Apple M3 and M3 Pro Performance Analysis: Should Intel, AMD, and Qualcomm be worried?

/

Less than two weeks ago Apple took the wraps off a new family of processors for its MacBook Pro and iMac devices, the M3. Now reviewers and consumers have them in their hands, at least ones powered by the M3 and the M3 Pro. (The M3 Max systems haven’t shipped quite yet.)

This launch is made even more interesting due to the pending onslaught of competition coming the PC space from Intel, AMD, and Qualcomm. Only days prior to the Apple announcement, Qualcomm had shown the Snapdragon X Elite SoC that will come in 2024, making some significant claims of performance relative to the best from its rivals. Heading out to my local retailer on Tuesday I picked up a pair of the new MacBook Pro laptops to do some testing and get a feel for how well these new devices compare to other systems powered by Intel and AMD, as well as my own M1 Pro-based MacBook Pro.

The results are incredibly interesting and paint a picture of where Apple silicon stands in the pantheon today, and also helps highlight some of the deficiencies (and highlights) of other processors. Consider this article a ‘quick look’ of sorts, consisting of a handful of tests that can be run consistently and repeatably, erring towards those that have cross-platform (Windows and Mac OS) capability.

System Configurations

  • M3

    • MacBook Pro

    • 4x performance core, 4x efficiency core CPU

    • 10 core GPU

    • 8GB system memory

    • 512GB SSD

  • M3 Pro

    • MacBook Pro

    • 6x performance core, 6x efficiency core CPU

    • 18 core GPU

    • 18GB system memory

    • 1TB SSD

  • M1 Pro

    • MacBook Pro

    • 8x performance core, 2x efficiency core CPU

    • 16 core GPU

    • 16GB system memory

    • 1TB SSD

Cinebench 2024

Cinebench 2024 is a new version of a well-known, and well-worn, benchmark that is primarily used to compare CPU performance. It offers both single threaded and multi-threaded results with a workload based on offline ray traced rendering from Cinema 4D. It runs for at least 10 minutes, looping the rendering workload to warm up the system and make sure it is measuring the sustained performance of the system, rather than any short-lived burst capability.

In the single threaded results, which test the capabilities of just a single core of the processor, the new M3 and M3 Pro are a solid step up from the M1 Pro. It’s interesting to see where the Snapdragon X Elite comes in, not quite at the performance of the M3 but very close, and well ahead of the options from this class of Intel’s 13th Gen and AMD’s 7000-series processors.

If we look at scores on a relative scale, using the M1 Pro system as the baseline, you can see the patterns a bit more easily. The Core i7-1360P result is the lowest performer, 12% behind the M1 Pro. The new M3/Pro SoCs are 30% faster than M1, and should provide a nice speed up for consumers in terms of responsiveness, where single threaded performance has the most impact.

If we open these systems up and enable Cinebench to use all the cores and threads on the system, things look very different. The winner here is the Snapdragon X Elite, though of course I realize we aren’t testing a real-world retail machine yet. It’s 12-core design is faster (at the higher TDP configuration) than the 12-cores of the M3 Pro. The M3 Pro does show off its capabilities compared to the M1 Pro and base M3. It’s eye catching to see the Intel Core i7-1360P, with its 4 P-cores and 8 E-cores, coming in at the back of the pack.

If we convert these to a relative scale, we can again more easily decipher some of the comparisons. The Snapdragon X Elite is more than 50% faster than the M1 Pro, and almost 20% ahead of the M3 Pro! The base M3 is 9% slower than the M1 Pro which will be a talking point for anyone considering an upgrade from any M1-based MacBook. I touched on it above, but seeing the Intel platform at just about half the performance of the M1 Pro system that was released in late 2021 is concerning.

Geekbench 6.2.1

Geekbench is another performance measurement tool that is extremely popular due to its ability to scale with CPUs and that it can test across platforms. It is quite different from Cinebench though in that it consists of 16 unique CPU sub-tests that look at algorithms common for file compression, PDF rendering, text processing, background blur, photo filters, and more. The GPU scores have 8 different sub-tests too, from face detection to gaussian blur to particle physics and more.

The results for the single threaded tests in Geekbench are showing a similar trend as the Cinebench scores. The new M3 and M3 Pro (using the same new core architecture design) are noticeably faster than the M1 Pro. The Snapdragon X Elite system scores aren’t quite as fast as the new Apple designs but are close, and still ahead of the performance of the x86 competition from Intel and AMD based on these results.

Moving to a relative performance view, Geekbench shows that the M3 and M3 Pro have a 30-35% single threaded advantage over the M1 Pro, again, a great indicator of responsiveness of a processor. The Snapdragon X Elite is as much as 27% faster than the M1 Pro here, depending on the TDP configuration.

The Geekbench multi-threaded results are interesting and differ quite a bit from the Cinebench nT results. The stacking of the M1 and M3 processors is basically the same, but in Geekbench the Intel and AMD platforms are much more competitive with the M1 Pro and the base M3 products, a good sign for the competitive landscape in 2024. The Snapdragon X Elite looks very good, not as fast as the M3 Pro, but outpacing the 1360P and the 7840U.

If we normalize to the M1 Pro again, you’ll see that the Intel Core i7-1360P is only 9% slower than it, NOT 50%! The M3 Pro is 33% faster than the M1 Pro AND the base M3 – a hefty uplift despite a lot of online concern.

