Adobe Reportedly Considering Manufacturing Own Processors
One of the most significant — and understated — transitions in enterprise computing in recent years has been the number of companies building and deploying their own hardware. A number of high-profile firms have designed custom solutions using a mixture of FPGAs, custom ASICs, and conventional x86 CPUs in customized server deployments. Some firms, like Amazon, have taken out an ARM license and deployed that company’s silicon, seeking efficiency advantages or cost savings in specific markets.
But according to Axios, Adobe is actually mulling over designing its own CPUs or at least positioning itself to write software much closer to the silicon its code runs on. Longtime reporter Ina Fried attended an Adobe internal innovation conference at which Adobe CTO Abhay Parasnis posed the following question:
Do we need to become an ARM licensee?” he said, referring to the company whose underlying chip design is used across a wide range of devices, including computers, servers and phones. “I don’t have the answer, but it is something we are going to have to pay attention to.”
Later, Parasnis referred to using ARM’s software model as a means to “package its technology much closer to silicon.”
This is not, at heart, a crazy idea. In fact, the concept has been explored for years. It’s possible to create dedicated circuits for specific functions or to pair dedicated processing hardware with a general purpose CPU. As a pioneer in the “Auction off an organ of your choice to pay our annual license fees” business, Adobe would undoubtedly love to link its rent-to-never-own software model with a hardware backend. The licensing options alone could make a lawyer’s heart pound.
Instead of just hooking people into Adobe’s Creative Cloud, the company could upsell them on cloud software instances with dedicated processing for AI, ML, or video editing tasks. If you want to run the software on your equipment, you pay one price. If you want to run it on a platform it’s actually been optimized for, you’d pay another. The gap between the two costs is unlikely to be larger than the yearly GDP of a small nation.
But the idea of Adobe launching its own CPU R&D division still seems like more of a pipe dream than an actual proposal. If the company wants to dedicate more resources to run on ARM processors — and according to Fried, Parasnis’ is firmly on Team ARM — it can do that without building a custom microarchitecture team. Even taking a bog-standard ARM SoC design from license to final product takes time and effort; building a custom architecture from scratch is a multi-year development process, even for a top-tier team. Building a custom chip to accelerate AI and ML workloads in Adobe’s specific business segments would likely be a multi-year affair as the company experimented with the types of processor design that would yield the best results over time.
Still, the fact that the company is examining the idea at all is a sign of how difficult the general-purpose market has gotten for CPU manufacturers like Intel and AMD. After years of anemic performance improvements, hardware and even software developers are looking for ways to engineer their own advances.