Nvidia GPU Cloud | Hacker News Reader

Nvidia GPU Cloud

nvidia.com

81 pointsjonbaer40 comments

It's called "Nvidia GPU Cloud", but it's actually just a set of container images.

Having ready-to-use, properly optimized TF images is convenient, but calling that "Nvidia GPU Cloud" is a little bit overdicked.

Welcome to the corporate world, where buzzwords are everything.

Basically a set of fine-tuned virtual machine images for Amazon Cloud.

Yeah, I was hoping that this would be an nvidia managed cloud platform as a new competitor in the space. With only AWS, Azure, and Google there is definitely room for another deep learning cloud competitor. But this announcement makes sense in the context of the new AWS P3 instances, so there is definitely a partnership going on that nvidia wouldnt break from.

IBM Bluemix?

Bluemix is SaaS/PaaS, not IaaS.

IBM Bluemix got an IaaS layer as well. It was previously called softlayer, but is now wrapped under the Bluemix brand.

https://www.ibm.com/cloud-computing/bluemix/info/softlayer-i...

Ah didn't know that, thanks!

<strike> Looking further though, I see Softlayer does offer GPU for bare metal provisioning [0]. But it is not available in Virtual machines. Also, even for the GPU offering, it doesn't mention the details at all, not even if it is OpenCL or CUDA.

[0] https://www.ibm.com/cloud-computing/bluemix/bare-metal-searc... </strike>

EDIT - nevermind, was looking in the wrong place! Found this: https://www.ibm.com/cloud-computing/bluemix/gpu-computing

Having had experience with SoftLayer in recent years (that is, currently still deal with them and have for years), I can’t emphasize enough that you should steer clear. Year after year support and just overall stability has fallen to shit after IBMs acquisition. I can’t wait to be free of it.

sorry you have had a bad experience. Do you mind sharing what could be improved? disclaimer: I am one of the core network engineers and my opinions expressed here are my own and not official statements

When you say there’s going to be four minutes of network outage in a four hour window, I relay that to our clients. It seems to consistently push into 30 minutes. Just say 30 minutes up front and it’ll be daisy. Also, it seems like I have to pull out teeth when I have a VSI go down because of host (dom0) failure. When I call in, I literally have to argue with the support engineer that the machine is down so he doesn’t waste 30 minutes of everyone’s time to “debug” what I already know to be true (I.e. if I can’t trigger a force restart on the VSI from control or API, and the VSI is down, and “KVM” to it is down, just fucking trust me and escalate). Also, why the hell cant a ticket auto open for that, or even just a manual “open tickets to all accounts affected” button. There’s zero communication with customers, compare that to Google Cloud or the much smaller Linode. It’s really pathetic. My list of grievances can go on and on. My current one is “why did you all bother migrating us all to hardware supposedly capable of live migration, just to not use it and tell me the earlier migrated machines need to be migrated again because you all used me as a guinea pig before putting them on LTS machines!”

EDIT: also, since this is your domain, what exactly has been happening in lon/ams with networking? SoftLayer used to pride itself with its network, but it feels subpar at best now.

in your opinion would it be better to overestimate as a precaution? I think people generally tend to naturally estimate a scenario where things go as planned and present that.

About lon02 we are actually currently going through a migration to new routers and there was some possible impact communicated to customers. Many apologies it didn't go as smoothly as hoped.

And AMS? It feels just like it’s constantly something or another. I have hundreds of emails for “events” that can sometimes be hard to constantly sift through too. Feels like we need to hire someone to babysit that.

Oracle OCI also has V100 support.

What about Paperspace?

(Really, I'm wondering: did you just overlook them or is there a reason they are not competitive with the other platforms?)

Cloud is hard business though, and requires far more than chips.

It says it works on-premise, but both the video and FAQ make it sound like it is only for amazon right now.

Why do I need an API key with NVIDIA?

But can we use it to play games? You know, for research..

There’s also GeForce now. https://www.nvidia.com/en-us/geforce/products/geforce-now/ma...

There’s a public beta for macs. It’s dope as fuck. Like smooth as the fucking butter. Integrates with steam and battle.net. So you can play your games. Free in beta. I’ve been telling everyone I know about it. I can play pubg on ultra settings. There’s an extremely occasional skip but playing natively would do that too. I know I sound like a shill but it’s been a while since a product let alone beta blew me away this much.

I wonder how they did it cause the latency is very small. But maybe I’m just a noob.

On an tangential note this product has insane potential. Like they could do their own steam and basically own the whole gaming pipeline from the publishing to the platform users play on.

Thanks for this, I had no idea it existed and it looks amazing!

Definitely trying it out tonight.

Post your experience here afterwards.

Will do!

I just tried it and am completely floored. It worked wonderfully. Just some expected mouse input lag. Otherwise smooth as butter.

I have gigabit fiber so that might help a little with the overall experience but I'd be willing to pay a good amount of money for this as a service.

Ok, just started it.

I'm ABSOLUTELY FLOORED. Like, to the point where I' debating selling my gaming PC if I can get a better internet connection to reduce the input lag a bit. But even with that, the input lag is barely noticeable.

I'm playing Destiny 2, highest graphics settings, at 1920x1080, totally smoothly.... on my 2012 MacBook.

This is a game changer. Thanks for the suggestion!!

I feel the exact same way. My productivity is probably gonna take a nosedive tho.

I feel I need to tell everyone I know about this.

Looks great. Installed and ran smoothly.

Steam refused to install DOOM (the recent reboot) and COD Black Ops II. Not a failed attempt - a dialog popped up declining to install. Maybe there's a license deal behind the scenes that governs that.

Very cool that it worked at all. I had a playable game during a screen-sharing telecon on the same machine. For science only, of course.

This is awesome.

I can't wait for UK/EU, thanks for this.

Thanks for sharing! This is really cool and has a pretty broad spectrum of Games ranging from Minecraft to PUBG!

"What is it?":

> 1. WHAT IS THE GEFORCE NOW™ FOR MAC® FREE BETA? GeForce NOW for Mac beta is a new service from NVIDIA that connects you to a powerful GeForce GTX gaming PC in the cloud, turning your Mac into a powerful gaming rig. We’ve launched a free beta of this service. so you can try it out at no charge. The GeForce NOW beta app will require you to log in to Steam to install games included in our list of supported games. You must already own or purchase these games to play them.

Other interesting note: 10. HOW MANY GAMES CAN I INSTALL? Install as many of your games as you like. GeForce NOW game storage is unlimited.

You can play anything you own on steam r battle.net.

Yes here is a guide to make your own. That's from 2015 but the idea is the same. https://lg.io/2015/07/05/revised-and-much-faster-run-your-ow...

But there are services too

https://liquidsky.com

https://parsec.tv

Not really seeing a material difference between this and just using one of the available machine learning AMI's on EC2. Relatively anticlimactic

From NVidia's perspective, the more of the GPU computing stack it's in control of, the better.

Paperspace also announced NVidia Volta VMs available today. As well as a contest for 100 compute hours credit ;)

https://blog.paperspace.com/tesla-v100-available-today/

This seems to redirect to a 404 at the moment?

The associated YouTube video is still up, however: https://www.youtube.com/watch?v=MXmFRhA1VTk

https://www.nvidia.com/en-us/gpu-cloud/ doesnt work

use https://ngc.nvidia.com/signup/register

thanks...

https://ngc.nvidia.com/signup/register

"This eliminates time-consuming and difficult do-it-yourself software integration..."

writing easy to integrate, open source, software would also eliminate that

I'll take "Things Nvidia would never do" for $200, Alex.

Reply on news.ycombinator.com