4 ms·
10GBASE-T has rather insane computational requirements: way more work per bit compared to predecessors. And 10x the bits. A pretty sizeable chunk of silicon is
by nousermane 5y ago
10GBASE-T has rather insane computational requirements: way more work per bit compared to predecessors. And 10x the bits. A pretty sizeable chunk of silicon is needed to do all of it. Hence high power consumption and price-tag. Even now, 15 years later. I guess standard's authors had a little bit too much faith in Moore's law.
> LDPC FEC coding [2048,1723], with the parity check matrix construction based on a generalized Reed–Solomon [32,2,31] code over GF(26).
> Tomlinson-Harashima precoding (THP)
[0] https://en.wikipedia.org/wiki/10GBASE-T#10GBASE-T https://en.wikipedia.org/wiki/10GBASE-T#10GBASE-T
- Spooky23 5y agoThere was a point in time circa 2014-16 where it would make sense in some datacenter scenarios. IIRC, top of rack switching and copper cable was cheaper than running fiber. Pretty sure that’s not true anymore, and with workloads moving to cloud, there’s probably no investment/scale.
- p_l 5y agoFor Tor, DAC was cheaper and more accessible. By 2016, the prices for 10Gbase-T and SFP+ switches was often close to the same, but SFP+ gave significant flexibility while 10Gbase-T meant you had to architect specifically for it. With SFP+ hardware we had coax DAC for cheap connections, precut cheap fiber with hardwired transceivers, and of course expensive transceivers for longer range connections on fiber, and we could mix and match. A 10Gbase-T device meant we had to build the entire segment for new type of wiring, ensuring right NICs and switches, because mixing 10Gbase-T with SFP+ always required expensive transceivers (they were also not trivial to acquire in 2016, speaking from experience trying to unfuck a purchase of Nexus 3k in 10Gbase-T variant when all servers had SFP+)