Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

I'm not aware of any reason nor modern computer system that would have these things clocked by a common clock line of any kind.

Today, GPUs connected via PCIe or the like use 8b/10b coding over differentially-signaled pairs. The signal itself has clock recovery.

(V)RAM is generally clocked at a different frequency than the CPU as well, and all DDR utilizes strobes to determine when data is valid because access time is variable.

In some SOCs/FPGA-based devices, a central clock generator will sometimes provide LVCMOS/HCSL/LVDS/etc. clock lines to each device, but these aren't often re-used. This allows for flexibility and later programmability. There's generally no assumed phase or frequency relationship between these derived clocks and the original source - especially after the signal has traveled 20cm across a board.

In the case of a CPU/GPU, though, a 20 cent crystal oscillator at each device feeding into internal PLLs is typically the go-to.



Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: