dkobran

Great idea. What do you recommend as a good standard/tool we could use to publish benchmarks? We have some CPU/memory-centric instances but our primary focus is on GPUs.

Sorry for the confusion here, we are in the midst of separating out the two products and this is not reflected everywhere yet. Under the CORE section of the interface, you'll find these instances and you can download the desktop app here: https://paperspace.com/download

Nice catch, typo (the values are correct but the text said "/hour" instead of "/month").

Disclosure: I work on Workstream

The US, like the UK, did a terrible job early on when containment was most critical. Several states are in lockdown now but arguably, the reason for our awful current new cases/death rate is due to the resistance to taking action in February when it first hit. There are also many places in the US that are not in lockdown. From another comment: “Approximately 44 of 50 states have lower per capita death rates than Sweden.“ The US cannot be compared directly to Sweden because it doesn’t have a single policy like Sweden. It’s also worth noting that Sweden has taken some lockdown measures ie your comment about “no lockdown” isn’t accurate. Finally, it’s early days and time will tell. The UK changed their policy because there is no credible evidence that herd immunity is effective. I believe Sweden will as well once the disease spirals out of control. And hopefully less people will die as a result. Unfortunately, there will also be the insufferable few that claim we could have survived without being cautious because you can’t prove that those measures were responsible for saving lives. Don’t be that person.

Sweden has by far the highest number of cases/deaths with respect to neighboring countries. Finland has 1/10th of the number of deaths per capita [0]. They are also faring much worse than Denmark and Norway (in terms of both cases and deaths) where much stricter measures are enforced.

[0] https://www.forbes.com/sites/davidnikel/2020/04/14/sweden-22...

Yes, many of the GCP products that were launched in the past six months were not shutdown.

GCP has been around for a fraction of the time of Google as a company. It is perfectly valid to have at least some concern here as Google as a company has a long track record of shutting down products. Conversely, it is tenuous at best to use the argument that since GCP has never shutdown a product that they won’t given how many of their products were launched very recently.

Clarifying a few inaccurate points:

The Colab model of "resources are free, just not guaranteed" model” that you mentioned is literally identical to Gradient: The free tier includes spare capacity at no cost just like Colab. You can upgrade to access high-end instances eg the NVIDIA V100. These additional instances do not have a max runtime as well. I can’t see how this would be a con given this additional option is simply not available in Colab.

Gradient does not have a $5 deposit for storage, the free plan does not cost anything. There is no credit card required. In fact, Gradient doesn’t charge for storage at all.

Here is a working GauGAN sample notebook you can fork: https://ml-showcase.paperspace.com/projects/gaugan

Disclosure: I work on Gradient

It does and we offer it. It's called the P6000 and it includes 24GB GPU RAM. It's one of the most popular chips we offer for any kind of CV task as you can fit a ton of images, video frames etc. in GPU memory. In any case, here's a link to the full lineup we offer: https://gradient.paperspace.com/instances

Paperspace team here. We offer a 24GB GPU option with minimal RAM. Check out the P6000 here: https://gradient.paperspace.com/instances Hope that's what you're looking for :)

Interesting to see another cloud provider go with Quadro chips. NVIDIA repackages the same silicon under several different brands (GeForce, Quadro, GRID, Tesla) and we (https://paperspace.com) have found Quadro to offer the best price/performance value. Despite minor performance characteristics, such as FP16 support in the Tesla family, Quadros can run all of the same workloads eg graphics, HPC, Deep Learning etc. If you’re interested in a similar instance for less $/hr, check out the Paperspace P6000.

This is good advice but we actually do pull from BofA. The mere act of using TransferWise is what triggered the freeze. Or at least that's what I was told.

We've definitely considered this and the only reason we haven't is that BofA offers a relatively nice mobile app, web interface, support (24/7), and permission management (roles) for our team. I agree in principle but it's a tool that we actually spend a non-trivial amount of time in so the experience of interacting with the service is something we, unfortunately, can't fully discount.

NVIDIA is attempting to separate enterprise/datacenter and consumer chips to justify the cost disparity. Specifically, they're introducing memory, precision etc. limits which have major performance implications to GeForce and there's also the EULA which was been mentioned here. That said, everything AWS comes at a premium as they're making the case that on-demand scale outweighs the pain of management/CapEx. This premium is especially noticeable with more expensive gear like GPUs. At Paperspace (https://paperspace.com), we're doing everything we can to bring the cost down of cloud and in particular, the cost of delivering a GPU. Not all cloud providers are the same :)

Disclosure: I work on Paperspace

In case you missed it, here's a link to the full training example that you can run yourself: https://www.paperspace.com/console/jobs/jvqssfqawv5zn/logs

Inference example: https://www.paperspace.com/console/jobs/js4mqzm91fj2lg

Disclosure: I work on Paperspace

Cool, just got downvoted a bunch. The Amazon mafia takeover of HN is ruining this community.

Also on HN today: “Amazon was just granted a patent for tracking Bitcoin transactions and selling the data to the government”

https://www.techspot.com/news/74246-amazon-granted-patent-tr...

Just that you’re limited to the number of saved and running experiments (Jobs) on the free plan. For most research/non-production environments, this is sufficient. Hope that helps.

There are referral codes floating all over the internet ;)

Paperspace Cofounder here. We just launched Gradient, a really simple and affordable DL platform: https://paperspace.com/gradient Feel free to ping me with any questions.

I agree, how we are tracked needs to be more transparent/upfront. Clearly these companies will make every possible effort to hide what’s going on behind the scenes so this stuff needs to be regulated. Companies like Google/Facebook are not upfront about what they capture and how they capture it because they know many people would not be comfortable with it. Unlike non-advert business models, the actual transaction that is happening is tucked away in a EULA. This to me is the sign that consumers, especially non techies, need protection.

Next year the AI will enter the mayoral race itself :)

How are you going to release a Broadwell based server in 2018? Sourcing Broadwell processors is already difficult today.

NVIDIA makes a chip called the P6000 that offers 24GB memory. It isn't high-throughput (HBM2) memory but regardless, the chip will benchmark relatively close to the Volta V100 and costs about half. Here's a deep learning benchmark for comparison: https://github.com/rejunity/cnn-benchmarks

HBM2 to be specific: https://en.wikipedia.org/wiki/High_Bandwidth_Memory#HBM_2

It's my understanding that fp16 (available on the previous generation P100) and mixed-precision (major innovation of V100) are different things and the speedup of TensorCores is entirely missing from this benchmark. Unlike the general purpose P100, the TPU is a heavily optimized chip built for Deep Learning, hence it's performance increase. However, the V100 is also heavily optimized for Deep Learning (arguably the first non-GPU chip) from NVIDIA. I'm in no position to defend NVIDIA here haha but it seems like the benchmark misses the point if this is indeed the case.

Just to clarify, is this benchmark leveraging mixed-precision mode on the Volta V100? The major innovation of the Volta generation is mixed-precision which NVIDIA claims is a huge performance increase over the Pascal generation (P100 in the case of your benchmark).

Link to NVIDIA documentation on mixed-precision TensorCores: https://devblogs.nvidia.com/inside-volta/

Dan from Paperspace here. Our fast.ai template is Linux based but you can spinup a GPU backed Windows instance on our cloud if you’re interested in a remote option. Our streaming tech is GPU accelerated so it feels snappy.

This makes me want to die