if I recall correctly, the geekbench score does run on small bursts and it's designed to find the peak performance without taking the thermal limitations in account.

SPEC, on the other hand, takes hours to run.

Of course the iPhone chip isn't as beefy as the M1, but the results still speak for themselves.

https://www.anandtech.com/show/16226/apple-silicon-m1-a14-de...

Apple has since explained that M1s are slightly different between the Air, Pro and Mini, accounting for the different thermal chassis. (In the case of the Pro they enable an 8th GPU core.) It sounds like they are three different chips rather than the same chip in different configurations --I think he said that in marketing speak. https://www.youtube.com/watch?v=2lK0ySxQyrs

Apple makes it clearer that in the real world, these machines are only going to offer their incredible performance on Metal, iPad/iPhone apps and for any Mac apps that happen to have been ported over to M1 by their developers (using Xcode). These machines will only offer similar performance to existing Intel Macs when running existing Intel Mac apps because the incredible performance will be reserved for Apple's Rosetta2 software to make those unmodified apps compatible.

But what went unsaid, except during the part where they say they 'learned from their experience in the past processor transitions', is that by introducing the chip at the low-end of the lineup first, they create a market for the (few remaining relevant) Mac developers to invest in porting their code over to ARM and likewise, because these new machines run iPad apps at full speed on upto 6K displays, there is incentive for the iPad/iOS-only devs to expand the functionality beyond what their wares can do on a tablet/phone. (Any Mac dev that drags their feet porting may find that there are 50 iPad apps that now run fullscreen performing 75% of their functionality, costing them sales in the big volume accounts where they buy licenses by the thousands.) Meanwhile, the type of users who can get by with two USB ports, 16GB of RAM and a single external monitor probably don't run many third-party Mac apps and are going to have an awesome experience with the iPad apps and Apple's native apps.

Hence why Cinebench is often used these days when evaluating real-world performance with sustained workloads.

Different tests. Different purposes.

GB deliberately avoids running up the heat because it is focused on testing the chip, not the machine's cooling ability.

Cinebench, as you say, tests "real-world" conditions, meaning the entire machine, not just the chip.

The chip's ability to run at sustained load is a part of its design also. Precisely because modern chips has to throttle in order to meet power and thermal envelopes, we should be looking at sustained performance as a more accurate measure.

In a majority of cases, burst performance only affects things like responsiveness, and those things should be measured instead for a better reflection of the benefits.

If you perform an integrated test, would you not perform unit tests? An unit test may show areas for easy improvement if other aspects of the total package are changed.

For example, if someone thought M1 was thermally constrained, they might decide to rip mini out of the case and attach a different cooling method.

Not saying that burst performance shouldn't be measured, but it shouldn't be the de-facto go-to performance measure like it is now with geekbench.

If you run only unit tests, you don't get useful data.

> they might decide to rip mini out of the case and attach a different cooling method.

99% of customers will never do this.

That's not true at all about Geekbench.

"Geekbench 5 is a cross-platform benchmark that measures your system's performance with the press of a button. How will your mobile device or desktop computer perform when push comes to crunch? How will it compare to the newest devices on the market? Find out today with Geekbench 5"

I don't see anything in this quote that discounts the parent.

The new R23 release even does multiple runs by default. Excitedly waiting for results for the M1 to start popping up now that its released and has support.