Announcement

Collapse
No announcement yet.

Benchmarking Amazon's New EC2 "C3" Instance Types

Collapse
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • Benchmarking Amazon's New EC2 "C3" Instance Types

    Phoronix: Benchmarking Amazon's New EC2 "C3" Instance Types

    Last week Amazon unveiled the new "C3" instance types for their Elastic Compute Cloud platform. The C3 instance types deliver their highest processor performance in Amazon's cloud. The C3 instances are backed by Intel Xeon E5 "Ivy Bridge" processors, use SSD-based storage, support AVX and Turbo as part of the Ivy Bridge feature set, and also deliver on improved network performance in the cloud. Coming out of Phoronix today for helping you measure cloud performance are benchmarks of all the new C3 instance types and compared to some bare metal systems running locally.

    http://www.phoronix.com/vr.php?view=19355

  • #2
    Effects of compilation flags / AVX instruction-set?

    For, e.g., the Rodina CFD benchmark, it would be interesting to see the difference between the generic "vanilla -O3" optimized executables and SandyBridge/IvyBridge "O3 AVX" optimized executables for both the 4960X and the (ought-to-be-comparable) c3.2xlarge. I suspect that "O3 AVX" optimized executables for 960X will really blow c3.2xlarge out of the water.

    I've done SandyBridge benchmarks for the WRF weather-forecast model and found that "on the bare metal": the "O3 AVX" optimized executables are about 40% faster than the "vanilla -O3" ones (which says something about the sloppy design of WRF ;-( ). However, on the SB based (also AVX) CC2-8xlarge AWS nodes, the two types of executables run within 0.2% of each other (yes, within one part in 500) -- which seems to indicate that the Amazon-cloud VM memory overhead overwhelms the computational advantages of the AVX instructions for my codes.

    For comparison, one expects a 70% performance improvement for AVX for well-structured CFD codes on the bare metal...

    Comment


    • #3
      Interesting to note:

      The spot prices for some of these instances can be rather important to note as well... I use them a fair bit when doing 3D rendering for non time critical tasks.

      For instance, the C3.8xlarge costs $0.53/hour at the current spot price, and I can 'bid' to use it and specify a max cost of $0.8 per hour. If someone else wants it and bid higher, then they will get it, mine will be closed (don't pay for the hour it gets shut down in )

      To note how variable the price is, the C3.2xlarge is currently costing $10/hour spot price (meaning someone is paying that) while it normally costs $0.6 per hour.

      Comment

      Working...
      X