HN user
https://il.moi/
Get out of the house. Doens't matter if you want to, force yourself to. Do classes/activities with other people. Quikckly you'll meet people and things will change. But the key key key thing is - don't sit at home alone. You'll just become depressed. And don't ask yourself if you want to get out. Just do it.
and? network effects. what's so surprising.
I think Karpathy did that in one of his videos!
very interesting, thanks for sharing
To clarify: by prosumer I mean products that are more advanced than what's sold on the mass market. Imagine if you're a cyclicst - a normal bicycle might cost you $1-3k, a "pro" one with advanced features can easily set you back $10-15k+.
this is actually awesome, thanks so much for sharing
I'm building https://gpu.land/. It's an online provider of dirt cheap GPUs for deep learning. It's 1/3 the price of major cloud providers such as AWS, GCP, etc.
This was a project for me to learn the full stack (non coder before) and been a really exciting journey. 6 months ago I thought it would take me 2 months to build and launch:D
Sorry I missed this. Yes, more than happy to. Email me at hi@gpu.land.
You are right in that it's cheaper, but not that much cheaper if you look at renting v100s. Actually, on the website there's a whole section that compares us to them - https://gpu.land/versus
Hey - for sure. Because I'm the only team member, the entity behind the business is registered as a "sole trader". If you email me on hi@gpu.land I'm happy to share the details.
Hey, when you say "whitelist" - are you using some sort of adblocker? If so, could you share which one? Or where are you whitelisting them? Trying to replicate the issue.
Hey - I'm just trying to replicate the issue. Could you share which adblock you're using? I downloaded Opera but the browser seems to open gpu.land fine.
Not externally (in the same way that vast.ai does it), but you could always boot the instance -> ssh in -> manually run nvidia-docker.
Added to feature requests!
Woah, that's not good. Will look into asap (today). Thanks for the feedback!
Both are valid points. And no, there's currently no programmatic access to machines - but that's high on the roadmap.
Regarding environments, it's funny, when I was starting out I would kill for a pre-configured instance because I wanted to focus on modeling and didn't care if package X was version x.x.y or x.x.z. But as you grow as an ML engineer and develop your own toolkit these things start to matter.
So when creating a machine on gpu.land you have the choice of going pre-configured or just having a clean Ubuntu image. The former is meant for newcommers while the latter for pros. That was my thingking!
Great questions.
1) There's a finite capacity in the high 10s of GPUs. Currently the service is utilized at <10% capacity, so there's a lot of room to grow before we run out. If that was to happen (or even if we get to 50%), I could go and request more and (hopefully) expand the capacity quickly.
2)When building gpu.land I specifically wanted to make it safe for users to upload / store sensitive data. Of course the service is more geared towards hobbyists / researchers, just because that's an easier market to reach for a solo dev - but there is nothing at the technical level that sacrifices data privacy. For example, data on instances is encrypted both at rest and in transit within the DC + you and only you have SSH keys to your instance and nobody else can access it. Check out our security & privacy section in our FAQ - https://gpu.land/faq
When you stop the instances (instead of killing it), you persist data. It costs $0.02/gb/mo, so $4/mo for a 200gb harddrive.
I will add to FAQ, thanks for pointing out.
Hey!
So the way I would describe your progression as a student (at least from my experience):
- Traditional ML -> probably can run on your laptop
- Simple DL -> colab is great. Cheapest there is.
- SOTA DL -> you're probably looking at training times well into 10s of hours / days, so you'll need something that can last longer than 12h of colab time. Plus at this point you're probably sophisticated enough that you want to setup your instance once and start/stop it rather than going through setup with colab every time. That's where https://gpu.land/ fits it.
So tl;dr; - absolutely use colab first (it's cheapest), but when you outgrow it, consider gpu.land.
Great point re storage, I'll be sure to add to the FAQ. To answer your question it's SSD block storage.
You can! All machines within a single account are inside a private VLAN, which means they all talk to each other (out of the box, no setup required), but nobody else on gpu.land sees them.
See the "Are instances within my account connected?" item in the FAQ - https://gpu.land/faq
If you're going to need 64 GPUs I'll have to increase your account limit (currently 16 GPUs per account). Email me at hi@gpu.land
Ok noted. That's for me to work on. Thanks for pointing out.
Yep - most people stop instances and come back whenever they're ready to train again. No charges for GPUs while the instance is stopped.
If you have a say 200GB harddrive you're only paying $4/month for storage.
CUDA develpment is totally fine. The service was designed with DL in mind, but that shouldn't stop CUDA dev in any way.
Wow thanks for posting this! I defo haven't thought about this use case. So interesting.
So we actually explicitly prohibit mining in the T&Cs, it's just that in the FAQ I wanted to be a bit more human and dissuade people.
Also, there's quite a few protections built-in at the network layer to make sure mining isn't possible. Ports, ips, dns, even DPI. I learnt a lot about the early days of bitcoin and mining protocols when I was building gpu.land:)
But again, thanks for sharing this. Really good to know.
Thanks! I used Tailwind CSS and built the theme from scratch, but I drew inspiration from:
Yep you're correct that V100s are available as spot instances. If you're willing to write some extra code for saving down weights you could easily go with them.
Actually, I'm renting V100s. Got lucky to know the right person at the right time:)
You are correct that vast.ai is pretty close.
The biggest difference is probably security / guaranteed uptime. With vast you're getting what it says on the tin - a machine from a marketplace. Could come from anyone / anywhere. No idea what else is running on it. Ours are hosted in a professional DC, managed and secured as they should be.
If anyone's curious, there's a detailed comparison page with other platforms here - https://gpu.land/versus
hah I am European indeed but not Dutch:) But I have spent a year living in the Netherlands.
Do you think it's the scripts from stripe that are blocking the view? I would guess it's auth0.
This was my first front-end project so any feedback 100% welcome.
Exactly the same as you would be had you been running an EC2 instance at AWS. The machines are hosted, maintained and managed in exactly the same way in a whitelabel DC.