Why does a 2014 MacBook Pro model beat out all the 2016 MacBook Pro models in benchmark scores?

[logicstudiouser]

https://browser.primatelabs.com/mac-benchmarks

The top end 2014 2.8GHz model beats out the top end 2016 2.9GHz model in both single core and multi core tests. Just curious as to why that would be? The scores are close, but 2016 still seems to score lower.

 

[Sanpete]

Because it's faster in that benchmark, probably due to higher turbo speeds. But it probably also gets hotter, so it may not run as fast over time due to throttling. That's what happens with the 2015, at least, in tests that last longer than a minute. The new machines have better heat management.

The new machines also have better dGPUs, if you do things like video editing.

 

[jerryk]

There is noise in those scores and the ranking move around, based on what was submitted. Anything within a few percentage is probably a wash. Also, when I look at the numbers the 2015 systems were on top.

 

[caramelpolice]

The Haswell 15" CPUs actually had higher Turbo Boost clocks, so in short bursts they are very slightly faster.

However, they throttle much harder under load, so in real-world performance (ie, not a 60-second Geekbench test) the new model typically wins out.

 

[ZapNZs]

IMO, because quad core laptop processor advancements in almost the past half-decade have been pretty marginal compared to the half decade before that. In 2015, Apple got a lot of grief for not updating the CPU in the 2015 15-inch MacBook Pro over the 2014...well, now we see why they didn't bother changing it.

From a pure processing standpoint, the 2016 doesn't really seem to make advancements over the 2015. Passmark shows a certain small degree of favorability to the older 2014/2015 CPUs, but some of this would be offset by the improvements in heat dissipation, and some of it could be margin of error.

How much does thermal throttling harm the 2014/2015? Well, I have a 2014 with a 2.5 GHZ i7 and the effect is a lot less than I thought. The computer can hold about 3.5 GHz continuously. But, that also means holding nearly 200F temps consistently, along with fans running at 6200 RPM consistently. With the much more capable (and reportedly hotter) GPU in the 2015, the effect could be different. With those who have the current 2016, the computers seem to hold a similarly high speed at lower temps and lower fan speeds. Even if the effect on performance is limited, less heat can mean less wear, and fans at 4000 RPM are a lot less annoying than 6200 RPM, so there are certainly some advantages with the newer design that a benchmark would not reflect.

However, at the same time, processor efficiency, (especially) the iGPU, the dGPU, and SSDs have made significant advancements...and Apple (among other Makers) has certainly made effort to integrate these advancements into their new products since they can't magically make the God of Intel poop out a faster piece of silicon.

On a positive note, AMD is about to light a firecracker under the butts of Intel execs, and hopefully will again be able to challenge Intel's long-standing status as industry leader. Hopefully, that will mean more rapid CPU advancements & falling CPU prices (at least until the limits of silicon are exhausted, and then we are forced to move to something else.) It could also mean a MacBook Pro with an AMD CPU and a NVIDIA GPU, which I think would be hilarious.

 

[macpot]

I think this is legit, not just random numbers in a benchmark,

My old Mid-2012 (Non-Retina) Macbook is faster than my new 2016 Macbook Pro 2016 15" (2.7GHz , Radeon Pro 455) in everything aspect except in Gaming (Graphics) and SSD Performance.

Same settings, both have FileVault enabled and Sierra 10.12.3.

When I click iTunes albums, they change in 2012 model quickly with smooth effects, on 2016 model it's very laggy (click on album > lag (like 0.2s) > suddenly album selected > lag > song list opened with glitch not with smooth effect)
Firefox also runs slower here for me with the same usage.
Almost everything is slightly slower, not noticeable alone, but when you work with them side-by-side you'll really see the difference!

I mean, Insane speed on SSD, Newer generation CPU with +0.1GHz (not much, but put it together with 3 generation upgrade), More Ram with Higher Frequency, I expected a huge increase in real life performance, but this is slightly slower!

 

[Sanpete]

I think this is legit, not just random numbers in a benchmark,

My old Mid-2012 (Non-Retina) Macbook is faster than my new 2016 Macbook Pro 2016 15" (2.7GHz , Radeon Pro 455) in everything aspect except in Gaming (Graphics) and SSD Performance.

Same settings, both have FileVault enabled and Sierra 10.12.3.

When I click iTunes albums, they change in 2012 model quickly with smooth effects, on 2016 model it's very laggy (click on album > lag (like 0.2s) > suddenly album selected > lag > song list opened with glitch not with smooth effect)
Firefox also runs slower here for me with the same usage.
Almost everything is slightly slower, not noticeable alone, but when you work with them side-by-side you'll really see the difference!

