Compute

General purpose: right balance between price and performance, vCPU and Memory ratio.
Compute optimized: for workloads that need more compute vs. memory such as HPC (high performance computing), game servers, latency-sensitive API serving.
Memory optimized: for workloads that need ore memory such as SAP HANA, database.
Accelerator optimized: NVIDIA GPU for workloads such as CUDA-enabled AIML training and inference, and HPC.
Custom machine types: custom ratio.

E2:
1. More cost effective specifically compared to N1, but with limitations on large instance sizes, GPUs, local SSD.
2. E2 already includes SUD, and is eligible for CUD. Can offer up to 31% savings compared to on-demand TCO or CUD. E2 together with CUD (3 years) can bring savings up to 55%.
N2: is the second generation vs. N1
1. Up to ~30% higher performance compared to same specs VMs on prem.
2. N2D:
  1. Offers largest general purpose VM with up to 224 vCPUs.
  2. Up to 13% price-performance compared to other N-series instances.
3. vs N1:
  1. Higher memory to vCPU ratio.
  2. Up to 20% price-performance improvement vs. N1.
  3. Up to 2.82x faster than N1 on AIML workloads.
N1: required for specific GPU workloads.

vs. N-series: smallest M-series instance offers ~900 GB of memory. For smaller workload, choose N-series instances instead.
Suitable for SAP HANA workloads.
M1 offers up to 4TB of memory, M2 offers up to 12TB of memory.

SUD offers discount while resources are being used, in term of active time per month, without any commitment. Discount starts from day 7-8 (25% of month), and goes up to 30% for a full-month active duration.
Sustained use discounts do not apply to VMs created using the App Engine flexible environment and Dataflow.
Sustained use discounts do not apply to E2 machine types.
SUD discount can be viewed from the bill at the end of the billing cycle.

CUD applies to GKE, Dataproc, and Compute Engine.
With CUD, resources such as vCPU, memory, GPU and local SSD will be at discounted rate.
Commitment term is for 1 year or 3 years.
The discount is up to 57% for most resources like machine types or GPUs. The discount is up to 70% for memory-optimized machine types.
CUD is regional.
CUD can be enabled to be shared across projects (or just for a specific project) under the same billing account.
CUD does not renew, a new CUD must be created to continue receiving discounts.
For SAP on SLES, CUD can offer further discount on pooled licences.
CUD can be purchased together with reserved resources, which applies to a specific zone.
Limitations:
1. CUDs do not apply to VMs created using the App Engine flexible environment and Dataflow.
2. CUD respects resource quota.
3. Purchasing commitment for GPU and SSD require a reservation with the same GPU or local SSD.