Shipping today

乙太網路是否已為 AI 做好準備?證據已在供貨

The four sections below each answer it with something already in production, not a roadmap: shipping 800G silicon, a hyperscale Ethernet cluster, the closed utilization gap, and a moving standard. Every figure is sourced to a hardware spec or a named third party.

51.2 TbpsTomahawk 5 per switch, 64 x 800G
24,576 GPUMeta's published Ethernet AI cluster
90%+Fabric utilization target with DLB
UEC 1.0開放乙太網路傳輸規範,2025 年 6 月
Is open Ethernet ready for production AI fabrics? 是的,證據已在交付。800G 晶片如今已在 HCL 收錄的開放交換器中,超大規模雲廠商已在 24,576 GPU 規模的乙太網路上執行 AI 訓練,負載平衡造成的利用率差距已經彌合,Ultra Ethernet 標準也在推進。OcNOS-DC 如今即執行無損 RoCEv2 fabric,以 OcNOS Systems 形式採購時,軟體與經過認證的硬體由一份 IP Infusion 合約覆蓋。
Proof 1 · The silicon

The hardware is here

Open 800G AI switches are not a preview. They are on the hardware compatibility list and shipping. OcNOS-DC runs on Broadcom Tomahawk 5 and Tomahawk 4 platforms from Edgecore and UfiSpace, the same merchant silicon the rest of the industry builds AI fabrics on, so you keep switch, NIC, and GPU choices open. These are on-chip, shared-buffer switches built for lossless RDMA at 800G.

BCM78900 · 51.2 Tbps · 64 x 800G

800G spine and leaf

Edgecore AIS800-64D and UfiSpace S9321-64E / S9321-64EO, the latter with 400G ZR+ coherent ports. On-chip shared buffering for lossless RoCEv2. Both platforms are HCL-listed.

Broadcom Tomahawk 5
BCM56990 · 25.6 Tbps · 64 x 400G

400G leaf

Edgecore AS9736-64D for 400G leaf roles and smaller fabrics, on the same lossless RoCEv2 primitives as the 800G tier, breaking out to 100G and 25G.

Broadcom Tomahawk 4
Software + qualified hardware

Supported as a system

藉助 OcNOS Systems,一份 IP Infusion 合約即可同時涵蓋 OcNOS-DC 與經過認證的交換器硬體,因此整個 fabric 只需一個 TAC 和一份 SLA。

Proof 2 · The scale

The scale is proven

The most common objection to Ethernet for AI is scale, and the industry has already answered it in production. You do not have to take a switch vendor's word for it, because the hyperscalers publish their designs.

Source: Meta's published AI cluster

Meta 在基於乙太網路的 fabric 上執行 24,576 GPU 的 AI 訓練叢集。 這是業界的證明:標準乙太網路如今已在正式環境中大規模承載超大規模 AI 訓練。它是回答「是否已就緒」這一問題的參照點:fabric 不是瓶頸。

在 OcNOS-DC 上,同樣的物理規律也適用於您的構建。雙層 Tomahawk 5 設計在每 GPU 800G、1:1 收斂比下可達 4,096 個 GPU。超出後,super-spine 層可擴展至 16,000 個以上。請參閱 AI fabric 拓撲 for the port math behind each tier.

Proof 3 · The utilization gap

The utilization gap is closed

The old knock on Ethernet was load balancing: standard ECMP pins each large flow to one uplink, so during a collective some links jam while parallel paths sit idle and the sync waits on the slowest link. That gap is now closed on the same switches, and the proof is a hardware feature, not a promise.

目標 90%+

How the gap closes. Static ECMP pins each large flow to one uplink; dynamic load balancing rebinds flows on real-time congestion so a well-run AI fabric targets utilization in the 90 percent range on Tomahawk 4 and Tomahawk 5, on the same switches, with no extra uplinks. OcNOS-DC ships DLB on those platforms.

DLB deep-dive →
Proof 4 · The standard

The standard is moving

適用於 AI 的乙太網路並非單一廠商的專有岔路。 Ultra Ethernet Consortium 發布了 UEC 1.0 規範,使乙太網路傳輸適配 AI 與 HPC 的規模,並獲得晶片、NIC 與雲端運算企業的支援。IP Infusion 是 Ultra Ethernet Consortium 的 成員 。

UEC 1.0 定義了一種多路徑傳輸,將一條流分散到多條路徑上,而不是固定在單個 ECMP 雜湊上,這一端點行為執行於支援 UEC 的 NIC 上。OcNOS-DC 如今已在 Broadcom Tomahawk 4 和 Tomahawk 5 交換器上,於生產環境中執行基於 PFC、ECN 標記和 DLB 的無損 RoCEv2。DCQCN 執行於 NIC 上,並對這些 ECN 標記做出反應。請參閱 Ultra Ethernet 深度解析 for how packet spray and multi-path RDMA change the transport.

Proof 5 · Ready to deploy

What OcNOS-DC ships today

Underneath the proof points is a fabric you can turn up now. OcNOS-DC provides the lossless Ethernet fabric that RoCEv2 requires, so your training collectives move GPU-to-GPU over RDMA without loss. DCQCN is the end-to-end loop between the NICs and the switches; the switch role in that loop is early ECN marking. Every primitive below is verified and shipping, not a roadmap item.

PFC

Priority flow control, including over L3, with DCBX and LLDP.

ECN and Dynamic ECN

Early congestion marking, the switch side of the DCQCN loop.

ETS

Enhanced transmission selection to keep RDMA flows ahead.

WRED

Weighted random early detection for queue management.

DLB

Dynamic load balancing with Reactive Path Rebalance and Random Flow.

