2020 i7 MBPro 1068NG7 benchmarks

[Spectrum]

I'm starting to think about the i5 differently. Essentially, it is already a CPU that is much more powerful than a regular stock i5 from intel, due to the higher base speed and higher TDP. Thus it probably makes far more sense why there may be such a small performance difference between the i5 and i7 model 2020 model. They are almost the same CPU, and are both limited by the thermals of the enclosure.

 

[magbarn]

Yes thermals and power. Really regret the $200 I spent on the i7, but love the machine so much I'm going to keep it

 

[cjdbet]

I averaged the single-core and multi-core score from the geekbench search result pages for the i5 (1038ng7) and i7 (1068ng7). These are from the first page of results, which has 25 results per page, sorted by Upload date, I believe.

	1038ng7 (i5)	1068ng7 (i7)
Single-Core	1155.32	1150.96
Multi-Core	4246.92	4272.92

I'm a bit dubious about some of the results on both pages, I'm guessing like KPOM said, people are running the benchmarks while other processes are eating cycles. If you want to try other pages, here's the code I used (open the Developers Tools Console in your browser and paste this in):

Single-core: Array.prototype.filter.call(document.getElementsByClassName("list-col-text-score"), (item, i) => { return i % 2 === 0 } ).map(item => Number(item.innerText)).reduce((acc, val) => acc + val ) / 25
Multi-core: Array.prototype.filter.call(document.getElementsByClassName("list-col-text-score"), (item, i) => { return i % 2 !== 0 } ).map(item => Number(item.innerText)).reduce((acc, val) => acc + val ) / 25

Thanks for this have been monitoring and using this to track and monitor as the sample set increases, I have calculated averages as they change. It appears in more recent results there is a more pronounced gap starting to appear, maybe as more of the background tasks etc. finish. Over the last 50 results there appears to be an 8.6% increase in single core and 6.25% increase in multi core.

May 15th	1038ng7 (i5) Last 100	1068ng7 (i7) Last 88
Single-Core	1136.85	1197.10
Multi-Core	4257.98	4418.25

May 15th	1038ng7 (i5) Last 50	1068ng7 (i7) Last 50
Single-Core	1147.83	1247.50
Multi-Core	4265.98	4532.96

 

[sdedalus1]

Thanks for this have been monitoring and using this to track and monitor as the sample set increases, I have calculated averages as they change. It appears in more recent results there is a more pronounced gap starting to appear, maybe as more of the background tasks etc. finish. Over the last 50 results there appears to be an 8.6% increase in single core and 6.25% increase in multi core.

I don't think recent entries are necessarily any more accurate, since they could just be new machines too—and I'd prefer a larger sample size over a smaller, more recent one.

Here's the current total with all results, and i7 has pulled away a bit (whereas i5 average stayed basically the same):

single core / multicore
i5 = 1142 / 4254 (309 results)
i7 = 1213 / 4447 (95 results)

i7 difference = +6.2% / +4.5%

I think we can have confidence in the numbers once we get enough i7 results that the average starts being consistent like the i5. And maybe i7s are just less consistent overall. Statisticians, weigh in!

And please keep posting Cinebench R20 scores, people!

 

[magbarn]

Here’s an interesting article Ice lake article
looks like ice lake 28w is underperforming compared to 15 watt models when power/performance is compared...

 

[sdedalus1]

Here’s an interesting article Ice lake article
looks like ice lake 28w is underperforming compared to 15 watt models when power/performance is compared...

Except that the article doesn't point out that those "15W" models can be run at *25W* according to manufacturer preference and cooling options. So I wouldn't expect a huge difference between 25W and 28W 10th gen chips.

It is disappointing that the Firestrike physics score of MBP's 28W i5-1038NG7 (9050) is 25% below 25W i5-1065G7 (11996). The 28W MBP chip should do at least the same or better. It could be thermally limited, but it's also only a sample size of one . . . And the claim of a similar 11th gen Tiger Lake CPU showing 40% gain is impressive, but also from an anonymous twitter post. I don't think we'd see those in MBP until spring 2021 in any case.

