Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
iMac Pro (IMP) All-in-One: 8 & 10 core Hands-on, Benchmarks, wipe/reinstall, & RAM upgrades
- Thread starter Trebuin
- Start date
- Sort by reaction score
That is one hot CPU. Glad to know that the machine is not loud.
Too bad the i7 didn't get this cooling setup.
Too bad the i7 didn't get this cooling setup.
True. I'm hoping the next generation Mac will get it given that Apple could take this shell & swap out the GPU & Motherboard for the next gen. The only thing that has to go if they adopt this is the standard Hard Disk Drive. If they want to keep that, they have to move things around again.
Also, my external speakers still will beat the iMac Pro's audio quality only when I use Boom 3D. They're old creative speakers (that should be powered) & a Sony portable speaker that acts as a subwoofer. It takes Boom 3D to be able to push enough sound quality to make the speakers really nice.
This Youtube video is interesting:
Looks at video editing with the IMP vs a 2017 iMac using Final Cut X, with a number of different head to head tests.
A quick summary: interestingly the iMac beat the IMP at H264 encoding by a small margin - the IMP was only 40% utilising its cores / GPU - but on H265 / 4K / 8K the IMP pulled ahead, being about 10x faster on one extreme task.
Looks at video editing with the IMP vs a 2017 iMac using Final Cut X, with a number of different head to head tests.
A quick summary: interestingly the iMac beat the IMP at H264 encoding by a small margin - the IMP was only 40% utilising its cores / GPU - but on H265 / 4K / 8K the IMP pulled ahead, being about 10x faster on one extreme task.
Nice summary of results, just one remark:
Cinebench isn't usable as GPU benchmark at all. It's 100% CPU limited these days, so it gives you a small idea about CPU overhead in graphics drivers (which is usually bad in macOS) and single core performance of the CPU, but it doesn't allow any conclusions about GPU performance.
It might have been demanding for a HD 5770 back in the days, but it certainly isn't for a Vega.
Cinebench isn't usable as GPU benchmark at all. It's 100% CPU limited these days, so it gives you a small idea about CPU overhead in graphics drivers (which is usually bad in macOS) and single core performance of the CPU, but it doesn't allow any conclusions about GPU performance.
It might have been demanding for a HD 5770 back in the days, but it certainly isn't for a Vega.
my 5 cents about the imac pro:
1600-1700 cinebench is ok, but you can get 2500cb for the same price if you buy a ready built pc
i guess significantly more of course if you can build it yourself
on the other hand that calculation doesnt regard that you get a screen and a beautyfully manufactured machine.
so if rendering is not you main task, then the price is actually not too bad, also compared to the pc world.
another factor is that you have to realize that not having any options to upgrade is a hidden price tag!
in 4-5 years your computer is outdated and there is no other option than to replace it completely.
an upgradeable machine lives much longer and saves you a lot of money.
so, if you are the type "i buy a computer and keep it 10yrs", then the imac pro is definitely a good choice.
if you always want to have the highest possible performance, then the imac pro is an expensive choice.
but well... there are not so many choices in the apple world anymore.
1600-1700 cinebench is ok, but you can get 2500cb for the same price if you buy a ready built pc
i guess significantly more of course if you can build it yourself
on the other hand that calculation doesnt regard that you get a screen and a beautyfully manufactured machine.
so if rendering is not you main task, then the price is actually not too bad, also compared to the pc world.
another factor is that you have to realize that not having any options to upgrade is a hidden price tag!
in 4-5 years your computer is outdated and there is no other option than to replace it completely.
an upgradeable machine lives much longer and saves you a lot of money.
so, if you are the type "i buy a computer and keep it 10yrs", then the imac pro is definitely a good choice.
if you always want to have the highest possible performance, then the imac pro is an expensive choice.
but well... there are not so many choices in the apple world anymore.
Exactly. I ended up ordering a high end regular iMac. I just don't how selling an iMac Pro 3 years down the line would work, in terms of price and buyers. Also 2018 could be the year of big changes to the iMac line, surely 2019. If I had the money and proper justification for it, I would buy the iMac Pro in a heartbeat. It is a beautiful machine, with several great upgrades from the regular iMac. As others here have said, this is a machine made for professionals with rather specific needs - not prosumers or consumers. Sure, if you have the money to spare, and want to pull the trigger anyway - I salute you.
Exactly. I ended up ordering a high end regular iMac. I just don't how selling an iMac Pro 3 years down the line would work, in terms of price and buyers. Also 2018 could be the year of big changes to the iMac line, surely 2019. If I had the money and proper justification for it, I would buy the iMac Pro in a heartbeat. It is a beautiful machine, with several great upgrades from the regular iMac. As others here have said, this is a machine made for professionals with rather specific needs - not prosumers or consumers. Sure, if you have the money to spare, and want to pull the trigger anyway - I salute you.
i m not that optimistic about the upcoming new "pro". what would be right is to deliver a simple workstation like the cheesegrater was. but i'm sure apple won't do that and invent something "totally great and new", but again ignoring industry standards.
best i expect from them are kind of physical modules like a core module with the cpu, a gpu module and a storage module that you can stack. this will work, but ofc it will be way more expensive than a simple standard tower.
i personally need dual xeons and the ability to add standard nvidia (cuda) gpus.
if that doesnt come, i will switch to a windows pc for desktop work in the future
.. we'll see.
Last edited:
Given the foxconn leak rumors there is little chance the next gen iMac will use the iMac Pro's design. While this design might be great for performance logically wouldn't make sense given how Apple markets devices and the current iMac design is way past due for an update(It will be 6 years since the slim body debut and the general unibody design has been around for much longer). The big question is will the next iMac be closer to a giant Macbook Pro internally in order for it to be very slim. I wouldn't put past Apple since the iMac Pro is out there to serve the power users.
