> ## Documentation Index
> Fetch the complete documentation index at: https://biznetgio.creations.ren/llms.txt
> Use this file to discover all available pages before exploring further.

# NEO GPU catalog and pricing

> The H200 GPU-as-a-Service catalog, subscription vs on-demand cost, and what is and is not confirmed about it

NEO GPU is Biznet GIO's NVIDIA H200 GPU-as-a-Service line, billed either by subscription cycle or by the hour. For the resource schema and full code, see the [Terraform](/terraform/resources/gpu) and [Pulumi](/pulumi/resources/gpu) reference pages.

<Note>
  This is a different product from NEO Metal's two GPU-equipped bare-metal SKUs (`a1.small.gpu.x86` with an NVIDIA T4, `a1.medium.gpu.x86` with an NVIDIA L4). Those are whole physical servers with a GPU card installed, ordered through `biznetgio_baremetal` - see the [NEO Metal catalog](/products/baremetal). NEO GPU, described on this page, is shared H200 GPU capacity ordered through `biznetgio_gpu_instance` / `GpuInstance`.
</Note>

## Pricing (source: [biznetgio.com/pricelist#neo-gpu](https://www.biznetgio.com/pricelist#neo-gpu))

| Package    | CPU       | RAM     | Disk    | GPU RAM | Price / hour | Price / month |
| ---------- | --------- | ------- | ------- | ------- | ------------ | ------------- |
| 1 GPU H200 | 22 cores  | 224 GB  | 750 GB  | 141 GB  | Rp74.000     | Rp52.999.000  |
| 2 GPU H200 | 44 cores  | 448 GB  | 1500 GB | 282 GB  | Rp147.000    | Rp105.999.000 |
| 3 GPU H200 | 66 cores  | 672 GB  | 2250 GB | 423 GB  | Rp219.000    | Rp158.999.000 |
| 4 GPU H200 | 88 cores  | 896 GB  | 3000 GB | 564 GB  | Rp299.000    | Rp211.999.000 |
| 5 GPU H200 | 110 cores | 1120 GB | 3750 GB | 705 GB  | Rp369.000    | Rp264.999.000 |
| 6 GPU H200 | 132 cores | 1344 GB | 4500 GB | 846 GB  | Rp439.000    | Rp317.999.000 |
| 7 GPU H200 | 154 cores | 1568 GB | 5250 GB | 987 GB  | Rp509.000    | Rp369.999.000 |
| 8 GPU H200 | 176 cores | 1792 GB | 6000 GB | 1128 GB | Rp589.000    | Rp423.999.000 |

<Note>
  Public list prices, current as of this writing. The **hourly** figure applies to `on_demand` instances (accrues continuously until deleted); the **monthly** figure applies to `subscription` instances billed at `cycle = "m"`. Confirm the actual charge against your own account's catalog output - see [the two price sources](/products/overview#two-different-prices-two-different-jobs).
</Note>

## What `biznetgio_gpu_products` gives you, and what is genuinely unclear

```
products[] = {
  product_id, name, description, category_name, raw,
  flavors[] = { flavor_id, name, raw }
}
```

`product_id` and `name` work the same way as every other catalog: `product_id` is what you pass to `biznetgio_gpu_instance.product_id`. What is **not** documented anywhere upstream, and not something this documentation will guess at, is exactly what a "flavor" represents. It could be a GPU-count tier matching the pricing table above, a region, a configuration variant, or something else entirely - the provider source only says it comes from `GET /neo-gpus/products/{product_id}/flavors` and passes the fields through unchanged.

Before you build anything that depends on `flavors`, print your own account's output and read it directly:

<Tabs>
  <Tab title="Terraform">
    ```hcl theme={null}
    data "biznetgio_gpu_products" "all" {}

    output "gpu_catalog" {
      value = [
        for p in data.biznetgio_gpu_products.all.products : {
          product_id = p.product_id
          name       = p.name
          raw        = jsondecode(p.raw)
          flavors = [
            for f in p.flavors : {
              flavor_id = f.flavor_id
              name      = f.name
              raw       = jsondecode(f.raw)
            }
          ]
        }
      ]
    }
    ```
  </Tab>

  <Tab title="Pulumi (TypeScript)">
    ```typescript theme={null}
    import * as biznetgio from "@shirasakaren/biznetgio";

    const products = biznetgio.gpuProductsOutput();
    export const gpuCatalog = products.products.apply((items) =>
      items.map((p) => ({
        productId: p.productId,
        name: p.name,
        raw: JSON.parse(p.raw),
        flavors: p.flavors.map((f) => ({ flavorId: f.flavorId, name: f.name, raw: JSON.parse(f.raw) })),
      })),
    );
    ```
  </Tab>
</Tabs>

Match `product_id` against the pricing table above by `name` once you have confirmed the exact string, the same way as [NEO Metal](/products/baremetal#match-on-name-the-only-typed-filterable-field). Note that `biznetgio_gpu_instance` / `GpuInstance` itself takes only `product_id`, not a `flavor_id` - the flavors list is exposed for you to read, but nothing in the current resource schema consumes it. Whatever a "flavor" changes about your order, it is not something you select through these providers today.

## Subscription vs on-demand

Exactly one of `subscription` or `on_demand` must be set on `biznetgio_gpu_instance` / `GpuInstance` - the provider rejects both being set, and rejects neither.

```hcl theme={null}
subscription = {
  cycle = "m" # or "a" - NEO GPU takes no other cycle value
}
```

```hcl theme={null}
on_demand = {
  additional_hours = 0
}
```

`subscription` bills the monthly or annual figure from the pricing table above. `on_demand` accrues the hourly figure continuously from creation until you delete the instance; `reserve_additional_hours_trigger` is a one-shot action that reserves additional discrete hour blocks on an already-running on-demand instance. See the [billing guide](/guides/billing) for the general cycle/payment model and the [Terraform](/terraform/resources/gpu) / [Pulumi](/pulumi/resources/gpu) reference for the full input schema.

## The OS list gap

Unlike NEO Lite, NEO Lite Pro, and NEO Metal, **NEO GPU has no OS list data source at all**, even though the upstream API has the endpoint: `GET /neo-gpus/products/{product_id}/select-os`. `select_os` on `biznetgio_gpu_instance` is a plain required string with no catalog to validate it against. Discover valid values manually:

```bash theme={null}
curl -sH "x-token: $BIZNETGIO_API_KEY" \
  https://api.portal.biznetgio.com/v1/neo-gpus/products/<product_id>/select-os | jq
```

Wrapping this endpoint as a real `biznetgio_gpu_os_list` data source is a good, contained first contribution - see the [Terraform provider guide](/contribute/terraform-guide) and [Pulumi provider guide](/contribute/pulumi-guide).

## Console and monitoring

* **`biznetgio_gpu_console` / `gpuConsole`** - mints a console access URL for a running instance. **Side-effecting**: every read may create a new session. Do not reference it anywhere that gets evaluated during routine plan diffing.
* **`biznetgio_gpu_graph` / `gpuGraph`** - monitoring graph data for an instance, over `hour` / `day` / `week` / `month` / `year`.

Full schemas in the [Terraform](/terraform/resources/gpu#data-sources) and [Pulumi](/pulumi/resources/gpu#functions) reference pages.