Should you enable hardware acceleration on your router? My childhood friend upgraded to Gigabit but only ran 400 Mbps, and was controlled by his parents for half a month

last month, my childhood friend upgraded our 500MB to Gigabit, with a plan costing over 30 yuan a month, all for the download speed. But on the night of the upgrade, the speed test showed 423, which is less than 470 at the original 500 Mbps. The broadband technician came to the site, plugged my laptop straight into the optical modem to test, 940, and said the line was fine, but your router was faulty, so you should switch to a gigabit one. Those two Mercury M6Gs were bought by me at the end of September, gigabit ports—how could they not work? I told him not to pay for now. That night, I remotely checked his backend and found the root cause in ten minutes. This article will break down the matter and explain it clearly.

first eliminate the most easily blamed: not Mesh, not network cables

my family has two M6G mesh setups. During last year's setup, I wrote an article Mercury Mesh networking , and the study only had wired backbound connections. So the first step is to have him plug the laptop into the main router's LAN port for testing, bypassing the wireless and sub-routers. Still 430.

second step is to observe light decay. Their optical modem receives a power of -22 km, within the health line, gigabit port negotiation is also 1000M, and the network cable is fully connected with eight cores. If the physical layer is completely clean, then only the software layer remains. Speed measurement methods also have their own rules. Don't use your phone's WiFi to test through a wall. I've written about this process in broadband speed test Plugged in this laptop is reliable.

third, I remotely accessed melogin.cn. I originally wanted to check if any devices were stealing data, but the device list was quiet. But two details stand out: the traffic statistics curve on the backend homepage is open and has never stopped since the day of installation; In device management, his son's tablet had a parental control rule: internet disconnection from 9 p.m. to 7 a.m., set by his wife during the September PC installation period.

What is

hardware acceleration: the ETC lanes and manual lanes at toll stations

every data packet passing through the router must be translated: the internal network address of the home device is replaced with a public network address, then the returned packet is claimed by the corresponding device. This process is called NAT, and the translation record serves as a comparison table. Normally, this table is stored in a dedicated forwarding chip for verification, so when a packet arrives, it passes directly, and the CPU rests nearby—this is hardware acceleration: in the TP interface, it's called hardware NAT, overseas versions are called NAT Boost, ASUS is NAT acceleration, OpenWrt is traffic sharing—it's the same thing.

analogy is a highway toll station. Hardware acceleration is for ETC lanes, allowing cars to pass directly without slowing down, and barrier lifting and lowering rely entirely on sensor devices; CPU soft forwarding is a manual lane, where each car rolls down the window to hand over the card and pay. Even in the manual lane, you can pass quite a few cars in an hour, but once the traffic volume increases, the queue immediately forms. The CPU of a hundred-yuan router is just a toll collector—one person per window, and gigabit bandwidth traffic really can't handle it.

I mentioned this in the article about router CPUs: flashed OpenWrt but didn't enable traffic sharing, gigabit dropped to 380MB, CPU soft interrupts occupied 97%, and when I switched it on, it instantly returned to 930. My childhood friend's M6G uses the same level of design—4230—the numbers match perfectly.

where do brands hide this switch

this switch is called by different names and positions, so I looked through them one by one:

TP official statement is that some models have it enabled by default and no longer show a switch on the interface. If you can't find it, just pretend it doesn't exist—don't force it. Xiaomi's one is even more straightforward. If you want to fix this switch, you can only flash OpenWrt first. I wrote about flashing methods in OpenWrt flashing my phone.

the real pitfall: once these features are enabled, acceleration quietly disappears

returned to her childhood friend's home. Why is hardware acceleration enabled, still 400 Mbps? Because once parental control is activated, acceleration automatically gives way. The logic is simple: the premise of ETC lanes is that each car doesn't need to be checked individually, but parental control must distinguish who can access each device, traffic statistics must be recorded for each device, and speed limits must monitor each device's bandwidth—all of these require opening the package to read the content, which ETC sensors can't do and must switch all lanes to manual lanes.

This isn't Mercury's fault; it's the rule of the entire industry. TP official FAQ original statement: The QoS feature cannot be enabled simultaneously with NAT acceleration; you must disable NAT acceleration before enabling QoS. ASUS does the same: bandwidth limitation means that when using the NAT, hardware acceleration will automatically turn off. There's an even more subtle story: some ASUS users noticed that NAT acceleration changed from 'enabled' itself to 'handled by the CPU.' After a long investigation, they found out IP traffic monitoring was enabled.

I listed the features that easily disable acceleration:

I also tried throttling for my home TV box. At that time, I didn't feel laggy because my home is gigabit and the router CPU in the box is strong enough. This was written about in about Mercury's speed limits. But when you apply the same operation to a budget phone with control features, it looks like a completely different face. When is QoS worth opening? I did the math separately in QoS , and this one is exactly two sides of the same coin.

how do you check if your own acceleration is still on

three methods, from convenience to hardcore.

ASUS users are the happiest—the NAT acceleration tab directly states the current status, whether it's hardware forwarding or CPU processing—it's decided at a glance. OpenWrt users go to SSH to get a top score. Looking at software interrupt rates, 20% is the acceleration working, and over 80% is the CPU working manual lanes. For regular brands that don't allow viewing, use a simple method: turn off all control functions and test speed once as a baseline; Then open the functions one by one, testing each one. Whenever it dropped more than 20%, the culprit was this one. There are also tricks to reading speed test numbers. Don't take a single web speed test seriously. I've written this reconciliation logic in my three ledgers: bandwidth and negotiated speed.

Send Off Little Home is positioned like this: turn off parental controls and data statistics, restart, and test speed at 910. Then open the parent control separately, 450. Ironclad evidence.

can you only choose between functionality and speed? Divided into three tiers based on broadband megawattage

wrap up with a discussion of choices and trade-offs. Solution for Sending Friends to Childhood: Parents control the router and move it away; the tablet itself has a youth mode, leaving time management to the device itself, and leaving only clean forwarding on the router. This is the first approach—if you can move control to the device side, don't take up the router's lane.

the second tier, households with broadband below 500 Mbps can easily activate it. CPU soft forwarding can run 400 Mbps with no problem; you won't even notice any speed drops. Control and control when needed—don't scare yourself. How much megabit of broadband should be installed to be sufficient? I calculated in my about how many megawatts to install in , and for most households, 300 Mbps is just a dessert.

the third tier, gigabit broadband still requires router-level control. There are two options: either accept dropping to around 400 Mbps, or switch to a machine with a strong CPU that lets it drive in manual traffic—ASUS's Adaptive QoS can coexist with acceleration in some scenarios, but it's expensive. When buying a router, check the parameters first. I compared three levels in CPU to help you avoid pitfalls.

finish with a sequence: keep the speed measurement base→ find control switches in the backend→ compare each switch against the speed monitor→ locate the culprit→ move the device side→ and make trade-offs based on the megameter if it can't. Don't switch routers just because you hear about slow speeds. First, make sure the software costs are clear. A 400-yuan router doesn't necessarily lose to a 1,400-yuan router; it depends on whether you let it focus on its work.