Web Hosting Benchmarks: 48 Hosts Tested on Real Websites

In my latest round of tests, the hosts whose servers answered fastest were AWS, Time4VPS and Impreza.host. I tested 186 real websites on 48 hosts with Grafana k6 in 2026 and ranked each host on how long its servers took to build an uncached page.

These are tests of live websites on each host, not of identical test sites on accounts I bought. Each host's figure is the median of its 2 to 6 sites (up to 3 per hosting type), so one unusually heavy or light site can't decide a host's result. Read the results as a snapshot of real-world sites on that host, not as a lab test of its servers.

The rest of this page shows the full ranking, how I tested, and what the numbers can and can't tell you.

Key findings

• VPS and cloud sites were the fastest group. Their median server time was 327 ms, against 534 ms for shared hosting sites and 671 ms for the 12 managed WordPress and website builder sites.
• The same host can look fast and slow. AlexHost's sites ranged from 151 ms to 15,913 ms. On many hosts, the site owner's own setup decides most of the speed, which is why each host's figure is a median.
• Distance adds a fixed cost. From Ashburn, a server in the US added a median 22 ms of round trip, Europe 81 to 148 ms and Singapore 232 ms. That's why the ranking removes it. On raw TTFB, some hosts look faster than others only because their servers are closer to Ashburn, not because the servers themselves are faster.
• Caching hides a slow server. On 60 of the 186 sites, the cached page arrived in less than half the uncached time. Cloudways' sites dropped from a median 1,163 ms uncached to 44 ms cached.
• Errors were rare. The 186 runs used here had 0 failed requests between them.

Provider report card

The grade compares each host with the other 47 on server time, the ranking figure explained below. I split the 48 hosts into five near-equal groups: A for the fastest fifth, then B, C, D and E for the slowest fifth. A grade is relative to this round of tests, not a pass or fail mark.

The rest of this page is the same data, sorted by other measurements.

How I tested

Every test used Grafana k6, run in Grafana Cloud from its Ashburn, US load zone. The same location, settings and test length applied to every site, so the numbers compare hosts rather than test locations.

• Load: 5 virtual users sharing a cap of 1 request per second, for 2 minutes, with a 1 to 3 second pause between page views. That's light traffic, not a stress test.
• Pages: one public page per site, requested with a normal GET. No logins, carts, forms or admin pages.
• Two kinds of request: every other request added a random query string, which most page caches treat as a new page. Those uncached requests reach the server itself; the normal ones show what visitors get from the cache.
• Requests: 26,447 requests across the 186 runs used here, all in 2026.

Server time: removing distance

A page requested from Ashburn always includes one network round trip: a server in Germany starts about 100 ms behind one in Virginia before it does any work. Ranking on raw TTFB would punish every host with servers outside the US.

So the ranking uses server time: the average uncached TTFB minus the round trip to that server. For 141 of the 186 sites the round trip is estimated from k6's connection times; for the other 45 it was measured directly. That's why the tables say "est.".

For sites behind a CDN, the round trip ends at the CDN's nearest location, so their server time still includes the trip from the CDN to the real server.

Which websites

The test sites are live websites, mostly run by other customers of each host; a few are client sites I manage. None of them is one of my own sites. Each site's hosting was checked from its DNS records, the network that owns its IP address (ASN) and the IP's registration data, not from its nameservers.

• Up to 3 sites per hosting type a host sells, so a host with shared and VPS plans has up to 6. Each host's figure is the median of its sites.
• Comparable pages only. Sites whose tested page was under 30 KB (almost empty, or static) or over 500 KB (very heavy page-builder output) were replaced, because they say more about the page than about the host.
• Built-in CDNs stay in. Kinsta (Cloudflare), Wix (Fastly) and one WPX site sit behind the CDN that comes with the product, so it counts as part of the host's service. Their server location is hidden behind it.
• Caching plugins stay in. The uncached request bypasses them, so they don't change the ranking.
• Not counted twice: Cloudways sites run on DigitalOcean and Vultr servers, but they count for Cloudways only.
• Slow runs kept: 4 sites (AlexHost, DigitalOcean, MochaHost and The Onion Host) were slow enough to trip k6's safety limit and stop early, but they had no errors, so their measurements stay in.
• Fewer sites: DigitalOcean, HostGator, The Onion Host and VPSDime have fewer than 3 usable sites so far.

Test bench

Server locations come from the geolocation of each site's server IP address.

Benchmark test 1: Server time

Server time is the closest figure to how fast a host's server built the page, with distance taken out. The median host came in at 494 ms. The fastest and slowest site of each host are listed too, so you can see how much a host's sites differ.