EDIT: Poking around the Firestrike results shows most i5-1065G7 physics scores in 7000–9500 range. Not sure what's up with that high score, seems an outlier.

 

[Spectrum]

I don't think recent entries are necessarily any more accurate, since they could just be new machines too—and I'd prefer a larger sample size over a smaller, more recent one.

Here's the current total with all results, and i7 has pulled away a bit (whereas i5 average stayed basically the same):

single core / multicore
i5 = 1142 / 4254 (309 results)
i7 = 1213 / 4447 (95 results)

i7 difference = +6.2% / +4.5%

I think we can have confidence in the numbers once we get enough i7 results that the average starts being consistent like the i5. And maybe i7s are just less consistent overall. Statisticians, weigh in!

And please keep posting Cinebench R20 scores, people!

Are you able to share a table of the raw values for each model?

 

[sdedalus1]

Are you able to share a table of the raw values for each model?

I only grabbed averages in groups of 25 using @supernaut42 's script, except when there were less than 25 on a page, then I just computed that odd number by hand. I then took the overall average of those sub-averages. Is that a valid method?

Spoiler: values

i7 averages:
1156.4 4304.6 (25 results)
1134.7 4291.1 (25 results)
1244.7 4584.8 (14 results)
1230.8 4384.4 (25 results)
1297.8 4671.8 (6 results)
average of averages:
1212.9 4447.3 (95 results)

i5:
1103.6 4048.2 (25 results)
1151.6 4235.2 (25 results)
1179 4320.8 (25 results)
1114.2 4184.2 (25 results)
1110.3 4219.6 (25 results)
1117.7 4267 (25 results)
1239 4448.3 (6 results)
1102.9 4111.7 (25 results)
1115.7 4165 (25 results)
1162.2 4333.9 (25 results)
1144.6 4266.8 (25 results)
1101.7 4224.1 (25 results)
1171.16 4299.4 (25 results)
1177.3 4436.3 (3 results)
average of averages:
1142.2 4254.3 (309 results)

 

[KPOM]

Yes thermals and power. Really regret the $200 I spent on the i7, but love the machine so much I'm going to keep it

It’s still about 10% faster in burst tasks. That’s about the same CPU difference between the 8th gen and 10th gen.

 

[wi11]

I pulled the last 50 benchmark scores for the i5 and i7 from Geekbench 5 yesterday morning and based on that:

Single core scores (i5/1038ng7 & i7/1068ng7)
Mean ± 1 SD: 1147.82±96.98 and 1247.48±90.81 (8.68% difference)
Median: 1165.50 and 1271.50 (9.09% difference)

For anyone who isn't familiar with these figures, the line in the boxes represents the median score and the "X" is the mean score. The edge of the box is the 25th and 75% percentile and the end of the whiskers are the most extreme scores that are not outliers, which are marked by dots.


Multi-core scores (i5/1038ng7 & i7/1068ng7)
Mean ± 1 SD: 4266.02±246.36 and 4472.96± (4.85% difference)
Median: 4357.00 and 4598.50 (5.54% difference)

The mean differences are statistically significant. The median scores will be influenced less by extreme scores/outliers than mean scores are; I am not up to speed on benchmarking, but it has been suggested above that some low scores could be due to indexing, therefore the median may be more useful for comparisons.


So the way I see it, the i7 is ~9% better than the i5 for single core tasks and ~5% better for multi-core tasks. Happy to be told otherwise as I don't really know much about hardware

 

[Spectrum]

I had some time on my hands during lockdown, so I just did a similar analysis but plotted as a rank-ordered CDF which gives a better illustration of the true 'shape' of the data distribution:
[i5 has 341 results; i7 has 99 results]

Pretty clear that on average an i7 has a ~100 point SC advantage;
but a much more variable MC advantage, ranging form zero to ~450 points.

