!