Several hosts have one fast and one very slow site: AlexHost, DreamHost, Hostens, Ikoula, InterServer, OrangeWebsite, RackNerd, Ultahost and Vultr each had a site under 350 ms and another over 2,000 ms. The median keeps that single slow site from deciding their place.

Benchmark test 2: Uncached TTFB from Ashburn

This is the raw figure before distance is removed: how long the first byte of an uncached page took to reach Ashburn. It's what a US visitor gets when the cache doesn't help, for example on a logged-in page or a fresh search result.

The gap between the two columns is mostly distance. HostArmada (servers in Singapore) drops from 17th on server time to 28th on raw TTFB and AlexHost (servers in Moldova, Netherlands, Switzerland) drops from 6th on server time to 13th on raw TTFB. For a US audience, a European server adds roughly 81 to 148 ms to every uncached page; for a European audience, the same server would be close.

Benchmark test 3: Cached TTFB

The cached figure is what most visitors get when a page cache or CDN answers. It mostly reflects the site owner's cache setup, not the host, so it doesn't count toward the ranking.

A "cache saved" figure near 0% means the cached and uncached requests took the same time. Either the sites have no page cache, or their cache also answered the uncached requests.

Benchmark test 4: Response time under light load

The p95 response time is the full response time (first byte plus the rest of the page's HTML) that 95% of requests beat during the 2-minute run. It shows slow spikes that an average hides. With about 1 request per second, this is light load, so it isn't a test of how a host copes with a traffic surge. Each host's figure is the median of its sites.

The failed-request column counts HTTP errors and timeouts during each host's runs; across all 186 runs there were 0.

Beyond the benchmarks: analysis

Shared vs VPS vs managed hosting

VPS and cloud sites had a lower median server time than shared hosting sites (327 ms against 534 ms). That fits how the products work: a VPS keeps its CPU and RAM for one customer's sites, while a shared server splits them between many accounts.

The managed group is small (12 sites on Kinsta, WP Engine, Nexcess (Liquid Web) and Wix), so its median is less reliable than the other two.

Consistency between a host's sites

A host where every site performs alike is easier to predict. These hosts had the biggest gaps between their fastest and slowest site:

And these had the smallest:

A narrow gap isn't automatically good. The Onion Host's sites were close together, but all slow on uncached pages; Impreza.host's were close and fast.

Distance and server location

Server location matters most for the uncached pages that skip the cache. A visitor in the US loading a page from a server in Singapore waits about 209 ms longer than one loading it from a US server, before the server does any work. A CDN can hide that for cached pages, but not for pages it can't cache.

If your audience is in one region, pick a host with a data center there first, then compare server times within that shortlist.

What these results don't tell you

• They aren't controlled tests. The sites run different themes, plugins and amounts of content, on plans I don't know. A host's result describes the sites tested, not every plan it sells.
• The samples are small. With 2 to 6 sites per host, the median limits the effect of one unusual site, but a different set of sites could still move a host up or down.
• One test location. Everything was measured from Ashburn, US. Server time removes most of the distance, but a host can still perform differently for visitors elsewhere.
• Light load only. At about 1 request per second, these runs don't show how a host copes with a traffic spike.
• No uptime yet. This round only covers speed.

Retests, corrections and changelog

If you run one of the hosts tested here and think a result is wrong, or you'd like a retest, contact me with the details. A host can't pay for a better place or a retest.

• November 2026: host figures changed from the average of 2 to 4 sites to the median of up to 3 sites per hosting type. Test pages under 30 KB or over 500 KB were replaced, Cloudways sites no longer count for DigitalOcean or Vultr, and Cherry Servers was added: 48 hosts and 186 websites.
• 2026: first edition. 47 hosts and 128 websites tested with k6 from Ashburn, US.

FAQ

Which web host is the fastest?

In this round, AWS had the lowest server time (64 ms), followed by Time4VPS (67 ms) and Impreza.host (121 ms)., and each host was tested on 2 to 6 websites, so the order can change between rounds.

Is a VPS faster than shared hosting?

In these tests it was: VPS and cloud sites had a median server time of 327 ms, shared hosting sites 534 ms. A VPS still needs a sensible setup; some unmanaged VPS sites in this test were slower than good shared hosting sites.

What is a good TTFB for a website?

Google's guideline for a good TTFB is 800 ms or less, measured on real visits. For the server's own work, the median host in this test needed 494 ms of server time on uncached pages; with a working page cache, most sites delivered the first byte much faster.

Why are some hosts with fast servers ranked low on TTFB?

Because of distance. All tests ran from Ashburn, US, so hosts with servers in Europe or Asia start 81 to 285 ms behind. Server time removes that round trip, which is why it's the ranking figure.