• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

Why just two hardware thread in core?

life24

Senior member
Hello,
Why just two hardware thread in core? why not four or eight thread?
Thanks
eacxyrnehybftlll65b7.jpg
 
Not sure myself actually, but I would imagine it would be along the lines of diminishing returns vs efficiency or someone would have done it by now I would think.
 
As the others have mentioned, there are architectures that do it. But those CPUs are exceptionally wide (more so than Skylake), which means it can be very hard to fill out all of their execution units. Meanwhile there is a hardware cost to supporting more threads in SMT, so for Intel to implement it there would need to be a appropriate performance benefit. I doubt the average PC workload spawns nearly enough threads to benefit from it.
 
Simple answer,it's very hard to find a program that is so badly written that it will use less than 50% of even the most modern core.
(outside of console ports that are being made for 1,5ghz almost arc cpus )

HT and SMT in general is made to take advantage of CPU commands that are not being used by a thread, so 50% is the only thing that makes sense,if there were programs that only ran with 25% on a modern core then there would be reason for more than 2 smt threads.

The programs that run on servers are not programs but mostly just queries and in general things that do not need much cpu power so more threads does make sense there.
 
The Xeon Phi uses 4 threads per core.

It's partly about balancing throughput and latency. Adding more threads to a core lets you use the core more efficiency and get increased throughput, but each individual task may take longer to complete (increasing latency). As the Phi is designed to be a throughput monster, 4-way SMT makes sense.
 
Adding threads is simply an attempt to make sure no clocks on the cpu are ever wasted. however you eventually reach a point to where the execution unit is %100 utilized and adding more threads simply backs up the cpu log.
 
Back
Top