Topic: VectorDash - Rent GPUs to AI researchers. $7.68 per 1080ti per day (Read 2967 times)

Hi guys,
I have about 25 GPUs configured for ML, as long as not using all of them now, I'd like to rent out to anyone doing ML, please contact me, I'm ready to give it away much cheaper than any Cloud like AWS.

Hi everyone! Just stumbled upon this topic. Could anyone tell me please how to start renting HW on this website? I filled in the form from the website... What's next? Are 2 Gb RAM per GPU strictly necessary? I've got like 5x 1080 Ti but only 8 Gb ram. Please advise. Thank you!

Most likely /var/vectordash/roms/ was saved as 0a:00.0.rom 0A:00.0.rom
So a simple
The same happened to me and I managed to fix it this way. Also it is possible that only 3/4 cards will be available for renting because some weird reason (please reply here if this it worked for you).

Did you manage to run this rig ? What is your experience ?

Hy,
I started to install the software package. I am in contact with their support, looks like they have a lots to do, try to find help here with my installation problem:

///
Running Vectordash tests. This may take a few minutes.
Testing KVM support...
Testing Libvirt virtualization support...
setlocale: No such file or directory
Testing IOMMU support...
setlocale: No such file or directory
Testing Nvidia GPU 0 on PCI 05:00.0...
Testing Nvidia GPU 1 on PCI 06:00.0...
Testing Nvidia GPU 2 on PCI 09:00.0...
Testing Nvidia GPU 3 on PCI 0a:00.0...
libvirt: Storage Driver error : Cannot access storage file '/var/vectordash/roms/0a:00.0.rom' (as uid:0, gid:0): No such file or directory
  File "virt_utils.py", line 199, in create_vm
  File "/usr/lib/python3/dist-packages/libvirt.py", line 1035, in create
    if ret == -1: raise libvirtError ('virDomainCreate() failed', dom=self)
libvirt: QEMU Driver error : Requested operation is not valid: domain is not running
Requested operation is not valid: domain is not running
libvirt: QEMU Driver error : Requested operation is not valid: domain is not running
Testing all Nvidia GPUs...
libvirt: Storage Driver error : Cannot access storage file '/var/vectordash/roms/0a:00.0.rom' (as uid:0, gid:0): No such file or directory
  File "virt_utils.py", line 199, in create_vm
  File "/usr/lib/python3/dist-packages/libvirt.py", line 1035, in create
    if ret == -1: raise libvirtError ('virDomainCreate() failed', dom=self)
libvirt: QEMU Driver error : Requested operation is not valid: domain is not running
Requested operation is not valid: domain is not running
libvirt: QEMU Driver error : Requested operation is not valid: domain is not running
Testing connection to Vectordash server...

Test results:
KVM support:                                  Passed
Libvirt virtualization support:               Passed
IOMMU Support:                                Passed
Testing Nvidia GPU 0 on PCI 05:00.0:          Failed
Testing Nvidia GPU 1 on PCI 06:00.0:          Passed
Testing Nvidia GPU 2 on PCI 09:00.0:          Passed
Testing Nvidia GPU 3 on PCI 0a:00.0:          Failed
Testing all Nvidia GPUs:                      Failed
Connection to Vectordash Server:              Passed
Not all tests have been passed. Please contact Vectordash support.

///

Any thoughts?

TIA betaminer

Hardware is
4x 1080 ti 11 GB
Asus X99-e ws USB 3.1
Xeon E5-1650 v4
64 GB Ram
2 TB SSD on M2
4 TB HD SATA
Had a chat about it with Sharif, should be the minimum hardware expectations

maybe you could check nebula ai, which is going to release end of September.
http://nebula-ai.com/

Looks like they are still working on a solution that will work on the majority of hardware:

https://www.youtube.com/watch?v=0G-27pkVL7w

Tried reaching out to them like 5 times now, no answer whatsoever if anyone gets a hold of them..let me know

Would love some more info myself..

