Monday, November 4th 2024
Apple M4 Max CPU Faster Than Intel and AMD in 1T/nT Benchmarks
Early benchmark results have revealed Apple's newest M4 Max processor as a serious competitor to Arm-based CPUs from Qualcomm and even the best of x86 from Intel and AMD. Recent Geekbench 6 tests conducted on the latest 16-inch MacBook Pro showcase considerable improvements over both its predecessor and rival chips from major competitors. The M4 Max achieved an impressive single-core score of 4,060 points and a multicore score of 26,675 points, marking significant advancements in processing capability. These results represent approximately 30% and 27% improvements in single-core and multicore performance, respectively, compared to the previous M3 Max. This is also much higher than something like Snapdragon X Elite, which tops out at twelve cores per SoC. When measured against x86 competitors, the M4 Max also demonstrates substantial advantages.
The chip outperforms Intel's Core Ultra 9 285K by 19% in single-core and 16% in multicore tests, surpassing AMD's Ryzen 9 9950X by 18% in single-core and 25% in multicore performance. Notably, these achievements come with significantly lower power consumption than traditional x86 processors. The flagship system-on-chip features a sophisticated 16-core CPU configuration, combining twelve performance and four efficiency cores. Additionally, it integrates 40 GPU cores and supports up to 128 GB of unified memory, shared between CPU and GPU operations. The new MacBook Pro line also introduces Thunderbolt 5 compatibility, enabling data transfer speeds up to 120 Gb/s. While the M4 Max presents an impressive response to the current market, we have yet to see its capabilities in real-world benchmarks, as these types of synthetic runs are only a part of the performance story that Apple has prepared. We need to see productivity, content creation, and even gaming benchmarks to fully crown it the king of performance. Below is a table comparing Geekbench v6 scores, courtesy of Tom's Hardware, and a random Snapdragon X Elite (X1E-00-1DE) run in top configuration.
Source:
Tom's Hardware
The chip outperforms Intel's Core Ultra 9 285K by 19% in single-core and 16% in multicore tests, surpassing AMD's Ryzen 9 9950X by 18% in single-core and 25% in multicore performance. Notably, these achievements come with significantly lower power consumption than traditional x86 processors. The flagship system-on-chip features a sophisticated 16-core CPU configuration, combining twelve performance and four efficiency cores. Additionally, it integrates 40 GPU cores and supports up to 128 GB of unified memory, shared between CPU and GPU operations. The new MacBook Pro line also introduces Thunderbolt 5 compatibility, enabling data transfer speeds up to 120 Gb/s. While the M4 Max presents an impressive response to the current market, we have yet to see its capabilities in real-world benchmarks, as these types of synthetic runs are only a part of the performance story that Apple has prepared. We need to see productivity, content creation, and even gaming benchmarks to fully crown it the king of performance. Below is a table comparing Geekbench v6 scores, courtesy of Tom's Hardware, and a random Snapdragon X Elite (X1E-00-1DE) run in top configuration.
20 Comments on Apple M4 Max CPU Faster Than Intel and AMD in 1T/nT Benchmarks
Also, Geekbench, lmao.
And of course, not the usual it can decode some 4k, 8k format, real world projects means heavy color grading with masks, multiple cameras editing and in photoshop real world editing with filters that usually have no idea what to do with M3 max or M4, filters that use GPU to detect faces, objects, denoise...etc.
Probably you won't see that.
Also, a full s.o.c. @ $2600. Nice try.
I would really like, if benchmarks would include all possible platforms, even if the results can be problematic. Sure, it's not fair if single platform gets lower score because code is not optimized for that platforms. But if that is the case in 10 benchmarks, that is valuable information to me as a user. It's also fun to know, if a new phone can match performance of 5 year old desktop PC. How long before we get docking stations for phones and phones that can run desktop OSes?
546 GB/s memory bandwidth of M4 Max is impressive. Apple is using 8 memory channels while desktop PCs are stuck with 2 memory channels. Big LLMs need very high bandwidth. In simplified terms, for each word LLM generates, it needs to read the whole model. So, if LLM size is 50 GB, a 500 GB/s of memory bandwidth will allow to generate up to ~10 words/s. That is, if you also have enough compute power.
From real world experience in Davinci Resolve i can tell you nothing touches an RTX 4090 coupled with some decent CPU, not important since GPU is used most.
On set editing is also a thing
You have to love the callous marketing.One can argue that 2 modules of DDR5 modules in PC, are in fact Quad Channel, since 1 module is Dual Channel internally...
- It's >500mm² 3nm N3E vs AMD's 140mm² 4nm N4P + 100mm² 6nm N6.
- It has 512-bit memory interface on 3nm with 546GB/s of bandwidth vs 128-bit memory interface on 6nm with 96GB/s.
- Geekbench 6.3 added niche ML SME extensions which slightly inflates the score relative to its lead in SPECint 2017 1T since no compiler will emit these instructions for SPECint tests. Compare, for example, Geekbench 5.5 where Zen 5's performance is overstated such that it matches some of the M4 series in 1T composite score.
- M4 Max cannot be overclocked. A memory-tuned 9950X can easily score >3600 on GB6.3 1T composite, since it's rather memory speed sensitive.
As a consequence of #1 and #2 the cheapest M4 Max machine is $3700. A price where you can buy an RTX 4090 and 9950X PC.The point I was trying to make is that M4 Max has about 4x more memory bandwidth than the latest AMD or Intel based PCs.
My guy. That editing is clearly happening on a Windows PC. just look at the taskbar. :laugh:
So you just proved the other guy's point. ;):D:toast:
I guess for cutting any NLE will do but when you get to serious color grading you need that very expensive Nvidia Quadro and RTX 4090's, those threadrippers even if you don't fully utilize them, just because they have a ton of money and don't have the time to mess around with ARM cpu's.
I was curios why apple is so energy efficient so i browsed the web for information, it's not because the 3nm node or whatever exclusive deal they have to be first with TSMC latest, or that all in one chip design, it's mostly because of ARM instruction set and while is very efficient, not everything works, they mostly don't work.
So x86 carry a large instruction set so everything works from the "beginning of time" while ARM can't do that unless someone adapts for ARM, if it can be done.
I even know a vfx studio that stuck with the Mac when everyone moved to Nvidia,and somehow made it works. Professional is such a loose term, depending on your field, and the level at which one work, you don’t need a « pro » beefy computer. Where I’m working we have a Mac Studio, to handle heavy graphic design tasks, but we also have bottom of the barrel’s Lenovo PC’s plugged to 200 000 € worth of printing gear because the software isn’t available for MacOS, and you don’t need that much power when you just have to prep a file to be printed on something not that big.