The 2013 Mac Pro, Five Years Later
On June 10, it will have been five years since Apple first showed off the iteration of the Mac Pro that has come to be known as The Trashcan.
To put that in a little context, it was the same WWDC keynote where iOS 7 and OS X Mavericks were introduced.
[…]
That comment about OpenCL is an important one. Apple had bet big that executing computational tasks on the GPU was going to be a big deal, but it never really took off on the Mac. Maybe that was due to OpenCL itself, or the high cost of entry to the Mac Pro, but the truth is that the CPU remained the heart of most workflows for high-end Mac users.
As this was going on, Mac Pro customers started complaining of GPU failures. In February 2016, Apple opened a Repair Program for the machine, as Joe Rossignol reported[…] Even with the GPU issues, Apple failed to revise the computer in any way.
And now OpenCL is deprecated.
Previously: New Mac Pro Won’t Arrive Until 2019.
2 Comments RSS · Twitter
GPGPU workflows have been massively successful. All deep learning uses it, and most Adobe graphics products use it too. The problem is Apple chose to eschew the industry standard (CUDA) as it forced them to a single provider, NVidia, and instead pitched an alternative.
That would have worked too, except neither they nor their partners ever invested enough time and money in bringing it up to par, particularly porting over the higher level libraries.
Now Apple is ditching OpenCL for the even more constrained Metal, which means the Mac is still not a reasonable option for machine learning folk (how do you write your kernel to work on your AWS GPGPU cluster).
The eGPU idea is taking dongle-life to an absurd extreme.
It’s all part of the prosumerisation of the professional range.
Here's another take on the 2013 Mac Pro:
Despite absolutely no update to the available systems, 5 years later these machines are still for multi-threaded apps as fast as the fastest Macs you can buy with the exception of the iMac Pro which just hit the streets, really. If you had purchased an $8k machine as soon as they went on sale, you're still at least current after 5 years, and the system still has value. Cost of ownership over the lifetime is really not that bad.
Personally, I bought a 6-core dual D700 32G 1TB machine refurb unit almost 4 years ago now. It had the GPU issue and was gone for a month (I still have my 2011 iMac i7, so I switched back to that) but was repaired for free and has been rock solid since. Although I'm disappointed that the dual GPU idea never took off, it's still been a good machine. And it remains good enough for me to remain content to use it as my workhorse until the 2019 Mac Pro is revealed. It also makes me think I'm not going to wait next time. The high up front cost is a bit painful, but I've only had a few systems in my life that have lasted 2-3 years before I just couldn't stand using it anymore.
Much of this speaks to the slowing of Intel's processor progression (Moore's law). I've considered building a PC because it does seem that the GPU side has continued to advance at about a 10-15% annual clip, which should be considerable. And because I'm intrigued by VR and machine learning. ML is going to be so big that Apple cannot afford to miss it and maintain it's position.
