A GPU interconnect is the link or fabric GPUs use to exchange data or access one another’s memory. It matters in AI systems because splitting work across GPUs also creates communication work: devices may need to exchange intermediate values, gradients, parameters, tokens, or collective results. A faster or better-matched fabric can ease that traffic, but it cannot guarantee proportional speedups; the workload, GPU placement, system topology, and software all affect scaling.
What a GPU interconnect does
When several GPUs cooperate on one job, each device processes only part of the work. The application must distribute inputs and work, move data between devices when needed, and sometimes combine partial results. The interconnect is the communication path that carries those transfers or enables peer-to-peer memory access.
The amount and pattern of communication depend on how the job is divided. Some workloads exchange data mainly between pairs of GPUs; others use collectives, such as reductions that combine values across multiple devices; still others need many GPUs to exchange data with many others. The key question is not just how much data moves, but which devices exchange it, how often, and whether the traffic is small and frequent or large and sustained.
How interconnect bandwidth, latency, and topology affect performance
Bandwidth sets a limit on sustained data movement
Bandwidth describes how much data a link or fabric can carry over time. If GPUs need to exchange large volumes of data, limited bandwidth can leave devices waiting for transfers instead of doing useful computation. A bandwidth figure is not, by itself, a prediction of application speed: its meaning depends on whether it is stated per GPU or for an entire system, whether it is unidirectional or bidirectional, and how the system routes traffic.
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Clear out junk files and repair common Windows errorsFree Scan →Scan for outdated or missing drivers - takes under a minuteDriver Scan →#1 Best Overall
- Read Before You Buy — No Video Output: These adapters support charging and USB 2.0 data transfer, but cannot transmit video signals. Except for standard USB webcams (which use USB data only), they are not compatible with HDMI/DisplayPort cables, video-capable USB-C hubs, or docking stations with video output.
- Convert USB-A Ports to USB-C: Designed to connect USB-C earphones, cables, flash drives, card readers, and other USB-C accessories to standard USB-A ports. Plug-and-play with no drivers or software required.
- Aluminum Alloy Housing: Built with a sturdy aluminum alloy shell that aids in heat dissipation and protects against daily wear and scratches. Designed to maintain a stable and secure connection.
- Compact & Travel-Friendly: The ultra-compact design allows the adapter to stay plugged into your device without blocking adjacent ports or adding bulk, reducing wear and tear on your original USB ports.
- 12-Month Warranty: Backed by a 12-month manufacturer warranty for peace of mind. Designed to meet strict quality control standards for reliable everyday performance.
Latency matters for frequent exchanges
Latency is the delay involved in a transfer. For workloads that make many small exchanges, delays can accumulate even when the total data volume is modest. A high headline bandwidth does not remove the cost of initiating or coordinating those transfers.
Topology determines which paths are available
Topology describes how GPUs, links, switches, and host interfaces are connected. Two GPUs may have a direct path, communicate through a switch, or depend on a path involving other system components. Consequently, the same workload can communicate differently depending on which GPUs are assigned to which parts of the job.
A 2019 evaluation of particular NVIDIA servers and HPC platforms reported communication NUMA effects associated with NVLink topology, connectivity, and routing, as well as an issue related to PCIe chipset design. Its results support the point that GPU pairing and system topology can affect communication efficiency; they are not a current benchmark for today’s products.
Rank #2
- 5-in-1 USB-C Hub: Experience comprehensive connectivity featuring a Power Delivery input, two USB-A 2.0 ports, a USB-A 3.0 port, and an HDMI port. (Note: The USB-C power delivery input port is only for connecting an external wall charger to power your laptop and cannot power peripheral devices.)
- 90W Pass-Through Charging: Achieve optimal charging with 90W pass-through power to your laptop, supported by a total input of 100W, with the hub reserving 10W for operational efficiency. (Note: Wall charger not included.)
- Quick Data Transfers: Accelerate your productivity with rapid data transfers using a high-speed 5Gbps USB 3.0 port and two 480Mbps USB 2.0 ports.
- 4K HDMI Display: Enhance your visual experience with a hub capable of delivering 4K resolution at 30Hz in both mirror and extend modes. Please note that this hub is compatible with MacBook (macOS 12 and newer), Windows 10 and 11, ChromeOS, and laptops equipped with DP Alt Mode and Power Delivery. Note: This device is not compatible with Linux.
- What You Get: Anker USB-C Hub (5-in-1, 4K HDMI), welcome guide, 18-month warranty, and our friendly customer service.