I mean, Insane speed on SSD, Newer generation CPU with +0.1GHz (not much, but put it together with 3 generation upgrade), More Ram with Higher Frequency, I expected a huge increase in real life performance, but this is slightly slower!

The differences you're seeing aren't due to a slower processor. Any difference in processor speed is too small to make those kinds of differences. Even MacBooks and Airs can do those things faster than some people are finding the new MBPs do (though some see no lags). The lags appear to be due to how the software drives the processors.

 

[Michael Scrip]

On a positive note, AMD is about to light a firecracker under the butts of Intel execs, and hopefully will again be able to challenge Intel's long-standing status as industry leader. Hopefully, that will mean more rapid CPU advancements & falling CPU prices (at least until the limits of silicon are exhausted, and then we are forced to move to something else.) It could also mean a MacBook Pro with an AMD CPU and a NVIDIA GPU, which I think would be hilarious.

It's (finally) an exciting time in CPU development again.

AMD is set to release a fleet of new desktop CPUs that appear to be very fast for not-a-lot of money. For instance... an 8-core 16-thread CPU for around $500 that is supposed to be as fast as an 8-core 16-thread CPU from Intel that goes for $1000.

If true... awesome.

What's not being discussed... yet... is AMD's mobile chips. If I remember correctly... AMD was never very good at laptop CPUs.

Hopefully they've fixed that.

Otherwise... you won't be seeing an AMD CPU in a Macbook for a while.

 

[macpot]

The differences you're seeing aren't due to a slower processor. Any difference in processor speed is too small to make those kinds of differences. Even MacBooks and Airs can do those things faster than some people are finding the new MBPs do (though some see no lags). The lags appear to be due to how the software drives the processors.

So if this is a software issue, why isn't Apple doing anything? many updates and still nothing.
also why some people don't see such lags? software is the same for all.

 

[MrGuder]

When I click iTunes albums, they change in 2012 model quickly with smooth effects, on 2016 model it's very laggy (click on album > lag (like 0.2s) > suddenly album selected > lag > song list opened with glitch not with smooth effect)

I have the exact same lag in iTunes album folder>song list. In fact I had both the 2015 and my now 2016 side by side running the same version of iTunes and Sierra and it was butter smooth on the 2015 and laggy on the 2016.

 

[Sanpete]

So if this is a software issue, why isn't Apple doing anything? many updates and still nothing.
also why some people don't see such lags? software is the same for all.

Good questions.

 

[macpot]

I have the exact same lag in iTunes album folder>song list. In fact I had both the 2015 and my now 2016 side by side running the same version of iTunes and Sierra and it was butter smooth on the 2015 and laggy on the 2016.

I saw it the very first day of getting the new one,
That feel of paying 3K for a slower device ruined my first experience

 

[itsamacthing]

You are benchmarking soooo wrong

 

[macpot]

You are benchmarking soooo wrong

Skip the benchmark, it's slower in real life usage and performance!

 

[rawweb]

https://browser.primatelabs.com/mac-benchmarks

The top end 2014 2.8GHz model beats out the top end 2016 2.9GHz model in both single core and multi core tests. Just curious as to why that would be? The scores are close, but 2016 still seems to score lower.

I own a maxed 2014 and 2016 15" and can tell you from first hand real world experience that the 2016 is just faster, for my needs. My 2014 just can't take video editing/encoding as well as the new model. I did some personal benchmarks with exporting a project out of Premiere on both and found the 2016 was several minutes faster even though my Geekbench page says otherwise. The 2014, meanwhile, sounded like a 747 preparing for takeoff. Is the 2014 a bad machine? Of course not. It's all in how you use the tool.

IMO, the processors between both machines are marginally different and offer no 'huge' differences outside of obvious power/performance/cooling. As others mention, the theoretical turbo is faster in the Haswel but the gains are with the GPU and SSD. The old GT 750M was outdated even in 2014, if memory serves, it was based around the GT 650M (more or less) first released in the 2012 retina. In fact the 750M was a carry over from the 2013 model.

I personally believe the 2015 release is the cream of the first 'retina generation (2012-2016)' with it's improved dGPU, SSD speed and force touch trackpad.

 

[logicstudiouser]

I own a maxed 2014 and 2016 15" and can tell you from first hand real world experience that the 2016 is just faster, for my needs. My 2014 just can't take video editing/encoding as well as the new model. I did some personal benchmarks with exporting a project out of Premiere on both and found the 2016 was several minutes faster even though my Geekbench page says otherwise. The 2014, meanwhile, sounded like a 747 preparing for takeoff. Is the 2014 a bad machine? Of course not. It's all in how you use the tool.

