Microsoft Readies Azure GPUs
virtualizationpractice.com
virtualizationpractice.com
Comparing specs:
AWS g2.2xlarge:
* DRAM: 15GB
* GPU processor: 1xGK104
* 1536 CUDA cores
* 2.29Tflops/s peak single precision throughput
* GPU memory: 4GB
* 160GB/s peak bandwidth
* Price: $0.65/hr (US East, Linux)
Azure NC6:
* DRAM: 56GB
* GPU processors: 2xGK210
* 4992 CUDA cores
* 8.73Tflops/s peak single precision throughput
* GPU memory: 24 GB
* 480GB/s peak bandwidth
* Price: $0.66/hr
This isn't the same as having 1 GPU with 4992 cores and 24GB of memory.
CUDA kernel scaling is pretty decent but it's not 1:1 with additional GPU's it's also very much task specific so tasks that cannot be easily parallelized will not scale well or at all at least not without you splitting the workload manually. As for the memory scaling CUDA supports P2P memcopy so you can easily move memory from 1 GPU to another (or between hosts in a network distributed configuration) but under best case scenarios this goes through a PCIE bridge (in the K80 it's onboard) but no NVLINK support on the K80 as it's 2 gens behind Pascal, if you are using P2P memcopy you do not double the memory capacity (even lose some since memcopy requires another buffer iirc). CUDA also allows you to share the memory directly between multiple GPU's (and the CPU/Host) NVIDIA calls this Unified Addressing (OpenCL 2.0 later implemented something similar as Shared Virtual Memory) but this also comes at some cost the latency is greatly increased for memory access and you really need stellar code to make everything work properly if 1 GPU tries to dereference a pointer on another GPU while it's still is in use whelp "don't panic and move on" probably won't work :P
It's really amazing just how much money NVIDIA seems to be squeezing out of the compute market (almost makes you feel bad for AMD) especially considering that Kepler is a 4 almost 5 year old architecture by now. The K80 is still going for like 4000-5000$ a pop I would love to know just how many GPU's Microsoft ended up buying.
Not so for machine learning, at least in my experience.
http://pjreddie.com/darknet/hardware-guide/
Passthrough:
Might not be applicable in your case, but NVIDIA hasn't released a desktop part with FP64 capabilities even remotely comparable to their Tesla's since the original Titan.
Both Titan X's (Maxwell and Pascal) and the [Insert Physicist's Name Here] Quadros are pretty gimped as far as it comes to HP/DP operations.
If anything between generations DP performance across desktop parts have been severely degraded from 3:1 for Kepler to 16:1 for Maxwell to bloody 32:1 for Pascal desktop GPU's.
As for the passthrough it is possible but it requires a few hacks and is technically in violations of NVIDIA's policies, and unlike the Tesla parts there is no full virtualization once the a device has been initialized it can only be assigned to 1 host/guest and a full host reboot is required to reinitialize the device again. You also need to do a few UEFI hacks to prevent the UEFI from initializing the GPU before your host OS loads and it can be passed through to the guest.
This isn't something that cloud providers would dealing with, NVIDIA will not sell them anything but Tesla/Grid parts and buying from AIB's/OEM's and hacking your way through it isn't an option, you'll get zero support from NVIDIA, you could not thin provision your GPU's and you'll need to do a full host reboot every time you want to reassign the GPU or try to figure out if you can write a custom BIOS for your card that would somehow allow you to reinitialize it.
P.S. I've actually built a KVM multi-user gaming rig with NVIDIA cards, I would not recommend anyone to try that, AMD cards are more pleasant to work with, too bad the performance are utter garbage for high end gaming.
Baidu has actually built Titan X -based servers for themselves as they saw it as the most cost-efficient way to build the system: http://www.nextplatform.com/2016/04/22/baidu-eyes-deep-learn...
Same logic would probably work in "deep learning" clouds if there would not be the licensing issue.
