Release: Amazon EC2 on 2010-11-14

Support for the new cluster GPU instance type (cg1.4xlarge).

Release Date: November 14, 2010
Latest Version: 2010-08-31
Created On: November 15, 2010
Last Updated: October 09, 2017

New Features

Feature Description
New Instance Type Amazon EC2 now supports a new instance type: Cluster GPU (cg1.4xlarge). This type is available only in the US-East (Northern Virginia) Region.

For more information about the new instance type, go to:

A CentOS 5.5 reference AMI for cluster instances is available: ami-42a2532b. You can customize an instance of the reference AMI and then create your own image.

The reference AMI includes the NVIDIA driver and CUDA toolkit, which enable the GPUs. For more details about the AMI, go to Cluster Instance Concepts.

Known Issues

Issue Description
Current Limitations for Cluster Instances Following are current limitations of cluster compute and cluster GPU instances:
  • Only the US-East (Northern Virginia) Region supports cluster instances
  • Microsoft Windows Server is not supported
  • Spot Instance requests are not supported
  • Amazon Virtual Private Cloud is not supported
  • Amazon DevPay is not supported
Current Limitations for Cluster Placement Groups Following are current limitations of cluster placement groups:
  • You can't currently use Auto Scaling to launch cluster instances into a cluster placement group
  • Reserved Instances are not currently available within a cluster placement group
Current Limitations for Tags and Filters You currently cannot tag the following resources:
  • Elastic IP addresses
  • Key pairs
  • Security groups
  • Placement groups
You can currently filter your resources in the AWS Management Console by tag, but not by the other filters that are available to use with Describe* actions through the command line tools or API.
Current Limitations for Micro Instances Following are current limitations of micro instances:
  • Amazon Virtual Private Cloud is not supported
  • Amazon Elastic MapReduce is not supported
  • Amazon DevPay is not supported
Instance Clock Drift Some instances (Windows instances in particular) can experience a system clock drift. The issue appears to be more severe in t1.micro Windows instances that run CPU-intensive workloads. If your application is sensitive to time drift, consider using other instance types until a resolution is available.
Setting the TCP_MAXSEG Socket Option The 2.6.18-164.15.1 kernel used in the CentOS 5.4 reference AMI (ami-7ea24a17) contains a known issue: Setting the TCP_MAXSEG socket option on TCP sockets to certain values (e.g., 1500) causes the kernel to generate TCP Segmentation Offload (TSO) packets with invalid sizes that the NIC driver then drops. The impact is significantly lower network throughput. As a workaround, don't set this socket option and let the kernel use the default settings to handle segmentation and Path Maximum Transmission Unit (PMTU) discovery.
Can't Use ModifyInstanceAttribute to Modify Instance's Block Device Mapping The ModifyInstanceAttribute action currently does not allow you to modify the block device mapping for the instance.
Paid AMIs Backed by Amazon EBS Amazon EBS-backed AMIs are not currently supported by Amazon DevPay.
Windows AMI launch times Windows AMIs take longer to launch than Linux/UNIX instances due to larger AMI sizes and multiple reboots.
Windows AMI sizes Installing software on Amazon S3-backed Windows AMIs can cause them to become large and easily reach the 10 GB limit. Before bundling, check the size of the C:\ volume.
Limitation on drive mapping There are limitations on devices available for storage attachment. For more information, go to How to Attach the Volume to an Instance in the Amazon Elastic Compute Cloud User Guide.
Instance limit New AWS accounts are limited to a maximum of 20 concurrent instances, but many of our customers use hundreds or thousands of instances. If you need a higher limit, go to http://aws.amazon.com/contact-us/ec2-request.