david_chisnall on Nostr: nprofile1q…tvdjy What do you want? I'm still pretty impressed with the graphics on ...
nprofile1qy2hwumn8ghj7un9d3shjtnddaehgu3wwp6kyqpq0mwq9c623ujjzpql7r4fk090yxrf458sz2egzprw0zn0ca669q9q8tvdjy (nprofile…vdjy) What do you want? I'm still pretty impressed with the graphics on the Xbox One. The Series X is amazing. The Apple GPU in my laptop is ample for all of the games I run on it. For boring use cases, such as compositing windows so my CPU doesn't need to redraw them on every expose event and giving them drop shadows so they stand out better, a ten-year-old Intel embedded GPU is fine.
This is the business problem with Moore's law. The rate at which compute power (including storage and bandwidth) increases is geometric. The rate at which human problems scale is linear. You can grow rapidly by increasing the number of users, but then you hit market saturation (there are only a couple of billion humans who can afford your things). After that, the only way to grow is to find new problems.
In the '80s, there was a lot of low-hanging fruit. A typewriter that allows editing? Amazing! A balance sheet that automatically recalculates totals? Fantastic! A typewriter that can spot spelling errors? I want one! Every time you find a new use case, you can sell more compute.
Even in the '90s and early 2000s there were still a load of these. I can have a picture frame that shows any of my photos? Wow! I can share all of my photos with my family? Great! I can have video calls with people across the planet? A bit late, I'm sure that was supposed to happen in 2001, but nice that you caught up!
But now? When was the last time you bought a computer because an exciting new feature needed a faster processor? The last time I upgraded my phone and may laptop were because the old hardware was physically wearing out (in the case of my phone, with some help from the washer-dryer). The old ones were basically able to do everything I needed / wanted to do.
This absolutely terrifies tech companies and, especially, cloud providers whose entire business model revolves around customers needing to be able to scale up compute to meet requirements, when those requirements are shrinking relative to available compute every hardware iteration.
Then along comes AI. 90% of uses don't actually solve any problems that users care about but it requires much faster computers. That's not a bug, it's a feature. Requiring fast computers is the entire reason that vendors are excited by AI. Apple and NVIDIA want to sell you faster computers. Google, Microsoft, and Amazon want to rent you faster computers. If they can convince you that you need AI then you need faster computers and line goes up. If you have enough compute power then you stop buying new computer and you gradually reduce the amount that you're renting (or move to a cheaper provider as you rent the same amount) and line goes down.
You don't want line to go down do you? What are you, some kind of communist?
This is the business problem with Moore's law. The rate at which compute power (including storage and bandwidth) increases is geometric. The rate at which human problems scale is linear. You can grow rapidly by increasing the number of users, but then you hit market saturation (there are only a couple of billion humans who can afford your things). After that, the only way to grow is to find new problems.
In the '80s, there was a lot of low-hanging fruit. A typewriter that allows editing? Amazing! A balance sheet that automatically recalculates totals? Fantastic! A typewriter that can spot spelling errors? I want one! Every time you find a new use case, you can sell more compute.
Even in the '90s and early 2000s there were still a load of these. I can have a picture frame that shows any of my photos? Wow! I can share all of my photos with my family? Great! I can have video calls with people across the planet? A bit late, I'm sure that was supposed to happen in 2001, but nice that you caught up!
But now? When was the last time you bought a computer because an exciting new feature needed a faster processor? The last time I upgraded my phone and may laptop were because the old hardware was physically wearing out (in the case of my phone, with some help from the washer-dryer). The old ones were basically able to do everything I needed / wanted to do.
This absolutely terrifies tech companies and, especially, cloud providers whose entire business model revolves around customers needing to be able to scale up compute to meet requirements, when those requirements are shrinking relative to available compute every hardware iteration.
Then along comes AI. 90% of uses don't actually solve any problems that users care about but it requires much faster computers. That's not a bug, it's a feature. Requiring fast computers is the entire reason that vendors are excited by AI. Apple and NVIDIA want to sell you faster computers. Google, Microsoft, and Amazon want to rent you faster computers. If they can convince you that you need AI then you need faster computers and line goes up. If you have enough compute power then you stop buying new computer and you gradually reduce the amount that you're renting (or move to a cheaper provider as you rent the same amount) and line goes down.
You don't want line to go down do you? What are you, some kind of communist?