armstrongvds
Member
OP
ARM Enjoyer
- Joined:
- Jun 2024
- Posts:
- 199
- From:
- Seoul, KR
Hey everyone! So I finally pulled the trigger on an Ampere Altra box from OVHcloud and WOW. 2 vCPU, 4GB RAM, €3.50/mo. Runs circles around the old x86 KVM I had at twice the price!
Been running my static blog (Hugo + nginx) for two weeks now. Load average 0.02, barely touching 80MB RAM. The insane value is real!
Anyone else actually using ARM daily for small stuff? Or still waiting for "better compatibility"? Curious about real world experience, not benchmarks!
Specs if it helps:
- 2x Ampere Altra @ 2.8GHz
- 4GB DDR4
- 40GB NVMe (yabs showed ~950 MB/s seq read)
- 1Gbit port, seeing ~600 Mbit in practice
one small ping for man...
wendy
Member
- Joined:
- Jun 2024
- Posts:
- 164
- From:
- Manchester, UK
Most of this debate assumes everyone wants to run Linux in the first place. If you're on Windows Server with IIS, ARM support is basically nonexistent for the full desktop experience stack. I ran a tiny ASP.NET blog on a 2-core x86 box from Hetzner for about a year. Worked fine, cost $5/mo. Never thought about my CPU architecture once. Is the ARM price difference really that dramatic once you factor in the time spent fixing builds?
Have you tried restarting it?
sarahjey
Member
- Joined:
- Jul 2024
- Posts:
- 149
- From:
- Ottawa, Canada
Does this mean if I get an arm vps I cant just apt install nodejs and have it work? I was looking at the cheap ampere ones on RackNerd for my portfolio site but I only know basic ubuntu stuff.
Also is 2 cores enough for like maybe 100 visitors a day? Thanks everyone!
learning on $5 VPS and prayers
hankels
Member
52 VPS and counting
- Joined:
- Jun 2024
- Posts:
- 301
- From:
- Phoenix, US
Ok so I run 52 boxes across 6 providers and heres my take
Arm:
Contabo ampere 4c4g €4/mo
X86:
Vultr ryzen 2c4g $5.50/mo
Dollar per geekbench point the arm wins but ONLY if you dont count your time as money 😂
My breakdown:
- 37 arm boxes (mostly docker stuff)
- 15 x86 boxes (legacy crap, windows, stuff that needs binary blobs)
Had to compile pihole-ftl manually on 3 of them. Took 20 min each. At my hourly rate thats... uh... lets not math that
For a blog tho? Who cares. Flip a coin. Or pick whichever logo you like better
seedbox, NAS, tape, and three offsite
tallinnying
Member
Night Shift
- Joined:
- Aug 2024
- Posts:
- 268
- From:
- Tallinn, EE
Anyway heres what I did at 3am last tuesday
Tried to run my usual caddy + miniflux + postgres stack on arm. Worked fine. Then I got curious and tried to build this weird go ffmpeg wrapper I found on github
$ go build ./...
# github.com/u2takey/ffmpeg-go
undefined: syscall.Syscall6
Idk why I do this. Its a syscall thing on arm64 apparently. Gave up. Went to bed.
For a hugo blog tho? Yeah itll work. Everything works until it doesnt
builds at 3AM, sleeps at noon
garykwh
Member
Inference on a Budget
- Joined:
- Jun 2024
- Posts:
- 173
- From:
- San Jose, US
Ngl yall are talking about blogs like thats the interesting part
The real question is inference per dollar lol. I squeezed llama.cpp onto a 4gb arm box from GreenCloudVPS — https://greencloudvps.com — and it runs... barely. 2 tok/s on q4_0 7B. Same price x86 with avx2 gets you 4 tok/s
For ACTUAL ai workloads arm is still trash unless you get the graviton3 or ampere with 128 cores. For a blog? Lol who cares. Flip a coin.
Specs if it helps: 2x Ampere Altra @ 2.8GHz 4GB DDR4 40GB NVMe (yabs showed ~950 MB/s seq read) 1Gbit port, seeing ~600 Mbit in practice
Im curious if the nvme bandwidth helps with model load times
CUDA cores are my love language
