One year on a Vultr ARM box — long-term review

Hey everyone! Thought I'd finally write this up since my renewal just hit.

Specs: Ampere Altra Q80-30, 4 cores @ 3.0 GHz, 8 GB RAM, 100 GB NVMe. Location: their EU-Central zone. Price: €4.20/month.

Uptime: 364 days 7 hours. ONE reboot the whole year, and that was me messing with a custom kernel build that panicked. Their BMC/IPMI thing let me get back in without a ticket.

IO consistency is what blew me away. I've had €6 x86 boxes from Hetzner and Hostinger that would spike to 2s latency under load. This thing? Steady 0.3ms. NVMe sits around 950 MB/s read, 800 MB/s write. Not insane on paper but it NEVER drops.

Geekbench 6 single: 1280, multi: 4850. For context that's basically a Ryzen 3700X thread for single-core, and the multi score runs circles around every cheap dedi I've rented.

Power draw they report in the panel: 12W ave

Hey armstrongvds, thanks so much for taking time to write this!!

Really glad the box treated you well. I remember when you opened ticket about kernel panic — was worried we would lose you as customer haha. But you figured it out faster than our tech could respond.

About Docker images: we are working on adding ARM-native mirrors for common stacks, should land in Q1 2026. For now building from source is best bet, or using their buildx with QEMU (slower but works).

If anyone wants to try same plan, we have small stock left in EU-Central. Can give code ARMREVIEW10 for 10% off first 3 months, just mention this thread in order notes.

Cheers

Just use Arch ARM and build everything yourself, skill issue

Seriously though "one reboot in a year" because YOU broke the kernel? That's not a flex that's a CONFESSION

My Vultr x86 box has 2 years uptime and I didn't have to touch it because I run STOCK ARCH like a NORMAL PERSON

ARM single core 1280 is cute. My €5 OVHcloud Ryzen box pulls 1650. Enjoy your translated x86 instructions or whatever ampere does under the hood 🐧

docker images no arm builds is the real story here. Whole ecosystem still catching up. Wake me when I can run my 47 container homelab without compiling half of it

Some additional data points for comparison.

• I have been running a mixed fleet since early 2024: 6 x86_64 KVM instances across Vultr, Contabo, and InterServer, plus 2 Ampere Altra instances on Vultr.
• The ARM nodes show measurably better IO consistency. Standard deviation on fio randread 4k over 30 days: 0.04 ms (ARM) versus 0.18 ms (x86_64, same price tier). This aligns with armstrongvds's observation.
• Kernel compatibility: Linux 6.6.52 and 6.10.8 both boot cleanly. No out-of-tree modules required for standard virtio-net and virtio-blk.
• Docker ecosystem gap is real but narrowing. As of December 2025, approximately 78% of images in my standard compose stack have published linux/arm64 manifests, up from 61% in January 2025.

Practical recommendation: For production workloads requiring specific closed-source monitoring agents or legacy databases, verify ARM compatibility bef

IMO the Docker situation is way better than it was even a year ago. YMMV depending on what stack you run though.

I moved a small Node.js app and Postgres 16 to a Vultr ARM box six months ago. Zero issues with images, everything had arm64 builds. Take it with a grain of salt since my setup is pretty boring — no custom monitoring, just prometheus node exporter which does have ARM builds.

The IO consistency thing matches my experience too. I came from a Hostinger NVMe plan that would tank to 200 MB/s during their "maintenance windows." Never seen that on this box.

Not sure I'd recommend ARM to a total beginner just because of the occasional gotcha, but for anyone comfortable reading build logs it's solid.