Learn how Palo Alto Networks is Transforming Platform Engineering with AI Agents. Register here

Attend a Live Product Tour to see Sedai in action.

Register now
More
Close
AWS

p2.16xlarge

EC2 Instance

Previous generation GPU compute instance with 64 vCPUs, 732 GiB memory, and 16 NVIDIA K80 GPUs. Highest GPU density in P2 family for large-scale parallel processing.

Coming Soon...

icon
Pricing of
p2.16xlarge

N/A

On Demand

N/A

Spot

N/A

1 Yr Reserved

N/A

3 Yr Reserved

Pricing Model
Price (USD)
% Discount vs On Demand
sedai

Let us help you choose the right instance

icon
Spot Pricing Details for
p2.16xlarge

Here's the latest prices for this instance across this region:

Availability Zone Current Spot Price (USD)
Frequency of Interruptions: n/a

Frequency of interruption represents the rate at which Spot has reclaimed capacity during the trailing month. They are in ranges of < 5%, 5-10%, 10-15%, 15-20% and >20%.

Last Updated On: December 17, 2024
icon
Compute features of
p2.16xlarge
FeatureSpecification
icon
Storage features of
p2.16xlarge
FeatureSpecification
icon
Networking features of
p2.16xlarge
FeatureSpecification
icon
Operating Systems Supported by
p2.16xlarge
Operating SystemSupported
icon
Security features of
p2.16xlarge
FeatureSupported
icon
General Information about
p2.16xlarge
FeatureSpecification
icon
Benchmark Test Results for
p2.16xlarge
CPU Encryption Speed Benchmarks

Cloud Mercato tested CPU performance using a range of encryption speed tests:

Encryption Algorithm Speed (1024 Block Size, 3 threads)
AES-128 CBC N/A
AES-256 CBC N/A
MD5 N/A
SHA256 N/A
SHA512 N/A
I/O Performance

Cloud Mercato's tested the I/O performance of this instance using a 100GB General Purpose SSD. Below are the results:

Read Write
Max N/A N/A
Average N/A N/A
Deviation N/A N/A
Min N/A N/A

I/O rate testing is conducted with local and block storages attached to the instance. Cloud Mercato uses the well-known open-source tool FIO. To express IOPS the following parametersare used: 4K block, random access, no filesystem (except for write access with root volume and avoidance of cache and buffer.

icon
Community Insights for
p2.16xlarge
AI-summarized insights
filter icon
Filter by:
All

Iraklis ran Hashcat (a popular password "recovery" tool) with GPU support against the largest instance size available... the p2.16xlarge.

19-03-2025
benchmarking

Iraklis ran Hashcat (a popular password "recovery" tool) with GPU support against the largest instance size available... the p2.16xlarge.

19-03-2025
benchmarking

MATLAB users moving their analytics and simulation workloads onto the AWS cloud require their analyses to be processed quickly.

19-03-2025
benchmarking

Engineers utilizing the AWS cloud for simulations will definitely appreciate the added computational horsepower.

19-03-2025
benchmarking, development

They have chips dedicated to video encoding/decoding with the latest cards doing h265. Nvidia has an ffmpeg patch here

30-09-2016
development

By giving these cloud-based engineers access to GPUs, their computations should speed up considerably.

19-03-2025
benchmarking, development

This depends on if you're doing live transcoding or VOD transcoding.I work with live transcode, and it can be beneficial to run on g2 instances. Running c4.4xlarge I can transcode a good number of 1080@30 in with 1080/720/480/360/240@30 out. With a proper g2 instance I can transcode more simultaneously.So cost efficiency really depends on sustained traffic levels. I scale out currently using haproxy and custom code that monitors my pool and scales appropriately. But I monitor sustained traffic levels to know when it makes financial sense to scale up.If your main concern is transcode speed, CPU is likely sufficient -- I am unable to transcode faster than real time with live transcode.

30-09-2016
development, cost_savings

Iraklis ran Hashcat (a popular password "recovery" tool) with GPU support against the largest instance size available... the p2.16xlarge.

