CopprLink could change the eGPU game: Performance nearly on par with native PCIe.

CopprLink is designed for enterprise systems requiring extremely high bandwidth. This standard supports a full PCIe 5.0 x16 interface, achieving a throughput of 32 GT/s per lane, equivalent to a total bandwidth of up to 64 GB/s.

For many years, discrete GPUs (eGPUs) have been seen as a 'compromise' solution – trading performance for flexibility and scalability. However, a new connection standard called CopprLink, developed by PCI-SIG, may soon completely change this perception.

New tests show that CopprLink is getting very close to a seemingly impossible goal: delivering performance nearly equivalent to a GPU plugged directly into the motherboard.

Full PCIe 5.0 x16 bandwidth: A significant advantage over Thunderbolt.

CopprLink is designed for enterprise systems requiring extremely high bandwidth. This standard supports a full PCIe 5.0 x16 interface, achieving a throughput of 32 GT/s per lane, equivalent to a total bandwidth of up to 64 GB/s.

Compared to current eGPU solutions like Thunderbolt or OCuLink (PCIe 4.0 x8), CopprLink offers significantly higher bandwidth – up to four times in some cases. This is a key factor in eliminating the long-standing bottleneck with external GPUs.

High-end test setup: RTX 5090 and enterprise hardware

In PCWorld's test, the system was equipped with an Nvidia GeForce RTX 5090 Founders Edition GPU connected via CopprLink.

The external components consist of a HighPoint RocketStor 8631D chassis – an enterprise-grade solution priced at around $1,300, complete with a 1,300W power supply. The server side utilizes a HighPoint Rocket 7634D converter card costing nearly $1,000.

The total cost for the connectivity system alone amounted to approximately $2,300, not including the GPU. This pushed the total setup cost over $5,000 – clearly not for the average user.

CDFP: A new connectivity standard for professional environments.

CopprLink uses the CDFP connector – a new physical connection standard designed specifically for server and data center environments.

Unlike standard connectors, CDFP prioritizes high stability and security. This connection is virtually impossible to accidentally disconnect, requiring specialized unlocking before removal. This is crucial in rack systems, where a lost connection can cause significant disruption.

Real-world performance: Only 2.3% slower than direct PCIe.

The most notable point lies in the performance results. In tests, GPUs connected via CopprLink were only about 2.3% slower than when plugged directly into the motherboard.

This difference is almost negligible, especially when compared to current eGPU solutions which are often significantly limited by bandwidth. The operating system also recognizes the GPU instantly, requiring no additional drivers or complex setup.

In other words, in system terms, the GPU functions almost like a 'native' component rather than an external device.

Implications for AI, workstations, and data centers

Performance nearly equivalent to native PCIe unlocks significant potential for data centers and AI systems. Businesses can scale GPU resources modularly, offering greater flexibility without sacrificing performance.

In the context of rapidly increasing demand for AI computing, the ability to 'decouple' the GPU from the motherboard while maintaining high performance is a huge advantage.

The future of eGPUs: From niche to the new standard?

Despite being very expensive and primarily aimed at the enterprise market, CopprLink shows a clear direction for the future of discrete GPUs.

Compared to Thunderbolt – which remains popular with individual users but is limited in bandwidth – CopprLink represents a technological leap forward. If costs decrease and the ecosystem expands, this standard could completely redefine how we use GPUs in the future.

Currently, CopprLink isn't aimed at gamers or mainstream users. But with performance nearly on par with direct PCIe, it has set a new benchmark for the entire eGPU industry.