Announcement

Collapse
No announcement yet.

Software Boosting Patches Volleyed For CPPC CPUFreq Driver (Boost / Turbo Frequency)

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • Software Boosting Patches Volleyed For CPPC CPUFreq Driver (Boost / Turbo Frequency)

    Phoronix: Software Boosting Patches Volleyed For CPPC CPUFreq Driver (Boost / Turbo Frequency)

    A new set of kernel patches out this morning are for implementing software boost support within the ACPI CPPC CPUFreq driver...

    Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite

  • #2
    Interesting, not sure how to feel about this, while the efforts are appreciated for sure and much thanks, I am a bit doubtful as to the actual advantages. Given that a frequency per IPC increase ratio for a given CPU has it's limit of performance per waste heat dissipation. While a difference of lets say, a max boost frequency of 4300Mhz and difference between a boost of 4225Mhz or even 4100Mhz—is for the most part not that big of an increase in performance output—very minimal in ideal scenarios.

    The newest AMD CPU's right now behave more like GPU's frequency-scaling-wise. The cooler the unit the more apt to boost frequencies higher. Thermal ranges and frequency are tied closely together on Zen 2. What I found most advantageous for performance, is limiting voltage draw by a minus offset. Sure I can leave the vcore on auto and it can and does achieve it's top boost frequency more often—at least that seems to be the case.

    However, the returns on a sustained max boost frequency from at least my testing are extra heat waste and less performance in a number of things even though a higher frequency has been sustained.

    It seems at least in my use scenarios there seems to be a power draw/frequency optimal range before heat waste can get out of control. Electromigration is a thing, I would rather limit the accessible voltage pool, especially considering some boards by default supply excess voltage to the socket than the VID/VID's are actually even asking for which in my experience is usually significantly more than needed for optimal operation within thermal constraints, this goes for both AMD and Intel.

    All things considered, I am a temperature monitoring freak and like to look at Zen Monitor and Psensor quite often for amusement. Thermal tuning is obviously a hobby of mine and a somewhat fun yet annoying past-time.

    Am sure many are also aware that the frequency boosted to is very contingent upon a specific workload as well as the period length that a certain such task is carried out.
    Last edited by creative; 08 May 2020, 05:28 PM.

    Comment

    Working...
    X