19-03-2025
benchmarking

> I am unable to transcode faster than real time with live transcodeWell, yeah. Maybe quantum computing will change that one day!What is it you're working on? Colour me intrigued.

30-09-2016
development

Anyone knows if video transcoding on GPUs (with FFMPEG) is viable nowadays? If yes, what are the gains?

30-09-2016

From 0.90-14.40 USD/hr.

30-09-2016
cost_savings

Yess thank you, thank you, thank you. I was just signing up for the Azure N Series preview but we're good to go now :).

30-09-2016

Pricing?

30-09-2016
cost_savings

Nvidia has been doing a lot of work on hardware virtualization of GPUs. A goal of which is to run dozens of low-requirements desktops (think users running MS Office in separate Windows VMs) off of a single high-end server with a single GPU.

30-09-2016

That's not mentioned in the article, did you work on this?Disclosure: I work on Google Cloud.

30-09-2016

Can't speak to what happens on a Windows box, but across AWS instances it's definitely the case.

30-09-2016

At the sort of scale these instances run I'd be surprised if you're sharing any physical hardware at all, my guess is that they provision a single VM on each GPU instance.

30-09-2016

> The GPU memory gets wiped on initialisationWell, an attempt is made. As we can see in the case of WebGL, there are a multitude of interesting effects.

30-09-2016
memory_usage

It is interesting to compare this to NVidia's DGX-1 system. That server is based on the new Tesla P100 and uses NVLink rather than PCIe (about 10x faster). It boasts about 170 Tflops vs the p2.16xlarge's 64 Tflops. If you run the p2.16xlarge full time for a year it would cost about the same as buying a DGX-1. Presumably Amazon releases their GPU instances on older hardware for cost savings.

30-09-2016
benchmarking, cost_savings

The GPUs aren't shared between VMs, each GPU is dedicated to the VM it is allocated to. The GPU memory gets wiped on initialisation, just like host memory, so there's no leakage between successive VMs allocated the same hardware.If you do absolutely care about not being multitennanted, you can stipulate exclusive hw allocation when requesting the VM.

30-09-2016
memory_usage

Stupid question: are GPU safe to be shared by two tenants in a datacenter? I read previously that there are very few security mechanisms in GPU, in particular that the memory is full of interesting garbage. So I would assume no hardware enforced separation between the VMs too.

30-09-2016
memory_usage

Sounds like a great way to build a custom render farm. My home comouter has a dirt cheap GPU but it works well enough for basic modeling & animation. Terrible for rendering, though. I've been thinking of using ECS to build a cluster of renderers for Maya that I can spin up when needed and scale to the appropriate size. I don't know for certain if it's cheaper than going with a service, but it sounds like it is(render farm subscriptions cost hundreds), and I would get complete control over the software being used. I am glad to hear that Amazon is doing this. Granted, I'm more of a hobbyist in this arena, so maybe it wouldn't work for someone more serious about creating graphics.

30-09-2016
cost_savings

Also, doesn't include electricity, cooling, data center space, all of which might be significant?

30-09-2016
cost_savings

50K$ server at maximum.

30-09-2016
cost_savings

I can start g2 but not p2 instances (limit 0).

30-09-2016

One thing I discovered recently is that for GPU machines your initial limit on AWS is 0 (meaning you have to ask support before you start one for yourself)

30-09-2016

I think this is the right talk by him, but there's one where he implies some differences. For instance, that because they can promise exact temperature ranges they can clock them differently.

30-09-2016
benchmarking

Get an GTX 1080 for your local box and use it to write and test the model, then run in cloud for training.

30-09-2016

For anyone who interested in ML/DL on cloud.

30-09-2016

This is great - we'll try to get our Tensorflow and Caffe AMI repo updated soon:

30-09-2016

I tried using google cloud (switching from AWS) a few months ago and it was the worst user experience I've ever had trying to use any software / dev environment.

30-09-2016
development

AWS offers a far far better experience and cost effectiveness.

30-09-2016
cost_savings

MeteoSwiss uses a 12 node cluster with 16 GPUs each to compute its weather forecasts (Piz Escha/Kesch).

