At the press conference that ended not long ago, Apple equipped the new 14-inch and 16-inch Macbook Pro models with M1 Pro and M1 Max chips. Although these two chips use TSMC's 5nm process as the M1, the chip-scale is several times higher.
The M1 Pro chip is equipped with up to 10-core CPU, up to 16-core GPU, plus up to 32GB of unified memory, and the maximum memory bandwidth is 200GB/s. The M1 Max is only equipped with a 10-core CPU, the GPU is up to 32 cores, and the unified memory is up to 64 cores. The maximum memory bandwidth is 400GB/s, which is close to 6 times the M1 integrated memory bandwidth.
Stacked 10-core CPU
Apple first compared the speed of its previous generation M1 chip, saying that the fully equipped M1 Pro and M1 Max CPUs have a 70% increase in performance compared to the M1, and the graphics processor has increased up to 2 times and 4 times, respectively.
Judging from the chip stacking alone, these two chips are really scary, and people have to guess whether Apple has drained TSMC's 5nm production capacity. The M1 Pro chip packs 33.7 billion transistors, while the M1 Max packs 57 billion transistors. If the official comparison chart is rendered in full scale, then the M1 Pro has a size of 12.8 x 18.5mm and an area of 236.8mm2, and the M1 Max has a size of 19.7 x 21.5 mm, which is 423.6mm2. Based on this estimate, the chip density of M1 Max and M1 are similar, both around 134 MTr/mm2, while the transistor density of M1 Pro is slightly higher, reaching 143 MTr/mm2. Judging from these figures alone, Apple has not reached the 173MTr/mm2 of TSMC’s N5 limit.
In the configuration of the CPU, the M1 Pro full configuration and M1 Max both use a combination of 8 high-performance cores and 2 high-efficiency cores. The data given by Apple’s official website is also compared with x86 8-core Intel chip notebooks. It is said that under the same power consumption level, the peak performance of this ten-core CPU is 1.8 times that of the former, and the power consumption to reach its peak level performance is less. 70%. Apple also gave the specific model of this comparative notebook in the remarks, which is MSI's GP66 Leopard notebook, the specific model is 11UG-018, which uses Intel's i7-11800H processor.
32-core GPU surpasses RTX3080?
The GPU is where the gap between the M1 Pro and the M1 Max widens. The M1 Pro's GPU has a maximum configuration of only 16 cores. Although the number of cores is twice that of the M1, it also achieves a performance ratio of three times. Compared with the aforementioned Intel i7-11800H processor core display, the fastest speed is more than 7 times. Compared with high-power notebook discrete graphics, Apple claims that its performance is not only stronger, but also has 70% lower power consumption. Here, Apple chose Lenovo's Legion 5 15ACH6 as a comparison. This notebook is equipped with Nvidia's RTX 3050 Ti graphics card.
M1 Max is not stingy about the number of GPU cores, stacking it up to 32 cores, and the graphics processing speed is 4 times that of M1. Compared with a large high-end laptop equipped with RTX 3080, the power consumption of a full M1 Max is 40% less under the same performance; compared with a thin and light professional laptop equipped with RTX 3080, the power consumption of a full M1 Max is under the same performance is less. 100W. This also results in better heating and fan control of the M1 Max, which greatly improves the battery life of the notebook.
It should be noted that when Apple conducts these performance comparison tests, it often tests post-production software such as Photoshop and Final Cut Pro, rather than complete graphics performance comparisons. Especially when it comes to video processing, both M1 Pro and M1 Max integrate dedicated acceleration hardware such as a media processing engine, so taking video transcoding speed as a comparison is somewhat "non-skillful".
Arm's lack of PC gaming ecology and the immaturity of Apple's Metal graphics interface also means that Apple is not enough to fully surpass Nvidia and AMD in terms of game performance. For general performance comparison, it is more appropriate to look at the computing power. Apple’s M1 Max GPU has a peak computing power of 10.4 TFLOPS, while Nvidia’s RTX 3060 has reached 12.7 TFLOPS, not to mention the RTX 3080. Therefore, the graphics performance of M1 Pro and M1 Max is relatively "specialized".
Memory bandwidth comparable to HBM
With the support of TSMC's 3D packaging technology, Apple once again integrated the memory in the SoC, achieving a unified memory of up to 64GB, which is common with video memory. Even more surprising is that Apple claims that the maximum memory bandwidth of the fully equipped M1 Max can reach 400GB/s. Such a large memory bandwidth can already be comparable to HBM.
Apple has prepared a very sincere chip lineup for the new Macbook Pro series, although most of this performance improvement is achieved through stacking chip scale. Under the dilemma of TSMC's postponement of the 3nm process, whether Apple can still play with 5nm depends on whether there are other killer features, otherwise, next year is likely to usher in a new product release of squeezing toothpaste.
Only one year has passed since the two-year Arm transition period set by Apple, and it will attack the second year with a strong attitude, and strive to complete the full replacement after the transition period. However, the performance ceiling Mac Pro series in Apple's lineup has not been updated to Arm processors. Whether this workstation-level product can also leave the market for Arm depends on whether Apple is willing to continue to stack materials.