NVLink and NVSwitch: a link versus a fabric component
In NVIDIA systems, NVLink is a direct GPU-to-GPU interconnect. NVSwitch connects multiple NVLinks to support all-to-all communication among GPUs on supported platforms. Put simply, NVLink is a link; NVSwitch is a switching component used to connect links into a broader fabric. They are not interchangeable names for the same thing.
The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →These capabilities depend on the supported GPU system and its configuration. NVIDIA’s Fabric Manager documentation describes NVSwitch-based HGX and DGX systems; it does not establish that NVLink or NVSwitch can be added to any GPU or server after the fact.
Scale-up inside a system and scale-out between systems
NVIDIA describes scale-up as connecting accelerators within a tightly coupled multi-GPU domain, and scale-out as networking separate systems across a data center. These address different communication distances and needs. A multi-node AI job may use a GPU fabric within each server and a separate network to communicate between servers. A fast local fabric does not replace the cluster network.
Rank #3
- Sleek 7-in-1 USB-C Hub: Features an HDMI port, two USB-A 3.0 ports, and a USB-C data port, each providing 5Gbps transfer speeds. It also includes a USB-C PD input port for charging up to 100W and dual SD and TF card slots, all in a compact design.
- Flawless 4K@60Hz Video with HDMI: Delivers exceptional clarity and smoothness with its 4K@60Hz HDMI port, making it ideal for high-definition presentations and entertainment. (Note: Only the HDMI port supports video projection; the USB-C port is for data transfer only.)
- Double Up on Efficiency: The two USB-A 3.0 ports and a USB-C port support a fast 5Gbps data rate, significantly boosting your transfer speeds and improving productivity.
- Fast and Reliable 85W Charging: Offers high-capacity, speedy charging for laptops up to 85W, so you spend less time tethered to an outlet and more time being productive.
- What You Get: Anker USB-C Hub (7-in-1), welcome guide, 18-month warranty, and our friendly customer service.
Why AI workloads can stress the fabric
Parallel work still needs coordination
Data, model, and other forms of parallelism divide computation across devices in different ways, but can all require communication. For example, GPUs may exchange intermediate activations or gradients, distribute parameters, or combine partial results. Communication becomes a potential bottleneck when it takes long enough to keep devices from making progress on their share of the computation.
Mixture-of-experts can require all-to-all traffic
NVIDIA describes mixture-of-experts (MoE) inference in which tokens are dispatched to experts on different GPUs, then results are gathered and reordered. That pattern can create intensive all-to-all communication. It is one example of an AI workload where fabric characteristics may matter substantially; it does not mean that every AI workload is interconnect-bound.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Clear out junk files and repair common Windows errors3Scan for outdated or missing drivers - takes under a minuteHow to read published NVLink bandwidth figures
NVIDIA’s current product specification page lists the following per-GPU figures by NVLink generation and platform. The page labels specifications preliminary and subject to change.
Rank #4
- Dual Converters, Infinite Potential：Includes 2× USB C male to USB A female adapters and 2× USB A male to USB C female adapters. Perfect for a wide range of uses—tablets with Bluetooth keyboards, expand USB ports on macbook, and more. Two different converters for all your daily needs
- Next-Level 10Gbps & 3A Charging: No more slow 480Mbps, this usb to usb c adapter has a transfer speed of up to 10Gbps, allowing you to do more transferring in less time. This usb adapter fits both USB A and USB C charger, supporting up to 3A fast charging
- Upgraded Exquisite Craftsmanship: With an aluminum alloy housing and metal connector, the usbc to usb adapter is extremely durable and sturdy. Rigorously tested to withstand more than 10,000 times of plugging and unplugging, ensuring long-lasting performance
- Broad Compatible: The usb c to usb adapter widely supports all USB C/ USB A devices like laptops, tablets, cellphones, car chargers, and phone chargers. Such as compatible with MacBook Pro/Air 2023/2022, Thunderbolt 4/3 Devices,Apple MagSafe Watch 9/8/7/SE/Ultra, iPad Pro 2022/2021, Samsung Galaxy S23/S20/S10, and iPhone 17/16/15 Pro. Plug and play
- Please Note: To reach 10Gbps speed, keep the cable under 3.3 ft. For USB A Male to USB C adapters, try flipping the USB C connector. USB C Male to USB A adapters support bidirectional 10Gbps transfer within 3.3 ft
| NVIDIA NVLink generation | Associated platform | Listed bandwidth | Qualification |
|---|---|---|---|
| Fourth generation | Hopper | 900 GB/s per GPU | Per-GPU figure listed on NVIDIA’s product page; directionality is not specified in the cited wording. |
| Fifth generation | Blackwell | 1,800 GB/s per GPU | Per-GPU figure listed on NVIDIA’s product page; directionality is not specified in the cited wording. |
| Sixth generation | Vera Rubin | 3,000 GB/s per GPU | Per-GPU figure listed on NVIDIA’s product page; directionality is not specified in the cited wording. The product page labels specifications preliminary and subject to change. |
In a separate July 20, 2026 technical blog, NVIDIA describes the 72-GPU Vera Rubin NVL72 domain as providing 3.6 TB/s bidirectional bandwidth per GPU and 260 TB/s at rack level. Those figures use the blog’s stated platform and definitions; they should not be merged with the product-page table as though the sources reported identical measures. When comparing published numbers, check the platform, directionality, per-device versus aggregate basis, and measurement definition.