30-09-2016

By the way, I own a 1080 since two weeks and I can't overstate how powerful this thing is. Even if you are not into gaming, getting a 1080 is still a considerable option if you want to experiment with deep learning.

30-09-2016

if you factor in the cost of keeping a ML team waiting for the network to finish training, then it might be cheaper to invest $7000 and have it run in a few hours instead of weeks.

30-09-2016
cost_savings

You can set up the volume a CPU only machine (even on a free instance), and then just launch that volume with a machine with these big expensive GPU's on it.

30-09-2016

$0.9 per K80 GPU per hour, while expensive, opens up so many opportunities - especially when you can get a properly connected machine.

30-09-2016
cost_savings

Still seems much better to buy your own gtx 1080 for $700 which you would have spent in a month playing with parameters on these instances.

30-09-2016
memory_usage, cost_savings

Using spot pricing you tend to get it far cheaper than that.

30-09-2016
cost_savings

The P2 instances were designed for "heavier GPU compute workloads" such as high-performance computing (HPC), artificial intelligence (AI) and Big Data processing.

Amazon EC2 Vice President
2016-04-10 00:00:00
benchmarking

Compared to the g2.8xlarge, the largest instance in AWS' G2 instance family for accelerated computing, the P2 provides up to "seven times the computational capacity for single precision floating point calculations and 60 times more for double precision floating point calculations,"

Amazon EC2 Vice President
2016-04-10 00:00:00
benchmarking

> I am unable to transcode faster than real time with live transcodeWell, yeah. Maybe quantum computing will change that one day!What is it you're working on? Colour me intrigued.

30-09-2016
development

This depends on if you're doing live transcoding or VOD transcoding.I work with live transcode, and it can be beneficial to run on g2 instances. Running c4.4xlarge I can transcode a good number of 1080@30 in with 1080/720/480/360/240@30 out. With a proper g2 instance I can transcode more simultaneously.So cost efficiency really depends on sustained traffic levels. I scale out currently using haproxy and custom code that monitors my pool and scales appropriately. But I monitor sustained traffic levels to know when it makes financial sense to scale up.If your main concern is transcode speed, CPU is likely sufficient -- I am unable to transcode faster than real time with live transcode.

30-09-2016
development, cost_savings

From 0.90-14.40 USD/hr.

30-09-2016
cost_savings

Anyone knows if video transcoding on GPUs (with FFMPEG) is viable nowadays? If yes, what are the gains?

30-09-2016

They have chips dedicated to video encoding/decoding with the latest cards doing h265. Nvidia has an ffmpeg patch here

30-09-2016
development

Pricing?

30-09-2016
cost_savings

Yess thank you, thank you, thank you. I was just signing up for the Azure N Series preview but we're good to go now :).

30-09-2016

> The GPU memory gets wiped on initialisationWell, an attempt is made. As we can see in the case of WebGL, there are a multitude of interesting effects.

30-09-2016
memory_usage

Nvidia has been doing a lot of work on hardware virtualization of GPUs. A goal of which is to run dozens of low-requirements desktops (think users running MS Office in separate Windows VMs) off of a single high-end server with a single GPU.

30-09-2016

At the sort of scale these instances run I'd be surprised if you're sharing any physical hardware at all, my guess is that they provision a single VM on each GPU instance.

30-09-2016

That's not mentioned in the article, did you work on this?Disclosure: I work on Google Cloud.

30-09-2016

Can't speak to what happens on a Windows box, but across AWS instances it's definitely the case.

30-09-2016

The GPUs aren't shared between VMs, each GPU is dedicated to the VM it is allocated to. The GPU memory gets wiped on initialisation, just like host memory, so there's no leakage between successive VMs allocated the same hardware.If you do absolutely care about not being multitennanted, you can stipulate exclusive hw allocation when requesting the VM.

30-09-2016
memory_usage

Stupid question: are GPU safe to be shared by two tenants in a datacenter? I read previously that there are very few security mechanisms in GPU, in particular that the memory is full of interesting garbage. So I would assume no hardware enforced separation between the VMs too.

