g5.24xlarge

Analysis

The g5.24xlarge is a gpu instance instance in the g5 series with 96 vCPUs and 384 GiB of memory on AMD EPYC 7R32 (up to 2.8 GHz), which makes it balanced at 4 GiB of memory per vCPU.

Running it in N. Virginia instead of São Paulo saves $4,160 a month per instance: $8.14/hour against $13.84/hour, a 70% spread. It is not offered in Singapore, so plan around that if you need a presence there.

The newer g6.24xlarge on AMD EPYC 7R13 is 18% cheaper at $6.68/hour, which makes the g5.24xlarge hard to justify for new deployments.

It carries 4 GPUs with 96 GB of GPU memory, so most of the price is the accelerator rather than the CPUs. It has local instance storage (1 x 3800 GB NVMe SSD), which is fast but lost when the instance stops. Network: 50 Gigabit.

It is size 7 of 8 in the g5 series. Stepping up to the g5.48xlarge gives 192 vCPUs and 768 GiB for $5,945 more a month. Dropping to the g5.16xlarge saves $2,955 a month. Unit price is not flat across the series ($0.0640 to $0.2515 per vCPU-hour), so the size changes the rate you pay.

On demand it costs $8.14/hour in US East (N. Virginia), $5,945 a month if left running, which puts it mid-table for gpu instance types at the 36th percentile per vCPU ($0.0848 per vCPU-hour) and the 47th percentile per GiB ($0.0212 per GiB-hour).