Sunday, June 25, 2017

AI / DL: PC GPU box vs cloud AWS, Azure

Which GPU(s) to Get for Deep Learning

The $1700 great Deep Learning box: Assembly, setup and benchmarks

"...Amazon Web Services (AWS), which offered P2 instances with Nvidia K80 GPUs. Lately, the AWS bills were around $60–70/month with a tendency to get larger. Also, it is expensive to store large datasets, like ImageNet.
Nvidia’s cards: GTX 1070 ($360?), GTX 1080 ($500), GTX 1080 Ti ($700) and finally the Titan X ($1320)."

Amazon EC2 - P2 Instances
"P2 instances provide up to 16 NVIDIA K80 GPUs, 64 vCPUs and 732 GiB of host memory, with a combined 192 GB of GPU memory, 40 thousand parallel processing cores, 70 teraflops of single precision floating point performance, and over 23 teraflops of double precision floating point performance."
from p2.xlarge: 1 GPU, 4 vCPU, 61 GiB RAM, $0.9 / hour (= $648/month)
($0.425/h on 3 year commitment = $306/month = $11 K / 3 years)

Azure N-Series: General availability on December 1 | Blog | Microsoft Azure

Virtual Machine series | Microsoft Azure
The N-series is a family of Azure Virtual Machines with GPU capabilities. GPUs are ideal for compute and graphics-intensive workloads, helping customers to fuel innovation through scenarios like high-end remote visualization, deep learning and predictive analytics. Available in preview today, the N-series will feature the NVIDIA Tesla accelerated platform as well as NVIDIA GRID 2.0 technology... starting from $669.60 / month.

NVIDIA Tesla K80 GPU Accelerator for Servers 900-22080-0000-000
  • $4000
  • 560 MHz Core - Boostable to 876 MHz
  • 4992 CUDA Cores
  • 24GB GDDR5 vRAM
  • 10 GHz Effective Memory Clock
MSI GeForce GTX 1070 DirectX 12 GTX 1070 GAMING X 8G 8GB 256-Bit GDDR5 PCI Express 3.0 x16 HDCP Ready SLI Support ATX Video Card -
  • $450
  • 8GB 256-Bit GDDR5
  • Core Clock 1607 MHz
  • 1920 CUDA Cores

Data Science: csv,conf;

conference: csv,conf,v3 - May 2-3 2017, Portland, OR

Ben Shneiderman

"Visualization is a tool for thinking and communicating."

"Programming is blindly manipulating symbols"

"Code is the most general tool we have"

podcast: Partially Derivative