As a tech enthusiast with years of experience under my belt, I must say that the AI landscape is akin to a wild frontier, brimming with potential yet shrouded in mystery. The latest addition to this frontier, the ThinkCentre neo Ultra, presents an intriguing case study in the world of AI PCs.
This week, I’ve been working on assessing Lenovo’s ThinkCentre neo Ultra mini PC. It’s an elegant, potent, and space-saving desktop that I’ll share my comprehensive thoughts about in the coming days. However, I’ve encountered a problem in the process.
In addition to its Intel desktop processor and NVIDIA graphics card, the ThinkCentre neo Ultra also features a Neural Processing Unit (NPU). Unlike most AI computers I’ve encountered previously, which are usually in laptop form and integrated into the System-on-Chip (SoC), the NPU is a separate component in this system.
In the ThinkCentre neo Ultra, the NPU originates from Kinara, a company that debuted its second-generation Ara-2 NPU in December 2023. According to Kinara, this NPU is engineered to drive “edge servers and laptops with superior performance, cost efficiency, and energy conservation for running applications like video analytics, Large Language Models (LLMs), Latent Diffusion Models (LDMs), and other advanced Generative AI models.
Instead of the Kinara Ara-2, I’ve found other Neural Processing Units (NPUs) to be distinct because it is built onto its own M.2 board to make use of PCIe lanes. On the contrary, in the neo Ultra, from what I can gather, it seems to be installed on a PCIe conversion board that fits into the GPU slots. This NPU has its separate heatsink and a compact integrated fan. Quite intriguing! It’s my first encounter with this type of NPU, and I was excited to put it through its paces and compare it to other AI computers we’ve tested.
Where, oh where, has my NPU gone?
It didn’t take much time to understand that the Kinara Ara-2 NPU was only visible in Device Manager, nowhere else was it apparent. However, despite the ThinkCentre neo Ultra product page boasting about its “built-in AI features and software for enhanced productivity and creativity,” I couldn’t manage to locate them.
Despite being challenging to measure precisely, I didn’t observe any significant variations in performance outcomes whether the Neural Processing Unit (NPU) was activated or deactivated in the system. Even Lenovo’s built-in Vantage app, designed for monitoring hardware, appeared unaware that the device incorporates a discrete NPU.
It’s not unusual for some Next-Generation Peripheral Units (NPUs), found in AMD and Intel processors, to not appear in Task Manager or benchmarking tools until updates are installed. As we’re still in the early stages of AI PCs, there are some issues that need addressing. However, as long as they’re making a positive impact behind the scenes, that’s perfectly fine.
I use relatively simple methods to evaluate AI PCs, at least compared to what AI experts might consider. Mainstream tests for AI Computer Vision and Geekbench AI, which are used to assess Neural Processing Units (NPUs), are UL Procyon’s tools. However, it’s important to note that these options are quite new and come with their own set of limitations.
It’s intriguing to observe instances where the Neural Processing Unit (NPU) activates in AI-equipped laptops, as it’s engineered to deliver AI enhancement using less power compared to CPUs or GPUs. This aspect is particularly significant for devices powered by batteries. However, in a desktop PC like the neo Ultra, this might not be as crucial given that it also boasts an RTX 4060 GPU equipped with Tensor cores, which can handle AI tasks effectively.
Disabling the Ara-2 NPU in Device Manager didn’t seem to make any difference in the benchmark tests I ran, and its exclusion from any hardware monitors meant I couldn’t see when or if it was attempting to step in to help other apps utilize AI tools. I ultimately reached out to Lenovo for support.
Lenovo’s answer isn’t what I was expecting
I reached out to Lenovo’s technical support regarding the hidden discrete NPUs on the ThinkCentre Neo Ultra, and their response, upon writing this piece for initial publication, surprised me.
Although the Kinara Ara-2 NPU comes with the pre-installed PC, you were told that it can only be utilized within applications following a forthcoming software upgrade. No specific release date has been announced for this update yet.
When I inquired about a statement from the product brief provided with my review unit, which highlights how the “combination of the GPU and standalone NPU propels the ThinkCentre neo Ultra desktop into a content creation powerhouse,” I was advised that the information I received is not current.
It is common for such misunderstandings to occur, yet it’s important to note that there seems to be a lack of information on Lenovo’s website about the fact that the NPU (Neural Processing Unit) will only function at a later time.
