Trying The Turbo Boost Max 3.0 Patches On Linux 4.9 With A Core i7 6800K Broadwell-E

[phoronix]

Phoronix: Trying The Turbo Boost Max 3.0 Patches On Linux 4.9 With A Core i7 6800K Broadwell-E

Last week the Intel Turbo Boost Max 3.0 patches were updated for the Linux 4.9-rc1 kernel. While it won't be mainlined until Linux 4.10 at least, I decided to try out these TBM 3.0 / ITMT patches with a Core i7 6800K Broadwell-E CPU...

http://www.phoronix.com/scan.php?pag...x-4.9-rc1-Test

[Up123]

What's up with your machine name Is that an error message`?

[Michael]

What's up with your machine name Is that an error message`?

Yes, Ubuntu's installer tries to fetch the name via DMI information and routinely on many modern motherboards it runs into problems doing so.

[andrei_me]

Conclusion: no regressions and possible speed-ups

[cj.wijtmans]

Conclusion: no regressions and possible speed-ups

Conclusion: good for ancient single threaded apps but otherwise useless.

[fguerraz]

I think it would be interesting to see the power usage as well as short tasks benefits (as I think CPUs can't sustain the boosted frequency for very long)

[schmidtbag]

Conclusion: good for ancient single threaded apps but otherwise useless.

To my understanding, "Turbo Boost 2.0" and newer operate on all 4 cores, not just one. Personally, I really like the idea of the first iteration - it allowed only one core to speed up in order to complete 1 specific task more quickly, without breaching the CPU's maximum wattage. This was a very clever way to maximize performance without thermal or power issues (especially for laptops).

Today, TB is a real scummy way for Intel to make their products seem better than they really are. The idea of TB 2.0+ is to allow the CPU to speed up all cores beyond the "default" clock speed until it reaches power or thermal issues. But this leads to:
* The CPU using more energy than advertised
* The CPU runs faster than advertised. This means that a reviewer could say "wow, look how much faster this 3.4GHz CPU is compared to previous generation!" when in actuality the CPU is running operating at 3.6GHz. Pretty misleading, but I guess Intel has to convince people to upgrade somehow.
* Intel can supply a puny heatsink, good enough for the "default" clock speed, but insufficient for the "turbo" speed.

In other words, what's actually going on is Intel made a 3.6GHz CPU with an anemic heatsink and slows down the CPU to 3.4 when it gets too hot, claiming that as a feature and it's "normal" behavior.

[labyrinth153]

I would agree with you if amd made any cpus that could compete. :/

[Thaodan]

Sadly the feature or at least the patchset depends on cfs. If anyone wants to use the feature with 4.8.x
See this branch of my of kernel, all sources except the the - pf stuff are required
https://gitlab.com/Thaodan/linux-pf.git

The branch is called, turboboost3.

I think this feature is more useful on cpu with more cores (8+).

[schmidtbag]

I would agree with you if amd made any cpus that could compete. :/

AMD is irrelevant. Intel needs to keep their sales up and it's pretty hard to convince people to buy a new product when you have maybe a 5% performance increase over your last generation.