TE
TechEcho
Home24h TopNewestBestAskShowJobs
GitHubTwitter
Home

TechEcho

A tech news platform built with Next.js, providing global tech news and discussions.

GitHubTwitter

Home

HomeNewestBestAskShowJobs

Resources

HackerNews APIOriginal HackerNewsNext.js

© 2025 TechEcho. All rights reserved.

Introducing Amazon EC2 P3 Instances

214 pointsby aseidlover 7 years ago

18 comments

Smerityover 7 years ago
The P3 instances are the first widely and easily accessible machines that use the NVIDIA Tesla V100 GPUs. These GPUs are straight up scary in terms of firepower. To give an understanding of the speed-up compared to the P2 instances for a research project of mine:<p>+ P2 (K80) with single GPU: ~95 seconds per epoch<p>+ P3 (V100) with single GPU: ~20 seconds per epoch<p>Admittedly this isn&#x27;t exactly fair for either GPU - the K80 cards are straight up ancient now and the Volta isn&#x27;t sitting at 100% GPU utilization as it burns through the data too quickly ([CUDA kernel, Python] overhead suddenly become major bottlenecks). This gives you an indication of what a leap this is if you&#x27;re using GPUs on AWS however. Oh, and the V100 comes with 16GB of (faster) RAM compared to the K80&#x27;s 12GB of RAM, so you win there too.<p>For anyone using the standard set of frameworks (Tensorflow, Keras, PyTorch, Chainer, MXNet, DyNet, DeepLearning4j, ...) this type of speed-up will likely require you to do nothing - except throw more money at the P3 instance :)<p>If you really want to get into the black magic of speed-ups, these cards also feature full FP16 support, which means you can double your TFLOPS by dropping to FP16 from FP32. You&#x27;ll run into a million problems during training due to the lower precision but these aren&#x27;t insurmountable and may well be worth the pain for the additional speed-up &#x2F; better RAM usage.<p>- Good overview of Volta&#x27;s advantages compared to event the recent P100: <a href="https:&#x2F;&#x2F;devblogs.nvidia.com&#x2F;parallelforall&#x2F;inside-volta&#x2F;" rel="nofollow">https:&#x2F;&#x2F;devblogs.nvidia.com&#x2F;parallelforall&#x2F;inside-volta&#x2F;</a><p>- Simple table comparing V100 &#x2F; P100 &#x2F; K40 &#x2F; M40: <a href="https:&#x2F;&#x2F;www.anandtech.com&#x2F;show&#x2F;11367&#x2F;nvidia-volta-unveiled-gv100-gpu-and-tesla-v100-accelerator-announced" rel="nofollow">https:&#x2F;&#x2F;www.anandtech.com&#x2F;show&#x2F;11367&#x2F;nvidia-volta-unveiled-g...</a><p>- NVIDIA&#x27;s V100 GPU architecture white paper: <a href="http:&#x2F;&#x2F;www.nvidia.com&#x2F;object&#x2F;volta-architecture-whitepaper.html" rel="nofollow">http:&#x2F;&#x2F;www.nvidia.com&#x2F;object&#x2F;volta-architecture-whitepaper.h...</a><p>- The numbers above were using my PyTorch code at <a href="https:&#x2F;&#x2F;github.com&#x2F;salesforce&#x2F;awd-lstm-lm" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;salesforce&#x2F;awd-lstm-lm</a> and the Quasi-Recurrent Neural Network (QRNN) at <a href="https:&#x2F;&#x2F;github.com&#x2F;salesforce&#x2F;pytorch-qrnn" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;salesforce&#x2F;pytorch-qrnn</a> which features a custom CUDA kernel for speed
评论 #15557325 未加载
评论 #15565920 未加载
评论 #15558631 未加载
评论 #15557923 未加载
评论 #15565938 未加载
DTEover 7 years ago
Hi guys, Dillon here from Paperspace (<a href="https:&#x2F;&#x2F;www.paperspace.com" rel="nofollow">https:&#x2F;&#x2F;www.paperspace.com</a>). We are a cloud that specializes in GPU infrastructure and software. We launched V100 instances a few days ago in our NY and CA regions and its much less expensive than AWS.<p>Think of us as the DigitalOcean for GPUs with a simple, transparent pricing and effortless setup &amp; configuration:<p>AWS: $3.06&#x2F;hr V100*<p>Paperspace: $2.30 &#x2F;hr or $980&#x2F;month for dedicated (effective hourly is only $1.3&#x2F;hr)<p>Learn more here: <a href="https:&#x2F;&#x2F;www.paperspace.com&#x2F;pricing" rel="nofollow">https:&#x2F;&#x2F;www.paperspace.com&#x2F;pricing</a><p>[Disclosure: I am one of the founders]
评论 #15561078 未加载
评论 #15561496 未加载