Notably, when presented in this way (as a fraction of the data):
~90% of the i7 SC scores are greater than the i5 SC scores.
~70% of the i7 MC scores are greater than the i5 MC scores.

RECALCULATING THIS See post below here for this analysis:
i.e. Based on these results, if you were to randomly test any pair of i5 and i7:
~90% of the time the i7 would be faster for SC.
~70% of the time the i7 would be faster for MC.

For those interested, the CDF helps illustrate that the data is very non-Normally distributed, meaning that describing it parametrically (i.e. as if it were Normally distributed around a Mean ±Standard Deviation), does not accurately describe the distribution. When plotted as a CDF, Normally-distributed data (i.e. clustered around a mean value) would have a sigmoidal shape. The data actually have almost the opposite shape to that.

For example, if we look at the i5 MC scores, the upper 50% fall in the tiny range of 4340-4498 (158 points), whereas the lower 50% range is from 2644-4340, over 10x greater (1696 points).

The i7 MC scores are also skewed, although less so.

Enjoy!

----

EDIT: All calculated and plotted on a 2014 13" MBPro lol!

 

[sdedalus1]

Thanks @wi11 and @Spectrum ! The visualizations really help, especially to show how variable the i7 multicore results are.

@Spectrum , since you used a larger data set, are you able to compute median, mean, and % difference to easily compare with @wi11 's figures? Just curious if 50 vs. 100 vs. 350 results moves the numbers.

 

[Spectrum]

@sdedalus1: This is perhaps more useful to you:

Another way to look at this is to ask: "Based on these results, if I were to pick, randomly, a pair of i5 and i7, and run the test, how often would the i7 outpace the i5, and by how much?"

Here is that result:

The red bars indicate instances where the i7 score will come out (on average) faster than the i5, and by how much.

Summing up the red bars:
For SC scores, in 66% of comparisons the i7 will be equal or faster (in 34% it will be slower)
For MC scores, in 69% of comparisons the i7 will be equal or faster (in 31% it will be slower)

It would be great if @geekbench could provide a built-in comparison analysis like this. It is far more informative than just comparing the means.

---

But if you do want to compare the mean/median i7 vs i5 differences here they are:
Singlecore, mean=+66 points, median=+73 points
Multicore, mean=+186 points, median=+223 points

 

[sdedalus1]

@Spectrum Amazing, thanks!

So just to note that the larger data pool shows i7 mean & median ~ +5% for both single core and multicore, compared to larger increases @wi11 saw with only 50 results.

Agree, some of these other tools would be good to have on GB, but but least Geekbench has a lot of results and a way to browse them! I still think Cinebench R20 is a lot more meaningful for heavy workloads, but there's no database of scores to consult.

@Mattyman reported CB i5 scores of 1860–1885 here and @KPOM agreed. Previously Max Tech had the i5 CB score on video at ~1900, and I've see two other reports of i7s also getting ~1900 on CB. So i7s are zero to +2% on CB so far, but we only have a handful of results, unlike Geekbench.

EDIT: In sum, either
1) Geekbench needs to add an "extended" heavy workload test, or
2) Cinebench needs a browsable results page.

 

[Fowl]

All you statisticians, can I request a comparison of the Compute benchmark scores (OpenCL and/or Metal)? They appear impressionistically equal, and I'd be curious to know if there is some slight advantage to the i7.

 

[Spectrum]

Perhaps...
But I calculated the mean and median differences of all the pairwise comparisons, which (for medians) is not quite the same.

If I just calculate the means and medians of all the data:
i5:
SC Mean=1135; Median=1163
MC Mean=4233; Median=4343

i7:
SC Mean=1202; Median=1209
MC Mean=4419; Median=4527

i7 vs i7:
SC Mean delta = +66; Median delta = +46 (4%)
MC Mean delta = +186; Median delta = +184 (4%)

---

Anyway, the result stands: Assuming that these scores are representative of a larger sample, there is a modest increase in performance from the i7 of around 4-6% (on average).

