Why is it important: The way calculations are constructed is changing. As everyone looks for ways to deal with the Moore’s Law slowdown, companies will have to move away from general-purpose chips like CPUs and GPUs. To squeeze more performance out of our hardware, we will need to create more integrated, complex solutions that tie hardware and software more closely together.

- Advertisement -

Once upon a time, all chip manufacturers specialized in the development of one type of chip: Intel made processors; Qualcomm made modems; Nvidia made GPUs; Broadcom (before Avago) made network chips. This age is over. The future of semi-finished products will be to develop more and more specific chips for even more specific applications. This change will take many years, but the transition has already begun. This will revolutionize the convenience food industry to the same degree that the consolidation over the past 20 years.

- Advertisement -

There are many reasons for this. It’s easiest to just say that Moore’s Law is slowing down, so everyone needs to find a new business model. But that doesn’t explain much, so let’s unpack it. In the hazy past before 2010, Moore’s law meant that chips got “faster” or “better” every two years or so. If some customer had a special-purpose chip they wanted, they could go and develop their own, but by the time they could put that chip into production, new processors were coming into production, and they usually turned out better than the intended purpose. built-in chip under design.

Editor’s note:

- Advertisement -

Guest Author Jonathan Goldberg is the founder of D2D Advisory, a multifunctional consulting firm. Jonathan has developed growth strategies and alliances for mobile, networking, gaming and software companies.

Then Moore’s law slowed down, we don’t have enough PhDs to say it’s over, but it definitely slowed down. So everyone now has to work a little harder to squeeze performance gains out of their silicon designs. Clearly, this has opened the door to all of the Roll Your Own chips made by hardware and hyperscaler companies, but the changes should go beyond that.

The whole point of a semiconductor is to run some kind of software. As we said, in the past we could get a performance boost for this software with denser chips, but now companies will have to take a closer look at the software side of the problem. Google rolled out his TPU because they wanted something that better ran their AI algorithms. They rolled out HCU for the same reason, and this chip was actually designed by software engineers. Same story with Apple and its M and A series processors. In all of these cases, the whole point is to optimize silicon for software.

Not everyone will be willing or able to make their own chips, and so we’re starting to see a lot of intermediate chips that aren’t the same type, generic for computing, and aren’t fully customizable. AMD’s recently acquired Pensando DPUs are a good example of such an intermediate step.

Once upon a time, data centers were warehouses filled with processors. Now they should contain GPUs, AI accelerators, unusual network loads, and a bunch of FPGAs. This is often referred to as heterogeneous computing, and is the opposite of the former homogeneity of the CPU.

These changes aren’t just happening in data centers. The very notion of “Edge Compute” is increasingly seen as an exercise in off-the-shelf and semi-custom chips appearing in a variety of places – in cars, factories and smart cities – to name but a few.

Ultimately, the big chip makers will have to decide how to respond to these changes. Making custom chips is not a good business, but developing semi-custom chips is full of risks, not least in choosing the right design, supporting them, and hoping they hit the mark.

Established companies are already beginning to position themselves for this, and for the first time in a decade, the door to startups is beginning to open.