Rendered at 23:38:29 GMT+0000 (Coordinated Universal Time) with Cloudflare Workers.
Rishabhsec 24 hours ago [-]
Hitting 5GHz on a mobile chip is impressive, but sustained thermal throttling and power draw under heavy load will be the real test for continuous performance.
surajrmal 10 hours ago [-]
Continuous load is not the use case the beefy cores on phones are designed for. They exist for bursty loads related to user interactions. Humans are fairly slow, but each touch can generate a substantial workload. Being able to handle that workload quickly and then sitting idle is what these cores do. Continuous load runs on the smallest cpus possible or ideally offloaded entirely from the CPU to others subsystems. The goal here is to minimize power usage while maintaining good user experience.
dmitrygr 21 hours ago [-]
Marketingslop.
> Video editing: Decode, effects and export hand frames from one stage to the next, often across different cores. Those handoffs stay inside the cache instead of routing through the interconnect.
Yeah... none of that needs 5GHz cores. Video encode and decode are hardware units nowadays, so the core will not matter much.
> Agentic AI: An agent doesn’t do one task; it manages a chain of them, and the steps hand off between cores. Every core works from the same place, so the dataset stays resident as tasks hand off, and agents migrate between cores.
No serious amount of a modern ML model will fit into the CPU cache. So... also nonsense...
afr0ck 12 hours ago [-]
The video editing stuff was about the FlexCache. Cores can pull each others cache. This is not novel btw, it's used by recent IBM CPUs.
terminalcommand 16 hours ago [-]
There is a huge gaming community who use these processors to emulate gaming consoles and windows games.
I am still amazed I can emulate PS2 and Switch games with upscaled resolution on my S25 Ultra.
> Video editing: Decode, effects and export hand frames from one stage to the next, often across different cores. Those handoffs stay inside the cache instead of routing through the interconnect.
Yeah... none of that needs 5GHz cores. Video encode and decode are hardware units nowadays, so the core will not matter much.
> Agentic AI: An agent doesn’t do one task; it manages a chain of them, and the steps hand off between cores. Every core works from the same place, so the dataset stays resident as tasks hand off, and agents migrate between cores.
No serious amount of a modern ML model will fit into the CPU cache. So... also nonsense...