Answer:
The memory and cpu on the new Macbook Pro is holding the machine back. Apple wanted to preserve battery life so they made some compromises at the expense of performance.
1- Memory is slower.
2- CPU has a lower max clock speed.
3- Skylake runs faster on DDR4 not DDR3.
**important point**
Yes they are different chipsets and there are many variables which is why this is just a starting point and possibly solved not actually solved. Over to you world!
Why?
Apple wanted to keep the battery life to 10hours as we have come to expect. Skylake doesn't have support for low powered DDR4 so they had to go with low powered DDR3. However Skylake also doesn't officially support 2133Mhz DDR3 so they had to lower the timings and do what they could to get the higher frequency running and keep the power usage meeting requirements they set. This is why you also can't get 32GB of ram, it will use double the power.
Proof: 1. Memory is slower
The memory on the 2016 model is CL16, the memory on the 2015 is CL11.
In time this means:
2015 DDR3 1600Mhz CL11 latency: 13.75 nanoseconds
2016 DDR3 2133Mhz CL16 latency: 15 nanoseconds
Source: DDR3 RAM ineffectiveness sheet.
(lower is better.)
Latency timing screenshots in CPU-z:
2016
(Source Macrumors member deadworlds Thank you!!)
2015
(Notebookcheck.info Not the most credible source so if someone can do another test that would be great)
Memory Performance:
2015 memory latency Geekbench 4: 12.5 Moperations/sec
2016 memory latency Geekbench 4: 7.64 Moperations/sec
2015 memory bandwidth Geekbench 4: 25.1 GB/sec
2016 memory bandwidth Geekbench 4: 24.9 GB/sec
2015 memory copy Geekbench 4: 14.4 GB/sec
2016 memory copy Geekbench 4: 13.8 GB/sec
Remember this is DDR3 1600Mhz vs 2133Mhz so the numbers shouldn't be this tight.
2015 source: https://browser.primatelabs.com/v4/cpu/942360
2016 source: https://browser.primatelabs.com/v4/cpu/1047696
Proof: 2- CPU has a lower max clock speed
2016 max speed: 3.8Ghz
2015 max speed: 4Ghz
Source: http://ark.intel.com/compare/88972,83503
*Note* I am not saying that the 2015 CPU is a faster CPU. I am saying that in combination with the ram compromises it makes the 2016 i7 perform slower.*
Proof: 3- Skylake runs faster on DDR4 not DDR3.
Source: http://www.corsair.com/en/blog/2015/november/ddr3-vs-ddr4
Suggestion to Apple:
Give your customers the option to choose. Bring out a fix.
When plugged in, we are happy for lower timing if we get faster speeds.
If you want 32GB you will have lower battery performance.
Thoughts moving forward:
Apple is going to release a new Macbook Pro in 2017 on a new Intel chipset which will support LPDDR4. This will be the must faster Macbook and what we have come to expect.
*****Requests to the community*****:
Someone who has a completely upgraded 2015 Macbook Pro 15" Retina, can you please do a CPU-z and get us the memory details?
Also if we can do some memory/cpu benchmarks. Open to suggestion as to what would confirm the above theory.
Special thanks:
Thank you to members of the forum, specifically deadworlds for doing the test to get us the data, without him we wouldn't have been able to confirm anything. And Peter the CTO of ServersAustralia for helping me ask the right questions and identifying where the issues were as it didn't make sense why it was slower.
The memory and cpu on the new Macbook Pro is holding the machine back. Apple wanted to preserve battery life so they made some compromises at the expense of performance.
1- Memory is slower.
2- CPU has a lower max clock speed.
3- Skylake runs faster on DDR4 not DDR3.
**important point**
Yes they are different chipsets and there are many variables which is why this is just a starting point and possibly solved not actually solved. Over to you world!
Why?
Apple wanted to keep the battery life to 10hours as we have come to expect. Skylake doesn't have support for low powered DDR4 so they had to go with low powered DDR3. However Skylake also doesn't officially support 2133Mhz DDR3 so they had to lower the timings and do what they could to get the higher frequency running and keep the power usage meeting requirements they set. This is why you also can't get 32GB of ram, it will use double the power.
Proof: 1. Memory is slower
The memory on the 2016 model is CL16, the memory on the 2015 is CL11.
In time this means:
2015 DDR3 1600Mhz CL11 latency: 13.75 nanoseconds
2016 DDR3 2133Mhz CL16 latency: 15 nanoseconds
Source: DDR3 RAM ineffectiveness sheet.
(lower is better.)
Latency timing screenshots in CPU-z:
2016
(Source Macrumors member deadworlds Thank you!!)
2015
(Notebookcheck.info Not the most credible source so if someone can do another test that would be great)
Memory Performance:
2015 memory latency Geekbench 4: 12.5 Moperations/sec
2016 memory latency Geekbench 4: 7.64 Moperations/sec
2015 memory bandwidth Geekbench 4: 25.1 GB/sec
2016 memory bandwidth Geekbench 4: 24.9 GB/sec
2015 memory copy Geekbench 4: 14.4 GB/sec
2016 memory copy Geekbench 4: 13.8 GB/sec
Remember this is DDR3 1600Mhz vs 2133Mhz so the numbers shouldn't be this tight.
2015 source: https://browser.primatelabs.com/v4/cpu/942360
2016 source: https://browser.primatelabs.com/v4/cpu/1047696
Proof: 2- CPU has a lower max clock speed
2016 max speed: 3.8Ghz
2015 max speed: 4Ghz
Source: http://ark.intel.com/compare/88972,83503
*Note* I am not saying that the 2015 CPU is a faster CPU. I am saying that in combination with the ram compromises it makes the 2016 i7 perform slower.*
Proof: 3- Skylake runs faster on DDR4 not DDR3.

Source: http://www.corsair.com/en/blog/2015/november/ddr3-vs-ddr4
Suggestion to Apple:
Give your customers the option to choose. Bring out a fix.
When plugged in, we are happy for lower timing if we get faster speeds.
If you want 32GB you will have lower battery performance.
Thoughts moving forward:
Apple is going to release a new Macbook Pro in 2017 on a new Intel chipset which will support LPDDR4. This will be the must faster Macbook and what we have come to expect.
*****Requests to the community*****:
Someone who has a completely upgraded 2015 Macbook Pro 15" Retina, can you please do a CPU-z and get us the memory details?
Also if we can do some memory/cpu benchmarks. Open to suggestion as to what would confirm the above theory.
Special thanks:
Thank you to members of the forum, specifically deadworlds for doing the test to get us the data, without him we wouldn't have been able to confirm anything. And Peter the CTO of ServersAustralia for helping me ask the right questions and identifying where the issues were as it didn't make sense why it was slower.
Last edited: