ADLink’s portable RTX A500 Thunderbolt GPU was recently reviewed and benchmarked. storage review We’ve found that it delivers excellent discrete GPU performance for the applications it was designed for. This pocket-sized discrete GPU, an RTX A500 GPU with 2048 CUDA cores, was found in benchmarks to be nearly 60% faster than integrated graphics solutions such as Intel’s Iris Xe graphics for GPU-accelerated non-gaming tasks such as Luxmark and AI.
Storage Review benchmarked small GPUs from Luxmark Hall, Luxmark Food, and Procyon AI Inference, and found Pocket GPUs to perform well for their intended role. Compared to the Intel Iris Xe iGPU with 96 EUs, the Pocket GPU was 54% faster at 3979 points in Luxmark Hall. In Luxmark Food, the Pocket GPU is 52% faster with a score of 1837 (vs. 869) and 77% faster with a score of 264 (vs. 59) in Procyon AI Inference.
Storage Review also benchmarked the RTX A5000 laptop GPU against the RTX A500 to see just how slow it is. The RTX A5000 laptop unsurprisingly lapped the Pocket GPU, performing about 65% better on average.
However, this is entirely to be expected, as the external form factor does not make mobile GPUs significantly faster. His Ampere GPU with 2048 cores performs just as well as his 2048 core Ampere GPU outside of the laptop chassis. Ironically, the Pocket GPU’s RTX A500 is actually an underclocked unit, with a base clock of just 435 MHz and a boost clock of a whopping 1335 Mhz. TechPowerUp reported that this GPU would feature his 1440MHz base clock and 1770MHz boost clock which are significantly faster for the reference design. The GPU makers probably dropped the clock speeds significantly to keep the GPUs at 25W TDP (a third of the reference GPU’s 60W rating).
ADLink Pocket AI GPU is one of the most unique external graphics solutions on the market. It focuses on providing discrete GPU power to laptops and other mobile devices that do not have a discrete GPU built-in. The unit is about the size of a palm and features a full-fledged RTX A500 GPU (professional equivalent to the RTX 3050) and connects to compatible devices via Thunderbolt 3.
The unit itself isn’t particularly great. It has the performance of an entry-level GPU for 2021. But for businesses and professionals who need a little extra power to run AI workloads, video editing applications, or other programs that may require a discrete GPU, it’s very useful.
The GPU specs include 2,048 CUDA Cores, 64 Tensor Cores, and just 16 RT Cores running at up to 96GB/s on a 64-bit wide bus. The unit consumes 25W of power and utilizes up to 4 PCIe Gen 3 lanes thanks to the Thunderbolt 3 interface.
Either way, Pocket GPUs are great for niche cases where you need discrete GPU power to run your AI applications in a pinch. It’s pretty slow compared to more powerful solutions and costs a whopping $429, but it’s basically the only GPU of its kind marketed specifically for AI acceleration. This feature is missing from most GPU-less laptops, with the exception of AMD’s new Phoenix chips.
