Now that AI is starting to seep into all debates, regardless of the level of knowledge of those discussing it, it’s time we see many of the concepts in common use. Players have become involved in processes such as intelligent upscaling of images, and more creative people use the same or cleaning routines. This year there has been a boom for chat and generative AI models, but there are and will be many more areas of use both in work environments and in everyday life that have not found much support when the process is created by the client and not the cloud.
Therefore, it has been very exciting to test and review the new laptop Medion E15443 (MD62621). Basically, this is a so-called “multimedia notebook”, a mid-range laptop for non-demanding use that can be used for office automation, multitasking and even image, audio and video editing. The most interesting thing is that it is presented as a device focused on AI processes thanks to the Intel chipset which, in addition to the Intel Core Ultra 5 processor and Intel Arc graphics, includes a “neural” unit (NPU) dedicated precisely to AI.
Without spending too much time on theory, this NPU is not only meant to take care of the client-based neural tasks, but also to support and lighten the load on the CPU and GPU. At the same time, it should make it possible for non-advanced computers to run processes that would previously have required more powerful and expensive devices. In fact, Intel hopes that up to 100 million computers in all price ranges will use AI accelerators by the end of next year.
This is an advertisement:
In other words, despite being designed as a laptop, the Medion MD62621 must deliver better performance in client AI than can be expected from a mid-sized gadget. We first tested the laptop based on what can be expected of it during general use from a traditional perspective, and finally we also ran some more specific AI tests.
Priced at around £650, this laptop has a fairly decent 15.6″ 16:9 HD display including Intel’s kit, 16GB of RAM and a 512GB SSD. It’s quite thin and compact, without it tries to compete with the lightest books, weighing in at 20.6mm and 1.8kg.
In terms of connectivity and other features, I’m quite grateful for the SD reader, which is becoming increasingly rare, but I miss the Ethernet port, although I understand that it’s not a priority on a device with Wi-Fi 6. It the only real gripe in terms of hardware is the length of the power cable, which is surprisingly short.
For just over a week, we had the pleasure of using a modern and fast computer for all kinds of tasks. The specs match well and the response time is fast even with a tabbed search engine, many automated office suites and some multimedia features that take up most of the RAM. The keyboard is responsive and soft, and fortunately has a numeric keypad. Similarly, the construction feels solid, although it is not luxurious in terms of design or cooling solutions.
This is an advertisement:
Overall, if you’re looking for a general-purpose multitasking laptop that’s lightweight for a 15-inch screen at the aforementioned price tag, this Medion is a real option to consider.
As for the AI, the results are very close to what I expected. Although the Intel Core Ultra 5 NPU can help in some processes by running them faster than other similar PCs without it, it is heavily dependent on software and compatibility, which I also expected. Some programs redirect these tasks to the CPU or GPU in a very specific way, and I don’t feel that Intel’s architecture can handle part of this workload on its own if it is assigned literally. In the same way, when there were already differences in graphics routines when comparing Nvidia and AMD, I think the same applies here when relying on the graphics card, in this case integrated on Intel’s SoC.
Since I loved the examples Medion used in the presentation of various AI applications, I tried three of them on my own. Leaving aside the overused generative AI, I chose to clean up a noisy photograph in Adobe Photoshop, split a song into tracks in Audacity, and run a couple of games to check if scaling could work on a laptop that was not a gaming machine.
To make an informal comparison, in the first and second cases I measured the runtime of two 2018 desktop PCs against the Medion MD62621. For the first test I used an Intel Core i7-8700K and Nvidia GeForce GTX 1080Ti 32 GB RAM (my work PC). In the second test, I used a compact desktop PC with Intel Core i5-8499 @ 2.8 GHz and Nvidia GeForce GTX 1060 16 GB RAM (my home PC).
I took a picture with a Nikon Z6 in RAW in a dark interior. That’s a cropped image of 5405×3603 pixels (from 6048×4042), taken at 300 dpi with ISO 2500, 14-bit color depth, and 27.5 MB file size, just to give you an idea. With the “noise reduction” function at level 50 (high on purpose) in the latest Adobe Camera Raw, the most powerful desktop took 27 seconds, the more modest 1:19 and the Medion 3:20.
Instead of using a lossless format, I converted Red Hot Chilli Peppers’ track “Goodbye Angels” from iTunes to a 45.2 MB WAV. Originally it was a 9.15 MB 256k AAC M4A with a duration of 4:28. I separated it using the OpenVINO plugin into 4 strains: Drum, Bass, Vocals and Others, and tried both CPU and GPU. The more powerful desktop took 1:18 with the CPU and failed with the GPU. The little sibling took 1:22 with the CPU and also failed with the GPU. And finally Medion took 1:32 with CPU and only 37 seconds with GPU. Actually it only took 58 seconds because it was the first time the model was loaded, but after that it only took 37 seconds as you read it.
As for the video games, I ran benchmarks for demanding games like Returnal and Avatar: Frontiers of Pandora, knowing that it was too much to ask of a laptop. Neither of them reached a comfortable frame rate to play on low settings (both reached 15 fps), but it was interesting, and perhaps logical, to see how Ubisoft’s game improved performance by changing AMD FidelityFX Super Resolution 3 to Intel Xe Super Sampling on video configuration and not graphics.
In conclusion, the results of the AI tests were quite interesting and different, and they reinforce the theory that there is not enough optimization or compatibility in terms of how much the NPU can do. I get the impression that the Adobe routines are supported on a dedicated Nvidia graphics card and that it doesn’t know how to properly utilize the NPU, whereas in generative we saw more convincing examples. The song, qualities aside, shows that Medion can face and even overcome more powerful computers. Although it’s important to note that OpenVINO’s learning comes from Intel and thus has compatibility issues if you’re trying to use an Nvidia GPU, ironically. Finally, while we saw the Medion run an older game like Rise of the Tomb Raider without a problem thanks to Intel Arc’s scaling, this may be the highest point a laptop like this can reach. In other words, it is not too bad for playing older or less demanding games.
That said, I still think the Medion 62621 is a well-balanced product with remarkable multitasking performance and a clever idea of AI integration that we’ll surely see perfected over the next not even years, but months. Today it is not fully optimized, as shown in the tests, but it is always better to have a computer that is ready, and if I were to buy a laptop in this price range, I would want this SoC.
 –){kind=link}