Skip to content

Scale-up is not scale-out: NVLink vs the fabric

S1·E3The commodity fabric · Video call with the customer's ML platform team, Wednesday, two days before the BOM freezes

S1·E3Analyze~25 minsources checked todayverified against NVIDIA NVLink and NVLink Switch page, HGX AI Factory components, NVL72 AI Factory components and overview and physical topologies, all re-fetched 2026-09-07

Builds on: Which reference architecture governs this deal, Reading node nomenclature: 2-8-9-400

Before you read: what do you already know?

3 quick questions. Wrong answers are fine and expected; trying first makes the lesson stick.

After this lesson you can

  • Distinguish the NVLink scale-up domain from the RDMA scale-out fabric by semantics, size and bandwidth.
  • Quote per-GPU NVLink bandwidth and NVLink domain size for Hopper, Blackwell and Rubin from the NVLink page.
  • Compute the scale-up to scale-out bandwidth ratio for a platform and label it as derived rather than published.
  • Decide from a job's coherent-GPU requirement whether the NVLink domain or the fabric is the binding constraint.

Episode 3 — The commodity fabric

The situation · Video call with the customer's ML platform team, Wednesday, two days before the BOM freezes

The platform lead opens with the decision already made. They are looking at rack-scale NVL72 instead, so the east-west fabric is a commodity and the fabric line can go to whoever is cheapest. The network lead has his notebook open and says, mildly, “show me the counter.” The SE stops typing.

You ask one question instead. How many GPUs must a single forecasting job see coherently? The answer is 512.

That number moves the whole meeting. NVLink is memory-semantic and bounded by hardware: NVIDIA’s NVLink Switch table lists domains of 8 GPUs on NVLink 4 and 8 or 72 GPUs on NVLink 5, and inside an NVL72 rack nine fifth-generation NVLink switch trays deliver a total aggregated bandwidth of 130 TB/s.[1][3] Above 72 GPUs the job leaves that domain, and every crossing pair talks RDMA over a ConnectX-8 SuperNIC at 800 Gb/s per GPU across leaf and spine.[2] Each NVL72 rack is one scalable unit, and a fully tested system scales up to 8 SUs.[5]

The split exists because the two problems are genuinely different. Sharing memory coherently wants a bounded load/store interconnect with no headers; communicating at cluster scale wants a routable, message-semantic network with congestion control. Confusing them is how a 512-GPU training run ends up on a fabric nobody sized.

NVLink stops at the domain boundary, so the largest single job - not the GPU total - decides whether the fabric is plumbing or performance. Start with the semantics, not the bandwidth.

1Two networks with different semantics

Customers conflate NVLink and the fabric constantly, and the confusion is expensive because the two behave differently at every level.

NVLink is memory-semantic. A GPU reaches a peer GPU’s memory. The domain is bounded by hardware: NVIDIA’s NVLink Switch table lists domains of 8 GPUs on NVLink 4, and 8 or 72 GPUs on NVLink 5 and NVLink 6.[1] Inside an NVL72 rack, nine fifth-generation NVLink switch trays deliver a total aggregated bandwidth of 130 TB/s.[3]

The fabric is message-semantic. A GPU’s SuperNIC sends RDMA messages to another node across leaf and spine switches. On a B300 node each GPU has 800 Gb/s (2 x 400 Gb/s Ethernet) through its own ConnectX-8 SuperNIC.[2] Under dual plane those two 400 Gb/s interfaces land on two different leaf switches, each part of an independent fabric that scales to 1024 interfaces of 400 Gb/s.[6]

Rendering diagram…
Diagram source (Mermaid)
flowchart LR
  subgraph SU["Scale-up: NVLink domain, 8 or 72 GPUs"]
    G1["GPU"] --- NVS["NVSwitch tray"]
    G2["GPU"] --- NVS
    G3["GPU"] --- NVS
  end
  G1 --> NIC["ConnectX-8 SuperNIC 800 Gb/s per GPU"]
  NIC --> P1["plane A leaf 400G"]
  NIC --> P2["plane B leaf 400G"]
  P1 --> SP["spine"]
  P2 --> SP2["spine plane B"]
  SP --> REM["GPUs in another rack"]
  SP2 --> REM
Left of the SuperNIC is memory semantics inside one domain. Right of it is message semantics between domains. The boundary is the SuperNIC and it is where the units change from GB/s to Gb/s.

2The numbers, with their units

Write them down once and stop guessing.

