3 ms·
The hype is getting out of control. They do have an interesting cpu, but the apple target market isn't engineers/scientists/gamers so they traditionally have n
by m463 6y ago
The hype is getting out of control.
They do have an interesting cpu, but the apple target market isn't engineers/scientists/gamers so they traditionally have not done anything interesting with respect to ml and gpu kinds of things. The closest is video.
I think things might get interesting with an arm mac pro if they can add a lot of cores. I wonder if they will add PCIe slots so they can collaborate with other hardware manufacturers, or if they will navel-gaze some more and close it off.
- JulianWasTaken 6y agoIt seems odd to phrase things as "hype" and say Apple doesn't target engineers as their target market. The machine literally has a component in it called a neural engine which Apple advertises as: > In fact, with a powerful 8‑core GPU, machine learning accelerators, and the Neural Engine, the entire M1 chip is designed to excel at machine learning. (https://www.apple.com/mac/m1/ https://www.apple.com/mac/m1/) So asking "how well does that really perform" seems like precisely the right question, and to me it'd seem very clear Apple wants a piece of that market.
- R0b0t1 6y agoUsually a mobile device isn't for training just application.
- comex 6y agoThe Neural Engine is intended solely or almost solely for inference, not training. For instance, in a post from last November [1], Apple mentioned their tensorflow_macos fork can use the "the GPU in both M1- and Intel-powered Macs for dramatically faster training performance", but didn't mention the Neural Engine. (Incidentally, I think that is the same fork used for the benchmark here.) [1] https://machinelearning.apple.com/updates/ml-compute-training-on-mac https://machinelearning.apple.com/updates/ml-compute-trainin...
- p1necone 6y agoThat's for running end user applications that rely on already trained neural nets.
- pilsetnieks 6y agoDoes any of the software mentioned in the article even utilize the so-called neural engine? If it's just generic software running on CPU/integrated GPU, especially if under Rosetta (no mention about that in the article,) the results represent only the current software support situation, and not much else.
- ActorNightly 6y agoAnyone who is serious about ML (and by serious I mean like actually training fairly big models with lots of data for a project or as part of their job) is going to use cloud resources or at least have a decent gpu(s) at home. At no point will they ever look at getting a laptop to do ML one. Not only are you going to pay extra for less performance, but even with CPUs (which are still important for ML since you need to not only do the matrix math on GPUS but also prepare the datasets and set up the GPUs), you are not going to reach even close to the power that desktop chips can put out in any laptop with any manufacturer. As for general targeting of engineers the history of MBP pretty shows that they don't really care. Things like virtual escape key on the touchbar, shitty keyboards with sticking keys, hardware designed to not be reparable, and more recently, releasing the M1 models with a spotty backwards compatibility of software out the door is really not something that would be done if you were targeting tech minded people.
- enos_feedler 6y agoCue the bitter non Apple fan who can’t handle their success. Of course they will go many core. Of course they will provide extension points. I am sure you saw the FPGA accelerator from the existing Mac Pro.
- qbasic_forever 6y agoThe M1 laptops already have Thunderbolt 3 which is 40Gb/s and actually faster than PCIe 4.0 x16 (only ~35Gb/s). Also these laptops have 8 CPU cores (4 high performance, 4 high efficiency) and 8 entirely separate GPU cores. It's a powerhouse of multiprocessing, especially when you consider it's in a laptop that gets 20 hours of real battery life.
- lavishlatern 6y agoThunderbolt 3 is 40GBits/s (5GB/s). PCIe 4.0x16 is over 6x the bandwidth.
- smoldesu 6y agoPowerhouse is a bit of an overstatement, especially with the GPU, which is honestly a pathetic offering for a device bearing the "Pro" name. Even still, the M1's multi-threaded performance can be topped by the cheaper Ryzen 7 4800u, which was shipping in $500 laptops long before the M1 even hit the market. It's actually kinda disappointing to me that Apple didn't take advantage of the 5nm node they spent so much time and money securing.
- ustolemyname 6y agoA PCIe 4.0 x16 link is 32GB/s in one direction. Capital B. Quite a bit faster. It's also fully bidirectional, allowing for the same speed in the other direction at the same time. Thunderbolt is capped at 40Gb/s total bandwidth.