Well, that would be a terrible decision met by backlash from regular users. I see arious reviewers of this year's iMac complaining about "dated design due for an overhaul". This is nonsense. Apple has found the perfect size and shape for an AIO computer (any smaller will hurt the performance), and should just continue to update it with a better screen and hardware, and yes a better thermal system. The only real improvement in design IMO would be a curved ultrawide 34'' 21:9 monitor, although I doubt that'll happen.Given the foxconn leak rumors there is little chance the next gen iMac will use the iMac Pro's design. While this design might be great for performance logically wouldn't make sense given how Apple markets devices and the current iMac design is way past due for an update(It will be 6 years since the slim body debut and the general unibody design has been around for much longer). The big question is will the next iMac be closer to a giant Macbook Pro internally in order for it to be very slim. I wouldn't put past Apple since the iMac Pro is out there to serve the power users.
Last edited:
While it can give a close analysis between different systems, you are correct about the GPU side. It doesn't stress the GPU or CPU during the test. 3DMark seems to be the standard, but it doesn't run on OSX. There are also only a handful of the Heaven benchmarks on that list, & I do know that will stress the GPU pretty well.
Why should it be always smaller? I dont care about the size in total , but they can do the bezels Thinner and the back dont has to be like that. I am shure 90% dont bother if its 1 cm thicker at the back ....
From a pure aesthetic point of view getting rid of the bezel would be the next logical step. Really tough to say what will actually happen but i'd venture any backlash would be no worse than backlash they've gotten from other decisions in the past(removing headphone jacks, usb ports, not updating Mac Pro, completely scrapping pro applications...etc).
It suggests there is no hardware h.264 acceleration in the iMac Pro, but other tests have suggested that perhaps there is (Vega?).
BTW, a lot of the graph labelling in that video is wrong. It shows the iMac Pro as a 2015 iMac.
Max is a good tester and focuses on real-world scenarios, but it's not that clear. I personally tested almost identical material to what Vincent Laforet posted here: http://blog.vincentlaforet.com/2017/12/12/apples-new-imacpro-has-an-impressive-200-300-speed-bump/
My results indicate the top-spec 2017 i7 iMac was about 1/2 as fast on transcoding 4k H264 to 4k ProRes as the 10-core Vega 64 iMac Pro.
A lot more 3rd party testing is needed before drawing definitive conclusions.
Is it a problem that his testing is using a Fusion drive for his iMac non-Pro?
Yes, he's using external Thunderbolt 3 SSDs, but his iMac doesn't have Thunderbolt 3. It is Thunderbolt 2.
I don't think so, H264 transcoding is not I/O limited. The I/O rates are not that high -- either in MB/sec or IOPS. He'd probably get about the same results on a USB 3 portable drive.
Also his 2017 iMac was *faster* than the iMac Pro in certain tests. If anything Fusion Drive would make the iMac slower, not faster than the SSD iMac Pro.
Max usually does a good job. It is tempting to draw confident conclusions from every little new test then change opinions every day, but in reality it takes a considerable time and lots of testing to be sure.
This could be a difference between the 8-core Vega 56 version and 10-core Vega 64 version on a highly specific FCPX workflow or a problem in test procedures. E.g, if FCPX is calling the AMD UVD/VCE transcode acceleration hardware, it could be a bug in the code path, library or AMD hardware that's specific to the Vega 56 version. We just don't know.
The 2017 iMac is very fast on H264, and it would be pretty dumb if Apple released an iMac Pro which was hugely slower using their own FCPX on the world's most common codec. Apple obviously knows this so there must be some explanation. Vincent Laforet's test results (limited as they were) indicate significantly faster iMP performance, at least on FCPX H264 transcoding on the 10-core Vega 64 version. We just don't know and must await further tests.
[doublepost=1514682862][/doublepost]Thanks for your well reasoned statements. It seems like every other day there's some new bit of information that swings the 8-core/10-core question one way or the other. Hopefully things will clear up soon.
To compare 8core with 10core the only difference between the two configurations must be the Processor's core count. Everything else must be the same; RAM size, SSD capacity, GPU and of course the macOS version and the User's optional configuration such as BT, Wi-Fi being ON or OFF, Menu bar being the same, and so on. The macOS has 100s of small processes running and each can pop in and demand CPU cycles that will disturb a multi-core application and cause it to be interrupted.
Are there other benchmarks showing lower scores for the 32GB config vs. the 64GB? Some of the simulated workload benchmarks should be able to detect whether the amount of installed memory actually has a real effect on performance. It could just be a geekbench artifact.
The only hardware difference I can think of that might cause this would be if the 8GB DIMMs in the 32GB config are single rank while the 16GB DIMMs in the 64GB config are dual rank. That would allow rank interleaving in addition to the (quad) channel interleaving that is common to both configs. That’s definitely possible.
Last edited:
I don't think that there really is a difference between the 32 and 64 GB scores. I averaged a bunch of the individual entries on Geekbench and those averages were very close.
That's actually what I was eluding to when saying not all memory modules are equal. I think you're right in regards to the ranks, especially since geekbench weights memory types in their benchmarks.
The performance difference is definitely architecture-dependent but there is a Lenovo white paper where they examined the effect of rank interleaving on a Xeon 5500.
"For instance, a system populated with 6 x 2GB dual-rank DIMMs outperforms a
system populated with 6 x 2GB single-rank DIMMs by 7% for SPECjbb2005."
http://www.crc.nd.edu/~rich/Nehalem/Nehalem Memory performance.pdf
It would be interesting to see some workload-based benchmarks of the 32GB vs. the 64GB, but of course even then mileage will vary based on your particular use.