Linux 4.11 Doesn't Change The Game For AMD's Ryzen

[NihilMomentum]

i had already updated it.

phoronix Michael, do some Ryzen tests with the Nvidia proprietary drivers as people were saying the CPU was working fine there... so the problem is most likely AMDGPU/RADV's maturity, for now.

[wargames]

What I find very interesting would be a comparison of cinebench R15 on windows 7, windows 10 and linux (wine) to show differences in scheduling performance. Ryzen is said to be faster on win 7 then on win 10 in some workloads, so maybe it is even faster on linux?

It doesn't surprise me at all. Windows 7 is not being updated anymore, so Micro$hit is not able to introduce code that makes AMD CPUs perform worse than Intel ones. The monopolistic association is called Wintel for something...

[plasma5]

I tested Cinebench R15 on my i7-6700HQ in win10 and linux with wine. I can say there is not much difference between both operating systems. I got 148 / 681 in windows and 146 / 689 on linux. Would be cool ich Michael could test cinebench with wine.

[wizard69]

new tech is not alway's great better wait a few day's....

Generally that is the case. However Ryzen has taken forever to get to market you would think that the BIOS or processor set up issues would be nailed down by now.

[bridgman]

Forever ? By industry standards Ryzen development was pretty typical for a new core architecture, and the time from first silicon to launch was quite short. There are limits to the amount of platform refinement you can do on simulators and emulators.

Once we start shipping server parts the timelines will be more typical, and a bit closer to "forever".

[birdie]

It doesn't surprise me at all. Windows 7 is not being updated anymore, so Micro$hit is not able to introduce code that makes AMD CPUs perform worse than Intel ones. The monopolistic association is called Wintel for something...

This is the most stupid comment I've ever seen. Windows 64 is called Amd64 internally if that says anything to you. Perhaps Windows was initially written for x86-64 AMD CPUs.

Secondly, if Microsoft had indeed pulled a trick like this and willfully slowed down AMD CPUs, that would have been an instant antitrust lawsuit.

[juno]

bridgman are you involved in Ryzen/CPU stuff, too?

[flubba86]

Windows 64 is called Amd64 internally if that says anything to you. Perhaps Windows was initially written for x86-64 AMD CPUs.

While what you said is technically correct, I think most readers will interpret this comment incorrectly. I will provide a little more background and clarification.

It is called AMD64 internally because of the history of the x86_64 architecture.

Back-in-the-day Microsoft made two 64bit versions of Windows, one targeting the IA64 architecture used on Intel Itanium Processors, and one targeting the AMD64 architecture used by the AMD Opteron.

The industry quickly decided that having two different 64-bit x86 architectures was a bad idea, and they needed to pick one to be the 64-bit x86 arch for everyone to use.

The Intel Itanium processors at the time were extremely expensive, not as fast as people expected, and were not selling very well. So more people were using the AMD Opteron and more 64bit software was being compiled for the AMD64 architecture than the IA64 arch.

So the industry decided to adopt the AMD64 architecture as the 64-bit x86 architecture, and it was renamed to x86-64 (or x86_64 or x64).

All of Intel's new 64bit chips abandoned IA64 and went with x86-64 (AMD64).

A lot of software (like Windows) and compilers that were around at the time still call x86-64 'AMD64' internally because there is really no reason to change it.

So while it is correct to say "Windows 64 was initially written for x86-64 AMD CPUs", it is more correct to say "Windows 64 was written to target the AMD64 architecture" and these days in theory should not matter whether it runs on an AMD or Intel processor.

[VikingGe]

What I find very interesting would be a comparison of cinebench R15 on windows 7, windows 10 and linux (wine) to show differences in scheduling performance.

Cinebench would be a terrible choice to test this because it stresses either
a) one
b) all
threads that a CPU has to offer. Not much to see there unless a scheduler is hopelessly broken.

Granted, you can select the number of threads manually, but that doesn't change much. Most core hopping and SMT issues are likely to be found in workloads like modern games that try to make use of all available CPU cores but don't scale very well.

Also, would wine have a big difference on cinebench perormance due to overhead?

I'm getting identical scores on both Win10 and Arch on my old AMD machine.

[bridgman]

bridgman are you involved in Ryzen/CPU stuff, too?

Not much... I got called in when the Ryzen validation teams ran into GPU issues when testing Linux apps, but so far all of the issues also repro'ed on Intel and other AMD processors, ie nothing Ryzen-specific. I was mostly the annoying guy asking "so does it fail on Intel CPUs as well ? Has this test ever passed on anything ?". I also get involved in "around the edges" things like IOMMUv2 functionality on new CPUs.

The last time I was really involved in CPU design was well before I joined ATI, designing specialized processors using AMD 29xx bit-slice hardware. That was how I first got involved with AMD (even got to meet Jerry Sanders during a factory visit) but that was a long time ago.