The available figures do not establish a broad, directly comparable current bandwidth ranking across GPU vendors. A configuration-specific number should not be treated as a universal vendor verdict.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What an AMD example does—and does not—show
A 2024 paper examined a particular four-physical-GPU AMD MI250X node, comprising eight GPU compute dies, using Infinity Fabric. In that tested setup, the authors reported that direct peer-to-peer access and RCCL outperformed MPI-based approaches for communication latency and bandwidth. They also described heterogeneous link counts and measured bandwidth tiers.
Free tools Windows power users keep installed
One-click scans. No signup required.
Best Value
- 5-in-1 Connectivity: Equipped with a 4K HDMI port, a 5 Gbps USB-C data port, two 5 Gbps USB-A ports, and a USB C 100W PD-IN port. Note: The USB C 100W PD-IN port supports only charging and does not support data transfer devices such as headphones or speakers.
- Powerful Pass-Through Charging: Supports up to 85W pass-through charging so you can power up your laptop while you use the hub. Note: Pass-through charging requires a charger (not included). Note: To achieve full power for iPad, we recommend using a 45W wall charger.
- Transfer Files in Seconds: Move files to and from your laptop at speeds of up to 5 Gbps via the USB-C and USB-A data ports. Note: The USB C 5Gbps Data port does not support video output.
- HD Display: Connect to the HDMI port to stream or mirror content to an external monitor in resolutions of up to 4K@30Hz. Note: The USB-C ports do not support video output.
- What You Get: Anker 332 USB-C Hub (5-in-1), welcome guide, our worry-free 18-month warranty, and friendly customer service.
That finding is evidence about the specific node and methods the authors evaluated, not a general comparison of AMD and NVIDIA systems. Results from one configuration cannot establish how different platforms will perform on another workload or topology.
How to evaluate an interconnect for an AI workload
Start with the communication graph of the workload: identify what information moves, between which GPUs, how frequently, and whether the pattern is pairwise, collective, or all-to-all. Then evaluate the complete system and software path rather than choosing by a bandwidth headline alone.
- Supported hardware and generation: Confirm which GPU and interconnect generations the system supports, and whether its design includes the needed links and switches.
- Bandwidth definition: Check whether a figure is per GPU or aggregate, and whether it is unidirectional or bidirectional.
- Latency for the target pattern: Frequent small exchanges can behave differently from large sustained transfers.
- GPU-to-GPU topology: Determine whether the GPU pairs your job uses have direct or switched paths, and whether some routes are less favorable than others.
- Peer access and software support: NVIDIA’s CUDA programming guide explains peer-to-peer access and transfers, and notes that device selection should account for hardware properties, CPU affinity, and peer connectivity. It also points to NCCL and NVSHMEM as higher-level communication libraries.
- Cluster network needs: For jobs spanning servers, assess the scale-out network as well as the local GPU fabric.
- End-to-end workload behavior: Measure or otherwise evaluate the target application; fabric specifications alone do not show how much time it spends computing versus communicating.
Why adding GPUs does not guarantee near-linear scaling
More GPUs add compute capacity, but they also divide the work into more pieces that may need to communicate and coordinate. If communication, synchronization, or an unfavorable path becomes a bottleneck, the additional devices may contribute less useful work than expected. Near-linear scaling from one GPU count to another is therefore a question to test for a particular system, software stack, and workload—not a general consequence of having a fast interconnect.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
Recommended Free Tools