30-09-2016
memory_usage

It is interesting to compare this to NVidia's DGX-1 system. That server is based on the new Tesla P100 and uses NVLink rather than PCIe (about 10x faster). It boasts about 170 Tflops vs the p2.16xlarge's 64 Tflops. If you run the p2.16xlarge full time for a year it would cost about the same as buying a DGX-1. Presumably Amazon releases their GPU instances on older hardware for cost savings.

30-09-2016
benchmarking, cost_savings

Also, doesn't include electricity, cooling, data center space, all of which might be significant?

30-09-2016
cost_savings

50K$ server at maximum.

30-09-2016
cost_savings

Sounds like a great way to build a custom render farm. My home comouter has a dirt cheap GPU but it works well enough for basic modeling & animation. Terrible for rendering, though. I've been thinking of using ECS to build a cluster of renderers for Maya that I can spin up when needed and scale to the appropriate size. I don't know for certain if it's cheaper than going with a service, but it sounds like it is(render farm subscriptions cost hundreds), and I would get complete control over the software being used. I am glad to hear that Amazon is doing this. Granted, I'm more of a hobbyist in this arena, so maybe it wouldn't work for someone more serious about creating graphics.

30-09-2016
cost_savings

I think this is the right talk by him, but there's one where he implies some differences. For instance, that because they can promise exact temperature ranges they can clock them differently.

30-09-2016
benchmarking

I can start g2 but not p2 instances (limit 0).

30-09-2016

One thing I discovered recently is that for GPU machines your initial limit on AWS is 0 (meaning you have to ask support before you start one for yourself)

30-09-2016

AWS offers a far far better experience and cost effectiveness.

30-09-2016
cost_savings

This is great - we'll try to get our Tensorflow and Caffe AMI repo updated soon:

30-09-2016

For anyone who interested in ML/DL on cloud.

30-09-2016

I tried using google cloud (switching from AWS) a few months ago and it was the worst user experience I've ever had trying to use any software / dev environment.

30-09-2016
development

MeteoSwiss uses a 12 node cluster with 16 GPUs each to compute its weather forecasts (Piz Escha/Kesch).

30-09-2016

Get an GTX 1080 for your local box and use it to write and test the model, then run in cloud for training.

30-09-2016

By the way, I own a 1080 since two weeks and I can't overstate how powerful this thing is. Even if you are not into gaming, getting a 1080 is still a considerable option if you want to experiment with deep learning.

30-09-2016

if you factor in the cost of keeping a ML team waiting for the network to finish training, then it might be cheaper to invest $7000 and have it run in a few hours instead of weeks.

30-09-2016
cost_savings

You can set up the volume a CPU only machine (even on a free instance), and then just launch that volume with a machine with these big expensive GPU's on it.

30-09-2016

Still seems much better to buy your own gtx 1080 for $700 which you would have spent in a month playing with parameters on these instances.

30-09-2016
memory_usage, cost_savings

Using spot pricing you tend to get it far cheaper than that.

30-09-2016
cost_savings

$0.9 per K80 GPU per hour, while expensive, opens up so many opportunities - especially when you can get a properly connected machine.

30-09-2016
cost_savings

Iraklis ran Hashcat (a popular password "recovery" tool) with GPU support against the largest instance size available... the p2.16xlarge.

19-03-2025
benchmarking

Yess thank you, thank you, thank you. I was just signing up for the Azure N Series preview but we're good to go now :).

30-09-2016

AWS has launched a new family of Elastic Compute Cloud (EC2) instance types called P2. Backed by the Tesla K80 GPU line from Nvidia, the new P2 instances were designed to chew through tough, large-scale machine learning, deep learning, computational fluid dynamics (CFD) seismic analysis, molecular modeling, genomics and computational finance workloads

Converge360
2016-04-10 00:00:00

Iraklis ran Hashcat (a popular password "recovery" tool) with GPU support against the largest instance size available... the p2.16xlarge.

19-03-2025
benchmarking

I'm currently looking for the correct number of 2018 budget for the AWS instances in preparation of more deep neural net training, but now a bit confused because of this chart below: p2.16xlarge costs 3 times more than g3.16xlarge with the same number of processors but slightly less memory. Is this diff just because of the memory difference? or is there any other key difference between p2 and g3 instances?

