Upgrade existing Mac or new? OWC offers a wide variety of upgrades for all Macs. The 2009-2012 Mac Pro remains particularly relevant for upgrades to memory, ultra-fast PCIe SSD, regular SSDs, hard drives, graphics cards, etc.
Now in use for about a month, my intensive photography work is benefitting from the 2013 Mac Pro in modest ways: my most time-consuming tasks are sped up by perhaps 20-30%. But those tasks are short duration.
To use a car analogy, the new engine has 500 horsepower instead of 325, but most of the time its like rush hour, so the commute doesn’t go a whole lot faster.
The lack of serious gains in productivity boils down to three things:
- Adobe Photoshop CC makes poor use of CPU cores on average, as it has forever. Really dumb stuff that has no reason to exist—pure design faults—like no ability to sharpen more than one layer at once.
- GPU speedups for common operations either don’t yet exist (hope springs eternal) or have been disabled until the graphics driver bugs are fixed (in 10.9.3). I expect this situation will last until late May or so just to get sharpening back to optimized speed (not that sharpening alone can really speed up total workflow).
- Save and open are I/O bound, and this was already fast on the old model with the optimized Photoshop preference settings. On the new model the dual-SSD speed is gratifying but it was plenty fast in the older Mac Pro also.
On the CPU utilization front for example, the generation of a lens rendering aperture series takes anywhere from 30 seconds to a minute. Photoshop uses at most 2 CPU cores (out of 8) for that entire script, and it is a CPU bound operation (very little disk I/O)! Photoshop CC just is not efficient for way too many things.
The most meaningful improvement I am seeing comes in converting raw files to 16-bit TIF for the aperture series and/or when previewing the raw files in the Photoshop ACR conversion window. This is an interactive part of my workflow so any speedup is welcome, and I see perhaps 30% faster responsiveness and that really does feel nicer.
But bottom line: am I getting my work done faster, saving even 1/2 hour per day? The answer is an unequivocal “no”. Things here and there are most definitely faster but in terms of measurable increased in productivity (or reduction in time required to do XYZ), the difference just doesn’t merit the huge cost bother.
Put another way, the new Mac Pro has the potential to consistently run ~20% to 40% faster (more for things that really exploit the GPUs), but software remains the gating factor, as it always has. The only question is whether the “constant factor”, —e.g., that ~30% speedup—is worth it for smoothing out the interactive portions of the a worflow where small delays are big irritations.
Dave S writes:
Just read your latest post, and I have to say that after reading your new Mac Pro assessments intently, I'd come to a similar conclusion. I'm on a 12 core Westmere, upgraded to 3.33 and 64gb, so essentially your prior machine.
Your machine has the Accelsior cards as well, though I noticed that you weren't booting from them.
When weighing the speed gains of the new Mac Pro against externalizing storage and PCIe to very expensive Thunderbolt or USB3 solutions, there is a huge cost premium for a marginal speed boost. The new Mac Pro is a beautiful piece of industrial design, but nested amongst a plethora of new black boxes and a tangle of cables, it's a tough sell.
As a guru on Mac performance, you don't have the luxury of sitting back and waiting / watching. Kudos to you for your honest and direct assessments. Personally, I'll need the new Mac Pro to offer more in the balance, before I make the leap.
MPG: I was booting off the Accelsior on the older 12-core, but it really doesn't matter for much of anything versus any other SSD. All that really counts is using the fastest SSD for the I/O sensitive tasks.
Whether a new Mac Pro is “worth it” is highly dependent on what one is doing. Certainly if running the 2008 or older model or video work or very large storage requirements there are strong arguments to be made for the new model.
The cable mess is there, but that pile of boxes sits under my desk so I just ignore it.
Robin K writes:
None of this is a big surprise, but watching you go through the process has been entertaining and valuable. I am happy with my 6-core 5,1 and will use it for a couple more years, I suspect.
Multi-threading can be difficult. Back in my days at Oracle, we paid a lot of attention to it, and reaped the rewards when systems gained more and more compute engines. But multithreading a database was relatively straightforward as we were able to separate user work into separate threads from the Oracle kernel, which itself was multithreaded. Plus this development work had the backing (indeed, insistence) of upper management (Larry).
I don't know too much about the inner workings of Adobe, but apparently they have other priorities.
Also, it seems to me if multi-threading is difficult, then moving code to GPUs may be even more difficult. Never done it myself, so I'll assume there is a bit of black magic involved. And a limited number of developers who understand it sufficiently well. We're talking about zillions of lines of code - a non-trivial exercise.
MPG: Everything depends on the workflow.
I wrote interrupt-level drives and server code for many years—yes it is difficult. But when very smart people optimize particular functions but supply a user interface which precludes the most basic efficiencies—operation on N layers with one command thus allowing parallelization—that’s low hanging fruit that it’s crazy to leave unpicked: why can’t I select 10 layers and choose Sharpen (for example)? Even if zero time benefit accrues (unrealistically pessimistic), it saves my time by letting me invoke once, rather than 10 times. It’s obvious, so why isn’t it done?
As for the GPU support, yes it’s hard, but with the graphics drivers rife with bugs (rushed out by Apple in the opinion of MPG), Adobe had to defeat the GPU optimization for sharpening. No vendor can work with the GPUs until and unless Apple gets its act together. The good news is that I expect 10.9.3 to be a lot better, but that’s far from saying it will be robust.