Pith Number

pith:LEPUH3LR

pith:2025:LEPUH3LRFIXTB5OR22JYS7OHJQ

not attested not anchored not stored refs resolved

MultiPath Memory Access: Breaking Host-GPU Bandwidth Bottlenecks in LLM Services

Chengguang Xu, Daoping Zhang, Feiqiang Sun, Feng Jin, Guo Chen, Junjie Chen, Lingfeng Tang, Peihao Huang, Yuxin Chen

Multipath Memory Access routes host-GPU copies over unused server links to raise bandwidth 4.6x.

arxiv:2512.16056 v2 · 2025-12-18 · cs.DC · cs.NI · cs.PF

Open paper page JSON Open Graph Bundle Merged state Verified badge What is a Pith Number?

Add to your LaTeX paper

\usepackage{pith}
\pithnumber{LEPUH3LRFIXTB5OR22JYS7OHJQ}

Prints a linked badge after your title and injects PDF metadata. Compiles on arXiv. Learn more · Embed verified badge

Record completeness

1 Bitcoin timestamp

2 Internet Archive

3 Author claim open · sign in to claim

4 Citations open

5 Replications open

✓ Portable graph bundle live · download bundle · merged state

The bundle contains the canonical record plus signed events. A mirror can host it anywhere and recompute the same current state with the deterministic merge algorithm.

Claims

C1strongest claim

On an 8-GPU NVIDIA H20 server, MMA achieves 245 GB/s peak host-to-GPU bandwidth, a 4.62x improvement over native CUDA copies, and reduces TTFT for KV cache fetching by 1.14-2.38x and model wake-up/switching latency by 1.12-2.48x.

C2weakest assumption

That the additional relay paths through peer GPUs and high-bandwidth interconnects can be utilized with low enough overhead and without breaking CUDA stream semantics or application correctness under real LLM workloads.

C3one line summary

MMA routes host-GPU transfers over multiple available paths to deliver 4.62x higher peak bandwidth and lower latencies in LLM serving without hardware or driver changes.

References

52 extracted · 52 resolved · 3 Pith anchors

[1] MMA code base will be opensourced after paper ac- cepted

[2] Advanced Micro Devices 2022

[3] Advanced Micro Devices 2024

[4] Deepspeed-inference: enabling efficient in- ference of transformer models at unprecedented scale 2022

[5] Qwen Technical Report 2023 · arXiv:2309.16609

Formal links

2 machine-checked theorem links

Receipt and verification

First computed	2026-05-18T02:44:32.141856Z
Builder	pith-number-builder-2026-05-17-v1
Signature	Pith Ed25519 (`pith-v1-2026-05`) · public key
Schema	pith-number/v1.0

Canonical hash

591f43ed712a2f30f5d1d693897dc74c245222d973015ad300b790f68fc4b306

Aliases

arxiv: 2512.16056 · arxiv_version: 2512.16056v2 · doi: 10.48550/arxiv.2512.16056 · pith_short_12: LEPUH3LRFIXT · pith_short_16: LEPUH3LRFIXTB5OR · pith_short_8: LEPUH3LR

Agent API

Resolver JSON Graph JSON Events JSON Schema Signing key

Verify this Pith Number yourself

curl -sH 'Accept: application/ld+json' https://pith.science/pith/LEPUH3LRFIXTB5OR22JYS7OHJQ \
  | jq -c '.canonical_record' \
  | python3 -c "import sys,json,hashlib; b=json.dumps(json.loads(sys.stdin.read()), sort_keys=True, separators=(',',':'), ensure_ascii=False).encode(); print(hashlib.sha256(b).hexdigest())"
# expect: 591f43ed712a2f30f5d1d693897dc74c245222d973015ad300b790f68fc4b306

Canonical record JSON

{
  "metadata": {
    "abstract_canon_sha256": "1c276ad2095a7b3e426b7bf9defb888ab9216db718737a0b3515565e9a1c3757",
    "cross_cats_sorted": [
      "cs.NI",
      "cs.PF"
    ],
    "license": "http://creativecommons.org/licenses/by-nc-nd/4.0/",
    "primary_cat": "cs.DC",
    "submitted_at": "2025-12-18T00:45:00Z",
    "title_canon_sha256": "bee3c6b1a798f883f47117bc5ddb7809fff2573bae5d05669f3b5ee9bca3ad8c"
  },
  "schema_version": "1.0",
  "source": {
    "id": "2512.16056",
    "kind": "arxiv",
    "version": 2
  }
}