You are currently viewing SemiWiki as a guest which gives you limited access to the site. To view blog comments and experience other SemiWiki features you must be a registered member. Registration is fast, simple, and absolutely free so please, join our community today!
With communications improving all the time will not most of the heavy lifting in processing be done in data centers with just the basics being done locally? Are we about to see a fundamental shift to data center that can keep up with the latest software in just about everything and make much of strictly local computing beyond anything simple a thing of the past? Will this slow down PC demand if it comes to pass and change how software is sold and used? Any thoughts would be appreciated.
With communication tech getting better all the time, it looks like a lot of heavy processing might move to data centers, leaving just the basics for local devices. This could definitely change things up, making high-end PCs less necessary and shifting focus to devices that are all about connectivity. Plus, we might see more software sold as subscriptions or cloud services.
It's not clear to me what AI PC can do and will do in the whole AI deployment. I do see a lot potential for AI assisted devices for work and for home. For example, a AI robot can help me to mow the lawn and pick up the weeds. But where the "thinking" or so called "inference" should be performed?
IMO, a lot of them will be done on the smart devices, not on the AI PCs, and a lot of them will be done in the backend data centers to simplify the deployment and operations.
It's not clear to me what AI PC can do and will do in the whole AI deployment. I do see a lot potential for AI assisted devices for work and for home. For example, a AI robot can help me to mow the lawn and pick up the weeds. But where the "thinking" or so called "inference" should be performed?
IMO, a lot of them will be done on the smart devices, not on the AI PCs, and a lot of them will be done in the backend data centers to simplify the deployment and operations.
I think the ‘hope’ is it’ll be like an 80s, 90s, 2000s situation, where more compute available means new applications organically arrive. Examples from the past:
- Good/useful GUI’s required a level of graphics horsepower (I.e. Motorola 68000 from 1979)
- Streaming music, multimedia applications (Intel MMX, C= Amiga)
- Voice recognition and input (SSE / Pentium 3)
- Photo and Video editing at full quality (I.e. late 1990s/early 2000s graphics/CPU horsepower)
- Ability to drive 3D printers (~ 2000-2005 hardware is good enough here)
- Ability To do all of these things in a handheld form factor (~ 2005-2012 CPUs)
I know you can do more primitive versions of above on even older hardware, but I think the speculation going on is that with enough AI compute locally, new usage models will arrive. There are some pretty useful denoising old photographs and ‘live video editing’ uses that “AI” (or at least the harrdware) can provide on mobile devices either now or on the way.
I think AI processing location will be cyclic. Right now the buzz about the heavy lifting gear is about the data centers. Within a few years we will see more of the popular apps move towards the edge and in everyone's pockets. Lather, rinse, repeat. Not unlike prior processing shifts, though imo the cycles will be getting shorter.
That said, I also think today's AI PCs and smartphones are mostly about grasping for a reason to drive the refresh cycle. This is important for the manufacturers (gotta keep the lights on) but not very important yet for the users. The next crop of AI PCs and phones will bring notable value to the users.
