The original Khadas Mind from 2023 is still one of my absolute favorite mini PCs. Its absolutely tiny size and the modular concept made it easy to recommend, at least among Windows mini PCs. The company recently refreshed the model with Meteor Lake processors,but at the same time, it also introduced the AI Dev Kit, a version of the PC that actually includes the newer (and much better) Lunar Lake processors.

While performance does benefit from the newer CPU, the much higher price tag makes this harder to justify unless you really want a PC with a more advanced NPU. Though keep in mind this also sacrifices the internal SSD. It's still a solid PC, but this new model only really makes sense for AI enthusiasts.

About this review: Khadas sent us the Mind AI Dev Kit for this review. The company had no input in its content.

Pricing and availability

Khadas originally announced the Mind AI Dev Kit at IFA 2024 and made it available for purchase soon after. It can only be bought directly from the company's website.

Unlike the more consumer-focused Mind models, the AI Dev Kit doesn't have any customization options at checkout. It includes a Core Ultra 7 258V, 32GB of memory, and a 1TB SSD for $1,199.

Design and ports

It's the same as before

Khadas has done basically nothing to change the exterior design of the Mind AI Dev kit compared to the original Mind, and for good reason. The entire conmcept of the modular ecosystem of peripherals Khadas wants to promote hinges on the core compute unit having the same design, so it's going to stay the same. It would be interesting to at least see some color variation, but the core design isn't going to change.

I don't think that's a bad thing, though, because the Mind is still the smallest mini PC I've ever seen. This thing is absolutely tiny, so much so that you could carry it in a pocket. It's also very light, at just under a pound. If you thought the M4 Mac Mini was small, this thing will definitely surprise you.

Despite the tiny size, the Khadas Mind features a few ports, including two USB4 connections at 40Gbps, two USB Type-A ports with 10Gbps each, and one HDMI output. That's about as much as one can hope to fit in a chassis this small. The big thing here, as usual, is the Mind Link port on the underside of the PC. This is what enables communication with the ecosystem of accessories such as the Mind Dock and Mind Graphics, adding more ports and capabilities to the mix. Oddly enough, this model has some limitations, where you can't get as much bandwidth out of the port as you can on the Mind 2 (it's limited to PCIe 4.0 x4), which the company blames on Intel's Lunar Lake processors. I've also noticed issues with the Mind Dock frequently disconnecting from the Mind AI Dev Kit, forcing me to disconnect it and try again to get things going. This didn't happen with the original Khadas Mind.

Performance

It's faster, but not much quieter

The Khadas Mind AI Dev Kit comes with an INtel Core Ultra 7 258V from Intel's Lunar Lake lineup, along with 32GB of RAM and a 1TB SSD. Unlike the standard Mind, there's no internal storage here, and the SSD is easily accessible without any tools. Previous models also had an easily accessible SSD slot, but the base storage was soldered, so this is actually a bit of downgrade.

Intel's Lunar Lake processors are a huge step forward for the company, bringing new levels of performance and efficiency to the table, and the Khadas Mind AI Dev Kit definitely benefits from increased performance for both the CPU and GPU. The GPU in particular seems to benefit a lot, with the best example during my workload being the AI Denoise filter in Adobe Lightroom, which takes about 40 seconds to run on a 24MP RAW photo. On the previous model, this would be close to three minutes without an external GPU.

Of course, I ran some benchmarks, too. here's how it looks:

While a lot of tests weren't run on the original Khadas Mind, you can see a big upgrade in both CPU performance and overall performance in PCMark 10. And comparing it to Meteor Lake, while CPU performance is more back and forth, the GPU is clearly better with Lunar Lake.

A big potential benefit of Lunar Lake, however, is the efficiency. That was the big goal with these processors, but unfortunately, it's hard to tell here. I can't measure power consumption, but I can listen to the cooling system, and the Mind AI Dev Kit rarely seems quieter than the previous model. It's definitely nowhere near the noise levels of the M4 Mac Mini (which is dead silent 99.9% of the time).

AI performance

Of course, the big draw of this model is the AI performance, so I actually went ahead and ran Geekbench AI to see how much better it performs than the original Mind and some other PCs. Here's the rundown. For each cell, the scores are presented as Single precision / Half precision / Quantized.

AI benchmarking is very confusing right now, so the results require a bit of explanation. Windows and macOS have different AI frameworks, so for the Mac Mini, I had to substitute OpenVINO with Core ML, which offers the same options for the backend. Qualcomm-powered Windows PCs don't run the ONNX benchmark using DirectML and instead use Qualcomm's own QNN, so there's another difference there, but it's the closest comparison. And for some reason, the OpenVINO framework seems to not support Snapdragon processors yet, at least for this test.

You can tell that the Intel Lunar Lake chip is far ahead of the original Mind in both CPU and GPU performance, and even the M4 Mac Mini is outclassed here. However, when it comes to tests using the NPU, Apple's chip pulls a comfortable win. The Snapdragon X Elite using QNN for ONNX also seems to handily beat the Intel chip in Quantized performance, though it loses in Single Precision testing.

The Mind Graphics

Getting a new unit of the Khadas Mind, I was curious to try out the Mind Graphics because I had assumed that, like the Mind 2, this new Dev Kit would support more bandwidth over the Mind Link connector. Unfortunately, that's not the case. When I ran the 3DMark PCIe test using the Mind Graphics accessory, I got bandwidth of 6.57GB/s, which is only slightly higher than the 6.12GB/s I got with the original Mind.

As for the effect this had on other benchmarks, it's even smaller. Here's how the Mind AI Dev Kit compares to the original Mind when both are paired with the Mind graphics (specifically the model with 8GB of VRAM).

There seems to be some kind of bug with 3DMark Wild Life, but even without that, the scores here don't favor the new hardware all that much. That being said, if you are into AI development, I feel like you'd be better off with a device that has a more powerful Nvidia GPU that isn't limited by this the RTX 4060 Ti or the Mind Link connection.

Should you buy the Khadas Mind Dev Kit?

The answer for most people is not really. There isn't enough of an upgrade here to justify the increased cost of the Mind AI Dev Kit. But Khadas is pretty upfront about that. This is meant for AI developers and enthusiasts who can use that additional horsepower from the built-in NPU. Otherwise, you're better off looking at the Mind 2.

You should buy the Khadas Mind AI Dev Kit if:

• You need a PC with an NPU for AI development
• You want that PC to be as compact as possible

You shouldn't buy the Khadas Mind AI Dev Kit if:

• You're looking for a normal computer
• You want to get the best experience with the Mind ecosystem