Yeah we have plenty of servers and access to nvidias DGX systems but it’s just frustrating I always have to VNC instead of firing up some quick tests on my local computer. The 9980HK is plenty fast so it’s worth it in that regard but still.
I have a RTX 2080 and GTX 1070 in my desktop system (windows/Linux) and still use the cloud instance more often. It is faster and does not sound like a 747 taking off for 1/2 hour (or maybe it does, in the datacenter) while it trains my models.
And, a Macbook Pro will never be a great ML machine regardless of GPUs. It cannot dissipate the heat and the puny 96 W Power supply limits how hard the GPU can be run.