Question Laptop SoCs: Qualcomm vs Apple vs AMD vs Intel

[mikegg]

Starting a new thread on laptop SoCs. These already happen in architecture specific threads but there isn't a general vs thread. I think this topic warrants a thread because competition has heated up. AMD and Intel are very serious about trying to push out better laptop SoCs, while Qualcomm has entered the competition in a strong way.

I've provided some baseline performance figures for the discussion.

Cinebench R24 ST

• M3: 12.7 points/watt, 141 score
• X Elite: 9.3 points/watt, 123 score
• AMD HX 370: 3.74 points/watt, 116 score
• AMD 8845HS: 3.1 points/watt, 102 score
• Intel 155H: 3.1 points/watt, 102 score

Cinebench R24 MT perf/watt

• M3: 28.3 points/watt, 598 score
• X Elite: 22.6 points/watt, 1033 score
• AMD HX 370: 19.7 points/watt, 1213 score
• AMD 8845HS: 14.8 points/watt, 912 score
• Intel 155H: 14.5 points/watt, 752 score

Sources:

1. Notebookcheck: X Elite Youtube review video (don't want to paste link here because forum automatically turns it into a media).
2. https://www.notebookcheck.net/AMD-Z...and-Qualcomm-Snapdragon-X-Elite.868641.0.html

 

[poke01]

I feel like this thread will get really active next month.

 

[jdubs03]

There are some others that we can consider like the R9 8945 H/HS, Core Ultra 7 185H. But it won’t be as neat as in the source link comparison.
I’m also assuming the X Elite is the X1E-84-100 SKU.

 

[WelshBloke]

I mean which one you get is going to depend what software you want to run.

The performance is close enough that you can stick with your preferred OS/software.

 

[FlameTail]

Cinebench R24 ST

• M3: 12.7 points/watt, 141 score
• X Elite: 9.3 points/watt, 123 score
• AMD HX 370: 3.74 points/watt, 116 score
• AMD 8845HS: 3.1 points/watt, 102 score
• Intel 155H: 3.1 points/watt, 102 score

Cinebench R24 MT perf/watt

• M3: 28.3 points/watt, 598 score
• X Elite: 22.6 points/watt, 1033 score
• AMD HX 370: 19.7 points/watt, 1213 score
• AMD 8845HS: 14.8 points/watt, 912 score
• Intel 155H: 14.5 points/watt, 752 score

Those numbers are very interesting. Apple/Qualcomm are stomping on AMD/Intel in ST Perf/Watt, but the delta is significantly narrowed in MT Perf/Watt.

I wish Geekerwan would test power curves for X Elite, so we can get a fuller picture.

Wish their website had power curves for laptop chips too:

SOCPK - 移动芯片排行

socpk.com

 

[mikegg]

Those numbers are very interesting. Apple/Qualcomm are stomping on AMD/Intel in ST Perf/Watt, but the delta is significantly narrowed in MT Perf/Watt.

You can make up for MT efficiency having more cores and lowering the wattage.

I'd curious what AMD's CPU die size is compared to Qualcomm's X Elite CPU die size.

 

[Hitman928]

Those numbers are very interesting. Apple/Qualcomm are stomping on AMD/Intel in ST Perf/Watt, but the delta is significantly narrowed in MT Perf/Watt.

I wish Geekerwan would test power curves for X Elite, so we can get a fuller picture.
View attachment 105337View attachment 105338View attachment 105339View attachment 105340View attachment 105341
Wish their website had power curves for laptop chips too:

SOCPK - 移动芯片排行

socpk.com

Here you go. I added the lines for X-Elite and 370 myself to better visualize the curve/trend, removed the M3 line, and added the CPU info to the M3 line. The reason some of the dots jump around so much is because they are the same CPU tested in different laptops, which also shows why trying to compare mobile CPU performance is so difficult, as the laptop makers have a decent amount of control over how the CPU performs and it can vary quite a bit by model.

 

[poke01]

We got some results of the latest Geekbench AI benchmark courtesy of Max based on the latest from AMD and Apple and Qualcomm so far.
Very interesting results

https://twitter.com/x/status/1824133490360914051

 

[Doug S]

A couple of those numbers jump out out me. How is the M3 Max doubling up the M3? Does it have a bigger NPU? If so, I wasn't aware that was the case. And what's with the large variation in the X Elite fp32 results when the fp16 and int8 seem much closer?

It also seems rather fishy that the M4's NPU is able to match the RTX 4090 in int8. I know Nvidia's strength is fp, but really??