IMO, the processors between both machines are marginally different and offer no 'huge' differences outside of obvious power/performance/cooling. As others mention, the theoretical turbo is faster in the Haswel but the gains are with the GPU and SSD. The old GT 750M was outdated even in 2014, if memory serves, it was based around the GT 650M (more or less) first released in the 2012 retina. In fact the 750M was a carry over from the 2013 model.

I personally believe the 2015 release is the cream of the first 'retina generation (2012-2016)' with it's improved dGPU, SSD speed and force touch trackpad.

Good info, although I am hoping the 750M isn't too terrible. I ordered a 2014 Apple refurb from OWC to replace my 2011 MBP. I was considering the 2016 option, unfortunately, it was out of my price range.

 

[Sanpete]

I personally believe the 2015 release is the cream of the first 'retina generation (2012-2016)' with it's improved dGPU, SSD speed and force touch trackpad.

There may be a typo there.

 

[Naimfan]

Imagine how amazing the 2016 models could have been if Apple had kept the form factor of the 2012-2015 - a 99.5 watt-hour battery, more space for cooling, could have offered 32 GB of non-LPDDR4 RAM, added USB-C while keeping magsafe and a USB-A port or two, etc.

I don't do video editing, but in running seti@home my 2015 15" (2.5/16/512/370) averages more work units by about 6% than the mid-level 2016 15" I had (2.7/16/512/455). That's using the CPU 30% of the time and letting it use the GPU. Make of that what you will - I found it somewhat surprising, as I would have expected the results to be reversed. I also think the fan noise is a bit overstated - the 2016 got just as loud as the 2015.

 

[Sanpete]

Imagine how amazing the 2016 models could have been if Apple had kept the form factor of the 2012-2015 - a 99.5 watt-hour battery, more space for cooling, could have offered 32 GB of non-LPDDR4 RAM, added USB-C while keeping magsafe and a USB-A port or two, etc.

I don't do video editing, but in running seti@home my 2015 15" (2.5/16/512/370) averages more work units by about 6% than the mid-level 2016 15" I had (2.7/16/512/455). That's using the CPU 30% of the time and letting it use the GPU. Make of that what you will - I found it somewhat surprising, as I would have expected the results to be reversed. I also think the fan noise is a bit overstated - the 2016 got just as loud as the 2015.

An extra 23.5 watt-hours probably wouldn't be nearly enough to make up for the extra power use of DDR4 RAM over a day's use. The new MBPs are already much improved in heat management over the 2012-2015 models. According to objective measurements, and the reports of the vast majority of those who have made direct comparisons, the 2015 runs hotter and noisier. Keeping MagSafe would remove the advantage of being able to charge on either side. With tiny cheap adapters, I have all the USB-A connectivity I can use.

So, if the 2016 had kept the same form factor as the older models, it would just be bigger and heavier, not better.

 

[leman]

Because Geekbench 4 is flawed, plain and simple. Not only does it report wrong results in many areas (e.g. memory bandwidth!), when you run a real-world test, the difference between Haswell and Skylake becomes significant.

You might also find this interesting: https://forums.macrumors.com/threads/an-opinionated-cpu-benchmark-of-the-2016-15-mbp.2026475/

 

[ZapNZs]

Because Geekbench 4 is flawed, plain and simple. Not only does it report wrong results in many areas (e.g. memory bandwidth!), when you run a real-world test, the difference between Haswell and Skylake becomes significant.

You might also find this interesting: https://forums.macrumors.com/threads/an-opinionated-cpu-benchmark-of-the-2016-15-mbp.2026475/

Forgive the length and if I make a vocabulary error.

Is there any way to quantify how much of that is explainable to the 2016's superior thermal efficiency? Unlike most synthetic benchmarks, you were running a sustained test (and dare I say a closer duplication to real world usage?) and I am guessing this is where a bigger difference would be seen?

Further, do you have any thoughts on the difference between sustained CPU loads and sustained CPU and sustained GPU loads? My work is mainly CPU-intensive ONLY...and the CPU alone will take my MBP to its thermal limits (it's as high as the internal temp can go, and the fan speeds are maxed, even though the GPU isn't very active - and I was still surprised it did that well.) Presumably, the additional heat from heavier GPU usage (or even the heat from charging a battery) would generate more total heat than the overall design can dissipate (consequently resulting in thermal throttling, and more so with a GPU that makes more heat.)

I assume this is where the 2016 really excels considering the GPU advancements + heat reduction? (and I base this assumption looking at the comparison of UHD rendering times, which presumably push the CPU and GPU hard, and are disproportionately faster on the 2016 than the CPU differences alone could explain.)

