US East vs EU West for a global audience — actual latency numbers

Thread start: 2026-05-22

I spent the last two weeks running pings from 12 VPS nodes I keep around for testing. Same target hosts in US-East (Contabo, OVHcloud) and EU-West (Vultr, Hetzner). Here's the raw matrix, all ICMP RTT in ms, 100 packets each, 2026-05-20:

vantage point          us-east-fm    us-east-nw    eu-west-ck    eu-west-bh
--------------------------------------------------------------------------
nyc-vps (fiber)        1.2           2.1           72.4          74.1
mia-vps (fiber)        18.3          19.7          89.2          91.5
chi-vps (fiber)        12.8          11.4          95.6          97.3
lax-vps (fiber)        62.1          58.9          143.2         145.8
sea-vps (fiber)        68.4          65.2          148.7         151.3
lon-vps (fiber)        74.2          76.8          2.8           3.5
ams-vps (fiber)        78.9          81.3          6.1           5.9
fra-vps (fiber)        82.1          84.7          8.3           9.1
sin-vps (fiber)        218.4         215.6         168.3         170.9
syd-vps (fiber)        234.7         231.2         256.1         258.4
tyo-vps (fiber)        168.9         166.3         245.8         248.2
gru-vps (fiber)        119.3         122.7         187.6         185.4
--------------------------------------------------------------------------

avg to US-East:  78.1 ms (global)
avg to EU-West:  111.4 ms (global)

Route via Cogent in NYC adds 12ms versus route via Lumen same path. EU-West wins for Africa and parts of Asia, US-East wins for Americas and Oceania. No surprise, but the magnitude of the TYO-EU gap surprised me. 245ms is rough.

Question is: if you're serving a global audience on a $6/mo budget with no CDN, which poison do you pick? I lean US-East for the lower global average, but the APAC numbers hurt.

I think the TYO numbers deserve more nuance. I run a small JP-facing site from a OVHcloud box in Beauharnois, not Singapore, and route via NTT to EU-West is often worse than 280ms. Your TYO-EU at 245ms is actually decent for that path.

For APAC audience, neither US-East nor EU-West is ideal. I keep a $4 NAT box at RackNerd in Amsterdam for JP+AU users. The US-East box handles Americas. I may be wrong, but splitting by continent seems unavoidable without CDN.

Your global average favors US-East, but what percentage of your users are in APAC? If under 15%, US-East is probably fine. If over 30%, the pain is real.

A few corrections and additions.

First, your "route via Cogent adds 12ms" framing is imprecise. Cogent's AS174 has multiple paths; the 12ms delta you observed is likely due to suboptimal hot-potato routing from your NYC vantage, not Cogent's transatlantic path itself. Check if the Cogent path is landing at 15454 (New York) versus 3356's 555 (New York). The difference is often peering location, not transit quality.

Second, the TYO-EU gap is structural. Most paths TYO-EU go via NTT (AS2914) or PCCW (AS3491) to either HE (AS6939) at LAX or direct to AMS/FRA. The great-circle distance is ~9200km; fiber distance is roughly 15-20% longer. At 200km/ms in fiber, theoretical minimum is ~55ms one way, 110ms RTT. Your 245ms suggests either a Pacific path with multiple segments or routing via North America. Run a traceroute and check for LAX or SEA hops.

Third, for the "no CDN, $6/mo" constraint:

• A single box is always a compromise. Two boxes (US-East + EU-West) is the minimum for tolerable global coverage.
• If forced to one box, US-East has better submarine cable diversity to APAC than EU-West does.
• Consider a $5 geodns service to split traffic if you add a second box later.

1. The latency matrix is useful but incomplete without packet loss and jitter data. For TCP-based services, consistent 180ms with 0.1% loss beats 120ms with 2% loss.

2. The "global average" metric is misleading. Weight by your actual user distribution. I manage a SRE team for a mid-size SaaS; our weighted average to US-East is 89ms, to EU-West is 94ms, because 40% of users are in India and route better to EU-West via Tata (AS4755).

3. Specific versions: I tested this with iperf3 3.17.1, 10 TCP streams, 30 seconds. US-East to MUM showed 187ms RTT, 940 Mbit effective (1G port), 0.00% retrans. EU-West to MUM: 142ms RTT, 891 Mbit, 0.02% retrans. The EU-West path wins for India despite higher nominal latency in your matrix.

4. For the $6/mo constraint: KnownHost's 512MB KVM in Singapore ($5/mo) plus Hostinger's 1GB in NYC ($6/mo) totals $11. Use Route53-style geodns or even simple split-horizon by continent. Don't optimize for global average; optimize for your user's 95th percentile.

Route to US-East usually 190-210ms, EU-West 170-190ms. Difference is small but EU-West slightly better for VN users.

Question: do you need low latency for web or for game server? For web, TLS handshake makes first hit slow anyway. For game, 200ms is unplayable.

I run small forum from RackNerd Amsterdam, cost $4.50. Visitors from VN, TH, ID. Average load 1.2s, acceptable. Never tried global audience, maybe too slow for Europe.

Your matrix missing HKG or SIN as destination. Why compare only US and EU?

52 boxes and counting

OK so I actually did this same test last month because I have... too many VPS. Here's my $/latency efficiency ranking for cheap boxes:

• Contabo NYC €5/mo (I got 3) — 1ms to NYC, obvious
• Vultr AMS $5/mo — 3ms to AMS
• OVHcloud LAX $7/mo — 58ms to NYC but 130ms to TYO
• RackNerd SIN $5/mo — 168ms to NYC, 245ms to AMS, 45ms to TYO

For global no-CDN I run:

• 1x US-East (Contabo) for Americas
• 1x EU-West (Vultr) for Europe/Africa
• 1x APAC (RackNerd SIN) for Asia

Total $15/mo. Could cut to $10 if I dropped EU and let US-East handle it but 180ms to GRU vs 120ms from EU-West... I tested it, EU-West wins for Brazil. Something about the Atlantic cable vs going through New York.

My yabs on the Contabo box: disk 847 MB/s, network 895 Mbit up/down, Geekbench 6 single 1243. Not amazing but for $5/mo who cares. I used https://github.com/masonr/yet-another-bench-script

The real answer is you need more boxes.

Where's your vietnam node in this test

I don't think they had one. LichunLars listed NYC, MIA, CHI, LAX, then the post cuts off. Probably had more rows.

I can add my own numbers though. From my office in HCMC (VNPT fiber, 300mbit):

• Us-east-fm (Contabo NYC): 198ms
• Us-east-nw (OVHcloud Beauharnois): 205ms
• Eu-west-ck (Vultr AMS): 178ms
• Eu-west-bh (Hetzner HEL): 186ms

EU-West wins for Vietnam by 15-20ms. Not huge but consistent over a week of testing.

I should have included a SEA vantage point, my bad. The table got truncated when I pasted it. Full matrix had TYO, SIN, SYD, BOM as sources too. I'll upload the complete CSV to my server and link it later.

The SIN numbers were roughly: 220ms to US-East, 160ms to EU-West. So EU-West advantage holds across most of APAC except maybe Japan where it's closer to a wash.

Can confirm from my own testing. I have a Vultr Frankfurt and a Contabo NYC. From my friend's VPS in Singapore (DigitalOcean):

• Frankfurt: 157ms
• NYC: 218ms
• Ashburn: 224ms

The EU-West advantage for SEA is real. But for me in Prague, EU-West is 12ms and US-East is 95ms, so the choice is obvious regardless.