30-10-2017
memory_usage, cost_savings

This depends on if you're doing live transcoding or VOD transcoding.I work with live transcode, and it can be beneficial to run on g2 instances. Running c4.4xlarge I can transcode a good number of 1080@30 in with 1080/720/480/360/240@30 out. With a proper g2 instance I can transcode more simultaneously.So cost efficiency really depends on sustained traffic levels. I scale out currently using haproxy and custom code that monitors my pool and scales appropriately. But I monitor sustained traffic levels to know when it makes financial sense to scale up.If your main concern is transcode speed, CPU is likely sufficient -- I am unable to transcode faster than real time with live transcode.

30-09-2016
development, cost_savings

> I am unable to transcode faster than real time with live transcodeWell, yeah. Maybe quantum computing will change that one day!What is it you're working on? Colour me intrigued.

30-09-2016
development

They have chips dedicated to video encoding/decoding with the latest cards doing h265. Nvidia has an ffmpeg patch here

30-09-2016
development

Anyone knows if video transcoding on GPUs (with FFMPEG) is viable nowadays? If yes, what are the gains?

30-09-2016

From 0.90-14.40 USD/hr.

30-09-2016
cost_savings

Pricing?

30-09-2016
cost_savings

Nvidia has been doing a lot of work on hardware virtualization of GPUs. A goal of which is to run dozens of low-requirements desktops (think users running MS Office in separate Windows VMs) off of a single high-end server with a single GPU.

30-09-2016

That's not mentioned in the article, did you work on this?Disclosure: I work on Google Cloud.

30-09-2016

At the sort of scale these instances run I'd be surprised if you're sharing any physical hardware at all, my guess is that they provision a single VM on each GPU instance.

30-09-2016

Can't speak to what happens on a Windows box, but across AWS instances it's definitely the case.

30-09-2016

It is interesting to compare this to NVidia's DGX-1 system. That server is based on the new Tesla P100 and uses NVLink rather than PCIe (about 10x faster). It boasts about 170 Tflops vs the p2.16xlarge's 64 Tflops. If you run the p2.16xlarge full time for a year it would cost about the same as buying a DGX-1. Presumably Amazon releases their GPU instances on older hardware for cost savings.

30-09-2016
benchmarking, cost_savings

Sounds like a great way to build a custom render farm. My home comouter has a dirt cheap GPU but it works well enough for basic modeling & animation. Terrible for rendering, though. I've been thinking of using ECS to build a cluster of renderers for Maya that I can spin up when needed and scale to the appropriate size. I don't know for certain if it's cheaper than going with a service, but it sounds like it is(render farm subscriptions cost hundreds), and I would get complete control over the software being used. I am glad to hear that Amazon is doing this. Granted, I'm more of a hobbyist in this arena, so maybe it wouldn't work for someone more serious about creating graphics.

30-09-2016
cost_savings

The GPUs aren't shared between VMs, each GPU is dedicated to the VM it is allocated to. The GPU memory gets wiped on initialisation, just like host memory, so there's no leakage between successive VMs allocated the same hardware.If you do absolutely care about not being multitennanted, you can stipulate exclusive hw allocation when requesting the VM.

30-09-2016
memory_usage

> The GPU memory gets wiped on initialisationWell, an attempt is made. As we can see in the case of WebGL, there are a multitude of interesting effects.

30-09-2016
memory_usage

Stupid question: are GPU safe to be shared by two tenants in a datacenter? I read previously that there are very few security mechanisms in GPU, in particular that the memory is full of interesting garbage. So I would assume no hardware enforced separation between the VMs too.

30-09-2016
memory_usage

Also, doesn't include electricity, cooling, data center space, all of which might be significant?

30-09-2016
cost_savings

50K$ server at maximum.

30-09-2016
cost_savings
Load More
Similar Instances to
p2.16xlarge

Consider these:

Feedback

We value your input! If you have any feedback or suggestions about this t4g.nano instance information page, please let us know.

Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.