I get that AI is all the rage right now. But this is an information-thin Microsoft press release, if not outright ad, with the OpenAI stamp on it. These things really should not be on the front page.
Naw, what they're talking about with infiniband networking ends up being important for distributed GPU workloads and is pretty cool to see in a cloud offering.
Being "impressed by their roadmap" is legitimate; Microsoft is putting a lot of energy behind competing with AWS. I've been hearing good things about Azure, but have been an AWS user since ~2009. I haven't heard they're much better or cheaper than AWS, so I haven't bothered to switch.
There's certainly PR value here, I just don't think it should be dismissed outright because of that.
Much of the cheaper comes when you get the large enterprise discounts (versus amazon). Those don't last forever though. Their offering is definitely far more immature, but they are making constant improvements. They tend to be leveraging lot of ML stuff in the security/management space as well, which is pretty interesting (they have a vast trove of threat intel from the millions of MS defender and O365 type deployments).
If it was some other company than OpenAI, I would agree. But just the fact that Microsoft is partnering with OpenAI is noteworthy to many here.
OpenAI is a non-profit, and I don't know how lucrative the deal is for Microsoft or how much business OpenAI has, but I don't mind Microsoft getting some good press for the move either.
Microsoft probably offered them a really sweet deal (read: tons of Azure credits), in exchange for the press/marketing.
Amazon is pretty stingy with those kinds of things, so they are very Jeff Bezos when it comes to startup credits and the like.
The "open" name is telling, OpenAI's work should be agnostic of platform, and they shouldn't be pitching others' proprietary systems in public like this.
I think it would have been better had they mentioned some financial costs for OpenAI to use Azure. Since OpenAI is a non-profit foundation, giving them a free or substantially cost reduced access would mean the Microsoft is doing their bit to promote AI research. There are various ways to make something whole.
I am actually getting more and more impressed with Azure. The rate of maturity of their offering is really impressive, and the number of world-wide regions is insane.
Wile we're primarily still AWS users, but we're hosting free global tools (check out https://teleconsole.com) on Azure for low latency.
I've been using Azure in various forms since beta days. Back then it felt very startup-like with half baked software and updates coming all the time. That is still how their new stuff is delivered, but I've rarely had issues with anything that is out of preview.
Here what really sold Azure to me, the various companies I've worked for, etc. Azure allows you to do just about anything these days, and you can mix & match all sorts of different pieces of tech as you see fit, and its all the in one place. It's quite important for businesses that I have, for example, probably half a dozen ways to open a connection into my corporate network in order for my front end Azure-hosted API can access some ancient database. Microsoft knows how to make Azure appeal to businesses, since that is their bread and butter.
One last thing: where I currently work we have been running or entire infrastructure on Azure since day 1. We have to architect our applications for the cloud, but the benefits have always been amazing. Whenever we get bigger customers that slam our infrastructure with traffic, we just push a button (or automatically) and we get more machines/bandwidth/storage, etc. Not once has there been feelings of regret for choosing Azure in the company. It's a safe bet.
I mostly use Azure for hobby/small/exploratory work, but I'm generally quite happy with it, sticking mostly to their PAAS offerings, and a Linux VM here and there. I never touch their UIs, which are terrible indeed. But the CLI tools do everything and I'm pretty happy with those.
I'm currently working on an azure based project where we have scripted all of our deploys in powershell. Pretty quick that way and that also means you dont have to deal with UI
What exactly did Google do to Apple in Mobile ? You mean marketshare ? I believe Apple is fine with that, they still make the majority of profits in the mobile business.
I just taught a class[1] where the idea was that everybody would learn to spin up a GPU-backed AWS instance and learn to train their own tensorflow models.
Unfortunately, I was the only person who was able to launch an instance, though almost everyone was able to log into AWS and go through the pre-launch instance configuration. I think maybe having everyone request a g2 instance at roughly the same time triggered some sort of fraud-prevention system.
Maybe there's some kind of priming people can do to prepare their accounts before the class, but maybe next time we should try Azure.
(The class went mostly OK, though. I just showed people what I intended to help them do themselves.)
Martin Wicke had a few hundred of us up and running in AWS-powered lab environments, each with our own GPU, at the NVidia GTC in April. Now, they also had a direct line of communication to somebody at AWS to make sure that many GPUs were available. Maybe a call to let them know what you're up to? I'm sure they want people using the gear.
I don't think the problem was resource availability. Some of my students attempted to start t2's because we weren't going to be doing particularly heavy calculations in this class, and were rejected.
How much better is Azure at cloud GPU compute? I remember they were first to roll out K80 instances, but Amazon followed suit shortly after, so it seems like it would be a wash. Is the interconnect story on Azure way better or something, or is this just fluff?
