Max multicore frequency m3/i5/i7

[tley]

I would like to know the maximum sustainable frequency for the m3/i5/i7. You can test this by opening a terminal and typing the following: yes > /dev/null & yes > /dev/null & yes > /dev/null & yes > /dev/null

Then open the Intel Power Gadget and have a look at the Frequency window. The temperature should jump up til 100 in a matter of seconds, and then we will see what frequency the macbook can sustain with full load.

My 2017 i7 can sustain 2,5 GHz with around 100 degrees celcius. Please post your test results!

---

(cpu-world.com)
TurboBoost speed for rMB 2017
i7: 3,6 GHz single core, 3,4 GHz multi core
i5: 3,2 GHz single core, 2,8 GHz multi core
m3: 3 GHz singe core, ? GHz multi core

 

[EugW]

Thanks for posting that.

Using yes > /dev/null just once, it can sustain 2.6 GHz with Power of 6. I never saw it hit 3.0 GHz. Did yours go to 3.6 GHz? That reminds me of the speed of the 7Y30 not the 7Y32.
Typing in your entire command it, it can sustain 2.5 GHz with power of 10 but throttling with a temperature usually hovering around 95 to 96C. Sometimes a touch higher, but it never quite hit 100C.

Core m3, 16 GB, 256 GB SSD.

Interestingly, Intel Power Gadget says the thing idles at 1.3 GHz, not 1.2 GHz. Power near 0 and temp in low 30s.

 

[tley]

The highest I've seen is around 3,2 GHz and only for 1 second.

It seems like the i7 and m3 can sustain the exact same multicore frequency at 2,5 GHz under full load. Mine did 2,6 GHz single core as well. Also, the i7 idles at 1.3 GHz, not 1.4 GHz.

If I understand this currently, the m3 and i7 will perform identically for sustained full load. Did I miss something, or is this correct?

 

[EugW]

The highest I've seen is around 3,2 GHz and only for 1 second.

It seems like the i7 and m3 can sustain the exact same multicore frequency at 2,5 GHz under full load. Mine did 2,6 GHz single core as well. Also, the i7 idles at 1.3 GHz, not 1.4 GHz.

If I understand this currently, the m3 and i7 will perform identically for sustained full load. Did I miss something, or is this correct?

Interesting. So according to the Intel application, both the m3 and the i7 idle at 1.3 GHz, both go up to 2.6 GHz at single-core load, and both throttle to 2.5 GHz.

Obviously then, Intel's application isn't telling us the whole story. For example, there are clear and repeatable differences between the m3, i5, and m7 for Geekbench 4.1 results. If you look at the scores near the top for the various models, this is what you get, roughly.

m3: 3800 / 7000

i5: 4000 / 7400 (+5-6%)

i7: 4500 / 8400 (+18-20%)

Furthermore, Geekbench tells us that the m3, i5, and i7 start at 1.2, 1.3, and 1.4 GHz respectively as expected, and also that the m3 is indeed the 7Y32 variant.

 

[tley]

As far as I've heard, GeekBench pauses for about a second between its many different runs. That does not represent a sustained 100% workload (like fx exporting video), but rather a large number of bursts.

It would make sense that GeekBench scores higher with i7, if the CPU gets a second to cool down between the different parts of the run, and then is able to boost higher for the first second before settling in a 2,5 GHz. Try doing a Geekbench run, while monitoring the Intel Power Gadget. I'm pretty sure it will show significant spikes.

---

We could do some tests with seeing how long it takes to complete the same intensive workload, like encoding the same video. That should represent the processing power of m3 2,5 GHz with the i7 2,5 GHz.

I have the i7, 16GB, 512 GB SSD.

 

[EugW]

As far as I've heard, GeekBench pauses for about a second between its many different runs. That does not represent a sustained 100% workload (like fx exporting video), but rather a large number of bursts.

It would make sense that GeekBench scores higher with i7, if the CPU gets a second to cool down between the different parts of the run, and then is able to boost higher for the first second before settling in a 2,5 GHz. Try doing a Geekbench run, while monitoring the Intel Power Gadget. I'm pretty sure it will show significant spikes.

