CiscoSeptember 5, 20269 min read

350-601 DCCOR Lab Guide 2026: The 64 GB Fabric

Everyone quotes 8 GB per Nexus node. That figure is from the 10.2 era - current images want 10 to 12 GB, and that is what turns this into a 64 GB build.

  • 350-601Exam
  • 120 minTime
  • $400Fee
  • 5Domains
  • containerlabCheapest lab
  • 48 GB hostRAM needed
350-601 DCCOR home lab topology diagram

The minimum lab

DCCOR is the heaviest lab in this batch, and the advice circulating about it is out of date in two specific ways. First, the free tier of Cisco's own emulator ships no NX-OS image at all, so "just use CML Free" is wrong here in a way it is not wrong for the enterprise exams. Second, the widely repeated "8 GB per Nexus 9000v" comes from the 10.2 release; current images ask for four vCPUs and 10 to 12 GB each. Four of those is where your 64 GB goes. Note also that the blueprint moved to v1.2 and domain four is now Automation and Artificial Intelligence, which adds AI-infrastructure networking to a domain most study material still calls plain automation.

BGP EVPNvPC peer-linkSpine 1Spine 2Leaf 1 VTEPLeaf 2 VTEPHost AHost BUCS emulator
Seven nodes. The two leaves form a vPC pair; the UCS emulator sits outside the fabric entirely, on the management network.

Two spines and two leaves is the smallest topology that is genuinely a fabric: it gives you an underlay to route, an EVPN overlay to advertise into, and a vPC pair to break. The two hosts prove distributed anycast gateway and VXLAN bridging actually move a packet. If the memory is out of reach, drop to just the vPC leaf pair and two hosts - that still covers vPC, port-channels, SVIs, OSPF or BGP, HSRP, NX-API and RBAC in about 24 to 32 GB, and it is a legitimate way to start.

Where to build it

The platform choice here is unusually consequential, because one popular free option simply cannot run the images this exam is about.

PlatformGood enough?CostWhat decides it
containerlabYes$0, open sourceThe highest-leverage option if memory is tight: its lite Nexus profile runs at 6 GB and 2 vCPU rather than 10 to 12. SSH, NETCONF and NX-API come pre-enabled, which suits the automation domain.
CML-PersonalYes$199, 20 nodesThe minimum CML tier that works at all here. The paid image set carries the NX-OS 9000 node definition at 12 GB.
CML-FreeNo$0, 5 nodesA trap for this exam. The free image set has no NX-OS image, only IOL, IOL-L2, ASAv and Linux, so the node count is beside the point.
EVE-NG CommunityYes$0 + a free Cisco downloadTechnically equal to CML for Nexus - the Nexus 9000v image is itself a free download with a Cisco account. Needs physical cores, not threads.
Cisco DevNet SandboxCheck first$0The only realistic way to touch ACI or Nexus Dashboard. The sandbox platform went offline on 1 August 2026 for a rebuild, so confirm availability before planning around it.
Real hardwareFor storageUsed marketThe only credible path for Fabric Interconnects, MDS and Fibre Channel. Most candidates skip it and accept the gap.
Worth knowing

Two things Cisco does not put on the pricing page and everyone gets wrong. CML-Free's image set contains no NX-OS at all, so the free tier is not a data-centre lab whatever the node count suggests. And Cisco does not state whether the paid tiers are annual or perpetual, so treat $199 as a price, not a subscription.

Images and horsepower

The Nexus node is the whole memory budget. Everything else on this list is rounding.

ComponentWhat it isWhere it comes from
Nexus 9300vThe spines and leavesFree download with a Cisco account. Current releases want 4 vCPU and 10-12 GB each
Nexus 9300v via containerlabThe same image, lighterThe lite profile runs at 6 GB and 2 vCPU - the single biggest saving available
UCS Platform EmulatorUCS Manager, service profiles, pools and policiesFree with a Cisco account, about 4 GB. Frozen at UCS Manager 4.2 since 2022, so no X-Series and no Intersight Managed Mode
Alpine or UbuntuTraffic endpoints on each leafBundled, about 1 GB each
Automation toolingAnsible, Python, Terraform, Nexus Dashboard conceptsOn your host, not in the fabric

On current images at 12 GB a node, a four-switch fabric plus hosts lands around 62 GB, so a 64 GB host is tight but works and 48 GB is the practical floor using older 9.3 images at 8 GB. CPU is a real constraint too, not just memory: budget eight or more physical cores, because these images want physical cores rather than threads. If none of that is available, containerlab's lite profile and a two-leaf topology bring it down to roughly 24 to 32 GB.

The build list

Work top to bottom. Each step is a thing you build and then break on purpose, and each one is tagged with the blueprint domain it covers, so by the end you have touched the parts of the syllabus a lab can reach.

  • Build the underlay and bring EVPN upNetwork 25%

    Point-to-point links from each leaf to each spine, an IGP underneath, then iBGP EVPN with the spines as route reflectors. Nothing else in the fabric works until this does, and it is the largest domain.

