The links between GPUs matter

Multi-GPU PCIe Topology, NVLink and Interconnects

Understand aggregate VRAM, PCIe paths, NUMA placement, NVLink and NVSwitch before choosing a multi-GPU AI workstation or rack server.

Three-quarter supplier render of a 4U OEM multi-GPU rack server
Three-quarter supplier render of a 4U OEM multi-GPU rack server
OEM platform reference render. It is not evidence of a completed customer build or final specification. OEM supplier reference image.
OEM platform reference render. It is not evidence of a completed customer build or final specification.

Several GPUs can provide independent workers, replicated services or one distributed model. Those patterns use the same hardware inventory in very different ways.

Aggregate memory is not automatic pooling

Adding the memory printed on each GPU gives an aggregate figure. Software still needs an explicit partitioning or sharding strategy to use several cards for one model.

If one stage or tensor shard exceeds the memory of its assigned GPU, a large aggregate total does not solve the placement problem.

Independent workers

One model or job per GPU reduces communication and can provide predictable isolation. It suits embedding, transcription, rendering, image generation and replicated smaller language-model endpoints.

The scheduler still needs to understand memory, queue priority and failure handling.

Sharded inference

Tensor and pipeline parallelism divide a model or its execution across devices. Transfers then sit on the critical path. The runtime, checkpoint architecture, GPU count and topology all affect the result.

A quoted “supports eight GPUs” chassis does not establish useful eight-GPU scaling.

PCIe lanes and paths

The physical slot can be wider than its electrical link. A path may cross a switch, host bridge or CPU-level interconnect. On dual-socket systems, CPU and memory affinity can add another boundary.

NVIDIA DCGM can report GPU-to-GPU paths and CPU affinity. That describes topology; it does not measure application bandwidth or prove performance.

NVLink provides supported GPUs with direct high-speed GPU communication. NVSwitch connects many NVLink endpoints into a fabric on suitable platforms.

These technologies do not make all software scale automatically. Link presence, link state, supported collectives and the application’s communication pattern still matter.

Consumer RTX 5090 and the controlled RTX PRO 6000 workstation references in this catalogue are not presented as NVLink systems. Their combined VRAM must not be marketed as a universal pooled-memory promise.

NUMA and host resources

Balance GPUs across CPU sockets and root ports, then place CPU threads, system memory, storage and network with the workload in mind. Model loading, CPU offload and network serving can expose an otherwise hidden host bottleneck.

Record system RAM population and storage path as part of the test, not as decorative specifications.

Multi-node is a different design

Once a model or service spans servers, network bandwidth, latency, topology, orchestration and failure behaviour become first-class requirements. A single-node benchmark does not predict a cluster.

What to request before purchase

Ask for:

  • a board and PCIe topology diagram;
  • negotiated link width and generation for every GPU;
  • CPU-socket and NUMA relationship;
  • exact GPU edition and interconnect support;
  • power and cooling limits at simultaneous load;
  • runtime topology output;
  • an application benchmark using the selected parallelism.

The correct design is the smallest topology that passes the accepted workload with operating headroom.

Technical context

See the physical and operating boundary

Use these views to connect the guide to the machine, its airflow and its operating environment. Captions state the limits of what each image shows.

Open 4U OEM GPU server chassis showing passive GPUs, cooling fans, processors and memory slots
Open 4U OEM GPU server chassis showing passive GPUs, cooling fans, processors and memory slots
OEM supplier render showing one possible internal layout. Components vary with the ordered build. OEM supplier reference image.
OEM supplier render showing one possible internal layout. Components vary with the ordered build.
Diagram combining model weights, context, cache and active requests into a memory headroom check
Diagram combining model weights, context, cache and active requests into a memory headroom check
Memory fit depends on the workload, context, cache and simultaneous demand, then needs a representative test. GPU Servers technical illustration.
Memory fit depends on the workload, context, cache and simultaneous demand, then needs a representative test.
Exploded supplier render of passive GPUs arranged above an open 4U rack chassis
Exploded supplier render of passive GPUs arranged above an open 4U rack chassis
Supplier layout render used to explain GPU density and airflow. It does not represent a confirmed package configuration. OEM supplier reference image.
Supplier layout render used to explain GPU density and airflow. It does not represent a confirmed package configuration.
Diagram showing an approved request, a local service, an approved store and a policy-controlled data path
Diagram showing an approved request, a local service, an approved store and a policy-controlled data path
A private deployment starts with the permitted data path, access policy and logging boundary. GPU Servers technical illustration.
A private deployment starts with the permitted data path, access policy and logging boundary.

Primary sources

Sources are checked at the review date. Platform terms, prices and public guidance can change; verify them at the point of decision.

Use the guidance

Make the next conversation specific.

Describe the workload without uploading confidential material.