So what happened to Vectordash? I heard about it 3 months ago and was quite excited, thought this might be a future path for my GPU mining farm.

Did the project get abandoned? Team left? Met some regulatory, legal, or technical hurdle? The last step I've got to with Vectordash is getting the technical requirement for the mining rigs through their email. I decided to see how things go first before I upgrade my mining rig CPU's from celeron to i5, for all of my hundreds of GPU's.

The Vectordash Twitter, Facebook page and official website has not been updated for months already. Is the project still developing or dead?

Do you plan to release the boards in time fo the acorn release?

I hope to be able to start verifying the performance difference between 16x and 8x on real AI workloads within a few weeks.
I'm also looking for boards with lots of well spaced slots with full bandwidth and I'm not alone.

I suggest Threadripper and PCI switches (e.g. PLX) like what was done on some X99 boards in the past.
Other important factors for AI use cases is provisioning for large amount of RAM. Typically, you want 2x the amount of memory on the GPU.
Storage is another important parameter so bandwidth has to be reserved for that. If the compute models can't be fed data from storage as fast as it's being processed, all is for nothing.

I am definitely interested in seeing what you can come up with for a mining motherboard with more PCIE lanes, for future mining needs, rendering, ai research, or whatever else.  The more options the better for all of us.

Appreciate the feedback!
The Epyc mobo should be around 100USD more than the X399 version. We are leaning towards the Epyc based platform ourself too due to the 128x PCIe lanes it has. I am also talking with the people behind the upcoming Acorn FPGAs also to get their feedback and to make sure our new motherboard will play well with the Acorns. Smiley

I'd definitely grab those boards if they were made. But I will prefer EPYC, you never know when a task may benefit from 16 rather than 8 lanes. Is it possible to estimate how much more will the EPYC mobo cost compared to threadripper? CPU wise it's about $100.

Here's a discussion on pcie lanes by OTOY users: https://www.youtube.com/watch?v=XrOBvz454fU&t=3485s

Support for pcie burification, and using the remainder lanes for m2 pcie slots may also be important for people getting the acorn fpga accelerator.

Well finally someone uses the lanes of the Threadripper.
I would also love to see one for Intel X299 as well.

Hey hey, Octominer here. We manufacture and sell specialised mining hardware for crypto miners.

We are currently working on developing our NEXT GEN riser free motherboard that will be more focused on AI and Rendering tasks also and not just mining for mining.   

I would love to hear some feedback from you guys on which kind of motherboard would you think would be perfect for Vectordash or AI/Rendering rigs? Currently we are considering making a riser free motherboard with X16 slots based on either AMD Threadripper X399 platform or AMD EPYC server CPU platform. Advantage of the threadripper is that it is cheaper and the CPU clock speeds are much higher. Advantage of the AMD EPYC is that it has 128 PCIe lanes compare to the threadrippers 64 PCIe lanes and more cores. Which makes it possible to run 7x GPUs at full 16x speed on the EPYC motherboard, while the threadripper motherboard can support a maximum of 3x GPU running at 16x full speed or 7x GPUs running at 8x speed.

I would love to hear some real life feedback on the topic of 16x vs 8x PCIe speed when it comes to rendering and AI tasks. From what I understand for OTOYs OctaneRender 8x speed is optimal and there is only a 1-2% difference in performance when comparing 8x to 16x PCIe speed. Does anybody know how is the 8x vs 16x PCIe performance difference when it comes to AI tasks and Vectordash?

Why did they not use something legit like https://boinc.berkeley.edu/ to submit their project there? Instead, they offer unrealistic gains.
"High-throughput computing with BOINC
Scientists : use BOINC to create a volunteer computing project , giving you the power of thousands of CPUs and GPUs.
Universities : use BOINC to create a Virtual Campus Supercomputing Center .
Companies : use BOINC for desktop Grid computing ."

I guess we know the answer.