Blender Render Benchmark 3.6

Blender is a very popular open-source rendering tool that has become another staple in the performance measurement and benchmarking communities. It allows you test both the CPU and GPU in a PC to measure the sample rate of the Cycles engine, the higher result the better. I couldn’t get reliable results for the 1360P and 7840U CPUs, so the Blender testing here will only show the Apple silicon comparisons.

Because the base M3 system that I have for testing here only has 8GB of memory (the others have at least 16GB), the Junkshop workload appears to be hitting some limits where it is thrashing the SSD, significantly hindering the performance. The M3 Pro is roughly 30% faster than the M1 Pro in the Monster workload, but is 80% faster in the Junkshop workload.

That memory limitation gets worse for the M3 MacBook Pro with 8GB here as it was unable to complete the benchmark process. Because of the unified memory design, if you are planning to do any kind of creative work on a new M3-based MBP, I would avoid any 8GB configurations. But look at the performance uplift for the M3 Pro over the M1 Pro! This is the promise of the included hardware ray tracing capability on Apple’s new GPU architecture, resulting in a nearly 3x increase in Blender performance. (Though I didn’t show it, the GPU test of Cinebench 2024 is over 3x faster on the M3 Pro than the M1 Pro because of hardware ray tracing acceleration on the new GPU.)

3DMark Wild Life Extreme and Solar Bay

Graphics and gaming testing continues to be difficult due to the lack of significant gaming software compatibility of the Mac ecosystem. That said, a quick, sanity checking using a pair of 3DMark tests shows some interesting results. These tests were run in their “unlimited” mode that enables consistent resolution and refresh rate comparisons, essentially letting us measure chip vs chip as opposed to platform vs platform.

Wild Life Extreme is a great cross-platform test that runs on the Vulkan API for Windows and the Metal API for Mac OS/iOS. We have results for it across the entire set of CPUs, and in these results the M3 and M3 Pro look to have a noticeably faster integrated GPU than the Intel Core i7-1360P or the AMD Ryzen 7840U. The Snapdragon X Elite is almost as fast as the base M3 processor, but is just about half the performance of the M3 Pro.

The Solar Bay test is a relatively new one that measures ray tracing performance, and we only have results from the Apple devices here. The M3 Pro is 90% faster than the M1 Pro in this test thanks to its bigger GPU and the new architecture that adds hardware ray tracing acceleration. Even the base M3 system is running out ahead of the M1 Pro here.

I expect the dynamics of the integrated graphics world to change in early 2024 as Intel releases its first Meteor Lake GPUs that will integrate Intel Arc graphics on-chip. How much performance that actually turns into has yet to be seen, but at the company’s architecture day it was making claims of “2x performance” which would put in the neighborhood of the Snapdragon X Elite and the Apple M3.

Closing Thoughts

This initial look at the performance of the Apple M3 and M3 Pro silicon shows us a couple of interesting trends. First, the single threaded performance of the new CPU cores of the M3 family are about 30% faster than the M1 cores in my testing across Cinebench and Geekbench. That is a considerable uplift in just a couple of years, and despite some online discourse, should bring value to new MacBook Pro buyers.

For the multi-threaded performance things are more complicated as Apple made some core count changes generation to generation. So, while the M3 Pro is roughly 35% faster than the M1 Pro in our testing, it does so with two fewer performance cores and four additional efficiency cores. The total core count goes from 10 to 12, and the net result is both faster single and multi-threaded performance, so I really don’t understand the scale of the complaints from some in the community.

And though I don’t have any M2 products in the testing here, the fact that the base Apple M3 products are basically on par with the multi-threaded CPU performance of the M1 Pro is a win for anyone looking to jump into an Apple silicon MacBook Pro for the first time. And if you care about single-threaded performance, which you should, the base M3 is roughly 30% faster than the M1 Pro.

The new GPU on the Apple M3 Pro is very impressive, showing 40% better standard rendering performance in 3DMark Wild Life Extreme and 90% faster performance when testing ray tracing with Solar Bay. And if you are using this M3 Pro for rendering work, the ray tracing performance in our Blender testing is nearly 3x faster than M1 Pro! Now if only the Mac ecosystem was more friendly to gamers…

Looking at competitive CPUs in the context of the new M3 and M3 Pro from Apple, I think we are setup for a really interesting next few quarters of action in the PC space. The Snapdragon X Elite from Qualcomm, though I have not been able to spend time with a system myself yet and we aren’t expecting retail availability until mid-2024, still looks VERY strong in a world with the Apple M3. Single threaded and multi-threaded performance appears to more than hold up to the new Apple products and it will be up to Microsoft and Windows take advantage of the chip’s capabilities. We haven’t even started diving into the performance of AI workloads, but that is something I expect to do soon, and it seems likely that the Hexagon NPU on the Snapdragon part will again lead.

The current market share leader is probably doing a bit of hang-wringing looking at these results. The performance of the Core i7-1360P is behind both the Snapdragon X Elite and the Apple M3 family. Yes, Intel makes higher performance CPUs that go in very different classifications of systems (H- and HX-series parts), but in the world of “thin and light but still powerful,” Intel has a lot riding on the CPU and GPU performance of Meteor Lake, now branded Intel Core Ultra, and set to launch in December.

*Results from PC World testing. **Results from NotebookCheck, XDA, mobile01, Cnet.