Thanks for suffering through an hour and 8 minutes of insufferable tech bros to summarize the highlights of the iMac vs iMac Pro bakeoff for us.
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
Also at about 6 minutes, Geekbench scores:
5452 single thread / 36928 multi thread
These are higher than even the first 10-core score in post #1. And no, I did not suffer through the whole thing like joema2!
Last edited:
I saw this video, but realized that these guys aren't fully smart on the technical side. IE, anyone who thinks blackmagic is a decent SSD benchmarker is fooling themselves. I'll go into why when I release the SSD benchmark after mine arrives on the 28th. The Cinebench benchmarks are actually useful. I'd like to run differences between the GPU and CPU.
[doublepost=1513967127][/doublepost]
I suspect the Geekbench benchmarks are failing due to being tailored this time...not by these guys, but by Apple, either intentionally or unintentionally. These guys picked up that the newer cooling system is far more effect as I'll probably see in my temp tests, and Geekbench doesn't run long enough to get realistic results on the CPU. One did pop up that dropped the 10c to where it matched single thread to the 8c. The 10c multithreaded results is right about where I thought it would be mainly because multiple cores running will bring the temp where it should be much, much sooner.
[doublepost=1513967127][/doublepost]
Also at about 6 minutes, Geekbench scores:
5452 single thread / 36928 multi thread
These are higher than even the first 10-core score post #1. And no, I did not suffer through the whole thing like joema2!
I suspect the Geekbench benchmarks are failing due to being tailored this time...not by these guys, but by Apple, either intentionally or unintentionally. These guys picked up that the newer cooling system is far more effect as I'll probably see in my temp tests, and Geekbench doesn't run long enough to get realistic results on the CPU. One did pop up that dropped the 10c to where it matched single thread to the 8c. The 10c multithreaded results is right about where I thought it would be mainly because multiple cores running will bring the temp where it should be much, much sooner.
This week's Accidental Tech podcast has a ton of really interesting information about iMac Pro:
http://atp.fm
http://atp.fm
For the first hour they discussed: why Marco ordered the configuration that he did, the advantages of ECC RAM, and Nvidia vs AMD GPUs in Macs. At the one-hour mark Marco began talking about a one hour Q&A that he had with Apple PR. Some of the topics:
1. Marco thinks Apple focussed heavily on needs of developers in designing the iMac Pro.
2. Apple did a good job of redesign of iMac internals so that it meets thermal requirements with workstation components. Had to slightly down clock 8 and 10-core chips to meet those goals. Can run CPUs at full load with no down clocking or fan noise. Even with CPU and GPU being hammered hard fans are not very loud (who knows what not very loud is, but probably somewhere between quiet and i7). All CPUs get dual ADX512 compute units. I don't know significance of that.
3. Discussion of T2, mostly things already known. File Vault encryption handled by T2, and this may be faster as a result. Overall, inclusion of T2 brings Mac closer to phone in terms of physical security.
4. Discussion of ramifications of boot security, i.e. full security adds requirement of using 2nd Mac for recovery.
5. Discussion of PCI lanes: iMac Pro seems to use 28 of 48 available lanes. i7 has only 16. SSD is on DMI running at equivalent of 4 PCI lanes bandwidth. Discussed how many PCI slots could be added to a Mac Pro using these chips (perhaps 2?)
6. The 5K display in iMac Pro still uses the same multiplexing designed for the original iMac 5K.
7. Discussion of relative bargain of AppleCare for iMac Pro, being same price as cheapest iMac (Marco asked if it was a mistake. It isn't).
8. Discussion of the "pro" designation. Marco concludes that it is truly a pro machine. Overall Marco is much more positive about the machine and he and John less concerned about thermal problems.
I'm sure I missed details, it was a 2-hour podcast!
