My intention is to buy the "best" 13" rMBP with Touch Bar for field use. I need to stick to 13" as that fits my motorcycle pannier bag and is good size anyway. Using Apple's 14 day return policy, I purchased 2 models to compare and then keep one and return the other (or return both and order a 3rd configuration).
rMBP 13" i5 processor, 8GB RAM and 512GB SSD
13" i7 processor, 16GB RAM and 1TB SSD
Geekbench 4. I realise that this may not be the best comparison, but at least It is easy to run.
The i7 machine was about 15% faster on single core and 11% faster on multi core tests. Not a huge difference to my mind.
Open the same FCPX project and render it. I didn't time this, but I was running both machine simultaniously and watching the progress. The total task took around 20 minutes, however the i5 machine finished when the i7 machine was still at 91% done! This was a shock to me.
I was also watching Activity Monitor for CPU and Memory Pressure. Bear in mind that not much else was running. Both machines showed Memory Pressure to be green, and the 8GB ram was obviously using more in terms of percentage.
Blackmagic Disk Speed Test. Read speeds were vey close, but write speed showed the i7 1TB SSD to be 50% faster than the 512GB i5 machine.
Next I timed copying 65GB FCPX project from a USB 3 connected SSD. Both machines were almost exactly the same.
These results raise some difficult questions for me:
1) How come the i5 machine with 8GB ram rendered 10% faster using FCPX than the i7 16GB ram rMBP?
2) Is it therefore questionable to spend an extra AU$480 for a cpu upgrade from i5 to i7?
3) Is it worth upgrading ram from 8GB to 16GB for AU$320?
4) Is it worth upgrading from 512GB to 1TB SSD for AU$600?
My "gut feeling" is that RAM and SSD upgrades should be taken, but not the cpu, so that would save me almost AU$500.
Your comments would be much appreciated. Note I have only 7 days left to return these MBP's.
The administrator has disabled public write access.
Choosing MacBook Pro 13" for field use
22 Feb 2019 12:23 #99004
In your rendering test, take into consideration GPU, also. GPU is your single most important component for an NLE today. There are some tasks CPU will matter, some it won't. I've seen over and over and over, many, many times where upgrading a sluggish, problematic Mac from 8GB RAM to 16GB RAM gave huge improvements with stability and performance. Because the more physical RAM the less the CPU has to cache to disk RAM (virtual RAM) and the faster things are.
I'd not sweat CPU, some tasks will be faster on the i7, some faster on the i6, that gets into such a gray area, and the differences are not astronomical.
Get the most RAM you can and the best GPU you can. Drives are drives, performance will be equal on either Mac model. One model won't make a drive faster than using it with another model. But I will say that for even a road-edit laptop, system drive size does matter, and I'd go with 1TB. I have always found on the road I used up system drive space for one thing or another. It all got reset when the production was done, but the extra space came in handy on occasion.
Open the same FCPX project and render it. I didn't time this, but I was running both machine simultaniously and watching the progress. The total task took around 20 minutes, however the i5 machine finished when the i7 machine was still at 91% done! This was a shock to me.....1) How come the i5 machine with 8GB ram rendered 10% faster using FCPX than the i7 16GB ram rMBP?....2) Is it therefore questionable to spend an extra AU$480 for a cpu upgrade from i5 to i7?....
Performance testing can be time consuming to do right. In general you should disable Spotlight indexing, Time Machine, and any other background tasks which might interfere. To disable Spotlight indexing, use the instructions in this document but do not add back the drives until you are finished testing: support.apple.com/en-us/HT201716
There are various cache state issues which might cause inconsistent results so it's often better to restart FCPX after each test pass, then run at least three versions of each timed test and take the average. You may also want to delete all render files after each test pass.
In general I suspect a test variation issue caused the difference, not a machine difference. However depending on the codec and duration of the export, there might be thermal throttling differences between the i5 and i7 which cause variation. Obviously for laptop testing both machines should be on AC power and with all power saving features disabled or set identically.
Both i5 and i7 2018 13" MBP have the same Iris Plus 655 GPU; neither have an available discrete GPU. So the difference is likely not the GPU.
Both i5 and i7 are the same Intel generation (Coffee Lake) so there is likely no difference in single-thread instruction execution efficiency.
The i7 uses hyperthreading so on tasks which favor that it might be up to 35% faster. If your export is to H264 it's probably using Quick Sync which I don't think uses hyperthreading effectively. If the export was from/to a codec which was not hardware accelerated, e.g, ProRes, then the i7 might show a bigger advantage.
There's a 3rd party utility called CPUSetter which allows disabling hyperthreading on an i7. I did some brief tests on my 2017 i7 iMac 27 running FCPX 10.4.5 and Mojave 10.14.3, and for exporting 4k/23.98 XAVC-S to 4k H264, I didn't see a difference with hyperthreading on vs off. When exporting that same material to 4k ProRes 422, hyperthreading made it about 10% faster. That's an average of three test runs, restarting FCPX and deleting render files each time, and with Spotlight indexing disabled.
Several years ago I did similar testing and back then hyperthreading seemed to make an approx 30% improvement in FCPX export. It's possible the code path has changed so hyperthreading now makes less difference. Since those tests there have been several security-related fixes involving Intel vulnerabilities, including a fairly recent one which apparently fixed an attack based on hyperthreading. It's possible those have reshaped the hyperthreading performance advantage, at least for FCPX code.
Both rMBP's were connected to the mains and had batteries 100% full. No other programs were running at the time. As I wanted to compare real world environment, I didn't touch spotlight or anything else. Basically I set out trying to decide which to keep and wasn't planning to do any "scientific" testing. I posted here because of the unexpected results I found in my use case.
It appears, that for whatever reason, using FCPX , upgrading the cpu from i5 to i7 with all other parameters being the same, is not a value for money option, especially with the insanely fast SSD, which makes swapping very efficient.
I am still undecided on exactly what to do, but wanted to share my findings and invite comments.
The administrator has disabled public write access.