HYDRA SQ AGENT — benchmark assistant for PBO2 and CO tuning!

What if a benchmark could tell you exactly which cores to nudge and by how much? That’s the idea behind SQ Agent. It’s a friendly co‑pilot for PBO2 & per‑core CO tuning that focuses on balance inside CCDs, not just raw thermals or max watts.

Run it once and, in about a minute, you get a clear picture of how your sample behaves under an all‑core load — where the VF curves converge, where they drift apart, and which knobs to turn next.

• One click, one glance, one verdict. A global CPU score and CCD scorecards show whether the current tune is better or worse. Does your friend have the same processor? Check who has configured their processor better.

• Actionable color map. Red tiles mean a core is already riding above the pack (CO likely saturated); blue tiles hint it wants a more negative CO.

• No stress to your hardware. Safe warm‑up to steady state and a short measurement window make it gentle for modest coolers and VRM.

Why misalignment is bad (and why CCD frequency drops):

• Shared voltage & limits. In an all‑core load the SMU must pick a common VID and keep PPT/EDC/temperature in check. The worst core (the one needing more V for the same f) dictates the VID or forces a lower f for the whole CCD.

• Hotspots. A few hot cores raise ΔT inside the CCD. Thermal/FIT guards react earlier, so the CCD loses frequency sooner.

• Efficiency loss. If VID is set by a weak core, other cores are effectively over‑volted for that moment — more watts for the same work, lower MHz/W, lower total score.

• Stability & jitter. The SMU keeps chasing limits, so frequency/voltage wobble grows — you feel it as small dips or micro‑stutter.

Aligning VF curves lets the CCD run a higher, steadier all‑core frequency at safer voltages and power.

What it measures

• Thermal, voltage and frequency consistency per CCD. The agent looks for VF‑curve convergence across cores during an all‑core workload.

• Two levels of results. A global CPU Total score and individual CCD scores show overall health and where misalignment lives.

• CPPC awareness. It reports how CPPC ranks match the actual per‑core performance under load.

• Safe for weak coolers or VRMs. The run uses moderate y‑cruncher load, a dynamic warm‑up until dT/dt ≈ 0, and a short 30s measurement window.

• Focuses on the AMD PBO2 tab. The score reacts exactly to PPT/EDC/TDC/THM/Fmax/eCLK and to your per‑core CO values.

Note: global scores correlate within the same CPU architecture. Do not compare Zen 3 with Zen 5 one‑to‑one; a cross‑gen preset will arrive later.

Color map: where to tighten CO

Each core’s tile is tinted by ΔEF/1V (its effective frequency per volt vs the median of its CCD). Red means the core’s VF curve sits notably above its siblings — CO is likely already saturated here. Blue means the core underperforms per volt and usually benefits from a more negative CO. Neutral gray means curves are aligned well.

Real examples : Ryzen 9 7950X — default

Under a 194 W all‑core load the agent finds very strong CPPC alignment (ρ≈0.877 → 93.9)* but a clear CCD imbalance: CCD0 ΔT ≈ 8.9°C (score 80.8) vs CCD1 ΔT ≈ 14.0°C (score 71.1). Core #10 reaches 75.2°C while others run cooler — a sign of uneven paste/contact or even a potential RMA case.

* Detailed statistics can be found in the LOGGING tab.

Real examples : Ryzen 9 9950X — before vs after tuning

Default: the CPU is hard (PPT limit ≈ 99.9%) and CCD0 shows poor VID uniformity; Overal Score ≈ 77.5. Tuned (CO + PBO2): frequency stability improves (CV 0.27% → 0.05%)*, efficiency rises (412.6 → 448.0 MHz/W)*, PPT limiting disappears (6.7% of time), and Overal Score climbs to 87. Some cores still need light CO rebalancing, which the color map highlights.

* Detailed statistics can be found in the LOGGING tab.

Release date: end of next week.