评论 #15561306 未加载
评论 #15559327 未加载
评论 #15560093 未加载
评论 #15622363 未加载
plantainover 7 years ago
But where are the C5 instances? It&#x27;s been 11 months since Amazon announced Skylake C5&#x27;s and we&#x27;re still waiting!<p><a href="https:&#x2F;&#x2F;aws.amazon.com&#x2F;about-aws&#x2F;whats-new&#x2F;2016&#x2F;11&#x2F;coming-soon-amazon-ec2-c5-instances-the-next-generation-of-compute-optimized-instances&#x2F;" rel="nofollow">https:&#x2F;&#x2F;aws.amazon.com&#x2F;about-aws&#x2F;whats-new&#x2F;2016&#x2F;11&#x2F;coming-so...</a>
评论 #15560334 未加载
jeffbarrover 7 years ago
More details in my blog post at <a href="https:&#x2F;&#x2F;aws.amazon.com&#x2F;blogs&#x2F;aws&#x2F;new-amazon-ec2-instances-with-up-to-8-nvidia-tesla-v100-gpus-p3&#x2F;" rel="nofollow">https:&#x2F;&#x2F;aws.amazon.com&#x2F;blogs&#x2F;aws&#x2F;new-amazon-ec2-instances-wi...</a>
评论 #15558109 未加载
评论 #15561573 未加载
评论 #15559187 未加载
eggie5over 7 years ago
Here&#x27;s my results:<p>Testing new Tesla V100 on AWS. Fine-tuning VGG on DeepSent dataset for 10 epochs.<p>GRID 520K (4GB) (baseline):<p>* 780s&#x2F;epoch @ minibatch 8 (GPU saturated)<p>V100(16Gb):<p>* 30s&#x2F;epoch @ minibatch 8 (GPU not saturated)<p>* 6s&#x2F;epoch @ minibatch 32 (GPU more saturated)<p>* 6s&#x2F;epoch @ minibatch 256 (GPU saturated)
评论 #15558265 未加载
mamonover 7 years ago
Slightly off-topic but I&#x27;m curious: Nvidia Volta is advertised as having &quot;tensor cores&quot; - what does it take for a programmer to use them? Will typical Tensorflow or Cafe code take advantage of it? Or should we wait for some new optimized version of ML frameworks?
评论 #15562306 未加载
corfordover 7 years ago
Hmm just tried to spool up a p3.2xlarge in Ireland but hit an instance limit check (it&#x27;s set at 0), went to request a service limit increase but P3 instances are not listed in the drop down box :(
评论 #15564644 未加载
dharma1over 7 years ago
Price: p3.2xlarge - $3&#x2F;hr, p3.8xlarge - $12&#x2F;hr, p3.16xlarge - $25&#x2F;hr<p>These look very good for half precision training
评论 #15557206 未加载
评论 #15557211 未加载
kshnellover 7 years ago
Looks like Paperspace announced Volta support yesterday: <a href="https:&#x2F;&#x2F;blog.paperspace.com&#x2F;tesla-v100-available-today&#x2F;" rel="nofollow">https:&#x2F;&#x2F;blog.paperspace.com&#x2F;tesla-v100-available-today&#x2F;</a> One nice thing here is you can do monthly plans instead of reserved on AWS which is a minimum $8-17k upfront. Really great to see the cloud providers adopting modern GPUs.
moconnorover 7 years ago
An exaflop of mixed-precision compute for $250M over 3 years. That’s ballpark what the HPC community is paying for their exaflop-class machines.<p>You’d still build your own for that money, I think, but it’s an interesting datapoint.
评论 #15557218 未加载
评论 #15558240 未加载
评论 #15560703 未加载
bprasannaover 7 years ago
...advanced workloads such as machine learning (ML), high performance computing (HPC), data compression, and crypto__________.
评论 #15558140 未加载
againaover 7 years ago
Use reserve instances or use spot. The price decrease substantially. Then when you don’t need it... you don’t pay it... it’s a good deal
评论 #15558479 未加载
sethgeckoover 7 years ago
Is there an AMI that comes with Tensorflow&#x2F;keras with GPU support preinstalled or you have to do it yourself?
评论 #15557605 未加载
science404over 7 years ago
Why Ireland and not the UK? I can imagine a lot of startups&#x2F;banks in London could use this... Brexit fears?
评论 #15557556 未加载
评论 #15557593 未加载
评论 #15557637 未加载
评论 #15560926 未加载
评论 #15557665 未加载
评论 #15557623 未加载
评论 #15560134 未加载
psychometryover 7 years ago
Random question: Why are we still using mostly GPUs for computation rather than CPUs custom-designed for ML tasks?
评论 #15561473 未加载
评论 #15558307 未加载
评论 #15564798 未加载
评论 #15559472 未加载
评论 #15561283 未加载
评论 #15565288 未加载
jerianasmithover 7 years ago
P3 instances no doubt provide a powerful platform and is going to be useful for data compression.
评论 #15559046 未加载
JeanMarcSover 7 years ago
If ever you&#x27;ve got password hash to decrypt :)
g105bover 7 years ago
Bitcoin?
评论 #15558256 未加载