Indeed, the realm of AI-focused PCs is advancing swiftly, and manufacturers strive to keep up with the pace. However, a person purchasing this PC with the expectation that the additional $125 NPU will function flawlessly may experience, at a minimum, disappointment.
AI PC benchmarking needs to improve
Allow me to continue by reiterating two points I made earlier, even prior to hearing back from Lenovo. Firstly, existing AI benchmarking tools appear insufficient, and secondly, the practical application of an AI computer for the typical user remains somewhat restricted. These observations remain relevant today.
It’s problematic when AI PCs can’t be fairly compared due to advertising hype and the rapid introduction of new hardware, as this doesn’t serve consumers well. If I hadn’t delved deeper to test a Neural Processing Unit’s (NPU) performance, I might have thought it was turned off, attributing any poor benchmark results to the testing software or my own lack of knowledge.
To help an ordinary consumer understand the utility of an AI-powered computer, a term frequently used in advertising “AI superiority” is Trillion Operations Per Second (TOPS). However, it’s essential to remember that TOPS serves more as a general comparison tool for Neural Processing Units (NPUs). An NPU optimized for one task won’t match the performance of another NPU designed for a different task, even if they have the same TOPS count. For instance, tests conducted using OpenVINO will favor Intel NPUs over their competitors.
As a tech enthusiast, I’m excited about the Kinara Ara-2, boasting approximately 40 TOPS performance. It’s not far behind heavyweights like AMD’s Ryzen AI 300 (50 TOPS), Intel’s Core Ultra Series 2 (48 TOPS), and Qualcomm’s Snapdragon X (45 TOPS). For now, while it has the power to run Copilot+ tools in Windows 11, access is restricted. However, we’ve seen changes with Intel and AMD in the past, so let’s hope for future updates that unlock its full potential! For now, the NPU in the neo Ultra will be tasked elsewhere, awaiting that vital software update to bring it to life.
The rapid progression of AI models is outpacing the development of standard tests used to evaluate all Neuromorphic Processing Units (NPUs), making it challenging for benchmarking software to keep pace. As a result, manufacturers like Lenovo can market AI-equipped PCs without a straightforward method for testing and comparing AI performance. While this may be advantageous for PC builders, it poses a disadvantage for consumers who seek accurate comparisons when considering AI capabilities in their purchases.
Do you really need an AI PC with NPU?
AI-powered PCs are increasingly trending, and many recently launched laptops incorporate Neural Processing Units (NPUs) within them. However, these advancements may not significantly impact daily usage for the average user. Most users will notice the NPU activating for features like Windows Studio Effects, but beyond that, you might struggle to find a practical application for it.
Speaking directly about the ThinkCentre neo Ultra model, it caters to AI experts who are seeking a compact computer equipped with a high-performance CPU and GPU for desktop use. The NVIDIA RTX 4060 offers approximately 242 TeraOps/second (TOPS) of power from its Tensor cores, which significantly surpasses the 40 TOPS provided by the Ara-2. This is a rough estimation of power, but it represents a considerable gap between the two.
It’s difficult to take advertising speech, prop it up against actual testing, and deliver accurate buying advice for the general PC user when everything is so obfuscated. That gets even tougher when huge companies like Lenovo start advertising and selling AI hardware that can’t be used because the software hasn’t yet caught up.
In this installment, it’s been made clear once again that AI-powered computers are essentially newborns in their developmental stages, and my viewpoint remains that most AI marketing is simply hype. I fervently urge Lenovo to include a disclaimer on their site, acknowledging that the NPU isn’t operational without a forthcoming software update. Such transparency is essential, even as the world of AI PCs retains an untamed Wild West atmosphere. Look out for my comprehensive ThinkCentre neo Ultra review in the near future.
Read More
- GBP EUR PREDICTION
- HBAR PREDICTION. HBAR cryptocurrency
- SEI PREDICTION. SEI cryptocurrency
- HOOK PREDICTION. HOOK cryptocurrency
- CNY RUB PREDICTION
- TRB PREDICTION. TRB cryptocurrency
- CTXC PREDICTION. CTXC cryptocurrency
- BCH PREDICTION. BCH cryptocurrency
- INJ PREDICTION. INJ cryptocurrency
- MNT PREDICTION. MNT cryptocurrency
2024-12-06 15:11