____________Apple_____Intel off the shelf (OTS)
8-core___3.2/4.2 (???W)___3.7/4.5 (140W)
10-core__3.0/4.5 (???W)___3.3/4.5 (140W)
14 core__2.5/4.3 (140W)___2.5/4.3 (140W)
18-core__2.3/4.3 (140W)___2.3/4.3 (140W)
The CPUs on the 8 and 10 core being downclocked due to thermals doesn’t make sense. The 14 and 18 (and single thread 10) aren’t, so the 140W TDP isn’t the issue.
Three possible reasons come to mind.
1) Simply a better price from Intel on lower binned parts. If Intel has significant yield at the Apple frequencies compared to their off the shelf parts, Apple could have gotten a pretty good deal, since Intel would likely have little other use for parts that couldn’t meet the OTS spec.
2) Harmonize (equalize) the performance within the iMac Pro line itself. If they’d used Intel OTS parts, the 8 & 10 core parts might have looked “too good”, stealing demand from the higher profit margin 14/18 core SKUs.
3) Create market segmentation between the iMac Pro and the upcoming Mac Pro. If Mac Pro uses these same W-series CPUs, they would want to keep the iMac Pro from being such a good substitute for the Mac Pro (which would presumably use the Intel OTS speeds). The 15% higher speed for the 8-core and 10% higher for the 10-core would help keep the lower profit margin iMac Pro from cannibalizing the 8/10 core Mac Pro SKUs.
Thoughts?
I think you hit the nail on the head. It's all about market segmentation. The 8-core would be a natural choice over the 10-core for MANY more people if it had 10% greater base speed AND equal Turbo Boost speed.
Does this mean you should be able to overclock the 8/10 core to its natural values without worrying about heat since the TDP should be 140 regardless?
It depends on the transistor differences between the 8 core retail & the 8 core apple versions. If they are identical & simply down clocked, then you could overclock it with almost no heat issues (as proven by higher cores & performance). The downside is how do you overclock a CPU on a Mac & is it speed locked?
The TDP is the max design, not necessarily what the chip will use. Likely the 8 core is limited to a bit lower than 140.
Very likely they’re locked. Some locked Xeons from previous generations have been overclocked using EFI hacks but I think it’s ill advised even if someone eventually figures out how.
I'm more interested in seeing ifixit tear it down & hopefully find that we can upgrade the CPU barring some pains. I wonder if Apple figured out how to properly use thermal cooling compound yet.
Looking forward to the tear down as well. I’m wondering about the SSD and also the video card. I’d be surprised if the 4 different CPUs weren’t interchangeable but who knows?
I discovered that the memory has a large impact on the 8 core benchmarks. A few more have popped up & those who have 32GB vs 64GB actually have sound 4870 single thread on average vs 5120. This was not something I expected.
Hmmm... the Xeon W-series are quad channel; we might infer that Apple’s using 2 x 16GB instead of 4 x 8GB for the 32GB machine, but that’s a suboptimal configuration that entails a performance hit. Memory bandwidth in this case is only 1/2 of the maximum, since only two of the four memory channels on the CPU are utilized. The upside for the user is it allows a subsequent upgrade to 64GB by just adding two 16GB DIMMs.
4 x 8GB is the way to configure the 32GB machine for best performance. That allows for maximum memory bandwidth, as all four memory channels are used. The downside is that an upgrade to 64GB would require replacing all four 8 GB DIMMs with four new 16 GB DIMMs.
Real world performance would be higher or lower than the 5% seen in the benchmark. It would depend on the extent you’re starving for memory bandwidth.
Last edited:
This is the kind of nonsense from Apple that really gets to me. The 8-core could easily have been at least 3.4/4.5, but Apple wants to force people to pay for those overpriced upgrades, and purposely weakens the most useful CPU configuration. Having said that, I wonder if the 14 and 18 core CPUs will throttle, and create a more noisy machine.
Apple isn’t forcing anyone to do anything. They’re offering a range of price/performance alternatives that are clearly stated, and buyers will choose what best suits their needs. Intel Xeon processors are expensive. Apple could have used faster parts, sure—but you would have paid for it.