With Geekbench (and, to my understanding, Passmark), when I run a benchmark, the GPU sits near idle. For real-world usage, I've read Makers are progressively moving to using the GPU for more everyday tasks if doing so improves efficiency - do you know if this is happening with the new MacBook Pro? To me, this would make sense since now the design seems to be able to better cool both under a sustained load. Do certain (updated versions of) Apps on the 2016s use the GPU in a fashion that previous MBP iterations did not?

If I had a 2014/2015 2.8, I'd send it to you to run the same test for poops and giggles.

 

[leman]

Is there any way to quantify how much of that is explainable to the 2016's superior thermal efficiency?

Frankly, I wouldn't even know how to start tackling this... I suggest we should just sweep this under the rug and test the system as a whole

Unlike most synthetic benchmarks, you were running a sustained test (and dare I say a closer duplication to real world usage?) and I am guessing this is where a bigger difference would be seen?

This was the idea, yes. But again, the entire setup was designed to closely match the kind work that I am doing. Other workflows will react very differently. For instance, I do not have any intuition about burst workflows. Skylake, with its more ALUs and more responsive turbo boost should have an upper edge here, but it would be interesting to look at it at detail. How does one test this though? I have no idea, I am certainly not an expert in this area.

Further, do you have any thoughts on the difference between sustained CPU loads and sustained CPU and sustained GPU loads? My work is mainly CPU-intensive ONLY...and the CPU alone will take my MBP to its thermal limits (it's as high as the internal temp can go, and the fan speeds are maxed, even though the GPU isn't very active - and I was still surprised it did that well.) Presumably, the additional heat from heavier GPU usage would generate more total heat than the overall design can dissipate (consequently resulting in thermal throttling, and more so with a GPU that makes more heat.)

I would certainly expect the 2016 model to perform better here, because of a much more efficient GPU. But again, this is a very complex domain. How many applications are there that will truly push the boundaries of the CPU and the GPU at the same time? The software that is most likely to do that are probably still games, because they usually consists of fairly well defined, semi-autonomous modules that engage at the same time (e.g. AI, resource loading, rendering, model logic). You can for instance send off your rendering commands to the GPU and deal with update logic and AI on the CPU while the GPU is busy drawing. In contrast, if I understand correctly, for professional applications, managing the available resources efficiently can be a major pain. The logistic effort involved in transferring the data here and forth and changing states is enormous. This could be one reason why for instance Photoshop and friends scale so badly with GPU performance...

Of course, it shouldn't be too difficult to write up a small torture app that loads the GPU with some angry kernels to crunch while also trying to fully load all the ALUs of the CPU with AVX-2 instructions. I am sure that this will easily overheat the laptop. But what purpose would such a test serve except testing the limits of the machine? Its certainly neither a realistic nor a useful usage scenario.

 

[itsamacthing]

Skip the benchmark, it's slower in real life usage and performance!

Benchmarks can really be misleading... agree, export some videos or something and then get back to us

 

[Charlesje]

There is a good review with extensive testing on notebookcheck on the 4870hq (2014 and 2015 macbook pro). It shows the throttling issue of the haswell processors when the four cores are used in max load settings clearly. And this does not occur after a minute but after a few seconds! I copy pasted the passage underneath. This has also already widely been covered on macrumors in the past.
Please note as well that if you compare skylake to haswell on a perf. per clock base you'll roughly have a 10% perf (per clock unit) increase for skylake

I use the 2,9 macbook pro myself everyday and it certainly performs faster cpu wise (by 15% at least in my tests) then the previous generation in condotions under heavy load (also within the first minute of testing). For the people still not believing this, please look up some real performance tests or show us some different results from clear and repeatable tests.

Notebookcheck quote: "According to the specifications, the Core i7-4870HQ reaches impressive maximum clocks of 3.7 GHz during single-core applications and 3.5 GHz when you stress all four cores, respectively – but it will require certain conditions (power consumption, core temperature). And this is actually a problem: The CPU does at least start at 3.3 GHz in the Cinebench Multithread tests (battery and mains identical), but consumes far more than 60 watts and reaches almost 100 °C (~212 °F). As a result, the clock drops to around 2.8 to 3.0 GHz after a few seconds and can maintain this level. "
http://www.notebookcheck.net/Apple-MacBook-Pro-Retina-15-Mid-2015-Review.144402.0.html

 

[MrGuder]

I know your mostly talking about the dGPU but because it runs cooler is this why the iCPU also runs cooler thus giving users longer battery life? I noticed when I had the 2015 along side the 2016 that the 2016 ran cooler temps that the 2015 and the battery discharge rate was higher on the 2015 than the 2016. So while the 2015 had a larger battery it discharged faster than the 2016, this has to do with temps and I was only using the internal graphics.