armstrongvds
Member
OP
ARM Enjoyer
- Joined:
- Jun 2024
- Posts:
- 199
- From:
- Seoul, KR
OVHcloud base is €5 not €6
You are right, I had the old Starter tier price. They dropped it last year. Updated my sheet.
But everyone is ignoring the actual question: when do you *want* block storage? Storage VPS is one disk, one failure domain. With OVHcloud I can snapshot the boot volume separately from the data volume. That separation is worth something.
Also Hetzner Storage VPS is ARM only in Nuremberg. X86 is €3 more. I only run ARM but some of you are still on amd64 for reasons I do not understand.
one small ping for man...
RAJ
Member
- Joined:
- Jul 2024
- Posts:
- 294
- From:
- Mumbai, IN
But everyone is ignoring the actual question: when do you *want* block storage?
When your provider suspends you for "excessive I/O" on the local disk and you need to move the data somewhere fast. Block storage on OVHcloud has separate I/O limits. Learned this the hard way with a Magento store in 2023.
Contabo at €0.008/GB is real, I have the invoice. But their Singapore DC routes through Nuremberg for some upstreams. 180ms to Mumbai. Use Mumbai DC instead, it exists now.
Also: nobody talks about snapshot costs. Hetzner snapshots are €0.01/GB. Contabo is free but limited to one. OVHcloud block snapshots are €0.01/GB too. Math is not just disk.
your margin is my opportunity
garykwh
Member
Inference on a Budget
- Joined:
- Jun 2024
- Posts:
- 173
- From:
- San Jose, US
Inference angle: if you are storing model weights, Hetzner Storage VPS is the cheapest warm storage. I pull 70B Q4_K_M to a local GPU box over wireguard. 2TB fits two models with headroom. €12/month is less than one hour of A100.
But I think Hetzner's ToS has something about mining or similar uses, so I wondered if inference falls under that. I asked support, they said "academic research is fine, commercial inference contact sales." I did not contact sales. YMMV.
Contabo would ban this instantly. Their abuse automation is hair-trigger.
CUDA cores are my love language