Is geekbench a bad metric?

Why does geekbench seem to say the new iPhone has higher single core performance than my workstation?

Other urls found in this thread:

browser.geekbench.com/v4/cpu/4634452
browser.geekbench.com/v4/cpu/4634367
browser.geekbench.com/v4/cpu/4634339
browser.geekbench.com/v4/cpu/4712899
geekbench.com/doc/geekbench4-cpu-workloads.pdf
twitter.com/SFWRedditVideos

Because the A11 Bionic is by far the most powerful processor in any consumer device

Yes

Now, that isn't true. My home machine's single core performance is still much better.

How do you figure?

yeah whatever
I HIGHLY doubt your "self built anime consumption machine" is better than a device that's had the best engineers in the world spend billions of dollars designing it

It must just be faster because there's no history of manufacturers tuning performance towards benchmarking software...

1/10, put more effort into it next time.

Troll much? The CPU in my desktop is of course faster then the CPU in a smartphone.

That's what I'm asking. Is it a bad metric?

geekbench cpu benchmarks are synthetic.

...

Of course it's a bad benchmark.

Fact remains, for anything which fits in the cache it's amazingly fast. Really shows how much Intel has dropped the ball without any real competition.

thistbhfam

iPhones have like 8MB L1/L2 cache or some shit which heavily lopsides benchmarks in their favor, even when they're being utterly shat on by Snapdragons that are half the die size with half the cache.

Yes it is. It is poorly optimized for big core processors, much of the actual die goes unused because the work units are so small and basic.

Scaling from single to multi core scores is less than 100%, so two cores is less than 2x the single core score, four cores is less than 4x.
The A10 processor was a quad core processor with two performance cores and two efficient cores. The A10 couldn't use all four cores at the same time so Geekbench only used two cores when testing multi core performance. You can see this from these results
browser.geekbench.com/v4/cpu/4634452

The A11 is a hex core processor with two performance cores and four efficient cores. The A11 can use all six cores at the same time and, importantly, it can use them all for performance applications if necessary (read, if iOS determines it is a benchmarking application), see
browser.geekbench.com/v4/cpu/4634367

If you look at your Galaxy Note 8s they either have an Exynos or Qualcomm octa core processor with four performance cores and four efficient cores.
These processors can use all eight cores at the same time but Android will separate applications to either performance or efficient cores, based on how demanding they are.
So Geekbench only utilizes four of the eight cores in the multi core benchmark, and you can easily see this in the results
browser.geekbench.com/v4/cpu/4634339

Looking at the single core value in Geekbench4 for the A11 and scaling it the same amount the A10 scales for its two performance cores (about 60%) gets a score of a tad under 7000.
So comparing the performance clusters of the A11 and the 835 the result is pretty close, the A11 is just getting another 3000 points from adding in the efficient cores which don't get tested on Android.

As an addendum to this, looking at the scores for my Sony Xperia X
browser.geekbench.com/v4/cpu/4712899
We see that the multi core score is more than 2x the single core score. The processor is a hex core with two performance cores and four efficient cores.
Looking at perfmon during the benchmark you do see the efficient cores kicking in during portions of the multi core benchmarking, but not all portions and not always very high utilization (usually over 60% but only really hitting 100% for a couple of the tests).

I would be really interesting to see the core utilization on the iPhone side during Geekbench, but the general conclusion is still sound. With a multi core processor that has core clusters, the result for the performance cluster should be about the 85% of the single core result multiplied by the number of cores in the cluster.

If the multi core score deviates from that then it is loading the efficient cores for some of the multi core benchmarks.

Basically, Geekbench is terrible because you don't even know how many cores it is utilizing and how consistently it is doing so.

So what you're saying is that only the Single Core Scores matter? If that, that means iPhone btfos the 835 even harder.
No matter how you spin this, the iPhone 8/X is going to come out on top no matter what.

geekbench is only good when apples dont top the scores

I'm not saying only the single core matters, because the simple fact is that both platforms will take advantage of the multiple cores.

The problem is that the single core scores are (possibly) """"accurate"""", but the multi core scores are unreliable so the relative performance of the devices is impossible to deduce from the results.

Oh, and to be clear I'm not concerned with the A11 coming out on top, it's just the difference in performance between it and the other current flagships is probably far less than Geekbench is presenting.

yeah ask apple to release the ISA so that we can see if they are meddling with the benchmarks

which pretty sure they are considering the fact that in real life every iphone is always behind

The ABSOLUTE state of iToddler mental gymnastics.

>The GEMM workload uses 512x512 single-precision matrices

geekbench.com/doc/geekbench4-cpu-workloads.pdf

Small workloads begins on cache, simple ALU operations.

It's a bad benchmark yes. Changes are made to show apple cpus in the best possible light.

>use a cryptography workload
>score it only 5% because all modern SoCs accelerate AES
>use AES for the workload anyway instead of something that isnt accelerated so it is just a battle of hardware accelerator implementations

Because Apple is the best evar

ARM.

Is this real?