Generation NVLink per GPU Links per GPU Domain Aggregate
NVLink 4 - Hopper 900 GB/s up to 18 8 GPUs 7.2 TB/s[1]
NVLink 5 - Blackwell 1800 GB/s up to 18 8 or 72 GPUs 130 TB/s in NVL72[1]
NVLink 6 - Rubin 3600 GB/s up to 36 8 or 72 GPUs 260 TB/s in NVL72[1]

The NVLink 6 row is marked as preliminary specifications that may be subject to change; repeat the disclaimer whenever you repeat the number.[1]

Two RA pages give the same figures from the platform side. The HGX B300 baseboard publishes a total aggregate bandwidth of 14.4 TB/s with GPU-to-GPU bandwidth of 1800 GB/s across its eight GPUs.[2] The GB300 NVL72 rack publishes 130 TB/s of total aggregated NVSwitch bandwidth and up to 1800 GB/s per GPU.[3]

Now the scale-out side, in different units. ConnectX-8 gives 800 Gb/s per GPU on both B300 and GB300.[2][3] One GB300 compute tray carries four ConnectX-8 Host Channel Adapters delivering 800 Gb/s plus one dual-port BlueField-3 B3240 at approximately 480 Gb/s aggregate.[3]

The derived ratio. 1800 GB/s is 14,400 Gb/s. Against 800 Gb/s of fabric per GPU that is about 18 to 1. Say that out loud as our arithmetic: both inputs are published on NVIDIA pages, the division is not.[1][2] An FAE who presents a derived ratio as a quoted figure loses the room the first time somebody searches for it.

Which RA governs this deal?

Four customer facts → one named document, its scope sentence and its SU arithmetic.

GPU count576 GPUsband 6 / 9
Platform
Who installs
Rack power
Enterprise Reference ArchitecturesDGX SuperPOD Reference ArchitecturesProgram Whitepaper32–256 GPUs · defines the node names2/4HGX AI Factory (B300)32–128 nodes · 256–1,024 GPUs3/4NVL72 AI Factory (GB300)up to 8 SUs · 18 trays per SU3/4HGX AI Factory (H100/H200/B200)the Dell installed base3/4SuperPOD B300 — Spectrum-4 / DC busbarSU = 64 nodes = 512 GPUs1/4SuperPOD B300 — Quantum-X800 / ACSU = 72 nodes = 576 GPUs2/4SuperPOD GB200 NVL72SU = 8 NVL72 racks · NDR-era1/4
Shortlist:
3/4 customer factsLast updated May 18, 2026

NVIDIA NVL72 AI Factory — Enterprise Reference Architecture (GB300)

Scope, in the document's own words

multi-user, single tenant workloads … users are all part of the same enterprise

NVL72 AI Factory — overview
Why this RA does not fully fit
  • Power: this RA is written for DC busbar (MGX), the site has AC rack PDU.
Scalable Unit

"fully tested" up to "8 SUs (Scalable Units)"; larger clusters are "possible based on customer requirements". One SU is 18 trays = one NVL72 rack.

  • Nodes at 4 GPUs each: 576 ÷ 4 = 144 nodes derived
  • SUs: 576 ÷ 72 GPUs per SU = 8.00 → 8 SUs to quote derived
Rack power — The rack demands "up to 142 kW", from "8 power shelves of 33 kW" with "six 5.5 kW PSUs" each. "liquid cooled, based on the MGX architecture".
Support path — Enterprise RA: partner/OEM-built. The RA assumes "12 management nodes" for HA and does not name an NVIDIA TAM.
Node nomenclature — 2-4-5-800 (dual plane). 2 Grace CPUs, 4 Blackwell Ultra GPUs, 5 NICs (4 ConnectX-8 Host Channel Adapters at 800 Gb/s + 1 BlueField-3 B3240 DPU at ~480 Gb/s from "two 400 Gb/s ports"), 800 Gb/s per GPU. The 4 CX-8 arrive as "two such boards, with two ConnectX-8 devices on each board". NVL72 AI Factory — overview
Scale-up (NVLink, inside the accelerator) — Rack-scale NVLink: "9 NVLink Fifth-Generation switch trays", "130 TB/s" aggregated NVSwitch bandwidth, "1800 GB/s per GPU" across all 72 GPUs.
Scale-out (the fabric, outside) — "ConnectX-8 SuperNIC: 800 Gb/s per GPU", usable with either Quantum-X800 InfiniBand or Spectrum-X Ethernet.
⚠ Derived, not published: 1,800 GB/s scale-up = 14,400 Gb/s, against 800 Gb/s scale-out per GPU on B300 — about 18x. This ratio is arithmetic over the NVLink page and the HGX B300 components page; neither prints it.
⚠ A search snippet phrases the top end as "Eight GB300 NVL72 scalable units with 144 trays/nodes and 576 Blackwell GPUs". The fetched overview page only says "up to 8 SUs" with 18 trays per SU — treat 576 GPUs as arithmetic, not as a published number.
What this RA does not publish
  • No BGP/EVPN, IP addressing, MTU, RoCE QoS (DSCP/PFC/ECN), adaptive-routing or congestion-control settings — those live in the Cumulus / Spectrum-X software docs, not in this RA.
  • No multi-tenant design. The RA is scoped to a single tenant; a service-provider deal falls outside it.
  • No AC-PDU variant: the rack is liquid-cooled MGX with 8 power shelves.

