Colder is not better?? Cooling a CPU on a WiFi router.

[63W]

This left me scratching my head and I cant seem to figure out why colder is not better.

As a fun project I removed the PCB board from its enclosure on an ASUS AC68U router, over clocked the CPU to 1,200Ghz and kept ram at 666Mhz. Ambient temp is around 75F / 23C degrees and routers CPU is hovering at around 167F / 75C.

Overclocking CPU did improve performance, when copying files to an external hard drive that is attached to the AC68U router by 5MB/s, from an average of 18MB/s to an average of 23MB/s, Its not much but still a great improvement.

So I decided to take it a step further and cool down the entire PCB and the CPU. I have reusable thermal packs, so I placed the PCB on top of it flat, waited about 10 minutes and CPU temp dropped to 86F / 30C. Performed another transfer test of the same large file that is about 10GB, and the speeds plummeted down to 4MB/s, few seconds later it would shoot up to 20MB/s and drop to 10MB/s. The speeds were inconsistent the entire time while the routers CPU temp was hovering at around 31C / 88F degrees. I performed the test several times just to be sure it's not a fluke and results were still just as poor.

Once I removed the thermal pack and let the CPU reach temps above 104F / 40C, the transfer speeds were better, but still not as consistent when CPU was at around 167F / 75C.

Ive overclocked CPUs on computers for many years, I’ve seen many results when people use liquid nitrogen to cooldown an overclocked CPU and GPU on computers and yet none of them had such significant adverse effects when it comes to a CPU performing calculations or benchmarks.

I know there are people who are a lot smarter than me and perhaps one of you can provide some insight as to why I’m seeing less performance from a CPU on a wifi router, that is cooldown to only 31C / 88F degrees.

 

[Sully]

For fun, I ran this through Deep Seek R1. Here’s what I got back:

This is a fascinating experiment, and your observations are indeed counterintuitive at first glance. However, there are several potential reasons why cooling the router's CPU to such a low temperature (31°C / 88°F) could lead to inconsistent performance and reduced transfer speeds. Let’s break it down:

---

### 1. **Thermal Throttling Mechanisms**
- **Inverse Throttling**: Some systems, especially embedded devices like routers, are designed with thermal management systems that assume a certain operating temperature range. If the CPU temperature drops too low, the firmware or hardware might interpret this as an error or abnormal condition, causing it to throttle performance to "warm up" the chip. This could explain the inconsistent speeds you observed.
- **Temperature Sensors**: The router's temperature sensors might not be calibrated for such low temperatures. If the sensors report unexpected values, the firmware might behave unpredictably, including reducing performance.

---

### 2. **Clock Stability at Low Temperatures**
- **Clock Signal Integrity**: CPUs and other components rely on stable clock signals to function properly. At very low temperatures, the clock signal integrity can degrade, leading to instability. This is especially true for overclocked systems, where the margins for error are already slim.
- **Voltage Regulation**: Many modern CPUs dynamically adjust voltage based on temperature and load. If the CPU is too cold, the voltage regulation system might not function as intended, causing instability or reduced performance.

---

### 3. **PCB and Component Behavior**
- **PCB Contraction**: When the PCB is cooled significantly, it can contract slightly. This might cause micro-level stress on solder joints or traces, potentially leading to intermittent connectivity issues or signal degradation.
- **Capacitor Performance**: Capacitors on the PCB are designed to operate within a specific temperature range. At very low temperatures, their performance can degrade, affecting power delivery and signal integrity.

---

### 4. **Firmware and Software Optimization**
- **Firmware Assumptions**: Router firmware is typically optimized for a specific operating temperature range. If the CPU is too cold, the firmware might not handle the situation gracefully, leading to performance issues.
- **Thermal Management Algorithms**: The router's thermal management algorithms might not be designed for such low temperatures. For example, it might reduce performance to "warm up" the CPU, even if cooling is not necessary.

---

### 5. **Comparison to Liquid Nitrogen Cooling**
- **Designed for Extreme Cooling**: Overclocking with liquid nitrogen (LN2) is typically done on desktop CPUs and GPUs that are explicitly tested and validated for such extreme conditions. These systems often have custom firmware, voltage adjustments, and other modifications to handle the unique challenges of LN2 cooling.
- **Router Limitations**: Consumer routers like the ASUS AC68U are not designed for extreme cooling or overclocking. Their components and firmware are optimized for typical operating conditions, making them less tolerant of extreme temperature changes.

