The fan control algorithm is a bit amateur. It oscillates back and forth in frequency at a high enough frequency (0.1-0.2 Hz) – and with significant range – that it’s quite noticeable, audibly, and sometimes distracting.
For some reason the full-load CPU frequency on my machine has dropped from 3.6 GHz to 3.4 GHz. I have no idea why. The inlet temperature to the case is actually lower now, too. The workload isn’t identical, so perhaps this is a natural result of subtle workload differences (e.g. use of AVX512 vs not, or something like that – most Skylake Xeons have significantly lower clock frequencies when the AVX512 unit is in use, though AFAIK the relevant data-sheets aren’t available publicly for the specific SKUs Apple use in the 8- and 10-core iMac Pros, so I can’t be sure).
Note the 4500 (Hz) limit on the Y axis. Still merely aspirational, as far as I can tell. Even under the lightest loads I’m still yet to see it exceed 4.2 GHz, according to iStat Menus (or Intel’s Power Gadget). Though I don’t put too much stock in those, as alluded to in my first impressions post – I doubt they’re relying on actual turbo bin residency counters, but rather just an average over a relatively large period (e.g. using MPERF & APERF). (I’m not actually sure, off-hand, if Skylake has proper residency counters for this purpose)
‘System’ power draw as reported by the the machine’s own “Total Power” sensor maxes out at about 350W, but (including some external hard drives and other such devices) my UPS says 550W is being drawn. For comparison my prior 2014 Retina iMac reported just over 200W for the approximately same sensor (though alas I never checked what the UPS reported). In any case it clearly, and as expected, generates significantly more heat than the non-Pro iMacs, as immediately evident by its much improved ability to heat the room. 😄
The login screen at boot is super sluggish and buggy – it lags behind keyboard input by up to several seconds, and often after you select a user and it transitions to showing just their picture + the password field, it’ll then inexplicably go back to showing all the users’ pictures – but with the password field still visible. The first ten keystrokes into the password field are almost always ignored & lost. And sometimes, upon hitting return in the password field, it just obtusely removes the field and goes back to showing just the initial list of users, requiring you to select your user account again and start over.
It’s a plausible hypothesis that whatever is implementing this under the hood is significantly different from on prior Macs. Perhaps due to integration with the T2 SoC for security & flash access. And it’s not implemented well.
Whatever the cause, it’s kind of infuriating and baffling, that such an obvious & egregious flaw exists, given this is literally the first thing you experience every time you turn an iMac Pro on.
10-core w/ Vega64. Upgrading from a 2014 Retina iMac.
Relatively briefly, and in no particular order:
I don’t see why the very slightly different colour scheme, vs the regular iMacs, is such a big deal to some people. Yes, it’s fairly obviously a different colour. No, it doesn’t really look any better (nor worse) than the regular iMac’s colour.
It’s disappointing that it comes with such crappy input devices (the mouse & keyboard at least). They’re the usual ergonomic & general usability disasters that Apple’s infamous for as of recent years.
Digression: I also recently got a new MacBook Pro 13″ with Touch Bar for my work machine, which has an even worse keyboard than the iMac Pro, if such a thing is possible. It’s literally painful to type on.
On the upside, it doesn’t seem to ever drop below that, despite nominally having a 3.0 GHz base frequency, even under the heaviest loads I can throw at it (including heavy, concurrent GPU use).
On the downside, it’s supposed to turbo up to 4.5 GHz, but I’ve never seen the tool report that. It does get up above 4.0 GHz if you only have one or two threads actually active, but only barely. Intel’s tool only has 20ms sampling resolution, so it’s quite possible it is bursting to 4.5 GHz in very short stints. In fairness, the regular iMacs exhibit basically the same behaviour – my 2014 Retina iMac nominally boosted up to 4.4 GHz, but in reality rarely if ever hit that. Under load, that iMac struggled to reach 4.0 GHz. Unless the ambient temperature was uncomfortably cold, it’d easily fall down to not much more than 3.0 GHz under any kind of sustained load, and sometimes even further, into the 2.x GHz range.
The fan is quite audible under any real load, even though I have some loud Thunderbolt disk bays and other things even closer to me than the iMac Pro. I have no idea what some reviewers have been talking about w.r.t. the fan being “whisper quiet” or outright “inaudible”, because it definitely is not quiet. It’s not loud, to be sure, but you can’t miss it.
Under basically no load, there is indeed very little fan noise, but that’s both an unrealistic use case and certainly no better than the regular iMacs.
It does feel dramatically faster than a non-Pro iMac. I did not expect this. Certainly I expected significant objective improvements in parallel workloads – mainly batch photo & video editing in my case – but in fact the speed improvement is very noticeable even in single-threaded workloads. I’m not sure why yet… the internal SSD is faster than the SATA SSD in my prior iMac, but the difference I’m seeing doesn’t seem plausibly explained by that [alone].
I’m also seemingly seeing it perform significantly better under load, w.r.t. user interaction. Even with all CPU cores completely busy, and the GPU likewise, interactive use remains basically as fast as when it’s idle. This is a pretty big difference – and very pleasant improvement – over the non-Pro iMacs. It’s really nice to not have to just walk away while CPU-intensive tasks are running.
The screen doesn’t immediately appear much different – in terms of colours, contrast, brightness, etc – to my old 2014 Retina iMac. But it’s very clear which is which, because the iMac Pro has no image retention issues, whereas the 2014 iMac has pretty severe ones.
Though when specifically looking at sRGB vs Display-P3 examples, the difference is quite a bit moreso than I expected, which is of course a pleasant discovery.
It’s so much better to have a proper, native VESA mount vs the hacks you had to do with prior iMacs.
iStat Menus can’t read any sensors (except CPU frequency, once Intel’s Power Gadget is installed), though I expect this is going to be fixed fairly soon, in a future version.
The ports on the back aren’t properly aligned with the case where they protrude, unlike non-Pro iMacs. Meaning when you plug a cable in, it doesn’t align relatively flatly against the curved case, but rather tilts upwards a bit. This is a really odd change – though obviously minor and practically insignificant.
I don’t yet understand why, but Lightroom Classic CC is noticeably snappier. Particularly in the Develop module as you make edits and then wait for the results to appear on screen. In some cases it’s an order of magnitude faster – e.g. less than a second instead of 5-10 seconds. It’s still not consistently fast by any means, but it’s no longer always infuriatingly slow.
I’m unconvinced, regardless, that the laws of physics will allow creation of a computer upon which Adobe’s software won’t run agonisingly slowly.
Officially it’s quite a bit heavier than the non-Pro iMacs, but I was surprised to find that it’s actually lighter for me… though that’s because with the stand removed – replaced by the VESA mount – it of course under-weighs the regular iMacs with their fixed stands still stuck in them plus a VESA mount adapter.