FAE angle: dual plane is not redundancy bonding. Each 800G ConnectX-8 port is broken out to 2x400G that land on two different leaf switches in two independent fabrics — leaf count doubles at the same per-GPU bandwidth and the customer must accept two IP fabrics. Ask whether their automation assumes one fabric per NIC.

Select the NVL72 AI Factory card and read the SU arithmetic: one rack is one scalable unit and the NVLink domain ends at that rack boundary. Change the GPU count to see when the fabric starts carrying the job.

3The sizing question that decides everything

There is one question that turns this into engineering: how many GPUs must a single job see coherently?

If the answer is 72 or fewer, an NVL72 rack keeps the whole job inside the NVLink domain and the fabric carries only what leaves the rack. Each scalable unit in the NVL72 AI Factory RA is a single NVL72 rack of 18 compute trays with 72 Blackwell Ultra GPUs and 36 Grace CPUs, and a fully tested system scales up to 8 SUs.[5]

If the answer is larger, every GPU pair that spans two racks talks over ConnectX-8 and the leaf-spine fabric, and the fabric technology becomes the performance decision rather than a plumbing choice. Both published options are large: Spectrum-X claims scaling up to 128,000 GPUs in two tiers with multiplane[8], and Quantum-X800 publishes 144 ports of 800 Gb/s per switch with SHARP v4, adaptive routing and telemetry-based congestion control.[9]

Note the asymmetry this creates in a conversation. You cannot buy your way out of a bad fabric with NVLink, because NVLink stops at the domain boundary. And you cannot buy your way out of a small NVLink domain with more fabric, because the fabric is roughly an order of magnitude slower per GPU and message-semantic besides.[1][2]

4Generations shift the numbers, not the framing

The framing survives every generation; the numbers do not.

GB200 NVL72 is a ConnectX-7 and NDR-era rack: each compute tray carries two GB200 Superchips, four ConnectX-7 NICs supporting InfiniBand NDR at 400 Gbps, and two BlueField-3 NICs supporting 2x200 Gbps connectivity.[10] GB300 NVL72 is ConnectX-8 and 800G: four ConnectX-8 at 800 Gb/s plus one BlueField-3 B3240 per tray.[3] Same rack shape, same NVLink 5 domain of 72 GPUs, double the fabric speed per adapter.

That is a two-times optics and leaf-port error waiting to happen. A customer holding a GB200 quote who then reads GB300 documentation - or the reverse - will size transceivers and leaf ports for the wrong generation, and the mistake surfaces at the cable delivery rather than at the design review.

On the scale-up side, NVLink 6 doubles per-GPU bandwidth again to 3600 GB/s with up to 36 links per GPU and 260 TB/s of aggregate in an NVL72-class domain, still on the preliminary label.[1] Note that the domain size in the published table does not grow between NVLink 5 and NVLink 6 - both list 8 or 72 GPUs.[1] Bandwidth per GPU grows; the coherent-GPU count that decides your design does not.

Two columns, one honest ratio

Request. “Build me the scale-up versus scale-out comparison for Blackwell Ultra so I can use it in a customer deck.”

1. Fix the platform. Blackwell Ultra appears in two shapes: HGX B300 (8 GPUs on a baseboard) and GB300 NVL72 (72 GPUs in a rack). Build both columns or you will be corrected.

