According to the great Apple news guru, Mark Gurman, Apple would prepare a Mac Pro where only the storage memory (SSD) would be modular. A risky approach in the case of a business PC where raw performance and scalability are key.
The Mac Pro currently in the making in Apple’s labs could be the least modular high-powered computer Apple has ever launched. According to the famous Apple specialist, Bloomberg journalist Mark Gurman, the Californian company has decided to apply the same “all-integrated” design of its current Macs to its future professional machine. Understand here: solder the random access memory (RAM) as rumors already reported. But also and be content with the graphics processor (GPU) integrated into the processor without being able to call on an external GPU. Which would make this hypothetical Mac Pro an extremely frozen machine in time.
The next Mac Pro may lack user upgradeable GPUs in addition to non-upgradeable RAM. Right now Apple Silicon Macs don’t support external GPUs and you have to use whatever configuration you buy on Apple’s website. But the Mac Pro GPU will be powerful with up to 76 cores.
— Mark Gurman (@markgurman) January 26, 2023
If this is information leaks, not an official company announcement. Nevertheless, this change in the “parallel” narration of the genesis of the future supercomputer Apple lets take a look at the pros and cons of Apple’s embedded design. While dissociating the needs of the general public from those of professionals.
The benefits of fully integrated
From a purely hardware perspective, the way Apple’s M chips are designed has several advantages. If we dismiss the benefits for Apple (better control of the machine, more margins, simplification of designs, etc.), there are also benefits for users. It is necessary here to recall the specificity of Apple chips. The first is the unified memory design soldered close (within mm) to the compute chip – an SoC that integrates CPU, GPU and NPU, like mobile chips from Intel and AMD. The all-in-one processor therefore shares the memory for all uses. Except that in the case of the Pro, Max and Ultra versions, the size of the GPU part is enormously larger than the chips of the competition. So much so that it amounts to having a dedicated RTX3060/3070 type graphics card. One of the current limits in the increase in performance being the latency of information transmission from one chip to another – the GPU loads an element in memory, the calculation, the sending in the memory of the CPU which takes it in account, etc – the unification of memory in a large pool where everyone can draw can speed up many transfers.
The other advantage is to be found on the side of energy consumption. If the current Macbooks are the machines with the best performance/watt ratio, it is also because the CPU+GPU entanglement and the memory alongside it avoids the time and thermal losses of the round trips to the memories. A very important plural here: a machine with a dedicated GPU has its dedicated RAM. Who says therefore more chips, also says more energy consumption. All of this is well and good in ultraportable, or even portable, machines. But the energy situation is less significant in large computing towers. A world with many fields of applications, some of which are very sensitive to the amount of RAM.
The (big) advantages of modular
If AMD, Intel and Nvidia have not yet succeeded in catching up with Apple in the area of performance per watt ratio, these three players still largely dominate Apple in pure performance. While highly efficient, even Apple’s biggest GPU can’t outrun an Nvidia RTX4090 or any other professional graphics processor. Ditto for the processors, where a huge AMD EPYC 9474F and its 96 cores and 250 MB of L3 cache memory (more than $7,000 including tax per processor!) at 360W peak is necessarily more powerful than the Apple chip. The limit being in particular to be sought on the side of the integrated whole, where the heat dissipation – of which the power is an expression – cannot be distributed within a system.
Read also: Mac Pro: by switching to Apple Silicon chips, will it lose its main assets? (January 2023)
Depriving yourself not only of a CPU, but above all of a modular GPU can be quite a handicap for the scalability of the machines. While CPU performance jumps from one generation to another are rarely greater than 15% at the same range, GPUs always benefit from performance jumps greater than 50%. If according to Mr. Gurman, the Apple chip could benefit from additional power by pushing the frequency further thanks to more efficient cooling than that of a Mac Studio, Apple’s room for maneuver remains limited.
That will leave storage as the main user-upgradeable component in the new Mac Pro, which will have the same design as the current, Intel model. The big difference between a Mac Pro and a Mac Studio — in addition to M1 Ultra to M2 Ultra — should be performance from more cooling.
— Mark Gurman (@markgurman) January 26, 2023
Next comes the inability to increase RAM. If some applications prefer less memory, but very fast, others are very (very) fond of RAM. Having enough room in memory to maintain models, databases and others in a space faster than the SSD is sometimes necessary, especially for simulations and scientific modeling. And as for the graphics card, having the margin to support a very expensive machine for two or three more years by changing the GPU, or by increasing the RAM, is an argument for buying a professional machine. . Much more than the beauty of the case or the performance-per-watt ratio.
If we must remember that this is an analysis of a rumor – and therefore take tweezers – we must also add that we know nothing of Apple’s strategy. A company that is generally very good at targeting customers with its machines. Maybe the volume and effort that goes into maintaining a line of upgradable machines is no longer in his interest. The answer during the official announcement of the future Mac Pro!
Bloomberg’s Mark Gurman (Twitter)
