dl2q.24xlarge

Analysis

The dl2q.24xlarge is a machine learning asic instances instance in the dl2q series with 96 vCPUs and 768 GiB of memory on Intel Xeon Platinum 8259 (Cascade Lake) (up to 2.5 GHz), which makes it memory-heavy at 8 GiB of memory per vCPU.

Running it in Oregon instead of Frankfurt saves $1,953 a month per instance: $8.92/hour against $11.60/hour, a 30% spread. It is not offered in N. Virginia, Ohio, Canada Central, São Paulo, Ireland, London, Mumbai, Singapore, Sydney and Tokyo, so plan around that if you need a presence there.

Nothing else with 96 vCPUs and similar memory is cheaper: among 3 current same-shape types it has the lowest on-demand price.

It carries 8 GPUs with 128 GiB of GPU memory, so most of the price is the accelerator rather than the CPUs. Storage is EBS only. Network: 100 Gigabit.

On demand it costs $8.92/hour in its cheapest tracked region, $6,511 a month if left running, which puts it mid-table for machine learning asic instances types at the 58th percentile per vCPU ($0.0929 per vCPU-hour) and the 4th percentile per GiB ($0.0116 per GiB-hour).

Reserving it for a year with no upfront payment cuts the Linux rate to $5.62/hour, which beats on demand once it runs more than 63% of the time; three years brings it to $3.85/hour, 57% below on demand.