I reached out to them a few times in March and April and received no response whatsoever. I have later  seen a youtube video from one guy who quoted their requirements for PCs they emailed to somepeople (I was not lucky to get this email). By the way they set a minimum requirement at one GPU per machine which means you would need  around 100 PCs alltogether for them to consider you. I though if they are serious I could make that investment provided there are any guarantees from their side that they even launch. I contacted them again saying I could meet those requirements and asked for more info. It was a month ago. Still no answer.
It does not look serious at all to me. With that type of attitude developers if there are any and their project simply do not deserve your time.

Anyone tried it out?

Will it support windows. I'm ready to geek out and build an ai mining rig to spec if this works. Mining coins is dead. This is the new frontier gold Rush

i didnt reciedved link for downloads yet coz:
We'll be reaching out to hosts individually once you've been chosen for a batch!

and i think i will give a try on 5 pcs at least .... problem is not in hw itself still can be sold as gaming pc with easy
problem in "my city" is in connectivity ... yes my home is under optic fiber but mining place has only 5Mb up now and if dev need 15Mb for 1 gpu ......

Anyone gave it a try yet ?

Sorry for my ignorance about the some new hardware and requirement.
Of course a better CPU, a better motherboard is needed for higher requirement.
But for increasing revenue, people will try to optimize their configuration.

The only prohibitive issue I can see is the amount of PCIE 3.0 x8 or x16 slots.... that rules out all of the mining specific motherboards.  The amount of hard drive space is a potential issue too, but that can be solved. This could possibly work well for our personal desktops since they all meet the requirements and could make some $ while doing office type work at the same time. I'm curious if they really only use 1 CPU core or not though...

I think the site is pretty clear about the income per card.  The requirements are not that extreme.  I'm not an AI or research expert but I have quite a bit of experience running Folding@Home for Curecoin.  In that instance each GPU would take up a whole CPU core and lower PCIE bandwidth equaled lower performance.  I don't think its unreasonable for this site to require basic specs that you would find a in a data center.  You can get an older Xeon cpu and mobo for dirt cheap on ebay.  This service is essentially renting your machine out as a cloud server.  Their aim isn't someone running 8 cards on a shitty Chinese motherboard from Alibaba with a celeron embedded chip.

Ever heard of Folding@Home?  Stanford University a very well off private college uses GPUs from regular people all over the world to research diseases.  Universities aren't exactly bottomless pits of money.

- i have full pcie x16 riser in my pc (original Lian-Li)
- full atx mb (actualy "ANY" mb) has x16 and there are some, that have more than that (its in board pcie lines ... so you can have mb with 2* x16 pcie + usb3 + etc ....my mb has 24pcie lines and its nothing super extreme)

- i dont think this method will work ... i think that "AI miners" are for total use of whole system and there will be no room for anything else ... so your cpu will lose in power vs XY gpus (1 for AI, rest for common mining) ... yes you can improve power of cpu for something like 8cores+8ht which "can handle both processes" but WHY??? the price of this cpu is absurd and if we dont know income per card per day, its investment as hell

There still no stable riser can provide PCIE 1.0 x8, not to mention PCIE 3.0.

And the normal motherboard can only provide PCIE 3.0 x8 *2.

Maybe a GPU miner have 2 GPU on motherboard, and other card just like old method, can gain more profit and possible?

Seems prohibitive at any sort of scale.

Exactly what I thought when I got their email today. Not many miners can provide this.

so i recieved minimum requirements (by mail):
i dont say its something expensive, but still its not what a common miner can offer in great numbers (like 30 gpu with 5 cpus .... this means i need to buy rests for 30 PCs ... or for 5 at least)
till i dont know income per day, its "a bit huge" investment


For each GPU you plan on hosting, your machine should have at least:

a PCIe 3.0 x8 or x16 slot
1 CPU Core from one of the following CPUs (or equivalent):
Intel i5 or above
Intel Xeon
AMD Ryzen or AMD EPYC
4+ GB RAM for Nvidia 1070 and below, 8+ GB of RAM for Nvidia 1070ti and above
50+ GB Storage per GPU
Internet upload & download speeds of at least 15 Mbps

Welcome to QQ group to discuss vectordash: 532749652

Or you don't mine when you can't pay for electricity.... Or you don't go to the store to buy something when you don't have the money... 