2. Scale-up column, HGX B300. GPU-to-GPU bandwidth 1800 GB/s; total aggregate bandwidth across the baseboard 14.4 TB/s; domain size 8 GPUs.[2][1]

3. Scale-up column, GB300 NVL72. Up to 1800 GB/s per GPU; 130 TB/s total aggregated NVSwitch bandwidth from nine fifth-generation NVLink switch trays; domain size 72 GPUs.[3][1]

4. Scale-out column, both. 800 Gb/s per GPU through ConnectX-8, expressed by the RAs as 2 x 400 Gb/s Ethernet under dual plane, each interface landing on a different leaf in an independent fabric.[2][7]

5. Convert before comparing. 1800 GB/s x 8 bits = 14,400 Gb/s. 14,400 / 800 = 18.

6. Label the ratio. On the slide write: scale-up is roughly 18x scale-out per GPU (derived: 1800 GB/s vs 800 Gb/s; NVIDIA does not publish this ratio).[1][2]

7. Add the line that makes it useful. “Which means the design question is how many GPUs one job must see coherently, not which number is bigger.”[5]

Two columns on the shared screen

How it ended

You write the two numbers side by side - 1800 GB/s per GPU inside the domain, 800 Gb/s per GPU outside it - and say out loud that the roughly eighteen-to-one ratio between them is your own division and not a figure NVIDIA prints.[1][2] Then the sizing sentence: at 512 GPUs the job spans eight racks, so the fabric carries the gradients, and the RA is tested to 8 SUs.[5] What you actually say: “NVLink stops at 72 GPUs. Your largest job is 512, so the fabric is the performance decision, not the plumbing.” The fabric line survives. Overnight the customer’s network architect writes in: he has read the RA twice and believes he was sent an incomplete copy.

Lab

Goal: confirm the scale-out column against real hardware in the Dell lab. Every step is read-only.

  1. Pre-flight inventory. List the adapters with mst status -v and map them to interfaces with ibdev2netdev. Record the output before you change your terminal scrollback.
  2. Read negotiated Ethernet link speed: ethtool <iface> | grep -E 'Speed|Duplex|Link detected'. Expected: a Speed line in Mb/s, for example Speed: 400000Mb/s for a 400G port. If it reads Unknown!, the port has no link - note it and move on rather than touching configuration.
  3. Read the RDMA side: ibstat and, per device, ibstatus. Expected: a rate line per port and a state: ACTIVE on any linked port. On an Ethernet-mode card the link layer reads Ethernet.
  4. Compare each port’s rate to the scale-out column you built. A dual-port 400G card that shows two 400 Gb/s ports is the 800 Gb/s per GPU figure in its physical form.[2]
  5. Read PCIe width and speed with lspci -vv -s <bdf> | grep -E 'LnkCap|LnkSta'. Expected: LnkSta matching LnkCap in width. A x16 card negotiated at x8 halves host bandwidth and is a common cause of a benchmark that misses the published number - note it, do not re-seat anything.
  6. If a multi-GPU host is available, run nvidia-smi topo -m and record the matrix. Expected: NV# entries between NVLink-connected GPU pairs and PIX, PXB, PHB or SYS for pairs that cross PCIe. Write one sentence naming which pairs are inside the scale-up domain and which are not.
  7. Optional and customer-lab only: on an NVL72 rack read the NVLink topology and compare the domain size to the published 72.[1] Do not attempt this on shared production hardware.
  8. Acceptance: for one real port you can say the negotiated rate, the PCIe width behind it, and which column of your table it belongs in.

Retrieval check

10 questions from memory. Answer before looking anything up; misses become flashcards.

Explain it to a Dell SE

Explain to a Dell SE in four sentences why a customer cannot buy their way out of a weak fabric with NVLink, using one number from each side.

14 flashcards for this lesson — 0 in deck. Spaced review lives at /review.

Sources

Facts in this lesson were checked against NVIDIA NVLink and NVLink Switch page, HGX AI Factory components, NVL72 AI Factory components and overview and physical topologies, all re-fetched 2026-09-07. Dates are when each page was fetched.

  1. NVLink and NVLink Switch · fetched 2026-09-07
  2. Components - NVIDIA HGX AI Factory · fetched 2026-09-07
  3. System Hardware and Components - NVIDIA NVL72 AI Factory · fetched 2026-09-07
  4. NVIDIA GB300 NVL72 · fetched 2026-09-07
  5. Overview - NVIDIA NVL72 AI Factory · fetched 2026-09-07
  6. Networking Physical Topologies - NVIDIA HGX AI Factory · fetched 2026-09-07
  7. Networking Physical Topologies - NVIDIA NVL72 AI Factory · fetched 2026-09-07
  8. NVIDIA Spectrum-X Ethernet Platform · fetched 2026-09-07
  9. NVIDIA Quantum-X800 InfiniBand Platform · fetched 2026-09-07
  10. Key Components of the DGX SuperPOD - DGX GB200 Reference Architecture · fetched 2026-09-07

The same idea elsewhere

Other lessons that cover this ground, sometimes from another course's angle.