# Inter-domain point-to-point transfer (`p2p_inter`) | ServingStudio Kernel Library

Modeled time for one transfer between GPUs in different NVLink domains.

The simulator prices each MoE dispatch and combine transfer between GPUs in different NVLink domains, that is, across nodes, by looking up its bytes on this curve; an fp8 payload is fewer bytes on the same curve. A single node cannot measure such a transfer, so the rows are modeled: H100, H200 and B200 read a stored inter-node latency table, and other GPUs use a fixed bandwidth.

Category: Communication. Interactive page: https://servingstudio.cs.washington.edu/kernels.html?kind=p2p_inter. All kernel kinds: https://servingstudio.cs.washington.edu/kernels.md.

## Formula

- `effective bytes = message_size_bytes (H100/H200/B200); otherwise max(message_size_bytes, 32 KiB)`
- `algbw = effective bytes / time`
- `busbw = algbw`

## How it is measured

No transfer is timed. H100, H200 and B200 times are linearly interpolated from a stored latency table; values beyond its last point are extrapolated from the final segment. Other GPUs use effective bytes divided by a fixed bandwidth.

## Caveats

- Both backend labels return the same modeled time; neither launches a communication library.
- The fabric label does not change the estimate.
- The latency table was ported from an earlier reference model; how it was measured is not recorded here.

## Arguments

| Argument | Type | Unit | Meaning |
| --- | --- | --- | --- |
| `message_size_bytes` | number | bytes | Payload bytes on one cross-domain transfer. |
| `dtype` | dtype |  | Element type recorded for the payload; bf16 is used for the sweep. |
| `fabric` | label |  | Interconnect label used to identify the modeled curve. |

## Metrics

- Time (`time_ms`, ms)
- Algorithm bandwidth (`algbw_gbps`, GB/s)
- Bus bandwidth (`busbw_gbps`, GB/s)
- Energy (`energy_j`, J)

## Backends

- `nccl`: The modeled inter-node curve under the nccl label; nothing is launched. any CUDA GPU.
- `nvshmem`: The same modeled curve as nccl, under the nvshmem label. any CUDA GPU.

## Measured rows

38 rows in all.

| GPU | Backend | Precision | Rows |
| --- | --- | --- | --- |
| NVIDIA H200 | `nccl` |  | 19 |
| NVIDIA H200 | `nvshmem` |  | 19 |

## Used by these deployments

- [Qwen3 235B-A22B FP8: FFN side of an attention–FFN split](https://servingstudio.cs.washington.edu/models/Qwen3-235B-A22B-FP8/qwen3_fp8_ffn_moe.md) (`Qwen3-235B-A22B-FP8/qwen3_fp8_ffn_moe`)
- [Qwen3 235B-A22B FP8: DP attention, EP MoE, DeepGEMM experts](https://servingstudio.cs.washington.edu/models/Qwen3-235B-A22B-FP8/qwen3_moe_fp8_dp_attn_ep_ffn.md) (`Qwen3-235B-A22B-FP8/qwen3_moe_fp8_dp_attn_ep_ffn`)
- [Qwen3 235B-A22B FP8: vLLM, DP attention, EP MoE](https://servingstudio.cs.washington.edu/models/Qwen3-235B-A22B-FP8/qwen3_vllm_moe_dp_attn_ep_ffn.md) (`Qwen3-235B-A22B-FP8/qwen3_vllm_moe_dp_attn_ep_ffn`)
- [Qwen3 235B-A22B Thinking 2507 FP8: DP attention, EP MoE, DeepGEMM experts](https://servingstudio.cs.washington.edu/models/Qwen3-235B-A22B-Thinking-2507-FP8/qwen3_moe_fp8_dp_attn_ep_ffn.md) (`Qwen3-235B-A22B-Thinking-2507-FP8/qwen3_moe_fp8_dp_attn_ep_ffn`)
- [Qwen3 235B-A22B Thinking 2507 FP8: vLLM, DP attention, EP MoE](https://servingstudio.cs.washington.edu/models/Qwen3-235B-A22B-Thinking-2507-FP8/qwen3_vllm_moe_dp_attn_ep_ffn.md) (`Qwen3-235B-A22B-Thinking-2507-FP8/qwen3_vllm_moe_dp_attn_ep_ffn`)
- [Qwen3 235B-A22B: FFN side of an attention–FFN split](https://servingstudio.cs.washington.edu/models/Qwen3-235B-A22B/qwen3_ffn_moe.md) (`Qwen3-235B-A22B/qwen3_ffn_moe`)
- [Qwen3 235B-A22B: DP attention, EP MoE](https://servingstudio.cs.washington.edu/models/Qwen3-235B-A22B/qwen3_moe_dp_attn_ep_ffn.md) (`Qwen3-235B-A22B/qwen3_moe_dp_attn_ep_ffn`)
- [Qwen3 Coder 480B-A35B: FFN side of an attention–FFN split](https://servingstudio.cs.washington.edu/models/Qwen3-Coder-480B-A35B-Instruct/qwen3_fp8_ffn_moe.md) (`Qwen3-Coder-480B-A35B-Instruct/qwen3_fp8_ffn_moe`)

## API

- This kind: https://servingstudio.cs.washington.edu/api/public/v1/kernels/p2p_inter
- Measured rows: https://servingstudio.cs.washington.edu/api/public/v1/kernels/p2p_inter/rows (filter by any column, such as `?gpu=NVIDIA%20H200`; `format=csv` for CSV)
- Configs the public deployments ask of it: https://servingstudio.cs.washington.edu/api/public/v1/kernels/p2p_inter/configs

---

This page as Markdown: https://servingstudio.cs.washington.edu/kernels/p2p_inter.md. A guide to the whole site for AI agents: [llms.txt](https://servingstudio.cs.washington.edu/llms.txt).
