has reached the end-of-life and is now read-only. Please see the EOL announcement for details


we're seeing non-binary communications starting to take over almost all high-speed connections: pcie6, gddr6x, 400Gbe. all PAM4. 4 different possible levels per tick. by adjusting the signal less, & listening more closely, we can send more data.

but it's still powers of two. there are now four possible levels, versus two before, so now we transmit two bits at once.

i wonder if there'll be >4 amplitudes in my lifetime. i wonder if it'll be 8 different levels. or if it'll be a non-power-of-2 number of levels. can you imagine a PAM6 transciever: hi, yes, i send ~2.6 bits per tick (someone better at logarithms please check my math), how are you today?

a minimal icmp packet on classic ethernet is 74 bytes, takes 592 ticks. pam4, 296 ticks.
pam6, 228 ticks. pam8, 149 ticks.

seems unlikely we'd actually try to stuff a non-power-of-two increase, a 50% increase in (over upcoming champion pam4). but also, it does make a certain sense.

here's another example. more sure of this one. take three "hexits", 6*6*6 (=216), and one could send an entire PCIe3+ frame (130b including checksum) and still have 86 bits left-over.

it almost perfectly fits in 4*4*4 PAM4 "tetrits" (=128) except no space for the checksum. so you'd need 4*4*4*4 (=256) tetrits and have 126 bits left over.


made a spreadsheet lol, calculating how many binary bits a hexit word of n length can fit, and what efficiency that is of the hexit's state space.

· · Web · 0 · 0 · 0
Sign in to participate in the conversation

the mastodon instance at is retired

see the end-of-life plan for details: