Minisforum MS-A2 9955HX temperature fix: Difference between revisions

From ETCwiki
Jump to navigationJump to search
Created page with "= CPU Thermal Performance Analysis = == What I Found == '''Bottom Line:''' Through systematic testing, I discovered that the AMD 9955HX's default thermal settings prioritize maximum performance at the cost of longevity. By adjusting the TJmax setting from 96°C to 78°C in the UEFI, the system runs 15 degrees cooler with only a 5% performance reduction while extending CPU lifespan by approximately 800%. This simple UEFI modification addresses a significant thermal mana..."
 
No edit summary
Line 15: Line 15:


Based on electromigration studies and the Arrhenius equation, higher temperatures dramatically reduce CPU lifespan. The following shows expected lifespan for continuous 24/7 operation:
Based on electromigration studies and the Arrhenius equation, higher temperatures dramatically reduce CPU lifespan. The following shows expected lifespan for continuous 24/7 operation:
[[Image:Cpu-lifespan-vs-temperature2.png|thumb|right|CPU Lifespan by temperature]]


{| class="wikitable"
{| class="wikitable"
Line 38: Line 40:


Comparison of four different CPUs showing the relationship between thermal management and performance. The '''78°C anomaly''' on the 9955HX shows improved performance with lower temperature.
Comparison of four different CPUs showing the relationship between thermal management and performance. The '''78°C anomaly''' on the 9955HX shows improved performance with lower temperature.
[[Image:File:Various-cpu-efficiency-heat.png|thumb|right|Various cpu performance by temperature chart]]


{| class="wikitable"
{| class="wikitable"
Line 47: Line 53:
| Intel i9-13900K || 80 || 57,005 || High-End Desktop
| Intel i9-13900K || 80 || 57,005 || High-End Desktop
|-
|-
| AMD AI370 || 95 || 33,036 || AI-Focused (Throttled)
| (Minisforums) AMD AI370 || 95 || 33,036 || AI-Focused (Throttled)
|-
|-
| Intel N150 || 71 || 6,076 || Ultra Low Power
| Intel N150 || 71 || 6,076 || Ultra Low Power
Line 55: Line 61:


=== TJmax Settings (Primary Focus) ===
=== TJmax Settings (Primary Focus) ===
[[Image:File:Minisforums-ms-a2-temperature-78c-tjmax.png|thumb|right|Minisforums ms-a2 TJmax vs performance]]


{| class="wikitable"
{| class="wikitable"
Line 108: Line 118:
| Intel i9-13900K || 80 || 57,005 || 6.30 || High-End Desktop
| Intel i9-13900K || 80 || 57,005 || 6.30 || High-End Desktop
|-
|-
| AMD AI370 || 95 || 33,036 || 3.54 || AI-Focused (Throttled)
| (Minisforums) AMD AI370 || 95 || 33,036 || 3.54 || AI-Focused (Throttled)
|-
|-
| Intel N150 || 71 || 6,076 || <0.2 || Ultra Low Power
| Intel N150 || 71 || 6,076 || <0.2 || Ultra Low Power

Revision as of 02:07, 2 July 2025

CPU Thermal Performance Analysis

What I Found

Bottom Line: Through systematic testing, I discovered that the AMD 9955HX's default thermal settings prioritize maximum performance at the cost of longevity. By adjusting the TJmax setting from 96°C to 78°C in the UEFI, the system runs 15 degrees cooler with only a 5% performance reduction while extending CPU lifespan by approximately 800%. This simple UEFI modification addresses a significant thermal management shortcoming in an otherwise excellent high-performance system.

  • Performance Loss: Only 5% reduction from default settings
  • Temperature Drop: Runs 15°C cooler (from 95°C to 81°C)
  • Lifespan Extension: From ~1 year to ~8 years under continuous operation

What is TJmax?

TJmax is the maximum temperature setting in the BIOS that tells the CPU when to start slowing down to avoid overheating - by lowering this from the default of 96°C to 78°C, we're essentially telling the CPU to be more conservative about heat.

CPU Lifespan vs Operating Temperature

Based on electromigration studies and the Arrhenius equation, higher temperatures dramatically reduce CPU lifespan. The following shows expected lifespan for continuous 24/7 operation:

CPU Lifespan by temperature
Temperature (°C) Expected Lifespan (Years)
50°C ~50 years
60°C ~30 years
70°C ~18 years
80°C ~8 years
90°C ~4 years
100°C ~1.5 years
110°C ~0.5 years

Performance vs Temperature Analysis

Comparison of four different CPUs showing the relationship between thermal management and performance. The 78°C anomaly on the 9955HX shows improved performance with lower temperature.


