24 comments

[ 4.8 ms ] story [ 66.8 ms ] thread
Both p2p GPU renting for AI training and location-based renting for gaming are very interesting business models.

Also glad they didn't drag blockchain into that.

While blockchain definitely has it's use cases, it's surprising how much more effective a $5 DigitalOcean VPS with Postgres installed can be. At Vectordash we couldn't think of a single way in which involving a blockchain would result in a better product.
How exactly does this work? As far as I know, most ISP's send all traffic from their CPE equipment to their main data center where it is then routed. Which means if you happen to know your neighbor's IP address, and try to ping it, the packet goes 100+ miles away before being routed back to your neighbors CPE interface.

And I don't game, but if I set up a gaming machine I can "rent" it out?

There's very little a consumer can do to alter the way their ISP routes packets, however as long as the network latencies between the two machines are below ~15 ms and bandwidth up/down is 15+ Mbps, a near-native gaming cloud gaming is very much possible!

In an ideal world though, ping times would just be a function of distance.

If you're interested in hosting machines, you can check out http://vectordash.com/hosting/

Planning on supporting AMD GPUs anytime soon?
Unfortunately AMD support isn't on the roadmap.
While I love this idea hosting flexibility is a major issue. What prevents you from being able to run simply in a process? Is it lack of virtualization of the GPU device?
I sent in an application awhile ago but it never got responded to :(

I also sent an email and that dead-ended as well

How does the client and security work? What protections are there against using a host's IP for malicious activity?

Would be nice to take advantage of my 1080ti and excess solar generation I have right now.

Man I would not trust the sandboxing on any modern game engine enough to allow that.
Interesting, so from a hosting POV you wouldn't be comfortable hosting your machine for others to game on?
I think what they mean is that games have to provide sharding GPU resources or something? I'm interpreting not sure if that's right tho
It depends on what you do with that computer. I have a gaming computer that I rarely do work on, and I bet there are a ton of kids willing to rent out their gaming computer when they aren't using it who don't care much about security.
Interesting! Just curious .. what happens if a machine suddenly goes offline?
Thank you! If a host goes offline then gaming session is free! Host uptime is taken into account while assigning gaming sessions, so if a machine goes offline before it’s set date then future sessions are less likely to occur on it.
I remember this from a little while back on HN! IIRC the creator of Vectordash was a student interested in ML but disenchanted with the prices of other cloud GPU compute services. The idea was so ingeniously simple I was surprised nobody really tried it before. Glad to see they're expanding into other stuff, too.

On a side note, I'd love to see a tech write up on how Vectordash operates.

Is there any tech write-up on how this works? I'm genuinely fascinated and curious.

How does it actually run on my PC? What sort of sandboxing? Does it take over my GPU completely or can I still be using my PC for non-GPU-intensive stuff? Does it run like a VM on my PC with GPU access, or does it run the game as though I'm running it, and just stream the input/output back to the player?

this is a repost from a few days ago although they didnt have the google form for signup then just a semi broken sign up on the site (to get it to work you had to sign up for a vectordash core account before it would work)

im hoping that the service can iron out the few issues i have with liquidsky which is the service i currently use for gaming till i can afford to replace my laptop which manages to be nearly good enough sadly the few issues i have are enough that it just isnt worth it, maybe for a single player game where you can pause or save etc but not for competitive gaming.

i just hope this manages to handle the task, its one thing doing it on a server that is constantly running with a set amount of resources but theres no guarantee these machines will stay on, those selling the power may need it themselves so disconnect just like when mining.

what happens if the GPU doing the lions share of the work drops offline, are the threads it was processing seamlessly passed to the next nearest card with no loss, dont get me wrong, i like the idea but ill want to see how well it works before i commit to it.

also if theyre looking for feature suggestions may i suggest an android client thats capable of using in app purchases to pay for the sub, currently only one cloud gaming service ive seen offers this its called vortex but rather than having a VM of your own where you can install games of your choice that one has a set list of games which sadly doesnt include the main one i care about, overwatch, so i havent tried it myself however i know im not the only one with a rather fat balance of google play credit earned from survey rewards going unused

Back in ~1998 or so, Dwango[1] provided a similar service, and for the time / capabilities it worked very well. There is still a latency penalty for distance, and I don't see cloud computing solving that, so having these locally makes sense (gamers typically buy < 5ms gear etc)

[1] https://en.wikipedia.org/wiki/DWANGO

What happen to your original idea of machine learning GPU rentals?
They're still doing that - this is an expansion, not a pivot. Vectordash Core (the ML service) is totally still a thing.
I bet it's using mesh networking a la ZeroTier, I have a similar setup, working on my home desktop while being at my customer's offices. I get 15-40ms latency, 1920x1080 at 60fps and no noticeable input lag. Gaming works really well, but I only really use it to code.

I keep saying I'll write on how to set this up for cheap, but I didn't find the time yet.

Basic recipe: fiber, zerotier, one raspberry pi and a steam link

Edit: latency above referrs the display latency, the latency between the machines is between 2 and 10 ms