Its a very viable platform, now the support for any GPU under a 1080/1070 is a little ridiculous. This is what Golem is doing, but being released a lot earlier and without millions in funding required. His cut is 50%, so its very viable for this to be applicable, and if done correctly this could take off. Ofcourse many places will not go to a company that just started, but if they were to use funding to advertise and contract with companies and schools at a lower rate than they are advertising it could easily take off and become like a nicehash but for rendering and machine learning, etc.

Its better than to mine and to owe for electricity. You go in jail that way too

Many good points have been made and I agree with most. My biggest concern would be that even if it gets going, it will flounder and be abandoned in a few months without focused, mature leadership. It is easy to have a good idea and show small scale proof of concept, much harder to scale it and stay with it in a competitive environment.

i think the reason for hdd space question is that very much of miners using linux runing on usb flash drive etc
but yes i understand to be carful anyway

Just read the original reddit post. It is highly unlikely that any legitimate research (who is willing to pay big dollars), would work on an untested unfinished platform built by a student. I earlier said I would try this but after reading the information available i'd likely stay away until maybe the founder graduates from school.

Just my 2 sats.

Economically it makes a lot of sense.... LIKE A LOT ... not many infrastructure have the firepower of miners... and this because IT infrastructures have to use  , thanks also to the new nvidia license agreement,  specifically designed hardware such as quadro and tesla which are the VERY SAME chipsets of gaming cards , maybe silicon quality is better to ensure 24/7 stability but performance wise that means nothing... too bad these chips cost 10 - 20x the price of the very same chipset specifically designed for gaming .....


the website unfortunately looks very  unprofessional , so the scam alert is high .... i registered anyway and looking forward to test this.

In the reddit post it says "I had a friend who was mining Ethereum on his Nvidia 1080 ti's. I would Venmo him double what he was making by mining Ethereum, and in return he would let me train my neural networks on his computer at significantly less than what I would have had to pay AWS."

And

"Compared to the GPU server I rented from AWS, my neural net (using TensorFlow) trained ~5x quicker on my friend's mining rig which had one 1080 ti."

Note the Nvidia K80 costs $20 usd a day and is 5x slower than a 1080ti.

To them it makes economic sense to find cheaper alternatives to AWS. BTW my university only has a handful of K20X Teslas (which has a tenth of a 1080ti Tflops) and we are considered very well off: http://www.its.hku.hk/services/research/hpc/hpc2015/configuration

Institutes resources pales in comparison to miner's hardware. Phishing is always possible, the brave ones will test first.

I'm starting to believe that their "idea" might be legit.

Seems like the Sharif Shameem profile on twitter is pretty solid.
Also found his LinkedIn profile: https://www.linkedin.com/in/sharif-shameem/

One of his connections on LinkedIn: https://www.linkedin.com/in/arbaz-khatib/

Looks like VectorDash is a project of a few college guys from Maryland College.


Even if it is legit (as in: you are really getting paid for offering your hashes to researchers), it's still good to know:
- Are they a real licensed/registered company?
- Where are they located?
- What about the money handling? Who do we contact in case of problems?

Probably because they do it as a hobby and nobody offers free access to resources without any personal gain (that includes universities). miners are a cheap alternative. That being said all the above applies as well.

Rent gpu to AI researchers?

Sorry but I call this BS phishing attempt, mainly because why would any serious AI researchers first turn to a miner forum - regular people, instead of going to some high class universities/ some other institutes...etc. anywhere in the world and collaborate with their departments there?
I'm sure lots of universities/institutes would be open for these kinds of things and are well funded for these kind of projects if there's a good initiative...

