g6.8xlarge

Analysis

The g6.8xlarge is a gpu instance instance in the g6 series with 32 vCPUs and 128 GiB of memory on AMD EPYC 7R13 (up to 2.6 GHz), which makes it balanced at 4 GiB of memory per vCPU.

It is 18% cheaper than the older g5.8xlarge on AMD EPYC 7R32, so staying on the previous generation costs more as well as being slower. The newer g7.8xlarge on Intel Xeon Scalable (Granite Rapids) is 103% dearer at $4.09/hour.

Running it in N. Virginia instead of São Paulo saves $1,029 a month per instance: $2.01/hour against $3.42/hour, a 70% spread. It is not offered in Ireland and Singapore, so plan around that if you need a presence there.

It carries 1 GPU with 24 GB of GPU memory, so most of the price is the accelerator rather than the CPUs. It has local instance storage (2 x 450 GB NVMe SSD), which is fast but lost when the instance stops. Network: 25 Gigabit.

It is the cheapest of the 2 processor variants AWS sells at this size and generation. The g6e.8xlarge on AMD EPYC 7R13 costs 125% more, and no others also sit above it.

At $2.01/hour on demand in US East (N. Virginia), about $1,471 a month, the g6.8xlarge is one of the cheaper ways to buy gpu instance capacity: only 14% of the 78 current gpu instance types cost less per vCPU-hour than its $0.0630.