Skip to content

Introduce accelerators.json as Single Source of Truth for GCE machine types #5426

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Draft
SwarnaBharathiMantena wants to merge 5 commits into
base: develop
Choose a base branch
from
Draft

Introduce accelerators.json as Single Source of Truth for GCE machine types #5426

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
3a63504
Introduce accelerators.json as Single Source of Truth for GKE machine…
SwarnaBharathiMantena Mar 30, 2026
7cf6c28
Fix: add missing g4-standard machine types to accelerators.json
SwarnaBharathiMantena Mar 30, 2026
fc41503
pass the map using accelerator_configs as input
SwarnaBharathiMantena Mar 30, 2026
395e116
chore: update auto-generated README.md for accelerator_configs
SwarnaBharathiMantena Mar 30, 2026
c0b5c99
restructure the gpu-definition module to include variables.tf file
SwarnaBharathiMantena Mar 30, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
43 changes: 4 additions & 39 deletions modules/internal/gpu-definition/main.tf
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -12,7 +12,7 @@
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
*/

variable "machine_type" {
description = "Machine type to use for the instance creation"
Expand Down Expand Up @@ -45,44 +45,9 @@ locals {
# type = "https://www.googleapis.com/compute/beta/projects/PROJECT/zones/ZONE/acceleratorTypes/nvidia-tesla-a100"
# },
# ]
accelerator_machines = {
"a2-highgpu-1g" = { type = "nvidia-tesla-a100", count = 1 },
"a2-highgpu-2g" = { type = "nvidia-tesla-a100", count = 2 },
"a2-highgpu-4g" = { type = "nvidia-tesla-a100", count = 4 },
"a2-highgpu-8g" = { type = "nvidia-tesla-a100", count = 8 },
"a2-megagpu-16g" = { type = "nvidia-tesla-a100", count = 16 },
"a2-ultragpu-1g" = { type = "nvidia-a100-80gb", count = 1 },
"a2-ultragpu-2g" = { type = "nvidia-a100-80gb", count = 2 },
"a2-ultragpu-4g" = { type = "nvidia-a100-80gb", count = 4 },
"a2-ultragpu-8g" = { type = "nvidia-a100-80gb", count = 8 },
"a3-highgpu-1g" = { type = "nvidia-h100-80gb", count = 1 },
"a3-highgpu-2g" = { type = "nvidia-h100-80gb", count = 2 },
"a3-highgpu-4g" = { type = "nvidia-h100-80gb", count = 4 },
"a3-highgpu-8g" = { type = "nvidia-h100-80gb", count = 8 },
"a3-megagpu-8g" = { type = "nvidia-h100-mega-80gb", count = 8 },
"a3-ultragpu-8g" = { type = "nvidia-h200-141gb", count = 8 },
"a4-highgpu-8g-lowmem" = { type = "nvidia-b200", count = 8 },
"a4-highgpu-8g" = { type = "nvidia-b200", count = 8 },
"a4x-highgpu-4g" = { type = "nvidia-gb200", count = 4 },
"a4x-highgpu-4g-nolssd" = { type = "nvidia-gb200", count = 4 },
"a4x-maxgpu-4g-metal" = { type = "nvidia-gb300", count = 4 },
"a4x-maxgpu-4g-metal-nolssd" = { type = "nvidia-gb300", count = 4 },
"g2-standard-4" = { type = "nvidia-l4", count = 1 },
"g2-standard-8" = { type = "nvidia-l4", count = 1 },
"g2-standard-12" = { type = "nvidia-l4", count = 1 },
"g2-standard-16" = { type = "nvidia-l4", count = 1 },
"g2-standard-24" = { type = "nvidia-l4", count = 2 },
"g2-standard-32" = { type = "nvidia-l4", count = 1 },
"g2-standard-48" = { type = "nvidia-l4", count = 4 },
"g2-standard-96" = { type = "nvidia-l4", count = 8 },
"g4-standard-6" = { type = "nvidia-rtx-pro-6000", count = 1 },
"g4-standard-12" = { type = "nvidia-rtx-pro-6000", count = 1 },
"g4-standard-24" = { type = "nvidia-rtx-pro-6000", count = 1 },
"g4-standard-48" = { type = "nvidia-rtx-pro-6000", count = 1 },
"g4-standard-96" = { type = "nvidia-rtx-pro-6000", count = 2 },
"g4-standard-192" = { type = "nvidia-rtx-pro-6000", count = 4 },
"g4-standard-384" = { type = "nvidia-rtx-pro-6000", count = 8 },
}
accelerators_json = jsondecode(file("${path.module}/../../../pkg/config/accelerators.json"))
accelerator_machines = local.accelerators_json.gpus

generated_guest_accelerator = try([local.accelerator_machines[var.machine_type]], [])

# Select in priority order:
Expand Down
30 changes: 7 additions & 23 deletions modules/internal/tpu-definition/main.tf
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -12,9 +12,12 @@
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
*/

locals {
# Load shared JSON
accelerators_json = jsondecode(file("${path.module}/../../../pkg/config/accelerators.json"))

# Determine if this is a TPU node pool by checking if the machine_type exists in our authoritative map of TPU machine types.
is_tpu = contains(keys(local.tpu_chip_count_map), var.machine_type)

