You bring up a good point. Amazon does give you a better "vertical scaling" story. I'll still challenge you on the "breadth" when it comes to EC2 - the philosophy is just very different. Why do you need a "IO optimized instance" if you want just fast disk - that notion just seems very foreign and arbitrarily-constrained on Google Cloud.

You bring up Local SSD. Google's Local SSD is just badass by comparison:

- 680,000 Read and 360,000 Write IOPS included in the cost [0]

- $0.218 per GB per month. Instance cost is separate.

- Again, you can attach these to any instance type (hence the point on fragmentation of instances on EC2)

- AWS goes up to 365,000 Read and 315,000 "First Write" IOPS. Only if you buy an i2.8xlarge [2]

- An i2.8xlarge is $6.82 per hour.

You do the math :)

And someone else did more comparisons here [1]

[0] https://cloud.google.com/compute/docs/disks/performance

[1] https://medium.com/google-cloud/new-google-cloud-ssds-have-a...

[2] http://docs.aws.amazon.com/AWSEC2/latest/UserGuide/i2-instan...

The real problem I have is the low network performance. Yes, yes, before everyone jumps all over me and points to Jupiter etc.. I understand the problems in Pb/s bisection bandwidth for the large datacenters. That doesn't change the fact that I don't need an entire datacenter worth of stuff.. but I do need an Amdahl-balanced cluster. So big machines with wimpy (20Gb non-RDMA) networks prevent me doing my HPCish workloads on GCE.

Followed by waiting on GPUs and other user accessible accelerators of course.

hit me up, i'll connect you with some folks

There are only two notable differences:

1) The highest memory instance on Google is about 200GB, on AWS it's 2TB.

2) Google doesn't have GPU instances yet. They're been announced for 2017.

Other than that, google instances have more options and more flexibility.

Other than the fact that AWS has more options and flexibility google instances have more options and flexibility... hmmm