I don't have the MacBook with me. However, I suspected as much, considering when I run Geekbench on my 4.2 GHz Core i7 iMac, the fans never speed up. They remain silent at 1200 rpm.

We could do some tests with seeing how long it takes to complete the same intensive workload, like encoding the same video. That should represent the processing power of m3 2,5 GHz with the i7 2,5 GHz.

I'm surprised none of the reviews have done this yet. Video encoding easily causes my iMac to go to max fan speed. But boy is it fast!

A good test for the MacBook would be to take one of the longer 4K HEVC videos posted online and transcode it to another HEVC or something and time how long it takes. If the m3, i5, and i7 all throttle to about 2.5 GHz, then the longer the video, the closer the performance will be for the three. Shorter tasks would show the clearer differences between the m3, i5, and m7.

Also, as I've said before, the speed improvement of the i5 over the i3 does not seem very impressive. According to Geekbench, if you really want better burst speed over the m3, you should go straight to the i7. It seems the only reason to get the i5 is to get the 512 GB SSD. It would be interesting to see if other tests support my conclusion.

---

One thing though that stuck out to me, with our admittedly limited number of chips tested: My m3 never actually hit 100C in the max load test. Maybe if I had run it longer it would have, but it kept on fluctuating between 94-97 degrees, usually around 95C. Maybe I lucked out and got a good m3, or maybe it's just because my room is cooler than yours.

 

[Graham Perks]

My 2016 m5, 1.2GHz base speed, runs at 90º for 2.2GHz for a minute or so, then drops to 2.1GHz, with occasional drops to 2.0GHz.

Edit: i5 -> m5, thanks EugW

 

[EugW]

My 2016 i5, 1.2GHz base speed, runs at 90º for 2.2GHz for a minute or so, then drops to 2.1GHz, with occasional drops to 2.0GHz.

m5 you mean.

Yes, come to think of it I had also read somewhere else the 2016 m series throttle threshold was at 90C. Perhaps that is part of the reason for the big improvement in performance, with the 2017 m3 faster than the 2016 m5. Ten extra degrees -> 2017 throttles at 2.5 GHz.

I’m not sure how this applies to Geekbench though because I don’t think that causes the CPU to throttle.

EDIT:

Confirmed.

I ran Geekbench with Intel's Power Gadget in the background. Temperature peak trend was gradually upwards for the most part but the temp never went past 70C. As for Turbo Boost, it seemed to usually peak out at about 2.6 GHz, but I did see it spike once to about 2.75 GHz. Never to 3.0 GHz, if the graph is to be trusted. Perhaps the granularity of the graph is not fine enough. I think the default is every 100 ms. I'm not sure how to make sure of its logging feature unfortunately.

Geekbench score was over 6900 multi-core and over 3700 single-core. Without the Intel application running one can get a bit higher. That is consistent with reported results, which usually max out at a little over 6800 / 3800 on the 2017 m3.

 

[glenb2]

This has been bugging me. Using power gadget I too have never seen mine go above 2.6 at all either. Can anyone verify if the m3 really clocks to 3.0 ever?

 

[EugW]

This has been bugging me. Using power gadget I too have never seen mine go above 2.6 at all either. Can anyone verify if the m3 really clocks to 3.0 ever?

As mentioned in my last post, I saw one spike to 2.7-2.8 GHz on my m3. Not 3.0, but higher than 2.6. Mind you, I wouldn't really worry too much about it, because other benches do show the vast improvement in speed of the 2017 m3 over the 2016 m3 and 2015 m3 models.

It would be interesting for academic purposes to understand why Intel's application is behaving this way. Also, can someone tell us how to turn on its logging function?

 

[EugW]

I just learned that Intel Power Gadget is not yet optimized for Kaby Lake. So perhaps that explains some of the anomalies.

 

[tley]

For the logging function, you just click Cmd+L to start logging, and then Cmd+L to stop and save the log to file. Set the sample time as low as possible before you start logging.

Hopefully Intel will optimize the Power Gadget for Kaby Lake soon, so we can satisfy our academic curiosity

(Btw, if you find a good video and conversion tool, let's do the time test of m3 vs i7)

 

[moshen]

As far as I've heard, GeekBench pauses for about a second between its many different runs. That does not represent a sustained 100% workload (like fx exporting video), but rather a large number of bursts.

It would make sense that GeekBench scores higher with i7, if the CPU gets a second to cool down between the different parts of the run, and then is able to boost higher for the first second before settling in a 2,5 GHz. Try doing a Geekbench run, while monitoring the Intel Power Gadget. I'm pretty sure it will show significant spikes.

---

We could do some tests with seeing how long it takes to complete the same intensive workload, like encoding the same video. That should represent the processing power of m3 2,5 GHz with the i7 2,5 GHz.

I have the i7, 16GB, 512 GB SSD.

I came up with a longer sustained benchmark for my new i7 that's simple to do: https://forums.macrumors.com/threads/simple-long-cpu-benchmark-to-test-m3-i5-i7.2055398/

 

[EugW]

Interesting. So according to the Intel application, both the m3 and the i7 idle at 1.3 GHz, both go up to 2.6 GHz at single-core load, and both throttle to 2.5 GHz.

I installed the latest 3.5 release of Intel Power Gadget and redid the test on my 2017 m3 MacBook. It jumps to 2.6 GHz, then throttles to 2.4-2.5 GHz. CPU package power utilization is 11 Watts, and 9 Watts for IA. Idle is 1.3 GHz.

 

[streug]

Hi! I have the same m3 2017. The max frequency I have seen 2.6. I have checked on Sierra and High Sierra too. I suppose they have an artificial limit for the m3 processor. One of the reason could be no difference in performance with more expensive processors like i5 and i7. Because of that they have made the limitation. Or another explanation, the processor become hot instantly and it's not efficient. Apple does something that nobody knows... like with iPhones with old batteries

PS. Maybe it's even possible to go in a court with it, because it's written on the website:
1,2 GHz Dual‑Core Intel Core m3 (Turbo Boost bis zu 3,0 GHz) mit 4 MB L3 Cache

and where my 3.0 GHz of crazy performance of the 4.5W processor's?

PSS. It's not related to the topic but I wanted to ask about battery capacity, after 34 cycles and around 3 months I have around 5100 mAh from designed 5550. Is it fine? On my macbook pro 15 2013 late I had around 93-94% of the designed capacity after 4 years of using and 350 cycles. Thanks.

 

[Mike Boreham]

PS. Maybe it's even possible to go in a court with it, because it's written on the website:
1,2 GHz Dual‑Core Intel Core m3 (Turbo Boost bis zu 3,0 GHz) mit 4 MB L3 Cache

and where my 3.0 GHz of crazy performance of the 4.5W processor's?

I seem to remember reading somewhere that the max turboboost values are what a single core does. With more than one core the max turboboost is less. Will see if I can find where I saw this.
[doublepost=1514405120][/doublepost]

I seem to remember reading somewhere that the max turboboost values are what a single core does. With more than one core the max turboboost is less. Will see if I can find where I saw this.

Here its one place that confirms the 3.0 GHz turboboost only applies to single core:

https://www.theverge.com/2017/6/7/1...keyboard-intel-core-i5-i7-kaby-lake-processor

 

[streug]

Yes, Mike, you absolutely right about that, it's possible only when one core under heavy load, then we have to find one thread application which uses only one core and do the testing the idea here that it should be possible somehow to achieve 3.0 GHz ...

I have loaded only one core... and still 2.6 max, it's just a bad snapshot PS. two cores on the CPU History because of HT

 

[frings]

…

PSS. It's not related to the topic but I wanted to ask about battery capacity, after 34 cycles and around 3 months I have around 5100 mAh from designed 5550. Is it fine? On my macbook pro 15 2013 late I had around 93-94% of the designed capacity after 4 years of using and 350 cycles. Thanks.

Coconut Battery has a feature where you can compare your battery to others online. The average for 30 cycles is 5288.

 

[streug]

Coconut Battery has a feature where you can compare your battery to others online. The average for 30 cycles is 5288.

Thank you, frings.