CPUFreq Governor Tuning For Better AMD Ryzen Linux Performance

[SaucyJack]

Why did you waste everyone's time with 4k gaming tests where we know ryzen does fine in and is capped by gpu performance?

[mastercoms]

No, latencies (CAS, RAS, ...) are basically fixed and specified by the manufacturer in terms of nano/picoseconds in the datasheets. The times are then calculated in terms of clock cycles and programmed into the RAM (DDR1, DDR2, ...) controller.

As the programmed latencies can only be multiples of full (sometimes half) clock cycles, there is some rounding involved (you always have to round up). This typically leads to latencies (in picoseconds) going up and down when changing controller frequencies. The latency may be slightly higher at the highest supported clock rate compared to the second fastest, but it may also go down.

yes but relative to the clock cycle, there are higher and lower latencies. it would interesting to see how this affects the comparisons

[bug77]

Has anybody done performance/watt comparisons of Ryzen with Broadwell-E? I haven't seen any so far.

True, these seem to be MIA everywhere (I smell some "review guidelines" from AMD). Like everybody else, I'm waiting to see what Anandtech says.

[Kendji]

Why did you waste everyone's time with 4k gaming tests where we know ryzen does fine in and is capped by gpu performance?

Pls don't troll, all resolutions has relevance, even if just to show those running 4k monitors comparable results. These tests seem to show the intended differences regardless of resolution.

Ps. Yes I am aware that some reviewers have supported Gamers nexus testing methology and some critized it, pls don't contaminate this forum with the hate revolving around that youtube spat.

[duby229]

The graphs in this article make me feel like it's flicking me off with a ring finger....

[SaucyJack]

Pls don't troll, all resolutions has relevance, even if just to show those running 4k monitors comparable results. These tests seem to show the intended differences regardless of resolution.

Ps. Yes I am aware that some reviewers have supported Gamers nexus testing methology and some critized it, pls don't contaminate this forum with the hate revolving around that youtube spat.

Seriously, what the fuck are you talking about? Benchmarks show ryzen underperforming in gaming at 1080p, it would have been by far the most useful resolution to test scheduler differences with. I have no idea what internet drama you're referring to.

[Kendji]

What ever. My point is still valid.

This 1080p ftw spat started from Gamers Nexus and has spread everywhere, just tired of it, I don't care where you got it into your head or if you where smart enough to think of it yourself. Again, my point is still valid from my last post. Nuff said.

Pro 1080p: https://youtu.be/j7UBHjtCXhU
Pro multi-thread: https://youtu.be/ylvdSnEbL50

[liam]

I read a Reddit post last night that was asking how to emulate a smarter Windows scheduler for ryzen. A few tools were mentioned but there didn't appear to be a real solution. It did, however, immediately bring to mind schedutil governor (though an amd equivalent to Intel's p state may make more sense given the great number of sensors ryzen has but p state has taken a long time to come together). At a min, the scheduler needs to know about the topology and numa cost. For better decisions, it would need to have up to date info about soc conditions.
Some of this probably makes more sense to just leave to hardware routing and presenting a virtual topography.

[Girolamo_Cavazzoni]

The power consumption over time charts are rather useless without the numeric integral, also known as energy used. Would you please add those, Michael?

[LinuxID10T]

True, these seem to be MIA everywhere (I smell some "review guidelines" from AMD). Like everybody else, I'm waiting to see what Anandtech says.

The perf/watt in all the benchmarks I've seen vastly outperform against Broadwell-E and still beat Kabylake. I think the X99 chipset takes a lot of the power though.