And this conversation got immortalized to that slideshow as well for selecting hardware parts for deep learning systems: http://tinyurl.com/zzhj9vr
Also funny story, I was 14 or 15 and was trying to figure out a name for a stupid little chat client I was trying to make in C#, and somehow azure popped into my head. Not surprisingly azure.com was taken and just had a generic microsoft landing page (this was a little before it was announced in 2008).
Here's how you do it with Tensorflow:
https://www.tensorflow.org/versions/r0.10/how_tos/variables/...
The sweet spot would be updated AWS GPU instances at spot prices. You could spin up XX instances for a relatively short time to train fast. Needs some Kubernetes etc cluster setup for distributed training but should be doable. If the spot price is $0.1-$0.2/hr that's pretty attractive, considering that is close to the price of electricity alone
Also, you tend to be getting more per instance hour on both Azure and GCE, assuming you do things those environments are good at. In Azure's case if you can get your software running on windows then it's actually quite good performance for I/O intensive things.
I should check that again, it's been awhile.
But it depends on your platform. Its an unpopular but somewhat verifiable fact that the Windows kernels have much more powerful async I/O support and when software targets that, it is very fast compared to linux counterparts. If your toolkit supports this, then a strong argument can be made that you don't really care so long as SSH works. Powershell is serviceable for maintenance and most deploys are fairly agnostic to other aspects of the platform.
I've been doing skunkworks projects. I tried to get some stuff working in GCE and their docs are a mess and it never worked. Azure has some warts in its git push deploy stuff, but it worked the first time and I managed to diagnose the other issues.
async DISK I/O
Pricing does go up near computer vision conference deadlines though. Many research labs are too poor to buy our own hardware. :-)
Which GPUs they offer and how much, nobody knows.
They might not offer GPUs at all. It may be that Google is planning to build a cloud using their own proprietary "Tensor Processing Unit" chip designed for TensorFlow (https://en.wikipedia.org/wiki/Tensor_processing_unit)
It looks like the NV6 Azure instances use a newer Maxwell based chip, which can handle much more (concurrent, or higher bitrate/resolution) videos, compared to the 2 generation older AWS GPUs.
And the NV6 is virtually the same price as the g2.2xlarge... time to do some testing. Hopefully Amazon can catch up soon.
K1 (AWS): 1 GPUs x 6 H.264 streams (max single stream 720p@30fps)
K2 (Azure K80): 1 GPUs x 6 H.264 streams max single stream 720p@30fps)
M6 (Don't know if it's used by anyone ;)): 1 GPU x 18 H.264 streams (max single stream 1080p@30fps)
M60 (Azure M60): 1 GPU x 18 H.264 streams (max single stream 1080p@30fps)
You can sacrifice encoder bandwidth (max streams) for higher resolutions or frame rates.
The quality on the other hand even for the same bit rate is considerably worse than highly optimized CPU compilers as the GPU does do a lot of hacks along the way.
Azure: 2 year-old GPUs
Wouldn't it be awesome if a cloud provider could keep up with the GPU roadmap and ship something within a year or so of its release?
Interestingly, for the first time ever, GP100 is unique (well, OK, K80 too but K80 was too late). And since Quadro P6000==Titan XP==GP102, it's probably just a software block here as well.
Also, for the first time ever, the high-end Quadro will be the best FP32 GPU of it's generation. That's the most interesting part for me.
To overcome this you need to disable UEFI boot/GPU boot in the BIOS and blacklist the card in the host OS and then create a PCI-stub device that will be used for the passthrough.
This is an utter and complete hack and you can't really use for production grade implementations.
I'm not sure that Quadro actually supports vGPU also AFAIK only Tesla and Grid parts do, Tesla does have some additional in GPU support for virtualization, i don't know how GRID handles it. GRID parts are less for compute and more for game/video streaming so they aren't fully virtualized, AFAIK they do not support P2P gpu communication or shared memory access, they are only more Hypervisor friendly so they can be thin provisioned.