iMac Pro testing shows 10-core model dramatically faster than any other Mac on intensive t...
Two more iMac Pro impressions have been posted, with benchmarks from both showing massive gains in processing power in the iMac Pro over older models -- plus the inclusion of AVX-512 vector processing optimization in the W-series Xeon processor giving an added push to properly optimized apps.
The first set, published on Tuesday by Theodolite developer Craig Hunter, is with the same configuration that Apple provided to YouTube channel MKBHD -- a 10-core iMac Pro with the W-2155 Xeon processor running at 3.0GHz, 128GB of RAM, the Vega 64 GPU, and 2TB of flash storage.
Performing a computational fluid dynamics study, used in aerodynamic design and development on a single core, the iMac Pro completed the task in 128 seconds, with the 2016 2.6GHz MacBook pro knocking it out in 175 seconds, and the late 2013 Mac Pro with a 3.5GHz processor completing it in 209 seconds.
When a flow simulation is performed on multiple cores, the performance advantage the iMac Pro has only gets wider.
In Xcode, Hunter saw reductions in compile times of apps with 20,000 to 30,000 lines of code spread out over 80 to 120 source files falling between 30 and 60 percent, when compared to the 2016 MacBook pro, and a 2013 iMac. He notes that there are still bottlenecks in the time it takes to fully deploy an app while installing and launching the app on iOS -- but the iMac Pro "makes a pretty noticeable improvement in repetitive code-compile-test cycles."
Hunter notes that Intel's Math Kernel Library is optimized for AVX-512, and says that he's been told that Apple's Accelerate Framework has been optimized for it as well -- but AppleInsider couldn't confirm Apple's optimizations.
When used in comparison with a 2017 iMac with a 4.2Ghz quad-core i7 processor, Hunter obsserves that the faster clock speed in the iMac edges out the iMac Pro in single-core operations -- but is completely blown out of the water at calculations involving more than four processors. Additionally, with calculations that use AVX-512, the iMac Pro "runs away" in performance.
One test involved 10.96 GB of 4K, H.264 footage being transcoded to ProRes 422 in Final Cut Pro X. The iMac Pro completed it in 7 minutes and 56 seconds, with a 4GHz iMac 5K with 32GB of RAM knocking it out in 15 minutes and 47 seconds.
Using DaVinci Resolve, one hour and 17 minutes of RED 8K Helium at 60 frames per second was exported to ProRes 422 in 2 hours and 6 minutes on the iMac Pro, and 7 hours and 15 minutes on the iMac.
With Lightroom Classic, LaForet imported 50.74GB of RAW images, with the application building 1:1 previews. The iMac Pro completed it in 25 minutes and 26 seconds, with the iMac taking twice as long at 50 minutes and 45 seconds.
BlackMagic's disk speed test found that the drive writes at 2996MB/sec, with a read speed of 2450MB/sec.
The first set, published on Tuesday by Theodolite developer Craig Hunter, is with the same configuration that Apple provided to YouTube channel MKBHD -- a 10-core iMac Pro with the W-2155 Xeon processor running at 3.0GHz, 128GB of RAM, the Vega 64 GPU, and 2TB of flash storage.
Performing a computational fluid dynamics study, used in aerodynamic design and development on a single core, the iMac Pro completed the task in 128 seconds, with the 2016 2.6GHz MacBook pro knocking it out in 175 seconds, and the late 2013 Mac Pro with a 3.5GHz processor completing it in 209 seconds.
When a flow simulation is performed on multiple cores, the performance advantage the iMac Pro has only gets wider.
In Xcode, Hunter saw reductions in compile times of apps with 20,000 to 30,000 lines of code spread out over 80 to 120 source files falling between 30 and 60 percent, when compared to the 2016 MacBook pro, and a 2013 iMac. He notes that there are still bottlenecks in the time it takes to fully deploy an app while installing and launching the app on iOS -- but the iMac Pro "makes a pretty noticeable improvement in repetitive code-compile-test cycles."
Vectorization
The W-2155 supports AVX-512, and the iMac Pro is the first Mac to support the wider vector registers. As a result, it gets a big boost in non-scalar calculations, where each instruction is run one at a time, versus multiple calculations in the same instruction.Hunter notes that Intel's Math Kernel Library is optimized for AVX-512, and says that he's been told that Apple's Accelerate Framework has been optimized for it as well -- but AppleInsider couldn't confirm Apple's optimizations.
When used in comparison with a 2017 iMac with a 4.2Ghz quad-core i7 processor, Hunter obsserves that the faster clock speed in the iMac edges out the iMac Pro in single-core operations -- but is completely blown out of the water at calculations involving more than four processors. Additionally, with calculations that use AVX-512, the iMac Pro "runs away" in performance.
Photography and video editing
Videographer Vincent LaForet also was also granted an iMac by Apple for early testing, with his work more or less focused on scientific calculation.One test involved 10.96 GB of 4K, H.264 footage being transcoded to ProRes 422 in Final Cut Pro X. The iMac Pro completed it in 7 minutes and 56 seconds, with a 4GHz iMac 5K with 32GB of RAM knocking it out in 15 minutes and 47 seconds.
Using DaVinci Resolve, one hour and 17 minutes of RED 8K Helium at 60 frames per second was exported to ProRes 422 in 2 hours and 6 minutes on the iMac Pro, and 7 hours and 15 minutes on the iMac.
With Lightroom Classic, LaForet imported 50.74GB of RAW images, with the application building 1:1 previews. The iMac Pro completed it in 25 minutes and 26 seconds, with the iMac taking twice as long at 50 minutes and 45 seconds.
BlackMagic's disk speed test found that the drive writes at 2996MB/sec, with a read speed of 2450MB/sec.
Comments
I wonder about multi-GB Photoshop documents with lots of layers, Smart Objects, effects, etc.
Get out of here!
Already addressed your comment spam here: https://forums.appleinsider.com/discussion/comment/3015574/#Comment_3015574
/s
I love how PC builders think everyone wants to be bothered tinkering around with or even capable of modifying their desktop computers. Well, I can't be bothered messing with a computer anymore. I just want to use it and hope it's reliable. End of story. If something goes wrong with it I can let Apple repair it or replace it. That's far more suitable for me as a senior citizen. If it costs me more then so be it.
When it comes to desktops and pro machines, that's what makes them pro, IMO. (That's why I ditched my previous iMac, because I didn't feel I wouldn't harm it by running heavy stuff for long periods of time. I've, unfortunately, harmed a couple of MBPs over the years, and don't want to repeat that again.)
Not that this is really the machine for me (as much as I'd love one), but I'd need to see some real-world testing first. Like, that it can truly handle getting rid of 500W and only generates 400W full-out. And, are there areas that get hot where the heat doesn't get drawn away efficiently enough to prevent damaging nearby components. Stuff like that. The old Mac Pros and even cylinder Mac Pro (I think) were pretty darn good at this. I'm wondering if this is in the same class in that regard, or just in terms of top performance.
The lack of video inputs got me thinking too... are there TB3 based video input methods that replace the lack of ports? I suppose you're always going through some kind of app or processing then, which might introduce lag. Otherwise, I suppose having other systems/devices input into a window wouldn't be necessarily a bad thing. I'm just trying to figure out what possessed Apple to build these all-in-one machines with such beautiful, central displays but then make them single-use.
Seeing that a few creatives get one of the 10-core machines in the mail I couldn't help but email Tim asking for the chance to test it out.
It's been unrealistic up to now for me to do much past the initial RAW conversion in 16-bit.
Another thing that has kept me out of 16-bit is that certain edits were not available in this mode. Are all the adjustments & filters available in 16-bit these days?