That seems to mostly be that the new chips aren't interacting well with Linux rather than Geekerwan's measurement methodology being bad in general (i.e. affecting iOS/Android). Unfortunately with the video's poor subtitles and Andrei's vague comments it's not clear to me if he's complaining that Geekerwan's measurements were off for Lunar Lake or Snapdragon or both. In the video, the Geekerwan host does indeed complain that the Linux version for Lunar Lake/Snapdragon wasn't fully ready and seemed to have problems with power states in particular which Andrei also brings up. My guess is that Andrei is complaining about missing Snapdragon performance. Looking at Geekerwan's graphs for Snapdragon, I have to say that the SPEC Int results in particular didn't track with my expectations for the Oryon core, but obviously my expectations aren't data so I'm not sure how meaningful that is. Also, some of the subtitles also suggested that Linux wasn't yet fully compatible with Lunar Lake either.
I'm just not sure how meaningful that is, compare M2 MacbookPro and M2 mini results:
| Power is subtracting idle | CB R23 ST (perf/W, W) | CB R23 MT (perf/W, W) | Witcher 3 (only W) |
| M2 MacbookPro | 219.6 pts/W, 7.2 W | 385.8 pts/W, 22.67 W | 31.8 W |
| M2 Mini | 213.8 pts/W, 7.7 W | 414.6 pts/W, 21.16 W | 26.3 W |
The Witcher 3 results are the furthest at 20%, but in favor of the Mini not the laptop (the M2 Air did better than both on the Witcher 3 though it has to be said it may have throttled, worst on CB R23 MT, and best on CB R23 ST). If the battery of the MacBook Pro was significantly impacting these results, this shouldn't be the case, the Mini should always be less efficient. Now obviously I can't rule out that every laptop behaves itself under every circumstance, but I don't often see anything that stands out to me. Of course it *could* happen as is always something to watch out for when testing a device on battery. To make extra sure you could test with the battery removed for laptops (that used to work anyway), but I don't know anyone who does that.
==============================================
[URL unfurl="true"]https://www.notebookcheck.net/Snapdragon-8-Elite-fails-to-match-A18-Pro-in-CPU-efficiency-Dimensity-9400-lags-even-further-behind.906617.0.html[/URL]
The user comment in this one is new (to me) ... I've heard Apple's version of Clang described as quirky given that it rarely matches a single mainline Clang version, but it is never been described as super-optimized relative to standard Clang. I've certainly never found it to be so. In general, my understanding and limited experience is that GCC will tend to produce faster code than Apple's Clang and occasionally so will mainline Clang because sometimes Apple's Clang can be a little more out of date than what you can get off the mainline branch. Obviously most of the time they'll be pretty similar. EDIT: Okay I mean I think there are a couple of defaults that are different and those can produce different, may more optimized results if you don't turn similar flags on for standard Clang, but not none of them are close to what Intel's ICC does, especially for SPEC.
==============================================
[URL unfurl="true"]https://www.notebookcheck.net/Snapdragon-8-Elite-crushes-A18-Pro-in-GPU-performance-while-trading-blows-with-Dimensity-9400.906935.0.html[/URL]
Adreno GPU and Dimensity outperform Apple's A18 Pro - both have 12 cores and my guess is clocked similarly (and unlike the rumors that the Dimensity would be clocked at 1.6Ghz, it must be closer to 1 in practice). They went with wider and slower designs. Better performance, better efficiency, more expensive. Interestingly the Adreno GPU outperforms the Dimensity in OpenCL which may have been the case the last generation too.
