Graphics cards via Oculink will be much slower

A few years ago, when Thunderbolt 3 became popular, external graphics cards suddenly became viable and external chassis with PCI-Express 3.0 slots emerged from many manufacturers. What started as a way for Mac users to be able to upgrade the poor integrated graphics in Apple devices at the time.

Since then, Thunderbolt development has more or less stalled, although Thunderbolt 5 is now finally on its way and promises significantly increased speeds. Some companies last year began making laptops that contain an alternative to Thunderbolt that was previously only used on server motherboards: Oculink.

Oculink delivers up to 64 Gbit/s of transfers, nearly 50 percent more than Thunderbolt 3 and 4 which cap at 40 Gbit/s. In modern games and programs that often push graphic memory with large textures and data to be processed, bandwidth is more important than before, so it is interesting to see how fast modern graphics cards are when connected via the connector. Tested by Chinese Golden Pig Upgrade Company (via video card).

Golden Pig Upgrade tested both the new RTX 4070 Ti Super and RTX 4090 in 3DMark's Time Spy test, partly live in a desktop and partly in an external dock with an Oculink connection. The cheapest card was 9.8 percent slower through Oculink, while the most expensive card was between 0 and 22.6 percent slower. The difference was noticeably smaller with Time Spy's “extreme” settings, with no difference at all on an external display connected directly to the graphics card and 5.2 percent on the laptop's internal display. At default settings, the difference was 16.6 and 22.6 percent, respectively.

when Hot devices Testing the RTX 4090 connected via Thunderbolt, the Time Spy score was 32 percent lower than the live score via PCI-Express. This means that the Oculink feels 15-20 percent faster than Thunderbolt 3/4.