I wanted to open a conversation and get some feedback on the seeming unavailability of a compute cloud featuring AMD GPUs. I have been working on an OpenCL project for over a year now and all the development has been done on my own Radeons. When I went to look for a public host, I was unable to find one. nVIDIA has a nice list on their website (GPU Cloud Computing Service Providers | NVIDIA) but following those links resulted in finding they were nVIDIA only shops. Because this code is intended to be used for free, I can't afford to stand up and maintain a number of servers. I was hoping to find a solid 'cloud' host featuring high performance AMD FirePro S class GPUs for a reasonable rate. Ideally less than $100 USD a month for a nominal amount of usage; which represents the maximum amount of money I'd pay give away free software 🙂 Unfortunately, it seems I will have to either run the OpenCL code in CPU only mode or hope that nVIDIA's OpenCL 1.2 support is complete on the ~2 year old GRID GPUs everyone seems to be offering with their service. Honestly, it boggles my mind given the raw compute advantage of Radeon/FirePro that there isn't an AMD cloud system. Which is why I am writing this, in hopes that someone, somewhere knows of a reasonable and reliable AMD based GPU Compute Cloud where I can host my web service.
Ooh, good question. I'm digging around here and will let you know what I find. But anyone else?
Seconded. I've been told AMD has been focusing on virtualization only "recently" (whatever that means; I don't really follow virtualization). Following discussion.
Maybe we can raise a flag to see if anyone can donate GPU card, node, we just need to find out a ISP to host those device.
I'm planning to working a project to scheduling application/OpenCL can run across GPU/CPU cluster as well, we can see what we can share here.
I did track down what's happening. Unfortunately, there is nothing I can discuss right now about this. We are working to make cloud compute on AMD GPUs happen with some cloud providers whose names you would recognize. I cannot say anything about whom, or when.
I'm just curious of you would be able to update us on this. I'm hopeful something is closer to being a reality.
Nothing I can point to yet. The world of big business turns slowly. It is turning.
Are there any developments on this? Finding a GPU Compute Cloud Service (hopefully ASAP) is crucial (imperative in fact) to my business. Are there any cloud or even dedicated service providers out there? Even a single one would be enough. If you know anything, please help.
It sounds like AMD has something in the works. There are several GPU Cloud Compute offerings featuring nVidia. However, it seems you are in the same boat as me in that we have developed our applications to work best on AMD hardware and moving to a different platform is not ideal (E.g. not even sure it will work without doing some recoding).
<bump/> Does anyone know of any new GPU cloud compute services that feature AMD hardware? Does AMD have any plans for this.
As the OP stated, all of the GPU cloud services feature nVIdia hardware. The only one using AMD cards that I am aware of is a beta service
from Stream Computing.
It has been about a year since I first asked this question and see that AMD has inked a deal with Alibaba to use Radeon virtualization in their cloud services. Anybody know if anyone in the North American continent has set up such a cloud server where I can buy time on Radeons?
AMD and Google revealed plans for Google Cloud to offer FirePro SC9300x2, Tesla P100, and Tesla K80.
(AMD's press release gives a vague "Starting in 2017", a 365-day date range. Google's press release gives a more optimistic "Early in 2017", but includes both AMD and NVidia GPU compute accelerators.)
When these become available, a non-shared Google cloud CPU instance will be connected to "up to 8 GPU dies", which presumably means up to 4 FirePro SC9300x2 boards with 2 "Fiji" GPU dies each.
Google has a server cloud gpu customer notification form and survey to gauge which and how much hardware customers can use, and the applications they need to run.
Amazon announced AppStream 2.0 enabling one or more users to share parts of a virtualized S7150x2 "for streaming graphics applications" "that rely on hardware acceleration of DirectX, OpenGL, or OpenCL".
Introducing Amazon AppStream 2.0 Graphics Design, a New, Lower Cost Instance Type for Streaming Grap...
The S7150x2 appears in the "Graphic Design" instance
AppStream 2.0 Instance Families - Amazon AppStream 2.0
Pricing seems to be $0.25/GB/Hr, or $2.00/Hr for 8GB (GPU memory). A 2-month free tier offers 40 hours per month.