  • Make the leaves a vPC pair and break itNetwork 25%

    Peer-link, peer-keepalive, consistency parameters. Then break consistency deliberately and read the error - vPC troubleshooting is exam-shaped in a way vPC configuration is not.

  • Bridge two hosts over VXLANNetwork 25%

    One L2 VNI, a distributed anycast gateway, and a ping that crosses the fabric. This is the moment the overlay stops being theory.

  • Run UCS Manager in the emulatorCompute 25%

    Service profiles, pools, policies and templates all work in the emulator, and that is most of what the compute domain asks. Accept up front that it shows you UCS Manager 4.2 and no X-Series.

  • Configure storage networking and watch it not light upStorage Network 20%

    Enable the FCoE feature, build VSANs and zonesets. The configuration is accepted; the data path is not supported on virtual Nexus. Treat this as syntax and concept practice and know that is what it is.

  • Drive the fabric from code, and read the AI additionsAutomation and AI 15%

    Ansible and Python against NX-API, Terraform for intent, then read the new sub-topic on high-performance networking for AI in the data centre. Domain four grew a name and a scope in v1.2.

What you can and cannot lab

Cisco publishes the domain weightings but not the number of items or the passing score, so treat the percentages as how the paper is built rather than as a target. The column that matters here is the last one.

DomainWeightLabbableNotes
Network25%PartlyTODO
Compute25%PartlyTODO
Storage Network20%PartlyTODO
Automation and Artificial Intelligence15%PartlyTODO
Security15%PartlyTODO

The walls you will hit

The walls here are the hardest in the batch, and the biggest one lands squarely on a 20% domain.

Cannot be labbed at home

  • Fibre Channel and FCoE have no data path. Cisco states that on virtual Nexus, FC link-up and FC and VFC packet control and data path are not supported. You can build zonesets and never see a login. That is the 20% Storage Network domain, half-labbed at best.
  • No virtual MDS switch exists. Real FC zoning, VSANs and NPV behaviour are hardware-only.
  • No real UCS compute. Fabric Interconnects, chassis, IOMs, VIC adapters and actual server power-on are outside the emulator, which models UCS Manager and a fake inventory.
  • No Intersight and no X-Series. The emulator froze at UCS Manager 4.2 in 2022, and Intersight's free tier arrives with real hardware.
  • ACI's data plane is simulated. Cisco says so directly - you cannot validate a data path in the ACI Simulator, so use the free sandbox for the API and accept the rest is conceptual.
  • Throughput is a toy. Virtual Nexus uses a software data plane measured in megabits, so nothing about performance, QoS under load or buffering transfers.

Read that list and the study plan writes itself: the 50% of the paper that is Network and Compute labs genuinely well, the 20% that is Storage Network labs only as far as the command line, and the rest is documentation plus question volume. That is not a failure of your lab - it is the shape of a data-centre exam sat by people who mostly do not own a Fabric Interconnect.

A lab proves you can build it. Questions prove you can pass.

The blueprint is only half the exam — the other half is answering under a clock. Start with the free question sets, then $5.99 unlocks the full 350-601 DCCOR bank.

Free 350-601 DCCOR practice test350-601 DCCOR exam page

FAQ

Can I use the free tier of Cisco Modeling Labs for 350-601 DCCOR?

No. The CML free-tier image set contains no NX-OS image - only IOL, IOL-L2, ASAv and Linux nodes - so it cannot run a Nexus fabric regardless of its five-node limit. CML-Personal, EVE-NG with the free Nexus 9000v download, or containerlab are the working options.

How much RAM does a DCCOR lab actually need?

About 48 to 64 GB for a two-spine two-leaf fabric with hosts. The commonly quoted 8 GB per Nexus node is from the 10.2 era; current images ask for 4 vCPU and 10 to 12 GB each. containerlab's lite Nexus profile at 6 GB is the cheapest way down, and a two-leaf vPC pair with two hosts fits in 24 to 32 GB.

Can I practise Fibre Channel and FCoE in a virtual lab?

Only the configuration. Cisco documents that FC link-up and FC and VFC packet control and data path are unsupported on virtual Nexus, so zonesets and VSANs can be built and never come up. Storage Network is 20% of the exam and this is its hard limit.

What changed in DCCOR v1.2?

The weights are unchanged at 25/25/20/15/15, but domain four was renamed from Automation to Automation and Artificial Intelligence and gained a sub-topic on high-performance networking for AI in the data centre, alongside Nexus Dashboard, Terraform and Intersight. Cisco's own exam page still displayed v1.1 in its facts box well after v1.2 went live, so trust the exam-topics document.

Sources

Exam facts come from Cisco. Lab guidance is marked as official or as practitioner consensus where the two differ:

Checked September 5, 2026. Cisco revises exam topics and changes image licensing without notice — confirm before you spend anything.