a2-ultragpu-4g as a GKE node
48 vCPU680 GiBx86_644 GPUGoogle Cloud us-central1A2 Ultragpu 4g
Where the a2-ultragpu-4g goes
Default GKE setup: GKE ≤ 1.36.
CPU47.61 of 48 vCPU left for your pods
Memory659.25 GiB of 680 GiB left for your pods
your podsDaemonSets (200m, 300 MiB assumed)kubelet reservedeviction
110max pods per node
$3,330.90on demand / month · $4.56/h
$1,994.96spot / month (varies)
a2-ultragpu-4g GKE max pods and allocatable memory — quick answers
- How many pods can a a2-ultragpu-4g run as a node with GKE ≤ 1.36? Up to 110; kubelet keeps 20949 MiB of memory and 190m CPU, leaving 659.54 GiB and 47.81 vCPU for pods.
- How many pods can a a2-ultragpu-4g run as a node with GKE 1.37+ (Container-Optimized OS)? Up to 110; kubelet keeps 2250 MiB of memory and 190m CPU, leaving 677.80 GiB and 47.81 vCPU for pods.
- How many 500m / 1 GiB pods fit? 95 per node after DaemonSets; 1 nodes for 10 replicas, about $3,330.90/month on demand.
What kubelet reserves on a2-ultragpu-4g
| Setup | Max pods | kube-reserved CPU | Memory reserved + eviction | Allocatable CPU | Allocatable memory | Memory usable |
|---|---|---|---|---|---|---|
| GKE ≤ 1.36 tiered memory reservation | 110 | 190m | 20849 Mi + 100 Mi | 47.81 | 659.54 GiB | 97.0% |
| GKE 1.37+ (Container-Optimized OS) 15 MiB × max pods + 500 MiB, CPU capped at 1 vCPU | 110 | 190m | 2150 Mi + 100 Mi | 47.81 | 677.80 GiB | 99.7% |
How these numbers are computed: reservation formulas. Check a live node with kubectl describe node → Allocatable.
How many pods fit on one a2-ultragpu-4g
Assumes DaemonSets (CNI, kube-proxy, log and metrics agents) request 200m CPU, 300 Mi and 3 pods per node. Try your own numbers in the node calculator.
| Pod requests (CPU / memory) | GKE ≤ 1.36 | GKE 1.37+ (Container-Optimized OS) | $ per pod / month |
|---|---|---|---|
| 100m / 128 Mi | 107 | 107 | $31.13 |
| 250m / 256 Mi | 107 | 107 | $31.13 |
| 250m / 512 Mi | 107 | 107 | $31.13 |
| 500m / 512 Mi | 95 | 95 | $35.06 |
| 500m / 1024 Mi | 95 | 95 | $35.06 |
| 1000m / 1024 Mi | 47 | 47 | $70.87 |
| 1000m / 2048 Mi | 47 | 47 | $70.87 |
| 2000m / 4096 Mi | 23 | 23 | $144.82 |
| 4000m / 8192 Mi | 11 | 11 | $302.81 |
Similar GKE node types
| Type | vCPU | GiB | Alloc CPU | Alloc memory | Max pods | $/month | $ per alloc GiB |
|---|---|---|---|---|---|---|---|
| a3-highgpu-2g | 52 | 468 | 51.80 | 451.78 GiB | 110 | $1,846.46 | $4.09 |
| a2-highgpu-4g | 48 | 340 | 47.81 | 326.34 GiB | 110 | $2,159.27 | $6.62 |
| a2-highgpu-8g | 96 | 680 | 95.69 | 659.54 GiB | 110 | $4,318.55 | $6.55 |
| a3-highgpu-4g | 104 | 936 | 103.67 | 910.42 GiB | 110 | $3,692.92 | $4.06 |
| g2-standard-48 | 48 | 192 | 47.81 | 181.30 GiB | 110 | $1,285.90 | $7.09 |
| g2-standard-96 | 96 | 384 | 95.69 | 369.46 GiB | 110 | $2,571.80 | $6.96 |
| a2-ultragpu-2g | 24 | 340 | 23.87 | 326.34 GiB | 110 | $1,665.45 | $5.10 |
| a2-megagpu-16g | 96 | 1360 | 95.69 | 1,325.94 GiB | 110 | $6,421.79 | $4.84 |
Closest match on other clouds
| Type | vCPU | GiB | Alloc CPU | Alloc memory | Max pods | $/month | $ per alloc GiB |
|---|---|---|---|---|---|---|---|
| g7e.12xlarge | 48 | 512 | 47.81 | 504.86 GiB | 632 | $6,048.84 | $11.98 |
| p2.16xlarge | 64 | 732 | 63.77 | 728.97 GiB | 250 | $10,512.00 | $14.42 |
| g6e.16xlarge | 64 | 512 | 63.77 | 503.74 GiB | 737 | $5,531.35 | $10.98 |
| g6e.12xlarge | 48 | 384 | 47.81 | 380.52 GiB | 292 | $7,659.63 | $20.13 |
| g3.16xlarge | 64 | 488 | 63.77 | 484.97 GiB | 250 | $3,328.80 | $6.86 |
| p3.16xlarge | 64 | 488 | 63.77 | 485.14 GiB | 234 | $17,870.40 | $36.84 |