File:File:Various-cpu-efficiency-heat.png
Various cpu performance by temperature chart


CPU Model Max Temp (°C) Passmark Score Performance Class
AMD 9955HX (Optimized) 81.1 56,711 High-End Mobile
Intel i9-13900K 80 57,005 High-End Desktop
(Minisforums) AMD AI370 95 33,036 AI-Focused (Throttled)
Intel N150 71 6,076 Ultra Low Power

AMD 9955HX Detailed Results

TJmax Settings (Primary Focus)

File:File:Minisforums-ms-a2-temperature-78c-tjmax.png
Minisforums ms-a2 TJmax vs performance


TJmax Setting Actual Temp (°C) Passmark Score Tokens/Sec (Qwen3 8B) Performance vs Stock Est. Lifespan
Stock (96°C) 95.5 59,862 4.1 100% ~1 year
90°C 92 56,947 - 95% ~4 years
85°C 86 57,037 - 95% ~8 years
80°C 81.5 56,130 - 94% ~8 years
78°C (Optimal) 81.1 56,711 4.63 95% ~8 years
75°C 76 44,199 - 74% ~18 years
70°C 71 34,281 - 57% ~30 years

Alternative Settings Testing

I tested various Windows and BIOS settings to see if they could control temperature and performance, but most had minimal impact compared to the TJmax setting.

Setting Type Max Temp (°C) Passmark Score Tokens/Sec (Qwen3 8B) Notes
Stock BIOS 95.5 59,862 4.1 Default settings
Windows Efficient 94.5 59,597 - Minimal difference
Silent Mode 96 58,949 - Slight performance drop
75W Power Limit 95 57,344 - Power limited
55W Power Limit 95 56,414 - Severely power limited
Turbo Off 67.4 32,206 4.6 Major performance loss
78°C TJmax (Optimal) 81.1 56,711 4.63 Best balance found

Reference CPU Comparison

CPU Model Max Temp (°C) Passmark Score Tokens/Sec (Qwen3 8B) Performance Class
AMD 9955HX (Optimized) 81.1 56,711 4.63 High-End Mobile
Intel i9-13900K 80 57,005 6.30 High-End Desktop
(Minisforums) AMD AI370 95 33,036 3.54 AI-Focused (Throttled)
Intel N150 71 6,076 <0.2 Ultra Low Power
Nvidia 4090 - - 79.4 GPU Reference

Key Findings

The 78°C Anomaly

  • Performance jump from 80°C setting
  • Temperature actually decreased
  • Suggests optimal thermal design point
  • Better than default configuration

Thermal Throttling Wall

  • Massive performance drop below 78°C
  • 75°C: 26% performance loss
  • 70°C: 43% performance loss
  • Clear thermal design limitation

Efficiency vs Performance

  • Default: Maximum performance, 1-year lifespan
  • 78°C: 95% performance, 8-year lifespan
  • Power profiles had minimal impact
  • TJmax setting is the key control

Final Analysis

Recommendations for AMD 9955HX

Optimal Setting: TJmax 78°C in BIOS provides the best balance of performance and longevity.

Why 78°C is Perfect:

  • Performance: Retains 95% of maximum performance (only 5% loss)
  • Temperature: Reduces operating temperature by 15°C
  • Longevity: Extends CPU lifespan from ~1 year to ~8 years under continuous load
  • Efficiency: Optimal thermal design point discovered through testing

Key Insights:

  • Thermal Optimization: Default 96°C setting prioritizes peak performance over longevity
  • Sweet Spot Discovery: The 78°C anomaly shows better performance than 80°C setting, indicating precise thermal optimization
  • Power Profiles Ineffective: Windows power management and BIOS power limits had minimal impact compared to TJmax setting
  • Thermal Wall: Performance drops dramatically below 78°C due to aggressive thermal throttling

Server/Long-term Use:

With 78°C TJmax setting, this CPU is now suitable for server applications with minimal performance sacrifice and dramatically extended lifespan. The temperature reduction also means lower cooling costs and reduced system stress.

Additional Benchmark Data

Complete Test Results

Test Configuration Prime95 Max (°C) Passmark Score Passmark Temp (°C) Corecrunch Time (s) Corecrunch Temp (°C) Qwen3 8B (tokens/sec)
Stock BIOS 95.8 59,862 95.5 52.9 83.6 4.1
78°C TJmax 82.6 56,711 81.1 52.9 79.4 4.63
Turbo Off - 32,206 67.4 117.2 - 4.6

Note: The 78°C TJmax setting shows improved AI inference performance (4.63 vs 4.1 tokens/sec) while running significantly cooler, demonstrating that optimal thermal settings can actually improve performance in sustained workloads.