dkobran

People are always talking about spot pricing when GPU compute comes up which is valid but not for the ML use-case (which this article is discussing). Spot is 100% incompatible with ML training. You need to be guaranteed that your many hour/multiple day job is going to complete which is antithetical to spot. At Paperspace, our goal is to provide a steady state low price that you can depend on because our primary audience is ML. Spot would be more ideal from an efficiency perspective but it’s just not viable in this context.

Dan here (also Paperspace team). Totally agree that transfer costs are a significant pain point which is why we do not charge for it. We can peer with other providers (eg with AWS we can leverage Direct Connect directly from our datacenters) but most of our customers don't implement this unless they're moving major traffic.

Totally agree though the concept of the uncanny valley is based on computer generated or robotic representations of humans specifically -- not other forms of life or inanimate objects. Meaning, not just realism in general.

Thinking that tech money can't destroy a respected hundred year old publication is naive: https://www.theatlantic.com/magazine/archive/2017/09/when-si...

While the job doesn't sound like it involves "cool things that matter" and paints a pretty believable picture of working at a large corporation, I do find it interesting that this was published by thr Washington Post. Being owned by Amazon, a direct competitor to Google on many fronts (the war for cloud computing market share is/will continue to be huge), it seems suspicious that this is truly unbiased coverage. It strikes me as a possible next iteration of native advertising. This is not an attack on WP -- just something I thought might be worth discussing. The tech giants are becoming increasingly more powerful and this could be a new frontier to assert that power.

Taking a stand against companies fighting to undermine privacy is a good reason to choose a product like Firefox. I don't think it's as much about being a for-profit [3] as it is about the intentions of the company, its influence, and the way it makes money.

Really interesting glimpse into this world.

It's unfortunate that this article is getting negative feedback on HN. I think we all need to be careful not to belittle/condemn issues that don't personally affect us.

If you want some current generation GPUs ;) check out https://paperspace.com

One of the interesting variables in calculating ML training costs is developer time. The cost of a Data Scientist (or similar role) on an hourly basis will far outweigh the most expensive compute resource by several orders of magnitude. When you factor in time, the GPU immediately becomes more attractive. Other industries with heavy/time consuming computational workloads like CGI rendering have understood this for decades. It's difficult to attach a dollar sign to the value of speeding something up because it's not only about simply saving time itself but also about the way we work: Waiting around for results limits our ability to work iteratively, scheduling jobs becomes a project of its own, the process becomes less predictable etc.

Disclaimer: Paperspace team.

The citation is we're building a GPU cloud and have tested almost every GPU in existence :) Just kidding, here are a few examples:

http://vfio.blogspot.com.au/2014/08/vfiovga-faq.html https://www.reddit.com/r/linux/comments/2twq7q/nvidia_appare... https://www.redhat.com/archives/libvirt-users/2014-October/m...

I just quickly googled this so there are probably better sources. Some of these are old but I can tell you firsthand that this is still the case.

There are workarounds for certain hypervisors (KVM mainly) but it's very unlikely that this would be deployed in a production environment.

GTX drivers become crippled when the card detects the presence of a virtual environment. This means you can't run GTX in the cloud, otherwise, cloud GPU prices would be much lower. Without the availability of GTX, we've been trying just about everything at Paperspace to bring prices down and make the cloud a viable option for GPU compute. The argument being, there are real benefits to running in the cloud like on-demand scalability, lack of upkeep, minimal upfront costs, and of course, running a production application :) There are other indirect cost saving eg power consumption which can be quite significant when training models for long periods of time. Would love to see a total cost of ownership figure added to this post.

100% agree with this article. The cloud is where heavy computational work belongs but it's not trivial to leverage these resources today given the complexity of existing offerings. We built https://paperspace.com to make this easy and affordable. Ping me if you want to test out the platform or if you have any questions!

Hi there, Paperspace Co-founder here. There's an in depth write-up here: https://medium.com/initialized-capital/benchmarking-tensorfl... We're about about a quarter of the cost of AWS. Moreover, the K80 GPU on AWS/GCP is very old and NVIDIA has made a ton of optimizations with the new Pascal architecture which we offer. Feel free to reach out if you have any questions.

Paperspace co-founder here -- happy to discuss our AWS alternative which is focused on affordability, simplicity and performance. Recently, we've had a huge influx of people in moocs offloading their work to our GPUs/cloud. Would love any feedback you have.

Kudos to Google for making moves here. Having spent the last year+ tackling GPUs in the datacenter, super curious how custom sizing works. It's a huge technical feat to get eight GPUs running (let alone, in a virtual environment), but the real challenge is making sure the blocks/puzzle pieces all fit together so there's no idle hardware sitting around There's a reason why Amazon's G/P instances require that you double the RAM/CPU if you double the GPU. Another example would be Digital Ocean's linear scale-up of instance types. In any case, we'll have to see what pricing comes out to.

Shameless plug, if you want raw access to a GPU in the cloud today, shoot me an email at daniel at paperspace.com We have people doing everything from image analysis to genomics to a whole lot of ML/AI.

Initialized is venture capital without the attitude. Garry has been an incredible advisor to us ever since our first YC interview (he and Alexis were both there actually). Could not imagine a more down to earth, insightful and energetic group. Congrats!

Paperspace founder here. One of the things we are most excited about is rapidly evolving cloud licensing models. Adobe (among others) has paved the way for a more on-demand version like what you are describing. I think a lot of other companies will move in this direction as well. We definitely share your vision and hope to offer pre-configured machines as well as on-demand apps in the future. Also, you can clone a VM and distribute it really easily which could be interesting (a dumber approach to what you're talking about).

Paperspace founder here. So sorry that the copy is confusing--we've been really struggling with this but that's exactly what is going on. Paperweight is just a really cheap/dumb computer that connects with our servers that do all the heavy lifting. We prototyped a lot on the Pi, very similar setup. Any ideas on how to improve the copy would be appreciated...

Thanks for the typo catch... Great use-case btw, it's surprising how well virtual desktops can perform on slow connections. I'm from the east coast but work in Mountain View now and I stream my desktop from Virginia everyday for work. Right now in fact haha.

Paperspace founder here. You can take snapshots at any time and we offer an instant rollback to a safe state in the event something happened. One of the neat benefits of virtual machines over traditional desktops.

Paperspace founder here. We offer an instant rollback to a safe state which is a neat thing that VMs can do that regular computers can't (or at least don't do very well). Everything is stored in the cloud so have your entire desktop available to you at all times meaning you wouldn't need to transfer your stuff to a cloud storage service. It's actually just like cloud storage but with the added benefit of having your whole operating system (apps, settings, shortcuts) on the go.