CPU slow with certain tasks it was fast with before!

[Eten]

My HP Omen with a core i9-13900HX processor (13th gen) 64gb ram, and a Nvidia GeForce RTX 4090 got almost 10-50x slower when training Machine learning algos (mostly transformers, more on that later), scaling data (even with saved scalers), or when I stress test with recursive integer addition (like calculating Fibonacci numbers). My CPU isn't utilized more than 10% in these scenarios, and in general, yet I put them under a decent load, and about 2-3months ago, scaling with saved scalers would take a couple of seconds with 500k or more data points. Now it takes about 35 seconds just to scale 50k data points with a saved scaler!! Thats almost as long as scaling from scratch! 

It also happens that stress testing by multiplying floats (dot products), utilizes around 30-40% of the CPU, and it runs pretty fast. I've updated all drivers, have done tune ups, have set all performance settings to best performance, cleaned out my fans and applied new thermal paste to the heatsink, used diagnostic tools to no avail, tried to optimize my code further, have monitored temperatures and had no issues there, I have more than a terabyte of storage left, and have even replaced my battery since it was at 15% ware. I have performance limiters on according to HWinfo64, they are:

            IA: electrical design point/other          (almost constantly)

            IA: Max Turbo Limit                                 (always when running code)

            GT: max VR Voltage, ICCmax, pl4        (about 33% of the time)

            RING: max VR Voltage, ICCmax, pl4    (almost always).
When training transformers (variations of deep learning), I use the GPU as one would, but I get the performance limiter 'Utilization' in HWinfo64. When I look at the CPU-busy, GPU-wait/busy times while training. The majority of the time the GPU is waiting on the CPU while it's busy, and these times are nearly the same (CPU-busy and GPU-wait). To give a sense of how long the CPU is taking, it's busy for an average of 450ms, while the GPU is busy for about 2ms. Given that scaling with a saved scaler takes a while, and the part of ML training that requires the CPU (data loading? I'm not sure even since when I transfer the data to the GPU beforehand and don't use the CPU at all in the training loop, I still run into this issue) the issue is definitely CPU/power related? 

Another hint is that when I try to reset CMOS I get 'invalid checksum error'.  Strangely though, my old battery held time properly even with this issue. My new battery is different though, the clock will reset in BIOS, and the checksum will be invalid too. I've been trying to fix this issue for several weeks, and it's rather annoying.  Any help would be appreciated, I hope I don't need to spend even more money trying to fix a laptop that's only a year old.  Thanks, and apologies for this info dump!

[Max3Aj]

@Eten, Welcome to HP Support Community. 

Thank you for posting your query, I will be glad to help you. 

You're correct to suspect that the issue you're experiencing is related to power or CPU management, especially given the symptoms and diagnostics you've already performed. Here’s a detailed breakdown of the likely causes and recommended solutions:

Likely Root Causes

Power Limit Throttling (PL1/PL2/IA Limits)

You’re observing:

• IA: Max Turbo Limit → The CPU is unable to boost even when there is thermal headroom.
• IA: Electrical Design Point → Indicates current/power delivery limits from VRMs or firmware.
• RING/GT: ICCmax, PL4, VR Voltage → Confirms power delivery or firmware-based throttling.

These indicators directly affect both sustained and peak CPU performance, particularly under workloads that require rapid memory access and short, bursty calculations, such as scaling, recursion, and machine learning preprocessing.

BIOS/CMOS Issues

The Invalid CMOS checksum and BIOS clock resets are significant concerns:

• CMOS errors suggest that BIOS settings are not being saved properly, which could include:
  • CPU Turbo/Power settings
  • Voltage/current limits
  • Windows ACPI power interface (affecting Windows' ability to fully utilize hardware)
• The new CMOS battery not preserving time indicates that the BIOS/UEFI chip may be faulty or corrupted.

Firmware-level Performance Caps

Since the issue persists despite optimizations, it’s likely that your firmware (BIOS) has imposed performance limits due to a persistent flag. Possible triggers include:

• A failed or corrupt BIOS update
• CMOS corruption
• Improperly set Intel Embedded Controller (EC) flags
• Power event or EDP/VRM faults

Recommended Solutions (Step-by-Step)

Step 1: Full BIOS Reset (Properly)

You’ve attempted CMOS resets, but let’s perform a full EC + CMOS reset:

1. Shutdown completely and unplug the AC adapter.
2. Press and hold the Power Button for 60 seconds.
3. Open the back panel, disconnect both the CMOS battery and the main battery.
4. Wait for 10–15 minutes.
5. Reconnect the batteries, boot into BIOS.
6. Load BIOS defaults, save, and exit.

If the error persists, the BIOS ROM or EC may need reflashing.

Step 2: Update/Re-flash BIOS

Given the checksum failure, try a USB BIOS recovery:

Prepare a USB recovery stick using the HP BIOS Recovery Utility.

Force recovery mode:

• Plug in the USB drive.
• Hold Win + B and press Power.
• Wait for the BIOS recovery screen.

This can help reset corrupted EC/firmware registers that normal resets cannot address.
 

Step 3: Remove PL Limiters (Advanced)

Once the BIOS is restored and error-free:

1. Use ThrottleStop or Intel XTU (ensure compatibility with your CPU generation) to:
   • Disable PL1/PL2 limits
   • Increase ICCmax
   • Enable Turbo Boost long duration
   • Monitor IA Limit Reasons in real-time

If PL1 is hard-locked by the BIOS/EC, software tools will not override it, and only a modified BIOS can help (not recommended unless you are experienced).

Step 4: Test with Linux Live USB

To rule out OS-level interference:

• Boot from a Linux Mint/Ubuntu Live USB (no installation required).
• Run a quick NumPy-based benchmark or ML scaler on the CPU.
• If performance is still slow, it indicates a firmware/hardware issue.
• If performance is fast, the problem lies with Windows-level drivers or power management (ACPI).

Check CPU Affinity & Core Parking

• Open Task Manager > Performance > CPU > check if all cores/threads are active.
• Run the following command to inspect power settings:

powercfg -qh > powerconfig.txt

• Review the output for parked cores or limits on min/max processor states.

I hope this helps. 

Take care and have a good day. 

Please click “Accepted Solution” if you feel my post solved your issue, it will help others find the solution. Select "Yes" on the bottom left to say “Thanks” for helping! 

Max3Aj

HP Support