Note, however, that whilst purchasing an i7 provides a 65-70% chance of it performing better than an i5. There is also a 30-35% chance that an i5 will outperform an i7.

Also...this all assumes we believe that the variation in these results are mostly due to variation in individual computers....

However, if the variation we see is more due to user error (background task etc), I'd be tempted to look more closely at the top end each distribution. At the top end it is clear that the i5 CPU maxes out at: SC=1270; MC=4500, whereas only the i7 has the ability to hit scores up to SC=1380 and MC=4900. Giving closer to a ~+10% performance delta.

Indeed ~35% of all i7 SC scores are greater than the highest i5 score.
And ~50% of all i7 MC scores are greater than the highest i5 score.

@Spectrum Amazing, thanks!

So just to note that the larger data pool shows i7 mean & median ~ +5% for both single core and multicore, compared to larger increases @wi11 saw with only 50 results.

[automerge]1589667855[/automerge]

All you statisticians, can I request a comparison of the Compute benchmark scores (OpenCL and/or Metal)? They appear impressionistically equal, and I'd be curious to know if there is some slight advantage to the i7.

Someone will have to write me a script to scrape the details of Geekbench...
I doubt there is much if any difference. It's the same GPU with a tiny difference in clock speed, so I understand.

Separate point to note: One place where the 13 inch MBPro will suffer in CPU, compared to the 16 inch, is in workloads that tax both CPU and GPU because they are on the same die on 13 inch, and will compete with each other for power and cooling.

 

[gl3lan]

This video compares last year's i7-8569u and this year's i7-1068NG7 with Geekbench and Cinebench. Geekbench scores are pretty consistent around 1350/4900 (3 runs). Cinebench scores (2 runs) reach 1992 and 2007 points respectively.

 

[JohnDoe12]

Hey everyone, I really appreciate all the discussion here. I think all the information and analysis is incredibly insightful and it helped me out a lot.

Previously I was dead-set on getting the i7, but now I’m not so sure.

I’m assuming from what’s being said here, I can take that i5=4500 and i7=4900. This isn’t a huge difference, but it’s significant enough for me to think “hmm, is this worth the $200 difference?”

However, I also realized that these benchmarks are testing turbo speeds. I don’t believe Geekbench does anything related to actual sustained performance. For my use case (photo editing, running terminal programs and scripts, some development work, running many applications simultaneously, light gaming such as Minecraft), I believe that sustained performance will be more important.

Are there any benchmarks that can describe anything about the sustained performance of these processors? From what I can tell, the only info we have about that is the 2.0 and 2.3 GHz clock speeds between the i5 and i7, which is a 15% difference. It’s bigger than the 9% difference between 4500 and 4900. I’m not really sure what to choose, and I’m helping that some of you with more insight can help me out here.

Once again, thank you all!
[automerge]1589705215[/automerge]
I just realized also that @sdedalus1 seems to be saying that the Cinebench results are more representative of full workloads, and it doesn’t seem like there‘s much of a difference between the i5 and i7. Well, looking at what @gl3lan just posted, it seems as though Cinebench results can go up to 2000 for i7. That’s a big difference from the 1900 that others posted here. I agree that it would be nice to have a database of results. Problem is, I’m hoping to order it later today. Shucks... looks like I’ll have to make a decision soon.

It’s also worthwhile mentioning that the i7 will take nearly two weeks longer to ship... I’m trying not to make that a factor in the decision, but it’s difficult after being stuck with my old laptop for so long, haha...

 

[gl3lan]

From what i could gather, it seems that both i5 and i7 are pretty similar when it comes to actual all cores turbo frequency (i5 seems to run around 3.0/3.1 GHz, while i7 seems to run around 3.2/3.1 GHz). Of course, the 15% base frequency (2.0 vs. 2.3 GHz) difference will only be seen when the GPU is 100% used (which is rare for software development).

