All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
GPU VPS - R&D
Hello,
I'm potentially going into the VPS market in Q1 next year with a small startup firm.
We’re looking at implementing GPU VPS’.
The tiers and prices would potentially be as follows if this is an avenue we want to go down:
Gpu1:
Low end hardware suitable for budget virtual workstations.
2 cores (dedicated)
4/8GB RAM
100GB SSD cached RAID 10 storage
GPU: GT710 or better (dedicated)
KVM virtualisation
Cost: ~£22 Monthly
Gpu2:
Mid range hardware, suited for low end game streaming, video transcoding or as a virtual workstation.
8 cores (dedicated)
8/16GB RAM
100GB NVME SSD
250GB RAID 10 HDD
GPU: GT1030 2GB
Cost: ~£40 monthly
Gpu3:
High end hardware (AMD EPYC)
12 cores (dedicated)
16GB RAM
250GB NVME SSD (dedicated)
GPU: GTX1060 6GB
Cost: ~£65 monthly
I’m just looking for feedback to see what the market would look like on this.
We have the potential for 4 locations and all hardware would be owned.
Unfortunately I can’t go into any further details as of yet but any feedback is appreciated!
All hosts would have 1Gb network shared from a 10Gb port with a single ipv4 address (ipv6 is available in 3 of the locations we're investigating). Nothing would be over sold, not even CPU cores. All virtualisation is KVM. There will be the possibility to use your own GPU (mail it in) for a fee.
Like I say this is all just hypothetical at the moment and before we start further work on developing a panel and investing in more test hardware I wanted to get peoples thoughts on this.
Thanks for any feedback!
Comments
You cannot legally use GTX/GT series in hosting/DC/commercial use/VM, you have to use their high tier line
Yeah that is something that we looked into and we’re currently looking at work around, higher end cards are absolutely an option but we want to keep the costs down for the end user.
As much as NVIDIA do ban the use of consumer grade cards within a data centre environment it never actually becomes an issue until the driver is installed. From my understanding this is not an issue with open source drivers. Although I am waiting for confirmation on this.
There is also the option for AMD GPU’s....
Those GPUs are relatively low-end imho. Could definitely see usage as storage VM transcoders, but maybe not as a workstation. Not even sure that Parsec supports such old hardware as the 710 series (for cloud gaming).
@James94 I don’t know of open source drivers in Windows, and for Linux, Nouveau drivers don’t support CUDA (Usually used in ML), so it’s an uphill battle. Docker does work, though!
That being said, if you are interested in hosting better GPUs, shoot me a private message. We at FluidStack run a GPU marketplace and would be happy to have more providers. Payouts range from ~15/month for servers with 1x 1050 to $600+/month for A100 machines. We rent the machines from you, so no matter if we have customers or not, you will be paid.
@servarica_hani has something similar going on over there..
Yeah the idea is to start off low end and branch into the higher end market. There is a limited amount of High end GPU VPS' available and practically no low end ones available. The GT710 will plan would more than likely be replaced with the gt1030 plan. I just wanted to see if anyone would have any interest.
That is true on the windows side, its very much a grey area as it is fine to install the windows drivers in a DC environment if crypto is being mined.... AMD might be the better option here, or investing in cards that are designed/allowed to be used in a DC. Its something we're still looking into.
If/when we start this we will absolutely have an abundance of GPU servers until they are sold, we would be starting out with a full rack and somewhere between 60-80 GPU's on the first run.
Maybe we could work something out?