add arrow-down arrow-left arrow-right arrow-up authorcheckmark clipboard combo comment delete discord dots drag-handle dropdown-arrow errorfacebook history inbox instagram issuelink lock markup-bbcode markup-html markup-pcpp markup-cyclingbuilder markup-plain-text markup-reddit menu pin radio-button save search settings share star-empty star-full star-half switch successtag twitch twitter user warningwattage weight youtube

Ryzen ipc beating skylake in linux

BigAll
  • 37 months ago

Context


As a lot of you know Ryzen has been doing oddly in windows 10. This is due to windows (in general, but especially 10) relying a lot on low level optimizations to make things run smoothly that just aren't there yet for Ryzen based cpus. It isn't as bad in windows 7 as it relies less on these optimizations, but on the complete other end of the scale Linux tries to keep this kind of thing as far away from the kernel as possible as it breaks cross platform support and causes hard to debug, practically unreadable code. Also on linux's side is the fact that the kernel code specific to ryzen was finished months ago coded by AMD themselves and seems to be preouction ready (whereas the windows code has to be written from microsoft themselves from amd's documentation). This code was actually mainlined into the kernel before amd released the whitepapers fror the zen architecture and the original block diagrams for the high level architecture was made from deconstructing this code. This means that the architecture specific code should already be mature and work as flawlaessly as you can expect on a brand new product.

Benchmarks


so how does it do on the relatively matured linux kernel driver? I found a benchmark here showing linux performance in server related tasks, but It should be a good indicator of raw performance regardless. Unfortunately all of the benchmarks are against xeon server chips, so you have to do a little work to relate it to consumer chips (mainly identifying clock, architecture and core configuration) but performance is stellar, and I'm going to pick out a few interesting ones to explain.

https://www.servethehome.com/wp-content/uploads/2017/03/AMD-Ryzen-7-1700X-c-ray-Benchmark-800x509.jpg
Here we see the R7 1700x nipping at the heels of a dual E5-2630V4 system, which while the 1700x has a good clockspeed advantage, this does not excuse the relatively small difference between it and the 20c40t broadwell system running at a reasonable 2.2ghz base/3.1ghz boost.

https://www.servethehome.com/wp-content/uploads/2017/03/AMD-Ryzen-7-1700X-Sysbench-Single-Threaded-800x456.jpg
In this one we see the reason for the clickbait title, the 3.4 ghz base/3.8ghz boost R7 1700x is beating the pants off of the E3-1275V5, a 3.6ghz base 4.0ghz boost Skylake processor in the only single threaded benchmark shown and leading by a large margin.

What this means


well, It seems that once Microsoft has a good kernel driver for ryzen general performance should improve by a good bit and making it an even more tantalising option for single threaded workloads like music production and gaming.

Comments

  • 37 months ago
  • 1 point

Where's the AMD hatewagon at?

  • 37 months ago
  • 3 points

probrably on reddit something like that, pcpp users are all closet amd fanboys, lol

  • 37 months ago
  • 1 point

Sounds about right to me. That, or overt fanboys.

[comment deleted]
  • 37 months ago
  • 1 point

hopefully the windows kernel driver will be comperable to the linux driver in the future, but it will be harder than making the linux driver was and is completely dependant on microsoft so who knows

[comment deleted]
  • 37 months ago
  • 1 point

XD amd and their documentation.

Intel is a big contributer to the linux kernel (probrably the best drivers I've seen). I have no doubt that Intel has a very well done kernel driver. Considering that the architecture hasn't changed a whole lot since sandy bridge (well before ms started relying so much on architecture specific code) I don't see how there could be very much gain to squeze out.

[comment deleted]
  • 37 months ago
  • 1 point

Why is this not surprising.

Sort

add arrow-down arrow-left arrow-right arrow-up authorcheckmark clipboard combo comment delete discord dots drag-handle dropdown-arrow errorfacebook history inbox instagram issuelink lock markup-bbcode markup-html markup-pcpp markup-cyclingbuilder markup-plain-text markup-reddit menu pin radio-button save search settings share star-empty star-full star-half switch successtag twitch twitter user warningwattage weight youtube