Apple wanted to offer 8-core/10-core machines as low as $5,000/5,800. If they had chosen the faster parts, those prices might have been $5,300/6,000 instead. Would you prefer that? Maybe so, but in any case Apple chose $5,000/5,800.
You mention overpriced upgrades, so let’s talk about it. Apple chose $5,000/5,800/6,600/7,400 for the base 8/10/14/18 core machines. Sure, Apple could have made the bumps smaller. Would you have preferred $5,600/6,000/6,400/6,800? Or $6,000/6,200/6,400/6,600? Maybe so. But that’s not what Apple chose to do.
Re: thermals, the cooling system has been completely re-designed to handle the higher load. The fans speed up (and get louder) to prevent thermal throttling. Under full load, it will be louder, by design.
There’s no reason to think the cooling system won’t be adequate, but we’ll know for sure soon enough, as the machines become available to be tested.
Last edited:
You make a fair point, but isn’t the 8 core option a downlocked 3.7/4.5 Xeon CPU? So then one should deduce that Apple is using cheaper binned CPUs, otherwise the downlock doesn’t make much sense (shouldn’t be necessary thermally, compared to the other more expensive CPU options). Wow, $5000 device using binned CPUs.
Anyways, merry Christmas.
Last edited:
Until we can find out the transistor count on it, we don't know if it's a down clocked chip, a lower end chip, or a chip that didn't make the cut. I would love to know if you can overclock these, though.You make a fair point, but isn’t the 8 core option a downlocked 3.7/4.7 Xeon CPU? So then one should deduce that Apple is using binned CPUs that otherwise wouldn’t make the cut, otherwise the downlock doesn’t make much sense from your perspective. Anyways, merry Christmas.
A little bird tells me that there are no official Apple procedures documented for removing the GPU in the iMac Pro. There is for the SSD pair.
The 3.2/4.2GHz 8-core is almost certainly the exact same part, transistor for transistor, as the 3.7/4.5 part, there’s no difference in features from the higher speed part. Intel just “programs” it post-manufacture to operate at lower frequencies. Does that mean it couldn’t operate at the higher speed? Maybe or maybe not (but it’s going to be locked to the lower speed when Intel gets done with it lol).You make a fair point, but isn’t the 8 core option a downlocked 3.7/4.5 Xeon CPU? So then one should deduce that Apple is using cheaper binned CPUs, otherwise the downlock doesn’t make much sense (shouldn’t be necessary thermally, compared to the other more expensive CPU options). Wow, $5000 device using binned CPUs.
There’s a natural variation in stable operating speeds of any processors on a production line. Look at the single core frequency for instance. For all we know, Apple’s CPUs could pass the 4.5 spec—maybe they all could—but maybe not. Maybe it’s only a 4.3- or 4.4-stable part. Yes, doesn’t meet the highest spec but that doesn’t mean it has to be thrown away. They can program it for 4.2 and sell it for less.
But even if it can meet the higher spec, Apple doesn’t want that part. It wants a slower (yes, cheaper) part, and Intel will program it that way and charge less for it.
(It’s worth noting that Intel’s tests at higher speeds than the rated speed when binning. To pass the 4.5GHz spec, the part may have to be stable at 4.65 (guesstimate). In other words, it’s not like the part can just barely do it’s rated speed, and it would fail at 4.51. There’s enough headroom in the testing requirements to be sure the part can operate at its rated spec.)
Intel has been doing this for decades. If there are otherwise identical parts but they have a 4.3GHz and a 4.5 part, you can be sure it’s due to binning. Binning isn’t a derogatory term, per se. Maybe only 25% of the parts on the wafer can meet 4.5. If it helps, think of the 4.5 part as having been “cherry picked”, not the 4.3 as somehow being “lesser”.
btw, Intel does the same thing with cores. For instance, all the 14 cores probably start life as 18-cores. But perfect wafers are tough to manufacture, and if there are 1, 2, 3 or 4 bad CPUs on-die, Intel can just program it appropriately and sell it as a 14-core.
PS Merry Christmas to you too!
Last edited: