Nice work! Did you choose a degree-four polynomial by doing a log-log plot and finding you got a good fit to a straight line with a slope of four? Would you be willing to share this data? I'd like to play with some curve fitting myself.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Power curve and efficiency of Apple N3
- Thread starter leman
- Start date
- Sort by reaction score
I simply fit a regression model, fourth degree polynomial was the best at minimizing the residuals and going higher didn’t improve the fit. But the model is likely over-fitted, it can certainly be done better (the very small amount of data doesn’t help either). A more informed model (e.g. one uses known electrical engineering facts) would be a better idea.
I’ll publish the data shortly, just want to clean up the scripts…
Based on @leman's curves.
Which was my point. If Apple didn't clock the M2 higher, maybe it's because the curves are not what we think they are.
The curve just shows a trend. It tells nothing about actual physical limitations of the chip. If you start getting leakage above certain threshold because your circuits are not designed to handle the voltage, or synchronization issues because the logical blocks require certain timing range, that’s your limit, even if the trend suggests you still have a lot of headroom.
The only reason why I think the fitted curve might be suggestive of a wider power range for this core is because it appears flatter than that of previous products. But even that might be wishful thinking, with overfitting and all.
This is one of the most interesting threads I've seen here in a long time. It's interesting to get a glimpse of what's happening beneath the surface of our devices and try to understand how things are developing. I think that people need to tone down their expectations for what gains can be made by node shrinks moving forward. The ever decreasing sizes are causing a lot of issues for the chip designs that make simple scaling gains obsolete. But it does indeed look like Apple has been optimizing the P-cores for higher performance with the A17.
I added a few more data points to the git repository and can add more SoC generations later if there is any interest in going further back in time. I know I'm interested in that too.
I added a few more data points to the git repository and can add more SoC generations later if there is any interest in going further back in time. I know I'm interested in that too.
I dont think these curves really mean much, it means in a limited performance scenario, but in a real world usage typically the A17 will boost to 3.78 Ghz the moment you open an app, or load a webpage, enter in an Instagram profile, etc. And I doubt the first to idle will win in those cases it will consume more power than an A15 for example.
Waiting for that DXOmark battery review
Waiting for that DXOmark battery review
1) Zen uses larger transistors, as does any design that is striving for higher frequency. Apple gets its IPC wins from using lots of transistors, which means density and smaller transistors.
2) What do you think critical path IS? Why can't I run critical path faster? Because critical path cycle time is determined by the sum of the switching times of the sequence of transistors that make up the critical path!
You can see some discussion of these elements here: https://www.realworldtech.com/fo4-metric/
although I think *everyone* would agree that 6 to 8 FO4 is insanely low, you do much better overall by using a slightly longer cycle length that allows for at least some degree of superscalar/OoO/speculation smarts, and taking the IPC win over the frequency loss.
I searched for some numbers, in Geekbench 5 and GFXBench doing a simple math my Snapdragon 870 scored better than A17 Pro in efficiency, but I dont think thats fair, It works in unlimited performance scenarios but in limited performance scenarios things are different, so I looked at 3DMarks Wild Life extreme stress test.
15 Pro Max average 2750 points, takes 10% battery
275 points/% battery
My SD 870 average 1250 points, takes 8% battery
156 points/% battery
15 PM numbers are From a review, the 870 numbers are mine
15 Pro Max average 2750 points, takes 10% battery
275 points/% battery
My SD 870 average 1250 points, takes 8% battery
156 points/% battery
15 PM numbers are From a review, the 870 numbers are mine
As explained multiple times, this is not how you work out efficiency.
Good grief, how many accounts does this guy have?
It’s as fast as 12900K using 6x times less power. If that’s not impressive I don’t know what’s supposed to impress you.
What EXACTLY do you want to calculate?
Work/Joule is the correct definition of "efficiency", but it's basically useless because doing something slower always uses less energy, but no-one is interested in 1Hz CPUs.
(Work/Second)/Joule gets you to something called the Energy-Delay product which IS useful, a frequent figure of merit, along with the Energy-Delay-Squared product (which emphasizes performance more than energy, unlike the more "balanced" Energy-Delay product; both have their place).
Many amateur metrics of the form "rate"/"energy" (like this one by PgR7) actually intuit this and essentially get it right; he's giving a (reciprocal) energy delay product, which is the thing a sensible person basically wants. Of course you can quibble about the issue of battery percentage, and there are issues there in terms of battery age or varying battery size. But in my experience the sort of people who obsess over the last decimal place of these details are people who miss the wood for the trees, who are basically worthless as engineers because they have no idea WHY they are measuring something, and what it does or doesn't tell you.
Judging by the sentiments of many people, it’s supposed to be faster than everything, power consumption be damned, (unless it uses more power than last gen, in which case Apple has lost their talent and is doomed)
The highest point might not matter much in real life on a portable device for the reasons you mention. But the curve definitely does. It's where you end up for a given power draw, the lower the better, and the newer chips are lower for all power draws as one would hope.
Nice cherrypicking. Not are you only cherrypicking an outdated Intel chip, the comparison doesn’t even make sense.
To judge the 3nm performance, you need to compare it to the 5nm chip of the previous Apple Silicon version. And it doesn’t look impressive.
I expected alot more from the 3nm die shrink.
It’s a smartphone CPU that’s faster than M2 Ultra while using 25% less power… like what did you expect?
We are discussing the efficiency/performance of the A17 on N3. This person is running 3dmark and measuring a battery percentage drop over that time. That’s fine as a rough measurement of phone efficiency, but doesn’t yield much information about the efficiency of the A17 unless we know what’s happening to the other components.
The CPU is only 10% faster going to 3nm. And the 3nm chip doesn't even consume less power. You call this impressive?
First you pick an outdated Intel chip. Now you pick a M2 Ultra. None of these comparison you make, make any sense.
Compare the same chip but then before and after the die shrink. That is the comparison you should look at.
The outdated flagship intel chip which is *checks notes*
12% slower in single core and one gen older. And uses more power to do it.
And the A17 is 15% faster than the A16… and is a disappointment.
According to this site, it's only 10% faster and they used geekbench. GPU is 20% faster, but Apple added an extra GPU core. So it's not really one-to-one comparison.
Geekbench shows how much faster the iPhone 15 Pro's A17 chip is
The new iPhones are finally here, and when it comes to the iPhone 15 Pro and iPhone 15 Pro Max,...
9to5mac.com
If only a 10% performance gain at the same power consumption is what you expected from the 3nm chip, then you are easy to please.
Reviewers of the iPhone 15 Pro expected also alot more from the 3nm chip, so I'm not the only one.
So power consumption is a factor for Apple, but not others?According to this site, it's only 10% faster and they used geekbench. GPU is 20% faster, but Apple added an extra GPU core. So it's not really one-to-one comparison.
Geekbench shows how much faster the iPhone 15 Pro's A17 chip is
The new iPhones are finally here, and when it comes to the iPhone 15 Pro and iPhone 15 Pro Max,...
If only a 10% performance gain at the same power consumption is what you expected from the 3nm chip, then you are easy to please.
Reviewers of the iPhone 15 Pro expected also alot more from the 3nm chip, so I'm not the only one.
Then we wouldnt have all those comments of people writing about overheating more than the 14 PM, people are using it the same way than a 14 PM and they say it gets hotter and the battery life is worse, how do you explain that?, the truth is that the person that loads lots of webpages, goes from Instagram to Twitter and Youtube constantly, are using the burst performance, not the limited performance figures where you get those efficiency numbers. Its like you are talking about a car that in the speedway going 60 mph have very low fuel consumption, but when there is traffic it burns a lot more fuel than a normal car with the constant accelerations, and the thing is your usage is 70% city, so the lowered fuel consumption in the highway dont make up for the increased consumption in the city.
Im 100% sure that for that extra frequency they increased voltages, if they just released an A17 with the same frequency than the A16, but with lower voltages thanks to the 3nm process, then yes, you would get lower power consumption in every scenario and we wouldnt be arguing here. It would be a superb chip