Impact of BIOS power/memory/CPU settings on stressapptest consistency and bandwidth

[Sandeep Raman]

Hi,

I’ve been running stressapptest and observed that BIOS configuration seems to have a major impact on test consistency and throughput.

After adjusting power, memory, and CPU-related BIOS options, my results on Sys1 (288 cores, 1 TB, 16 channels) became both more consistent and higher overall in bandwidth.
I plan to run more controlled tests to isolate which BIOS parameters matter the most.

Before I do that, I’d like to understand better how these BIOS settings might interact with stressapptest behavior.

Could you please clarify:

a. Which system factors most influence the measured memory bandwidth — e.g., CPU frequency scaling, C-states, memory interleaving, or prefetching?
b. Does stressapptest benefit from “Performance” power profiles (i.e., all cores locked at max turbo) or does it already saturate memory regardless of CPU frequency?
c. Could BIOS options like NUMA interleaving, Memory Patrol Scrub, SMT on/off, or LLC prefetchers significantly affect results?
d. How sensitive is the tool to dynamic clock scaling (P-states / CPPC) or memory power-saving modes?
e. Does thread scheduling or memory affinity depend on BIOS topology reporting (e.g., ACPI tables, SRAT/MADT)?

I’d like to ensure my test results are representative of raw hardware capability rather than firmware configuration side effects.
Any guidance on best BIOS settings for stable and consistent stressapptest benchmarking would be greatly appreciated.

Thanks.

[Nick Sanders]

On Thu, Oct 23, 2025 at 3:22 AM Sandeep Raman <sandee...@gmail.com> wrote:

Hi,

I’ve been running stressapptest and observed that BIOS configuration seems to have a major impact on test consistency and throughput.

Stressapptest is just doing system memory copies, so stressapptest performance reflects general system performance.

 

After adjusting power, memory, and CPU-related BIOS options, my results on Sys1 (288 cores, 1 TB, 16 channels) became both more consistent and higher overall in bandwidth.
I plan to run more controlled tests to isolate which BIOS parameters matter the most.

Before I do that, I’d like to understand better how these BIOS settings might interact with stressapptest behavior.

Could you please clarify:

a. Which system factors most influence the measured memory bandwidth — e.g., CPU frequency scaling, C-states, memory interleaving, or prefetching?

Yes, all those.

 

b. Does stressapptest benefit from “Performance” power profiles (i.e., all cores locked at max turbo) or does it already saturate memory regardless of CPU frequency?

Yes, generally, as does any performance measurement.

 

c. Could BIOS options like NUMA interleaving, Memory Patrol Scrub, SMT on/off, or LLC prefetchers significantly affect results?

Yes, absolutely.

 

d. How sensitive is the tool to dynamic clock scaling (P-states / CPPC) or memory power-saving modes?

Very sensitive. stressapptest (as woudl any load) will stress the system with thermals and power. Optimizing for thermal or power use is generally a trade-off with performance.

 

e. Does thread scheduling or memory affinity depend on BIOS topology reporting (e.g., ACPI tables, SRAT/MADT)?

Yes, probably. It depends on the underlying drivers for your system.

 

I’d like to ensure my test results are representative of raw hardware capability rather than firmware configuration side effects.
Any guidance on best BIOS settings for stable and consistent stressapptest benchmarking would be greatly appreciated.

If you are looking for maximum stressapptest numbers, just generally try to optimize for overall system performance.

Note that the best test reflects the maximum real world use case. Disabling thermal management features may provide higher numbers, but that 

may move the system outside of its operating parameters, and cause failures or damage.

 

Thanks.

--

---
You received this message because you are subscribed to the Google Groups "stressapptest-discuss" group.
To unsubscribe from this group and stop receiving emails from it, send an email to stressapptest-di...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/stressapptest-discuss/21bf2628-5cd3-42c6-9256-bd46a1a7f592n%40googlegroups.com.