 

[Hitman928]

A couple of those numbers jump out out me. How is the M3 Max doubling up the M3? Does it have a bigger NPU? If so, I wasn't aware that was the case. And what's with the large variation in the X Elite fp32 results when the fp16 and int8 seem much closer?

It also seems rather fishy that the M4's NPU is able to match the RTX 4090 in int8. I know Nvidia's strength is fp, but really??

No, you are correct, the 4090 should be blowing the M4 out of the water in all 3 tests. The M3 max results are peculiar as well, it increases performance on all 3 tests, but only fp32 sees a huge increase which doesn't really make sense.

Edit: It looks like the OS is making a huge difference. The top M3 entry is on the same OS as the M3 Max and only the fp32 is faster and now by just over 10%. Maybe some kind of bug in the other OS or in the test on that OS? I don't know.

 

[poke01]

Edit: It looks like the OS is making a huge difference. The top M3 entry is on the same OS as the M3 Max and only the fp32 is faster and now by just over 10%. Maybe some kind of bug in the other OS or in the test on that OS? I don't know.

macOS 15 and iOS 18 had optimisations done so NPU performance is faster in these versions.

iPhone 15 Pro Max Neural Engine up to 25% faster with iOS 18 in early benchmarks - 9to5Mac

We learned that Apple Intelligence will require an A17 Pro chip or later at WWDC last week. And now that...

9to5mac.com

https://twitter.com/x/status/1803072948632469857

 

[poke01]

On iOS 18.1 beta 2 the M4 is even faster. Really interesting to see the difference updates make. It does make sense since iOS 18.1 is the AI update.

https://twitter.com/x/status/1824187885517750573

 

[Doug S]

Wow I wasn't aware the OS was making that big of a difference in the AI results. Obviously the ML libraries are still seeing major optimizations (maybe not as much for Nvidia, since they've been putting a lot more resources into this for longer than anyone else)

Makes it kind of hard to make any sort of realistic comparison between platforms when they might see a 50-100% jump from an update, so I suggest people should temper their AMD vs Intel and Apple vs Qualcomm and so forth horn tooting if their platform of choice is in the lead, as it sounds like that can change overnight from a software update.

Gonna have to scan these GB AI benchmark results when cited/compared pretty carefully to insure it is apples to apples (NPU vs NPU, not NPU vs GPU or CPU) and watch the OS versions as no doubt there will be plenty of both accidentally and deliberately unfair comparisons that aren't giving each "side" the same benefit of showing itself in the best light.

 

[ikjadoon]

I'd love to see standby battery life comparisons, too, though it pushes into SoC / platform (vs CPU), but I don't think the distinction is too relevant for consumers.

Disappointed not one reviewer has added it to their methodology. An overnight is more than enough and I assume most reviewers sleep. Just mark it before you go to bed, wake up, mark it again, divide, voila.

 

[Doug S]

I'd love to see standby battery life comparisons, too, though it pushes into SoC / platform (vs CPU), but I don't think the distinction is too relevant for consumers.

Disappointed not one reviewer has added it to their methodology. An overnight is more than enough and I assume most reviewers sleep. Just mark it before you go to bed, wake up, mark it again, divide, voila.

Who cares how much battery AI uses? It isn't relevant for running AI models, and if you're training a model you won't be using a phone or a laptop running off battery.

 

[FlameTail]

Who cares how much battery AI uses? It isn't relevant for running AI models, and if you're training a model you won't be using a phone or a laptop running off battery.

I don't think they were talking about an AI battery test...

 

[mikegg]

I'd love to see standby battery life comparisons, too, though it pushes into SoC / platform (vs CPU), but I don't think the distinction is too relevant for consumers.

Problem with any battery life comparisons is that all laptops are built differently with different components. Furthermore, most AMD and Intel SoCs will throttle as soon as you unplug it. You can run a loop browsing websites and come up with a battery figure but if the websites load much slower under battery life, you won't see it in the battery-only figures.

 

[mikegg]

4k Aztec High GFX

* AMD 890M: 39.1fps
* M3: 51.8fps
3DMark Wild Life Extreme

* AMD 890M: 7623
* M3: 8286
Power:

* AMD 890M: 46w
* M3: 8286: 17w

M3 about ~253% more efficient.

But of course, if your goal is gaming, AMD's GPU will still be better because of Vulkan, DirectX, and Windows support. In pure architecture, AMD is quite a bit behind Apple.