Geekbench is pretty much useless, especially across architectures and OSes. It claims a maybe three-watt iPhone CPU can beat a 40-watt Ryzen 1200, which is plainly ridiculous. Apple makes a good CPU, but you'd have to be crazy to think they have beaten the competition by a factor of ten in power consumption with no compromise. When you see something incredible, be incredulous.
Does anyone know of a decent iPhone benchmark? I'd be very interested in comparing their hardware to other CPUs, but none of the benchmarks I know run on iPhones. And the people who know what they're doing, e.g. Phoronix, don't test much mobile hardware.
> It claims a maybe three-watt iPhone CPU can beat a 40-watt Ryzen 1200, which is plainly ridiculous.
I've seen this claim before. The only way it makes sense to me is that the Apple chip has a much more powerful integrated GPU than then AMD one.
I suspect that the GPU in CPUs meant for computers are more optimized for things like yield, since anyone looking for performance can just buy an expansion card. Phone SOCs don't have this luxury, so they're probably pushed to the performance limit when it comes to graphics
To be clear - this is just a guess - but it's the only one I can think of that makes sense.
Many ARM SOCs use a "BIG.little"[0] CPU technique where they can swap from a high-power dual- or quad-core section of the CPU to a low-power quad-core section.
> Inside the iPhone 8 and iPhone 8 Plus, there's a new six-core A11 Bionic chip... There are two performance cores that are 25 percent faster than the A10 Fusion chip in the iPhone 7, and four efficiency cores that are 70 percent faster.[1]
They've added 2 low-power cores, and made the cores faster than the previous version. Normally, only one set of cores could operate at once.
> The A11 features an Apple-designed 64-bit ARMv8-A six-core CPU, with two high-performance cores at 2.39 GHz, called Monsoon, and four energy-efficient cores, called Mistral.[1][6][4] The A11 uses a new second-generation performance controller, which permits the A11 to use all six cores simultaneously,[8] unlike its predecessor the A10.[2]
Here, the 2 high-power and 4 low-power cores can all operate at once, alongside the 3-core GPU - 2 cores of which can assist ML tasks too.
> There's a new Apple-designed 3-core GPU that's 30 percent faster than the previous-generation GPU, and two of the cores, the Neural Engine, make machine learning tasks faster than ever.[1,ibid]
Not sure how much power it consumes in that mode, but it's reported to use half the energy of the A10.
> In terms of graphics performance, the GPU is said to be 30% faster than last year’s, while consuming half the power when working at the same rate as the A10.[3]
The transistor count of an A11 is 4.3 million, the Ryzen has about 4.8 million depending on the type. And x86 is a bit wasteful in terms of what it does effectively with those transistors because the front-end required to translate x86 instructions.
And x86 is a bit wasteful in terms of what it does effectively with those transistors because the front-end required to translate x86 instructions.
That's a tiny amount in comparison to all the other pieces like cache, ultrawide SIMD units, and bus interfaces; and given that it acts as a sort of "decompressor" interfacing the slower memory containing denser instructions with the faster core and its wide uops, I'd say those transistors are quite well spent.
AArch32 is about as compilcated as x86 to decode. About 1200 instructions with dozens of different encodings, and instructions that aren't aligned to instruction width and can straddle a page and cache line boundary.
This may be part of the reason why no A11-based devices support AArch32. iOS 11 dropped support for 32-bit apps, and its highly possible the chip doesn’t support the ISA at all.
You claim GeekBench's scores are wrong, but you provide no evidence that A) they are wrong or B) an alternative benchmark that contradicts them. So, why again do you know that they are wrong besides the fact that you find them "incredible" and therefore they must be "incredulous?"
Geek bench does not perform long running tasks where you would observe thermal throttling . So a gaming benchmark would work. 3dmark ? https://benchmarks.ul.com/3dmark
Anandtech test with Basemark OS II which is cross platform, although they only compare Android and iOS devices and haven’t published an article for the iPhones 8 / X yet sadly. Here’s the iPhone 7 review they did though and comparisons against similar gen Android devices: https://www.anandtech.com/show/10685/the-iphone-7-and-iphone...
Are geekbench scores meant to be used comparing different CPUs architectures like that? I thought they only made sense when comparing already similar processors.
https://browser.geekbench.com/processors/2051 https://browser.geekbench.com/ios_devices/50
Does anyone know of a decent iPhone benchmark? I'd be very interested in comparing their hardware to other CPUs, but none of the benchmarks I know run on iPhones. And the people who know what they're doing, e.g. Phoronix, don't test much mobile hardware.