It would be interesting to find out which sustained frequency is reached when it comes to 1 thread/2 threads turbo. Is the i5 (respectively i7) able to sustain 3.8 GHz (resp. 4.1 GHz) when only a single thread is used. Same question for 2 threads. According to cpu-monkey, the theoretical all cores turbo is of 3.4 for i5 vs. 3.6 for i7. It is obviously not reached when 8 threads are at 100% load (which is somehow disappointing), but I am curious about 1/2 threads actual turbo frequency. If 3.8/4.1 and 3.4/3.6 turbo frequencies can be sustained for fewer threads, the actual difference between i5 and i7 would be around 8% for 1 sustained thread and 6% for few more threads.

Could someone check actual sustained boost frequencies using Intel Power Gadget test menu for all possible configurations (Maximum Frequency vs. 2 Thread frequency vs. All Tread frequency, with or without GPU Maximum Frequency), for both i5 and i7 ?

 

[Spectrum]

This video compares last year's i7-8569u and this year's i7-1068NG7 with Geekbench and Cinebench. Geekbench scores are pretty consistent around 1350/4900 (3 runs). Cinebench scores (2 runs) reach 1992 and 2007 points respectively.

Great video. It So very useful to finally see someone doing a review where they repeat a benchmark. So often people report results (especially Geekbench test results) and assume they are stable. Geekbench has terrible consistency even on the same machine for some reason.

Anyway, this is a very good demonstration of the difference between the top end 8th Gen CPU (2.8/4.7 GHz turbo) and the new 10th gen CPU (2.3/4.1 GHz turbo). Very clear 10-15% performance advantage despite the 15% lower base and turbo clock speeds On the 10th gen.

The metal GPU scores are also a lot stronger (although OpenCL, not so much).

Not sure how much this helps decide between a 10th gen i5 or i7 though...🤪

 

[Spectrum]

Could someone check actual sustained boost frequencies using Intel Power Gadget test menu for all possible configurations (Maximum Frequency vs. 2 Thread frequency vs. All Tread frequency, with or without GPU Maximum Frequency), for both i5 and i7 ?

One way to test this would be to run Cinebench with varying numbers of threads (you can specify number of threads in preferences). And check the scores and stable CPU frequencies.

Alternatively, the cores can be disabled using the Instruments app, part of Xcode.

 

[JohnDoe12]

From what i could gather, it seems that both i5 and i7 are pretty similar when it comes to actual all cores turbo frequency (i5 seems to run around 3.0/3.1 GHz, while i7 seems to run around 3.2/3.1 GHz). Of course, the 15% base frequency (2.0 vs. 2.3 GHz) difference will only be seen when the GPU is 100% used (which is rare for software development).

Thanks! Where did you get that the all cored turbo frequency max out there for those CPUs? All I’ve seen so far are these Geekbench benchmarks.

 

[gl3lan]

Thanks! Where did you get that the all cored turbo frequency max out there for those CPUs? All I’ve seen so far are these Geekbench benchmarks.

here you go, 4 runs back to back
what do you make of it?

From that earlier post for i7, and from Max Tech YT video for i5. Now it seems that some i5s can sustain 3.3 GHz (1966 points on Cinebench R20 !):

 

[sdedalus1]

From that earlier post for i7, and from Max Tech YT video for i5. Now it seems that some i5s can sustain 3.3 GHz (1966 points on Cinebench R20 !):

Just when I thought the i7 is starting to pull ahead, the i5 bounces back!

 

[JohnDoe12]

Ended up getting the i7. For anyone reading out there: do make an informed decision about which one to buy. This was a gift for me and I begged the person gifting it not to get the i7 because it wasn’t worth it, but they ended up getting the i7 anyway. Wow...

In any case hopefully the i7 is somewhat worth the extra money they spent...