Normal
That's the intended behaviour for Turbo Boost on all CPUs: when one core is loaded, Turbo will hit the maximum frequency, when a small handful of cores are loaded there is a lower Turbo frequency they will hit (this was implemented on later CPUs, very early Turbo Boost implementations only had the single-core turbo), and when many cores are loaded the Turbo frequency will drop even further. AMD use the exact same method on their chips too, using the brand name Turbo Core instead.The exploit used in that thread is to remove the clockdown for AVX execution. If anyone remembers the issue with certain Prime95 versions that could heavily load AVX and Haswell where the clock-down functionality was missing and CPUs could fry due to overtemp/overcurrent, that's what the AVX clockdown on the larger Xeons is intended to prevent. This failure can occur regardless of cooling because the failure mode is blowing power traces rather than electron tunnelling failures in the silicon. It's like a more extreme version of laptop SoCs where using the GPU at higher clocks causes the CPU to clock down, but instead of just being two areas of the chip a few mm apart drawing power, the components are mixed in with each other and sharing power lanes, so the current draw is local and concentrated.If you have a Xeon E3-xxxx or a Xeon E5-1xxx (both single-socket) then exploiting the bug is unnecessary and you can just use regular old overclocking to do the same thing without the penalty of running on a weird mix of old microcode versions and courting isntability. It's only on Xeon E5-2xxx or higher (dual/quad/octa-socket) where things are really locked down due to inter-chip QPI link stability.
That's the intended behaviour for Turbo Boost on all CPUs: when one core is loaded, Turbo will hit the maximum frequency, when a small handful of cores are loaded there is a lower Turbo frequency they will hit (this was implemented on later CPUs, very early Turbo Boost implementations only had the single-core turbo), and when many cores are loaded the Turbo frequency will drop even further. AMD use the exact same method on their chips too, using the brand name Turbo Core instead.
The exploit used in that thread is to remove the clockdown for AVX execution. If anyone remembers the issue with certain Prime95 versions that could heavily load AVX and Haswell where the clock-down functionality was missing and CPUs could fry due to overtemp/overcurrent, that's what the AVX clockdown on the larger Xeons is intended to prevent. This failure can occur regardless of cooling because the failure mode is blowing power traces rather than electron tunnelling failures in the silicon. It's like a more extreme version of laptop SoCs where using the GPU at higher clocks causes the CPU to clock down, but instead of just being two areas of the chip a few mm apart drawing power, the components are mixed in with each other and sharing power lanes, so the current draw is local and concentrated.
If you have a Xeon E3-xxxx or a Xeon E5-1xxx (both single-socket) then exploiting the bug is unnecessary and you can just use regular old overclocking to do the same thing without the penalty of running on a weird mix of old microcode versions and courting isntability. It's only on Xeon E5-2xxx or higher (dual/quad/octa-socket) where things are really locked down due to inter-chip QPI link stability.
