Power vs Performance CPU and GPU

[dada_dave]

Hi all, I know it’s been awhile but I’ve been working hard on this project and I think it’s finally in a pretty good shape to share. My previous bubble graphs were done in Numbers and getting difficult to manage. Updating with new data was a pain as was changing which data to display. The graphs themselves were thus becoming visually cluttered and each graph had to be hand crafted. To remedy this, I decided to learn Plotly, which is a Python API for creating html-based interactive graphs. Since to share those interactive graphs requires html, I also decided to learn some html and create a website so everyone could play with the figures themselves:

https://powervsperformance.tiiny.site/

The first part of the website explains the graphs and why I created them and at the bottom are the plots themselves (or just the link to them if viewing on a small window/screen). I find the plots work best on the computer, but an iPad, tablet, and even phone will work. I really wouldn't recommend the phone as the primary way to interact with the plots, but it does technically work.

I'm publishing an article analyzing the latest Apple M5 and Qualcomm X2E chips using these plots here on techboard. Link below for when it goes live:

Apple M5 and Qualcomm X2E Analysis

In this article, I’m going to mostly focus on the power and performance of the latest Apple M5 (especially Pro/Max) and Qualcomm Snapdragon Elite X2E chips across three benchmarks: Cinebench R24 ST (CPU), Cinebench R24 MT (CPU), and Cyberpunk 2077 1080p Ultra (GPU). The data as always comes from...

• dada_dave
• May 22, 2026
• Category: Apple

 

[Cmaier]

wow!

 

[Jimmyjames]

Amazing work.

 

[leman]

Excellent stuff and a very interesting writeup! I think it's very interesting how the data illustrates Apple's design goals.

 

[dada_dave]

One interesting thing to note about the M5 results. For the CB R24 ST efficiency the M3 to M5 has remained pretty steady, maybe a slight decrease in the M5, but at least from the M4 generation the Pro and Max seem unaltered (no data for the M3 generation). If we were to look at clock speed however, the clock normalized performance, the Score per GHz, has increased by roughly 25%.

M5 score, clock = 200, 4.61Ghz
M3 score, clock = 141, 4.05GHz

M5 score/clock = 200/4.61 = 43.384
M3 score/clock = 141/4.05 = 34.815
43.384/34.815 = 1.246

There are a few possible factors here, in order of what I think is the most likely effect size:

1) These are the processors top clock speeds. That doesn't mean they stay at them the entire run time and the M5 may be staying closer to its top clock speed for longer than the M3 does. Thus, I am undercounting the clock speed ratio above.

2a) The M5 core is likely bigger in silicon than the M3. Despite the advances in TSMC nodes, the M5 may be drawing more power per clock than the M3 is.

2b) The differences in clocks (top or realized) is drawing a bigger power draw than we might expect given the advances in TSMC nodes between them, especially if 1) is true and the difference in actual realized clocks during the benchmark are larger than the difference in top speed clocks*.

Not sure about whether 2a) or 2b) is more likely to have a larger effect.

What do you guys think? Am I missing another possible explanation?

*EDIT: Actually node difference between TSMC's N3P and N3B seems relatively modest for increasing performance at the same leakage, even according to TSMC's own estimates:

TSMC's 3nm update: N3P in production, N3X on track

The last and final FinFET node from TSMC lives on.

www.tomshardware.com

So maybe that's just it, whether realized or top, Apple tacked on extra GHz than the N3P could deliver without also requiring a boost to power draw which set back Apple's efficiency to baseline (and indeed maybe that's a target ST efficiency for them, just as there as been clearly a target MT efficiency the last few generations).