So there will have been no M3! The 2024 iPad Pros that Apple has just presented have skipped this generation of chips to directly benefit from the brand new M4 chip from the Cupertino company. It is in fact the first product of the brand to integrate this SoC (system on-a-chip in other words “all-in-one chip”), while it traditionally inherited it from MacBooks. With two generations of iPad Pro having M chips (M1 and M2), Apple has hindsight on the integration of this chip in a tablet and has not feared integrating it into a device of only 5 mm d ‘thickness.
This recklessness comes on the one hand from the method of manufacturing the chip: designed around the second generation of N3 (3 nm) engraving from TSMC, the logic part of the M4 is much more finely engraved than the M2 chip that it replaces. A miniaturization which allows Apple to put more transistors – 28 billion compared to 20 in the M2 and 25 in the M3. And to promise not only that the CPU part of the M4 is 50% faster than that of the M2, but also and above all that it can deploy the same power for half the energy consumed.
A quick tour from the owner of Apple’s new chip – and undoubtedly that of the next MacBooks!
A much more efficient CPU
We mentioned it above: the CPU is more powerful and more efficient than the previous generation. But its design is a little different from the M2 and M3. While these two previous iterations were organized around two quartets – x4 P-Core dedicated to high performance, x4 E-Core with lower energy consumption – Apple pushed the number of efficient cores to x6, thus boosting the most parallelized applications.
In both cases (and as usual), Apple does not provide any information on frequencies, cache memories or the internal organization of the CPU cores. But the firm does, however, specify that these two types of cores have units for accelerating algorithms from machine learning (ML for Machine Learning). Understand here certain classes of algorithms pompously called AI by companies. Because yes, and as we will see later, it is not just the NPU or the GPU that work for the AI algorithms!
GPU: arrival of ray tracing in iPads
The graphics chip (GPU) of this SoC contains as many graphics cores as the highest-end variants of the previous M1 and M2, namely ten cores (the MacBooks and iPads with less RAM had only eight GPU cores). For the moment, it is impossible to know if each of its cores is, as for the M1 and M2, always made up of 16 execution units (EU) each. But Apple royally claims up to a fourfold increase in performance! What is this miracle about? A word : ray tracing (ray tracing in French).
For the first time in fact, an iPad integrates hardware units of ray tracing, a portmanteau containing several methods of photorealistic 3D image rendering. If the ray tracing has been known for ages in the professional 3D rendering industry, particularly for films and animations, it is only in recent years that it has found its way into our electronic devices. But in the case of the M4’s GPU – as in that of PC chips – it’s actually a range of tricks to improve performance. Invented and popularized by Nvidia’s GeForce graphics cards, these tools are generally based on hybrid calculations which mix rendering of a scene at a higher level of quality, but in a much lower definition.
The A17 Pro chip in the iPhone 15 Pro is Apple’s first chip to support ray tracing material. A month later, the M3/M3Pro/M3Max followed suit. © Apple
Behind it, we find a whole suite of tools for enlarging and improving image quality based on the famous AI. Given the example displayed by Apple – a 3D rendering engine called Octane rasterization pure (the front-end rendering method), but rather a use of rendering units ray tracing and support for any algorithm accelerator machine learning (oh, as luck would have it, there are some freshly integrated into the CPU!) to ensure this level of performance.
All this to tell you that specific applications, notably 3D rendering engines and games, will be able to see their execution speed greatly accelerated by this new GPU. However, developers will need to implement their support in their software.
AI on the rise, but not on the cutting edge
With a displayed power of 38 TOPS (tera operations per second, or trillions), the NPU of the M4 is twice as efficient in matrix calculations as the NPU of the previous M2 (and the M2 Pro and Max), which does not showed only 15.8 TOPS on the clock.
While the M3s only brought a minor improvement in the area (18 TOPS), here Apple goes the second to (a little more than) double the number of operations per second.
Qualcomm’s Snapdragon © Adrian Branco for Les Numériques
The optimization work was very intense on this side, since if the NPU is still made up of 16 execution cores like the M3, doubling the performance per core is a real challenge that the engineers at Apple. But this only allows the company to maintain itself, not to dominate. Because, although a pioneer in the integration of NPUs in SoCs, Apple is for once a little behind in terms of TOPS announced compared to Qualcomm.
Its Snapdragon X Elite will indeed display no less than 45 TOPS on its own, compared to 75 for the entire chip!
Don’t worry about a performance deficiency, though. NPUs are in fact only one of the AI accelerators of SoCs. In addition to certain tasks that CPUs handle better, it is mainly GPUs – sometimes with a helping hand from the NPU – that are the workhorses of AI acceleration. NPUs are currently more oriented towards the execution of algorithms in a sustained manner (blurring and microphone noise reduction in real time for example).
What about the M4 Pro/Max/Ultra processors?
You don’t need to be a great cleric to suspect that the new iPad Pros should offer a very noticeable extra power compared to the previous generation. Many journalists and professionals have questioned the relevance of integrating such a powerful chip into a tablet with an OS more limited than a classic macOS for example. But the continued progress of iPad OS, the presence of numerous professional software in the ecosystem and the fact that the iPad range has other less powerful references validate Apple’s strategy of offering a professional tablet.
How does the tablet version of this chip foreshadow the chips integrated into future MacBooks? If we refer to the previous versions of the classic M1 and M2, we are entitled to expect exactly the same chip, pushed higher in frequencies, undoubtedly available with one or two fewer GPU cores for entry-level machines. of range. Apple could even disable one or two CPU cores for these same purposes – which could improve manufacturing yields, by allowing the use of dies (i.e. the raw chip) that have defects.
Regarding the M4 Pro and Max versions, if we refer to what Apple has done in the past, we can already count on the same NPU, without any change – only the Ultra chips, which are literally two Max chips stuck together, see an arithmetic doubling of units. The real difference will therefore be, in addition to the frequencies undoubtedly increasing thanks to the better heat dissipation of computers, an increasing number of CPU and GPU cores. But in what proportions?