---

### Recommendations
If you want to continue experimenting, here are some suggestions:
1. **Moderate Cooling**: Instead of cooling the CPU to very low temperatures, try keeping it in a more moderate range (e.g., 40–50°C / 104–122°F). This might provide the benefits of cooling without triggering instability.
2. **Monitor Firmware Behavior**: Check if the router's firmware logs any errors or warnings related to temperature or performance. This could provide clues about why performance drops at low temperatures.
3. **Test Without Overclocking**: Try cooling the CPU without overclocking it to see if the performance issues persist. This could help isolate whether the problem is related to cooling, overclocking, or a combination of both.
4. **Check Power Delivery**: Ensure that the power supply and voltage regulation components are functioning correctly at low temperatures. Inconsistent power delivery could explain the performance drops.

---

### Conclusion
Your experiment highlights the complexities of thermal management in embedded systems. While cooling can improve performance in some cases, extreme cooling can introduce new challenges, especially in devices not designed for such conditions. By carefully controlling the temperature and monitoring the system's behavior, you might find a sweet spot that balances cooling and performance. Let me know if you try any of these suggestions—I’d love to hear how it goes!

 

[63W]

For fun, I ran this through Deep Seek R1. Here’s what I got back:

This is a fascinating experiment, and your observations are indeed counterintuitive at first glance. However, there are several potential reasons why cooling the router's CPU to such a low temperature (31°C / 88°F) could lead to inconsistent performance and reduced transfer speeds. Let’s break it down:

---

### 1. **Thermal Throttling Mechanisms**
- **Inverse Throttling**: Some systems, especially embedded devices like routers, are designed with thermal management systems that assume a certain operating temperature range. If the CPU temperature drops too low, the firmware or hardware might interpret this as an error or abnormal condition, causing it to throttle performance to "warm up" the chip. This could explain the inconsistent speeds you observed.
- **Temperature Sensors**: The router's temperature sensors might not be calibrated for such low temperatures. If the sensors report unexpected values, the firmware might behave unpredictably, including reducing performance.

---

### 2. **Clock Stability at Low Temperatures**
- **Clock Signal Integrity**: CPUs and other components rely on stable clock signals to function properly. At very low temperatures, the clock signal integrity can degrade, leading to instability. This is especially true for overclocked systems, where the margins for error are already slim.
- **Voltage Regulation**: Many modern CPUs dynamically adjust voltage based on temperature and load. If the CPU is too cold, the voltage regulation system might not function as intended, causing instability or reduced performance.

---

### 3. **PCB and Component Behavior**
- **PCB Contraction**: When the PCB is cooled significantly, it can contract slightly. This might cause micro-level stress on solder joints or traces, potentially leading to intermittent connectivity issues or signal degradation.
- **Capacitor Performance**: Capacitors on the PCB are designed to operate within a specific temperature range. At very low temperatures, their performance can degrade, affecting power delivery and signal integrity.

---

### 4. **Firmware and Software Optimization**
- **Firmware Assumptions**: Router firmware is typically optimized for a specific operating temperature range. If the CPU is too cold, the firmware might not handle the situation gracefully, leading to performance issues.
- **Thermal Management Algorithms**: The router's thermal management algorithms might not be designed for such low temperatures. For example, it might reduce performance to "warm up" the CPU, even if cooling is not necessary.

---

### 5. **Comparison to Liquid Nitrogen Cooling**
- **Designed for Extreme Cooling**: Overclocking with liquid nitrogen (LN2) is typically done on desktop CPUs and GPUs that are explicitly tested and validated for such extreme conditions. These systems often have custom firmware, voltage adjustments, and other modifications to handle the unique challenges of LN2 cooling.
- **Router Limitations**: Consumer routers like the ASUS AC68U are not designed for extreme cooling or overclocking. Their components and firmware are optimized for typical operating conditions, making them less tolerant of extreme temperature changes.

