olespete
Member
Trust No One
- Joined:
- Jun 2024
- Posts:
- 270
- From:
- Unknown
WARNING: automated stock bots are a SECURITY NIGHTMARE. You're giving some random Telegram bot your payment info, your address, maybe your ssh keys if it "helps" deploy.
What if the bot author is logging everything? What if it's a BUYVM COMPETITOR trying to harvest demand data? I don't even trust their actual checkout without fail2ban on my local machine.
And if 50 people hammer their API simultaneously that's a DDoS by definition, you could get your entire ASN blacklisted!
airgapped, encrypted, faraday'd, still worried
lee_mcr
Member
- Joined:
- Sep 2024
- Posts:
- 242
- From:
- Manchester, UK
Someone in discord said they restocked at 3am and sold out in 90 seconds
Right then, I was in that telegram group. Proper chaos when the bot fired. Fifty-odd people hitting vultr.com's stock page at once, site threw 502s for twenty minutes. Cheers lads, we broke it.
The bot itself was just a scraper on a 30-second loop, nothing fancy. Hosted on a free oracle tier ironically. Worked until it didn't, then everyone panic-refreshed manually.
I've got a Hostinger slot I'm not using if someone's desperate. Not BuyVM but the network's honest.
Honey badger don't care... about downtime
garykwh
Member
Inference on a Budget
- Joined:
- Jun 2024
- Posts:
- 173
- From:
- San Jose, US
Ngl I just wanted the Vegas slot for a 4GB slice to run llama.cpp. Inference on cpu is pain, vram math doesn't work out on anything under 8GB anyway.
By the time I got through the bot chaos the only stock left was some Los Angeles node with 200ms latency to me.
CloudCone has rtx a4000s in stock last I checked, not the same price tier tho
CUDA cores are my love language
svendeal
Member
Deal Sniper
- Joined:
- Jul 2024
- Posts:
- 301
- From:
- Miami, US
Which llama.cpp quant were you aiming for on 4GB
world record: 4min Arch install
garykwh
Member
Inference on a Budget
- Joined:
- Jun 2024
- Posts:
- 173
- From:
- San Jose, US
Which llama.cpp quant were you aiming for on 4GB
Q4_K_M on 7B, barely. Context length kills you before the model does. I was hoping BuyVM's Vegas network would offset the slowness since it's single-digit ms to cloudflare.
Ended up on a RunPod serverless endpoint for actual work. Not the same cost structure but at least it exists when I need it.
CUDA cores are my love language
svendeal
Member
Deal Sniper
- Joined:
- Jul 2024
- Posts:
- 301
- From:
- Miami, US
RunPod's fine until you hit cold start. Five minute spin-up on a 4090 isn't "serverless" it's "email me when it's ready."
I got my 4GB BuyVM slice back in 2021, still grandfathered at the old price. Haven't needed anything bigger because I don't do local LLMs, I do nginx and a wireguard tunnel. The stock situation's been ridiculous since the TikTok crowd discovered "cheap GPU" though.
world record: 4min Arch install