Last week I traded some tips with a neighbor who signed up for AT&T Gigabit on my recommendation. His speeds just weren’t as fast as mine. Is your AT&T Gigabit not fast? Here’s some advice that may help.
I’m able to consistently get speeds in both directions above 900 megabits on my AT&T gigabit connection, at least on my best hardware. Some of my older hardware lags behind that a bit, but still does much better than 500 megabits. So I’ll share my secrets.
Use a wired connection, not wireless
First things first: Test your Internet speed over a wired connection, not wireless. Even if you have an 802.11ac card in your computer, the speed probably isn’t stable enough to conduct a good Internet speed test. Make sure your computer has a gigabit card in it, and make sure you’re using a CAT5e or CAT6 Ethernet cable. While I have gotten gigabit speeds out of cables labeled CAT5 before, it was luck. CAT5e is the minimum.
Also note that if you have mesh networking, mesh networking increases reliability at the cost of more overhead. Your wireless connection speed will be more consistent throughout the house with mesh networking, but with current technology, you won’t get anywhere near gigabit speeds over wireless with it.
If you’re getting less than 100 megabits and you’re paying for a plan that’s faster than that, check your LAN connection speed. Try a different Ethernet cable if you can. I have occasionally had cables stop working at gigabit, so make sure your local connection speed is up to par first. Here are some more tips for dealing with cards refusing to connect at gigabit.
What good is gigabit Internet if wireless tops out at 300 megs?
What good is a gigabit upstream connection if your wireless can’t keep up with it? Any wired devices you have will still benefit from the highest possible speed. And multiple wireless devices in use will fight for bandwidth less, since there’s more than enough available to serve all of them.
I ran wired connections throughout my house, so I have connections almost everywhere I want them. But if you don’t want to do that, and you’re going to use wireless for everything, you might consider getting AT&T Fiber 500 mbps or even 300 mbps Internet service instead of gigabit, especially if the cost is significantly less.
Disconnect from your VPN during testing
If you’re connected to a corporate VPN and/or network security device such as a Zscaler, disconnect from that for testing. These route your traffic through their network connection, so if you run a test while you’re connected, you’re testing their network connection, not yours. And adding overhead in the process.
I couldn’t figure out why my employer’s Mac was benchmarking so terribly until I realized I was connected both to their VPN and their Zscaler.
By all means use those things when you’re working, of course. They protect you and your data. But if you’re going to use an employer’s device to test your Internet connection, disconnect beforehand. Otherwise you’re going to get some pretty weird results, and you’re also going to send a ton of data through their network that they probably don’t want.
If you’ve bought a private VPN to protect yourself from third party snoops, either disconnect during testing, or keep in mind that it’s lowering your result, at least due to overhead and possibly because of the speed of their upstream connection as well.
The benchmark you use matters
Browser-based benchmarks have a hard time with gigabit connections. Some come right out and say available CPU power and memory will affect the results. But even those who don’t say that seem to have the problem. The answer is to use a native app if you can.
Speedtest.net native app
Speedtest.net’s native Windows 10 app benchmarks my connection in the mid-900s pretty consistently. You normally won’t get a full 1024 megabits due to overhead. But running the native Speedtest by Ookla on your platform is the most reliable way to measure your speed, since it’s not affected by the limits of your web browser.
If possible, download Speedtest by Ookla from your operating system’s built in app store, to ensure you’re getting an untampered copy.
If your download speed is slower than your upload speed, this will usually solve it.
Browser-based speed tests
The standard Speedtest.net benchmarks AT&T Gigabit extremely low. I’ve noticed when I run that test that my CPU goes to 100% and I just don’t seem to have a powerful enough CPU to get a good benchmark from it. The faster the system I run it on, the better result I get, but if I believe Speedtest.net, I’m only getting 600 megabits down and 800 megabits up. If you run it on a freshly booted system with only one browser tab open, you’ll benchmark better, but unless you have a very high-end computer, you’ll probably benchmark low. The upload tests seem to be less CPU intensive than the download tests, so those benchmark quite a bit higher.
Some tests, like Verizon’s speed test, come right out and say your computer’s CPU and available memory will affect the results. I wish more of them would do that.
DSL Reports’ Speedtest is less CPU intensive. With it, I consistently benchmark over 900 megabits in both directions. Upload speed is usually a bit higher than download speed.
AT&T’s own Speed Test gives me similar results to DSL Reports. AT&T’s test is more likely to starve my CPU than DSL Reports’ test, but the speeds I get are generally within 75 megabits of each other. Often the AT&T test lags behind the DSL Reports test a bit, so that gives me some confidence AT&T isn’t inflating the results.
Over time, I expect the browser-based benchmarks to get better at measuring high-speed connections, but for now, your best bet is to use the tests from DSL Reports and AT&T itself, or better yet, a browser-based app.
PCs benchmark better than Macs
When I run benchmarks using my employer’s Mac, it consistently benchmarks 10-30% slower than my PC does. My neighbor experienced something similar. When I plugged my PC into his network and benchmarked it, I got speeds similar to what I get at home, within about 5 percent.
I think I know why. PCs usually have a network card embedded on the motherboard or sitting directly on the PCI bus somehow. Modern Mac laptops eschewed wired connections to make the machines smaller, so you have to connect a network card via Thunderbolt or USB. This adds overhead. And AT&T Gigapower seems to be very sensitive to overhead. An AT&T employee told my neighbor he’s observed people getting different speeds over Thunderbolt, USB 3.0, and USB-C, even. Same adapter, same computer, but connect it with an adapter to a different port, and sometimes it speeds up.
If there’s a weak link anywhere in your system, these benchmarks will find it. Is your AT&T Gigabit not fast? If it isn’t, that probably has something to do with it.
Your network hardware makes a difference
Whether you have a Mac or a PC, the network hardware you use and the path it has to take to reach your CPU will affect your benchmarks.
On browser-based tests, my PCs with Intel-based network cards consistently score about 20 megabits higher downstream and 40 megabits higher upstream than my PCs with Broadcom cards. But my Broadcom card gets an A on DSL Reports’ bufferbloat score while my Intels get a D+. I don’t have a working Realtek card anymore to use to test, but since I can see slight differences between Broadcom and Intel, I would expect Realtek to have its quirks, including lower throughput.
When I use a native OS-based app, which eliminates the browser overhead, the difference is less pronounced. But I would still expect an Intel or Broadcom card on a PCI Express bus to benchmark better than a cheap USB dongle containing a lower-tier chipset.