3 ms·

Frankly I’m surprised no one hasn’t built an external core/ram expansion docking station yet. With OS support for Numa and a broker to schedule tasks on Big/Li

by eecc 7y ago

Frankly I’m surprised no one hasn’t built an external core/ram expansion docking station yet. With OS support for Numa and a broker to schedule tasks on Big/Little cores all the pieces are there. My guess is the halo effect around the cloud that put these devices back into the drawers; when you can sync files across two conventional independent devices, risk aversion wins. But people might become aware and weary of routing everything through Facebook/Google so the value of cloudless mobility might tip the scale. Plus, if AR/VR ever takes off, the hardware for high bandwidth wired or not PAN will become available

GordonS 7y ago
> Frankly I’m surprised no one hasn’t built an external core/ram expansion docking station yet I imagine that the distance and all the connectivity required between the CPU and RAM sitting in an external docking station would destroy performance.
- PetahNZ 7y ago
  How would this be much different to a GPU running over the PCIE bus?
- eecc 7y ago
  Well, the point is that current interconnects are reducing this latency significantly, people are buying GPUs in external boxes today
yjftsjthsd-h 7y ago
Suddenly disconnecting a dock today is a minor annoyance; worst case, you lose some file descriptors/mountpoints, and network connections. If you plug in processors, accidentally unplugging outright drops processes, and likely panics the kernel.
- eecc 7y ago
  Likely? Depends how the kernel mediates access to the detachable CPU facility. Hot plug CPU has been around forever, as well as Apple's GCD. Access to the pluggable CPU/RAM complex doesn't have to be transparent and irrecoverable on error.