If it's fishy, it's usually because it is...

just my 2 cents.

Yeah. Too fishy for me.

I agree this sounds a little fishy, but an interesting idea, not a lot on Google.

Domain Name: Registered 3rd October 2017, so not brand new.

Reddit Page: https://www.reddit.com/r/gpumining/comments/86ofw2/rent_out_your_gpu_compute_to_ai_researchers_and/

Facebook Page: https://www.facebook.com/vectordashGPU/

Will be interesting to hear from anyone who has had a positive experience with this system.

sounds fishy a little, actually
I guess I'll pass this one. though the offer seem extremely profitable cause now gtx1080ti is barely making 1$ per day so 7$ is a way better, but I'd prefer it was some kind of a lab from some university, I guess
that would be more trustful for me  and those AI things frights me. actually, i really think terminator stuff can happen  and that is no fun at all

Hosting anything on a tiny rig's SSD/HDD doesn't make sense. However, I agree with you on other potential risks like botnets etc.

I am looking forward to more information about this project/.

Keep us up to date please. Thanks

Sounds promising if real, but i agree with the other posts, a little caution is best with new popup projects like this. I imagine they are funded by research grants and have the money to spend but until they provided more information on who they are and what they are doing, i could imagine a number of ways they could compromise your hardware and system.

I may try and install ubuntu on one of my rigs and update this post once i get more info.

Guys be careful... There's some info missing on that site. Company information, min payout etc etc.. Are you really giving some anonymous person access to your system?
They also wanna know how much disk space you got etc.

Just think of it. With access to your system, they could easily:

- host warez, illegal porn, illegal software, illegal services, ...
- install mallware
- use your server as a proxy
- use your server as part of a botnet
...

Be careful folks, that's all I'm saying.

This is really interesting. Regrettably I have G4400s and just 8gb ram in Onda B250s with single RAM slot and no SSD running SMOS on USB, though if this project has legs I'd consider upgrading.

I'm a quantitative researcher and use some ML, though locally on my own machine where CPU speed and core count are the rate limiting factors, and not with datasets in the terrabytes like these marketing projects. I would think for a cloud-based operation application run through people's GPUs they wouldn't need too much in the way of a user's CPU, though we'll see as I contacted them.

And again all company info etc is missing.. Who rents out their rigs to companies -or better individuals- like this?

I think I will give it a try and upgrade my system memory.

Vectordash: https://vectordash.com/hosting/
Reddit thread: https://www.reddit.com/r/gpumining/comments/86ofw2/rent_out_your_gpu_compute_to_ai_researchers_and/

Info I gathered so far:

How it works
- AI Researchers rent your GPUs on an hourly basis.
- The moment your rig is rented, all other GPU compute tasks (like mining) is paused till the rental period is over.
- Supported on Ubuntu. May work on other linux distros but no confirmation yet.
- No support on Windows 10 as GPU VM pass through isn't available on win 10.



Hardware requirements
- Nvidia Pascal GPU and Titans Maxwell, Pascal, Volta only. No AMD support.
- Minimum 2gb ram required per GPU.
- Slight reduction in performance using Pcie x1 lanes, but works.
- No concrete info on ssd space or CPU requirement.


Other info
- You can host as many AI researchers as you have GPUs. So if your rig has 8 gpus, it may be rented by 8 different researchers running their own AI training on a different GPU.
- If your rig went offline before the rental period is up, you aren't paid.
- There's a feedback system. If your system is super slow or unstable, bad feedback may be left on your rig.


Get started
- Upgrade to 16gb ram
- Install Ubuntu
- Fill up the form on the reddit thread and wait for a reply. Creator of this app said he'll reply tomorrow.

Thoughts? I'm switching one rig over to Ubuntu for now and will see how it progresses. I really like how our rigs will be doing useful work AND we will get paid 3x more too (if it's rented).