News:

Willkommen im Notebookcheck.com Forum! Hier können sie über alle unsere Artikel und allgemein über Notebook relevante Dinge disuktieren. Viel Spass!

Main Menu

Codename Vermeer: AMD's Ryzen 4000 desktop processors could be released in September

Started by Redaktion, April 09, 2020, 11:16:27

Previous topic - Next topic

Redaktion

A new report has stated that AMD's highly anticipated Vermeer family of Ryzen 4000 desktop processors could be released around September this year. The latest series of AMD desktop CPUs is based on Zen 3 microarchitecture, with the chips expected to be manufactured on an improved 7nm+ process. AMD CEO Lisa Su has already hinted at this release timeframe.

https://www.notebookcheck.net/Codename-Vermeer-AMD-s-Ryzen-4000-desktop-processors-could-be-released-in-September.460339.0.html

heffeque

Zen 4 for me, please.
Can't wait for SMT-4 to have higher GHz by lowering core count while maintaining multi-thread performance with 4 threads per core.

qtnt

Zen 4 is going to have SMT4 ?

> lowering core count while maintaining multi-thread performance with 4 threads per core

That's not really how it works though. you really get dimishing returns with SMT.
SMT2 give you a 1.6x performance improvement  over SMT1 in the most favorable cases. But it can also go down to 10% performance loss in the unfavorable cases. Most applications lie between 1x and 1.4x,
SMT4 shows diminishing returns. with 4x the threads, you end up with a 2.5x performance improvements over SMT1 in the most favorable cases, but up to 60% (!!) performance loss in unfavorable cases. Most applications lie between 1.9x and 0.8x.

In particular when executing parallel application, having more parallelism can lead to worse results whenever shared resources need to be accessed. You have more threads fighting for the same resource, leading to contention and wasted cpu cycles. This can be especially the case when some threads execute "faster" than others, like in SMT (SMT threads run instructions during the "wasted" CPU cycles of another "main" thread).

So not only is a SMT thread not as "fast" as a hardware thread, but also sometimes throwing more, slower threads leads to performance degradations.

Source: "An SMT-Selection Metric to Improve Multithreaded Applications' Performance", page 9, available on researchgate


Quick Reply

Warning: this topic has not been posted in for at least 120 days.
Unless you're sure you want to reply, please consider starting a new topic.

Name:
Email:
Verification:
Please leave this box empty:

Shortcuts: ALT+S post or ALT+P preview