PFC 死結偵測與復原(Deadlock Detection and Recovery)

Detects paused-queue cycles and auto-drains before jobs hang.

Buffer tuning

在晶片內建共用緩衝區上進行交換器封包緩衝區調校。

RTAG7 hashing

Fine-grained hashing for even spread across the fabric.

gNMI / OpenConfig telemetry

Streaming telemetry over OpenConfig YANG for closed-loop tuning.

Verified, shipping in OcNOS-DC
On the roadmap

What is coming next

The items below are roadmap and upcoming-silicon, labeled as such. They are not shipping capabilities today, and none of the figures are IP Infusion performance numbers.

全域負載平衡 (GLB)

Extends DLB fabric-wide: spines publish path-quality telemetry back to ingress leaves so routing scores the full multi-hop path.

Roadmap: OcNOS 7.1

Latency-based ECN

Congestion marking driven by measured latency, adding to the existing ECN and Dynamic ECN behavior.

Roadmap: OcNOS 7.1.0

LLR, CBFS, packet trimming

鏈路級重試、基於信用的流量控制和封包裁剪,用於在鏈路和交換器層面進行丟包恢復與擁塞處理。

藍圖

Broadcom Tomahawk 6

BCM78910 / BCM78914, 102.4 Tbps, 3nm. Announced by Broadcom. OcNOS support planned.

藍圖

Follow the evidence further

Each proof point above has a deeper technical page. Compare the transports head to head, size a fabric for your GPU count, or read how the lossless behavior is built.

AI fabric 實證常見問題

乙太網路是否已可用於正式環境的 AI fabric?
是的。證據已經部署落地,而不只是承諾。基於 Broadcom Tomahawk 5(51.2 Tbps,64 x 800G)的開放 800G 交換器已列入硬體相容性清單,目前已在供貨。超大規模雲端業者已在正式環境中以乙太網路執行大型 AI 訓練叢集,其中包括 Meta 公開的 24,576 GPU 乙太網路叢集。OcNOS-DC 提供這些工作負載所需的無損 RoCEv2 fabric,並具備可立即部署的 PFC、ECN 與動態負載平衡。
在 AI 工作負載上,乙太網路能否與 InfiniBand 相當?
就 AI 訓練所依賴的 fabric 行為而言,乙太網路提供同樣的無損 RDMA 傳輸,並彌合了過去區分兩者的負載平衡差距。Meta 公開的 24,576 GPU 叢集是業界證明乙太網路可在正式環境中承載超大規模 AI 訓練的例證。在使用率方面,動態負載平衡會根據即時壅塞重新分配流量,因此運作良好的 fabric 可在相同交換器上將使用率維持在 90% 左右,而靜態雜湊則會讓容量閒置。現在的區別在於,乙太網路維持開放與多廠商,而 InfiniBand 則是單一廠商。
乙太網路在 AI 訓練中已驗證的規模有多大?
Meta 公開的基於乙太網路的 AI 訓練叢集執行 24,576 個 GPU,這是標準乙太網路可承載超大規模 AI 訓練的直接業界證明。在 OcNOS-DC 上,雙層 Tomahawk 5 設計在每 GPU 800G、1:1 收斂比下可達 4,096 個 GPU。超出後,super-spine 層可擴展至 16,000 個以上。
OcNOS-DC 目前為 AI fabric 支援哪些功能?
OcNOS-DC 提供 2026 年 AI 工作負載所需的無損 RoCEv2 基礎功能,可立即部署:PFC(包括 L3 上的 PFC,支援 DCBX 與 LLDP)、ECN 與 Dynamic ECN、ETS、WRED、具 Reactive Path Rebalance 與 Random Flow 的動態負載平衡(DLB)、PFC Deadlock Detection and Recovery、緩衝區調校、RTAG7 雜湊,以及 gNMI 與 OpenConfig 遙測。這些都是經過驗證、已在供貨的功能,而非路線圖項目。
OcNOS-DC 在 Ultra Ethernet(UEC)方面處於什麼位置?
IP Infusion 是 Ultra Ethernet Consortium(UEC)的成員。OcNOS-DC 如今已在 Broadcom Tomahawk 4 和 Tomahawk 5 交換器上,於生產環境中執行基於 PFC、ECN 標記和 DLB 的無損 RoCEv2。DCQCN 執行於 NIC 上,並對這些 ECN 標記做出反應。Ultra Ethernet 傳輸執行於支援 UEC 的 NIC 上。如需規劃 UEC 在具體部署中的定位,請諮詢 IP Infusion 工程師。
現在可以部署開放的乙太網路 AI fabric 嗎?
可以。交換器已收錄於 HCL 並在出貨,無損 RoCEv2 fabric 如今已在 OcNOS-DC 中執行;藉助 OcNOS Systems,一份 IP Infusion 合約即可涵蓋軟體與經過認證的硬體,因此只需一個 TAC 和一份 SLA。您在 GPU、NIC 與交換器上的選擇保持開放。

深入了解,隨身帶走。

兩份簡短的技術下載,比本頁更深入:完整的 OcNOS-DC 資料表,以及無損 800G AI fabric 架構。

規格書

OcNOS-DC 規格書

完整的 OcNOS-DC 規格:EVPN-VXLAN 與 Ethernet for AI 功能集、軟體 SKU、支援的硬體平台,以及解決方案訂購指南。

取得規格書
解決方案簡報

OcNOS 800G 無損 AI Fabric

基於 Tomahawk 5 spine 的無損 RoCEv2 fabric:SKU 等級、經認證的平台與部署架構。

取得簡報

Ready to put the proof to work? Let's design your fabric.

Book an Architecture Review →