---

### Recommendations
If you want to continue experimenting, here are some suggestions:
1. **Moderate Cooling**: Instead of cooling the CPU to very low temperatures, try keeping it in a more moderate range (e.g., 40–50°C / 104–122°F). This might provide the benefits of cooling without triggering instability.
2. **Monitor Firmware Behavior**: Check if the router's firmware logs any errors or warnings related to temperature or performance. This could provide clues about why performance drops at low temperatures.
3. **Test Without Overclocking**: Try cooling the CPU without overclocking it to see if the performance issues persist. This could help isolate whether the problem is related to cooling, overclocking, or a combination of both.
4. **Check Power Delivery**: Ensure that the power supply and voltage regulation components are functioning correctly at low temperatures. Inconsistent power delivery could explain the performance drops.

---

### Conclusion
Your experiment highlights the complexities of thermal management in embedded systems. While cooling can improve performance in some cases, extreme cooling can introduce new challenges, especially in devices not designed for such conditions. By carefully controlling the temperature and monitoring the system's behavior, you might find a sweet spot that balances cooling and performance. Let me know if you try any of these suggestions—I’d love to hear how it goes!

You brought some interesting points.

I haven't tried cooling CPU on the router while it's operating at stock speed and testing side effects/improvements. That will be my next step.

Just to clarify, I'm not using "extreme" cooling. Even though I was using a thermal pack, the CPU temp was in the positive temperature range: +31C / +88F.

As you stated, it might be a CPU design from the get go. I'll retest it with default CPU clock speeds while using a thermal pack and get back to you with results.

I found a detailed PDF spec sheet for the CPU from 2011, it's 269 pages and it will take time to go through it. Once I find operating temp range and how CPU was designed, it should provide some clues.

Cheers.

 

[63W]

From what I have gathered so far, CPU ARM Cortex-A7, Dual-core, @ 1.2GHz, is designed to work at -40C to 105C.

It comes down to CPU manufacturer and router manufacturer to dictate what and how CPU operates and how its used.

Furthermore, the CPU in a AC68U router, its happy place is around 60C- 70C. Thats my understanding after reading through various forums going back a decade.

In my case, I have no control when it comes to voltage and I can't monitor voltage when CPU is overclocked.

@Sully

Looks like I found answers I was looking for. Thanks again for your time and providing some hints.

Cheers.

 

[Puonti]

As an aside, some of those LLM answers don't align with my understanding of how things work. That could be just me of course, but…

If the CPU temperature drops too low, the firmware or hardware might interpret this as an error or abnormal condition, causing it to throttle performance to "warm up" the chip.

If the intent is to warm up a chip, I'm not sure why anyone would design the system to throttle performance down? (the answer doesn't say down, but I imagine the original prompt specified what was happening)

Overclocking with liquid nitrogen (LN2) is typically done on desktop CPUs and GPUs that are explicitly tested and validated for such extreme conditions.

Do Intel, AMD and nVidia for example actually validate their desktop CPUs / GPUs for liquid nitrogen cooling? As in, badging them "Approved for LN2 cooling", etc?

 

[seggy]

You brought some interesting points.

I haven't tried cooling CPU on the router while it's operating at stock speed and testing side effects/improvements. That will be my next step.

Just to clarify, I'm not using "extreme" cooling. Even though I was using a thermal pack, the CPU temp was in the positive temperature range: +31C / +88F.

As you stated, it might be a CPU design from the get go. I'll retest it with default CPU clock speeds while using a thermal pack and get back to you with results.

I found a detailed PDF spec sheet for the CPU from 2011, it's 269 pages and it will take time to go through it. Once I find operating temp range and how CPU was designed, it should provide some clues.

Cheers.

Seems to be he brought up some AI slop points that have nothing actually to do with what you're experiencing, since the cooling is ostensibly down to room temp.

The thing I would verify is, is everything that needs cooling being cooled (CPU aside), as well as contact patches being consistent?

 

[63W]

Seems to be he brought up some AI slop points that have nothing actually to do with what you're experiencing, since the cooling is ostensibly down to room temp.

The thing I would verify is, is everything that needs cooling being cooled (CPU aside), as well as contact patches being consistent?

Entire PCB and everything that is soldered to the PCB is being equally cooled. PCB is seating flat on the thermal pack/ice pack.

Unfortunately, there is no way for me to isolate every single component from being cooled, unless I make the PCB the size of a foot ball field.

 

[seggy]

I just looked again at the original picture, adn I realized I grossly overestimated how you were doing what you were trying to do.

Can't offer much help I'm afraid 🤣