Expand All @@ -33,31 +36,12 @@ locals {
"tpu7x" = "tpu7x" # TPU v7x
}

# Map specific GCE machine types to the number of TPU chips per node (VM).
# The machine-type map must be updated to reflect new TPU releases with reference to public documentation: https://docs.cloud.google.com/tpu/docs/intro-to-tpu
# Project shared JSON into the expected format for tpu_chip_count_map (machine_type -> count)
tpu_chip_count_map = {
# v4 - ct4p
"ct4p-hightpu-4t" = 4

# v5e - ct5lp
"ct5lp-hightpu-1t" = 1
"ct5lp-hightpu-4t" = 4
"ct5lp-hightpu-8t" = 8

# v5p - ct5p
"ct5p-hightpu-1t" = 1
"ct5p-hightpu-2t" = 2
"ct5p-hightpu-4t" = 4

# v6e - ct6e
"ct6e-standard-1t" = 1
"ct6e-standard-4t" = 4
"ct6e-standard-8t" = 8

# v7x - tpu7x
"tpu7x-standard-4t" = 4
for k, v in local.accelerators_json.tpus : k => v.count
}


# Robustly extract the machine family prefix (e.g., "ct6e").
tpu_machine_family = local.is_tpu ? element(split("-", var.machine_type), 0) : ""
tpu_accelerator_type = local.is_tpu ? lookup(local.tpu_accelerator_map, local.tpu_machine_family, null) : null
Expand Down
50 changes: 50 additions & 0 deletions pkg/config/accelerators.json
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,50 @@
{
"gpus": {
"a2-highgpu-1g": { "count": 1, "type": "nvidia-tesla-a100" },
"a2-highgpu-2g": { "count": 2, "type": "nvidia-tesla-a100" },
"a2-highgpu-4g": { "count": 4, "type": "nvidia-tesla-a100" },
"a2-highgpu-8g": { "count": 8, "type": "nvidia-tesla-a100" },
"a2-megagpu-16g": { "count": 16, "type": "nvidia-tesla-a100" },
"a2-ultragpu-1g": { "count": 1, "type": "nvidia-a100-80gb" },
"a2-ultragpu-2g": { "count": 2, "type": "nvidia-a100-80gb" },
"a2-ultragpu-4g": { "count": 4, "type": "nvidia-a100-80gb" },
"a2-ultragpu-8g": { "count": 8, "type": "nvidia-a100-80gb" },
"a3-highgpu-1g": { "count": 1, "type": "nvidia-h100-80gb" },
"a3-highgpu-2g": { "count": 2, "type": "nvidia-h100-80gb" },
"a3-highgpu-4g": { "count": 4, "type": "nvidia-h100-80gb" },
"a3-highgpu-8g": { "count": 8, "type": "nvidia-h100-80gb" },
"a3-megagpu-8g": { "count": 8, "type": "nvidia-h100-mega-80gb" },
"a3-ultragpu-8g": { "count": 8, "type": "nvidia-h200-141gb" },
"a4-highgpu-8g-lowmem": { "count": 8, "type": "nvidia-b200" },
"a4-highgpu-8g": { "count": 8, "type": "nvidia-b200" },
"a4x-highgpu-4g": { "count": 4, "type": "nvidia-gb200" },
"a4x-highgpu-4g-nolssd": { "count": 4, "type": "nvidia-gb200" },
"a4x-maxgpu-4g-metal": { "count": 4, "type": "nvidia-gb300" },
"a4x-maxgpu-4g-metal-nolssd": { "count": 4, "type": "nvidia-gb300" },
"g2-standard-4": { "count": 1, "type": "nvidia-l4" },
"g2-standard-8": { "count": 1, "type": "nvidia-l4" },
"g2-standard-12": { "count": 1, "type": "nvidia-l4" },
"g2-standard-16": { "count": 1, "type": "nvidia-l4" },
"g2-standard-24": { "count": 2, "type": "nvidia-l4" },
"g2-standard-32": { "count": 1, "type": "nvidia-l4" },
"g2-standard-48": { "count": 4, "type": "nvidia-l4" },
"g2-standard-96": { "count": 8, "type": "nvidia-l4" },
"g4-standard-48": { "count": 1, "type": "nvidia-rtx-pro-6000" },
"g4-standard-96": { "count": 2, "type": "nvidia-rtx-pro-6000" },
"g4-standard-192": { "count": 4, "type": "nvidia-rtx-pro-6000" },
"g4-standard-384": { "count": 8, "type": "nvidia-rtx-pro-6000" }
},
"tpus": {
"ct4p-hightpu-4t": { "count": 4, "type": "tpu-v4-podslice" },
"ct5lp-hightpu-1t": { "count": 1, "type": "tpu-v5-lite-podslice" },
"ct5lp-hightpu-4t": { "count": 4, "type": "tpu-v5-lite-podslice" },
"ct5lp-hightpu-8t": { "count": 8, "type": "tpu-v5-lite-podslice" },
"ct5p-hightpu-1t": { "count": 1, "type": "tpu-v5p-slice" },
"ct5p-hightpu-2t": { "count": 2, "type": "tpu-v5p-slice" },
"ct5p-hightpu-4t": { "count": 4, "type": "tpu-v5p-slice" },
"ct6e-standard-1t": { "count": 1, "type": "tpu-v6e-slice" },
"ct6e-standard-4t": { "count": 4, "type": "tpu-v6e-slice" },
"ct6e-standard-8t": { "count": 8, "type": "tpu-v6e-slice" },
"tpu7x-standard-4t": { "count": 4, "type": "tpu7x" }
}
}
Loading