Is the router's CPU processor important? After flashing, only 380 Mbps of gigabit remained. I calculated the core frequency clearly

last month, I flashed OpenWrt on an old machine at home as the main router. That night, I tested the speed and got 380 Mbps of gigabit broadband. I changed the network cable, restarted the optical modem, and the computer's direct connection to the modem tested at 940Mbps. I only thought to check CPU usage until almost midnight—wow, 100% red wire is soldered shut. This article will clarify the question of whether a router's CPU is really important. After reading, you'll know whether your machine should be spent on a CPU.

clarify first: the router's CPU is usually slacking off

many people think the CPU in the router is desperately forwarding every data packet, but that's not the case. In home routers, there is a dedicated forwarding component called the hardware NAT engine. Different companies call it differently: OpenWrt calls it flow offloading, while MediaTek's solution calls it HNAT. All your internet traffic flows from the WAN port to the LAN port, going through this dedicated channel, so there's no need for the CPU to go all in.

this is why the old MT7621 from ten years ago, with dual-core 880MHz, can still run full gigabit broadband today—the forwarding issue is left to hardware, with CPU usage consistently around 5%, basically slacking off. Look at the CPU usage curve in the background of the original firmware—when FamilyMart watches videos or plays games, it only drops to just over ten percent.

so the question of "CPU clock speed and number of cores" is almost nonexistent for ordinary home users. Both running gigabit broadband, a 200-yuan Mercury and a 200-yuan ASUS unit have no difference purely in forwarding speed; the difference lies elsewhere—WiFi specs and memory size I've written before ( router memory capacity is sufficient with actual test data).

my night's failure: gigabit broadband down to only 380 Mbps

back to the 380 trillion at the beginning. That machine is a Phicomm K2P MT7621 CPU, which worked perfectly as an AP before flashing OpenWrt. After flashing and dialing up, the network speed was cut in half and halved again. I followed my usual approach and checked the order, which was similar to the in the gigabit speed test that didn't meet standards: replace the network cable, change the LAN port, and connect the computer directly to the optical modem. Direct connection can run 940Mbps, which means both broadband and cable are fine—the fault lies with the router.

SSH went to the backend and checked the top; the CPU section for soft interrupts (softIRQ) took up 97%. What does that mean? All the tasks like PPPoE dial-up and NAT conversion are all focused on CPU software forwarding, with hardware acceleration not enabled. OpenWrt does not enable data sharing by default, which is a common reason many users experience slow speeds after flashing.

go to the firewall settings and check "Software Traffic Loading" (some models also have an "Hardware Traffic Loading" option; check all options you can), save, restart, and test: 930MB, it's back. CPU usage dropped from 100% to 20%. The same machine, same CPU, a one-minute difference in setup can lead to a 2.5-fold difference in network speed—whether the CPU is heavy or not depends on whether the work is actually done for it.

CPU four scenes where you truly work personally

aside from the mishaps like not enabling loading, there are several types of tasks that the hardware acceleration engine can't help and can only be handled by the CPU bag by bag. If you frequently enable these features, CPU performance directly determines the network speed limit:

mention two details I've stepped on. One is SQM, which is effective at managing game latency, but it trades CPU for order. The CAKE algorithm doesn't work at all on the MT7621 gigabit broadband; only CPUs at the MT7981 level can barely keep up. The other is WireGuard, which has an acceleration kernel. The MT7621 can run over 100 Mbps, much more respectable than OpenVPN, but to run hundreds of Mbps, you still need a stronger CPU.

there's another hidden issue: the plugin itself isn't heavy, but its rules are attached to the firewall. Once traffic passes the rules, it's pulled back to the soft forwarding path, making hardware acceleration essentially free. My machine later crashed with two plugins—Ad Removal and Multi-Thread Splitting—so even with split loading, it only had just over 700MB left—this is the hidden tax on CPU performance.

How to read

core count and clock speed: Don't just count a few cores

sellers label "quad-core CPU" as a selling point. Before looking at the specs, remember one thing: a lot of work on routers is single-threaded, and single-core clock speed is more valuable than cores. A dual-core 1.8GHz CPU may run a VPN faster than a quad-core 1.0GHz one.

the Redmi at home AX3000T is the MT7981, dual-core at 1.3GHz, and running gigabit with three or four plugins is no problem. But if you want to max out the gigabit on SQM or run high data on OpenVPN, you'll need an x86 soft router like the N100, with quad-core speeds up to 3.4GHz, delivering performance more than ten times that of the MT7621. The trade-off is that you have to build your own machine, and the power consumption is much higher. Whether it's worth it can be seen in my article about soft routers being a waste of money three types of people advise against it, and three types of people buy it with a blind eye—it's very straightforward.

three simple ways to determine if your CPU is a bottleneck

you don't need to disassemble the device or check the specs—just three ways to locate it. Step one: Check the occupancy. The management page of the original firmware usually shows CPU usage, with the OpenWrt command line tapping the top. If FamilyMart's internet usage is consistently above 80% during peak periods, it's a bit risky.

Step two: compare with speed tests. Test once in the morning when no one is using the internet, and again at eight or nine in the evening when the whole family is watching videos. The huge gap between the two times and the nighttime drop was absurd. Aside from bandwidth being stolen ( devices were checked for bandwidth), the rest was most likely the CPU was fully loaded by a certain feature.

Step 3: Close the plugin comparison. Stop all the hanging plugins one by one, and test the speed each time. When it stops and recovers, it's the culprit. That's exactly how I checked out the ad removal plugin. Once it stopped, the speed dropped from 700MB to 930MB, confirming it pulled traffic back to soft forwarding.

Who should spend money on CPUs, and who doesn't

take it personally. The optical modem for dial-up and router are only used as APs (your home is probably like this), so you don't need to look at the CPU at all, and even weak players don't mind. 100Mbps broadband, no plugins, just original firmware — also ignore it. If you build a device for your parents' home, don't spend a penny more on the CPU; that money is more practical for WiFi specs.

the main cost: above gigabit broadband, which still requires SQM pipe latency; Heavy VPN users who stay on for several hours every day; If you stack more than five plugins and still want to keep them open; There are also Mesh main routers running up, but if the main router's CPU weakens, the whole family suffers.

give a sequence at the end to save money: first confirm if data separation is enabled (special attention for flashers→ then follow the above method to find plugins→ Only after testing can you consider switching machines or switching routers. Two more pieces of advice: don't be fooled by the "quad-core" when buying a machine. First, ask about the single-core clock and whether it has hardware acceleration; Before flashing, make sure to calculate the differences between flash and memory. the difference between flash and memory that article has a comparison table. With CPU, memory, and flash memory all together, you'll have a complete understanding of the basics of a router.