Tested: Multi-gig fiber internet is too fast for your PC:
If you’re about to bite the bullet on a blazing-fast multi-gig internet deal, you might just have to upgrade your PC’s storage drive and CPU too.
I know this because besides having to upgrade my networking hardware, which was too slow for my new Sonic 10Gb internet, I discovered that parts of my PC were also too slow to keep up with the firehose of data. And yours will probably be too — even if you have a reasonably powerful gaming rig with an SSD and a solid 6- or 8-core processor.
[ Further reading: The hidden costs of multi-gigabit fiber Internet: A cautionary tale ]
How it started:
It all started when a friend, who was himself contemplating an upgrade to 10Gb internet, and the potentially costly upgrade of his networking gear, asked me to run a test on my own multi-gig network.
“Do me a favor,” Greg Vederman asked one morning. “Download a 20GB Steam game — something large enough that there’s enough time to ramp up to the max speed, and then tell me what you actually max out at, and whether it’s just a spike or if it’s sustained. I think that’ll be the deciding factor on whether or not I deal with this upgrade to multi-gigabit internet right now. ”
While running those tests for “The Vede” I realized that although I knew downloading a game could easily outstrip the raw write performance of a SATA SSD’s theoretical 6Gbps write speeds, I did not know how much of an impact it would have in practice.
To quantify that, I sat down and measured download times using four different targets: a 480GB NVMe Optane drive, which Crystal Disk Mark measured at about 2.7GBps sequential write speeds; a very old NVMe PCIe SSD with a write speed in the 1GBps range; an older Sandisk SATA SSD, which had a measured write speed of 450MBps; and finally, as a reference, a 10TB Seagate hard drive with a measured sequential write speed of 171MBps. The hard drive should be of particular interest to the many gamers who continue to store their game libraries on hard drives due to their price-to-capacity advantages.
During testing, I switched off real-time antivirus protection to prevent that from being a potential bottleneck. Because downloading from a live server across the open internet can be unpredictable — similar to judging how fast a car can go during rush hour traffic — I decided to try to reduce traffic interference by simply doing all of my testing between midnight and 3 am Pacific Time when most people are asleep.
One final element that could be a factor is the network adapter in the machine. In this case, it’s a 10Gtek PCIe SFP + network card using an Intel 82599EN controller chip with the default adapter settings.
Steam’s servers usually topped out at 2Gbps to 2.7Gbps for me, which is below the theoretical write speed of most SATA SSDs. However, Steam actually seems to be sending compressed game assets to you to reduce bandwidth loads because, despite Steam reporting a maximum of around 260MBps during downloads, the actual writes to the drive were around 800MBps to 1.1GBps according to the OS. That’s likely why we’re seeing such a pronounced difference between the SATA SSD versus the two NVMe SSDs despite 260MBps being within SATA’s speeds.
Besides downloading a game using the Steam client, I also downloaded the 5GB Windows 11 ISO from Microsoft’s servers, which concurred that a hard drive is too slow for ultra-fast internet. But the difference between the SATA SSD and NVMe SSDs was fairly small. I should also mention that 5GB over a 10Gbps connection is ridiculously small, and the margins of error and variability can be vast. Still, the upshot is that whether it’s a large file transfer or a sustained game download, a hard drive is largely the bottleneck when you’re running a multi-gig connection.
If you’re wondering why this doesn’t match the results you’re getting on your 10Gb or 5Gb connection in the speed tests (where your machine might push 8Gbps downloads, as the in the image below), that’s because those tests are intentionally very small and designed to test your potential download (and upload) speeds, not test the potential bottlenecks that exist elsewhere in your system. In fact, the result below is from using the Windows Store client of Ookla’s popular Speedtest, which aggregates info from four different Ookla servers to measure the download performance. The browser version, for example, would put the performance closer to 2Gbps since it relies only on one server.
Your CPU might be too slow, too:
In addition to your drive being a potential bottleneck, your PC’s CPU could be a limiting factor. While that was not the case in the machine I used for write-speed testing, not everyone has an 18-core CPU on board. So to test different scenarios, I kept all of the hardware the same and manually disabled CPU cores in the BIOS for an older and slower Core i9-7980XE to simulate the impact of using a 2-core, 4-core, 6-core, 8-core, 12-core, and 16-core CPU while downloading the same 23GB Steam game. For the target drive, I used the fastest SSD for all of the tests and, again, switched off real-time third-party AV monitoring.
So is a 2-core, 4-core, or even 6-core CPU too slow for multi-gig internet? The answer, surprisingly, is yes to all three.
You can see the results above. The blue lines represent the total time it took to download the 23GB Steam game, while the red line represents Steam’s reported peak download bandwidth. Surprisingly I saw performance scaling at 8-cores and up, with the highest peaks using all 16-cores. If you look at the actual time it takes though, it’s really no big deal. At times, I’d see the 16-core configuration taking more time than the 12-core, which I’d chalk up to what happens when you’re using a network you have no control over. It could also be Steam’s servers frowning at me for downloading half a terabyte of the same game over and over and over again. Valve, after all, does have to pay the bill for the amount of bandwidth its consumers use on downloads.
What I can say without a doubt is that if you’re going to sign up for a shiny-new 10G or 5G plan, your 10-year old dual-core or quad-core is unlikely to really push it for real-world downloads . Yes, it might be fine for a short speed test, but in reality that CPU is a serious bottleneck.
Perhaps the best news about using an older dual-core or quad-core machine to download files over a 10G internet connection is that it means the speed of the target drive just won’t matter as much either. In fact, an older dual-core or quad-core with a SATA SSD is fine — until you realize you basically aren’t exceeding the speeds of even a gigabit connection.
Do not overreact:
Still, this shouldn’t cause you to overreact and shy away from multi-gig internet. First, this situation applies to a single wired desktop hunting for the absolute fastest downloads speeds and connected to the router using a 10Gbps connection. Yes, that lets me download a 100GB game in just a few minutes, but for most people, moving to a multi-gig setup will help you no matter what, even if every PC can not take full advantage of it.
For example, with a multi-gig internet connection to your router (assuming it can at least hit 2.5Gbps through its WAN port), wired clients such as gaming desktops can download at up to gigabit speeds, while wireless clients such as laptops, tablets , and media streaming still have plenty of bandwidth left over for them.
So if you can get multi-gig internet for the same price, or not much more, it’s well worth it. Just know that if you are not getting the speeds you expected, you can not just blame your router. It might be the relatively slow CPU and slow storage you’re using. If so, our guides to the best CPUs and best SSDs for any budget can help.