Transcription
This is what it looks like when a company that's widely regarded as a top mini PC maker makes, well, not a mini PC. Durability test. Kidding. That might work for a Geek Mini PC I tried to destroy last year and couldn't. But with laptops, you got to be a little more careful.
It's just a box. I already unboxed it. A few things stood out. The packaging is actually kind of fun. Ooh. I don't know if they meant it or not, but the cutouts on the box are actually shaped like smiley faces, or more like a "me" face. Intentional? I don't know. But this is GeekCom's Geek Book X14. It's a really cool name if you're a geek.
Inside, you get a 65-watt power brick, braided cable. And okay, this is the part I love. They just included a dock for you. I use these constantly to add USB-A ports and Ethernet when traveling. So, the fact that this is just in the box is a plus.
Now, the laptop itself. Full CNC magnesium alloy case, and it feels pretty premium. I weighed this next to my MacBook Air. The X14, 1,025g. MacBook Air, 1,217g. The X14 is almost 200g lighter.
As for the I/O, they've managed to add all the important bits. Dual USB 4. Yeah, a full-size HDMI, a USB-A, and a headphone jack. For a laptop this thin, that's actually impressive. Especially the full-size HDMI. That's something you still don't get in a MacBook Air.
Inside, you get an Intel Core Ultra 9 185H. That's 16 cores, 22 threads, and boosts up to 5.1 GHz. 32GB of LPDDR5X at 7500 MHz, 2TB of PCIe Gen 4 NVMe SSD, and Intel Arc integrated graphics with ray tracing and XSS support. So, yeah, it can game. But what we care about is AI running LLMs locally. This chip will share the memory with the GPU and allow pretty decently sized models to be loaded up.
There's also an NPU on board, which is what we got to have these days for AI tasks, but that's probably the last I'm going to mention of it right now because there isn't a general solution for NPU usage yet. Still waiting.
It's got all the other bells and whistles. Wi-Fi 6E, Bluetooth 5.4, and a 72-wh battery in a tiny, thin laptop like this. GeekCom claims it could go 16 hours. Oh, and it ships with Windows 11 Pro, not Home. Pro.
"What about Linux?"
"Where's Linux support?"
"Linux."
"I want Linux."
"I know. I know. Okay, you Linux nuts, you."
But the real headline here might actually be the display. This is a 14-inch 2.8K OLED. That's 2880 by 1800 running at 120 Hz. For a laptop of this weight class, that's a lot of screen.
Now, at 450 nits, the screen is not as bright as a MacBook, and I would not use this outside on a sunny day, especially because the screen is so glossy. But indoors, it's a whole different story. The pure blacks that the OLED screen reproduces just look really nice in a coding environment.
As for the 120 Hz screen, I'll admit that while I don't really care about this during coding sessions, going from 120 Hz to 60 Hz does become quite noticeable to me when I switch between my daily drive, or MacBook Pro to the MacBook Air, for example, which runs at 60. And same here during normal use, like switching windows or dragging windows and animations, 120 Hz is definitely quite pleasing.
Now, the keyboard. It's good. Minimal flex, decent travel, nice backlight that you can adjust with a dedicated key. I do like that they kept the full function row and dedicated Escape and Delete keys. May all keyboards never lose this again.
There is a strange extension on the backslash key that I've never seen before. I have no idea what it does. It doesn't affect my typing once you ignore it. Kind of like how the MacBook notch is basically invisible to me at this point, but I do have to wonder, why is that there?
The fingerprint sensor on the power button works well, but a warning: if you press it a little too hard, the thing goes to sleep immediately. You can change that in settings, though.
As for the trackpad, look, if you're used to a diving board-style trackpad, it's fine. But I found myself having to really force tap into the lower left corner for any feedback of my click to be registered. This is probably the laptop's weakest physical attribute. But for me, I don't care much because I just prefer using an external mouse anyway, even on my MacBook Pro, which has the best trackpad in the business. So, take my trackpad opinions with a grain of salt.
Let's start with the storage because for developers, this matters more than most people think. I actually did a video in the past showing how compilation times get affected by the SSD. The drive in here is a 2TB Crucial P310. That's a PCIe Gen 4 NVMe. And in my testing, I'm seeing reads close to 7,000 megabytes per second, writes around 6, just over that. And that's actually really good for most workflows. That's like MacBook Pro level. In comparison, here's the most recent MacBook Air at about a third of that. at about half to a third of that.
Where you actually feel SSD speeds as a developer is stuff like cloning large repositories, spinning up Docker containers, compiling projects with tons of small files, or loading up a big model repo in your IDE. Gen 4 handles all that without breaking a sweat.
Now, one thing to note, this is a QLC drive with no DRAM cache. So, sustained writes on really large transfers can slow down once that SLC cache fills up. For day-to-day dev work, you're probably never going to even notice that. But if you're moving around massive data sets or AI model weights, it's something you should consider.
The CPU is an Intel Core Ultra 9 185H here. And at first, I thought, wait a minute, Panther Lake CPUs just dropped and GeekCom is coming out with a Meteor Lake-based laptop now? But given the pricing of some of the new laptops with Panther Lake, I can see why GeekCom wanted to still keep a 16-core, top-of-the-line chip from a previous generation to keep the pricing a little bit more sane for this ultrabook. And to be fair, it's still a capable chip. 16 cores, 22 threads, boosts up to 5.1 GHz. I already mentioned all this.
In this chassis, it runs at a 35-watt power limit with burst up to 45. And of course, on battery, we still have that drop. Don't forget the the drop in performance because the power has to drop to 30 watts on battery and all the way down to 18 watts when you're below 20%. So, no, it's not going to give you the fastest compile times, but for everyday dev work, IDE, terminal, browser tabs, local servers, it moves.
Now, a couple of benchmarks. Geekbench. Yeah, the uh Apple is really far ahead as far as laptop. And this is without a fan, by the way. The single-core score on that M4 is 3696, 14,728 on the multi-core. And this Meteor Lake is it's not the best. Even though the M4 has only 10 CPU cores compared to the 16 that are here, we're still beating it by quite a bit. 1,800 single-core score on the GCOM and 10,400 on the multi-core. Single core is going to matter in things like JavaScript applications, for example. Uh, those are single-threaded usually. And multi-core matters when you're doing big compilations and pretty much everything else.
We got 29.1, which is actually a pretty decent score for a Windows-based machine. This is for Speedometer 3.1, which represents single-core operations. And if you take a look at the M4 MacBook Air, those pretty much lead the way in single-core performance on laptops and SOCs. 54 over there, which is crazy.
I usually show this dock or example voting app, which you can download. It's open source and run it yourself on your own machine to compare. And this is basically using five containers that work together: a Python web app, Redis cache, a networker, Postgres database, and a Node.js web app as well. This is a fresh build right here. And we'll see how long that takes. So that's about 9 seconds total, and that's one container per instance. So it brings up a total of five. You can try that on your own as well and see what numbers you get.
Another test I like to run is the Mandelbrot algorithm in Python that can be found right here on benchmarks.game. Check it out if you haven't been to this site before. Basically, it just compares a bunch of different languages and different algorithms to see who wins. One of my favorite things to do. And I do this one. Python number three. You just grab that code and run it. And I use Measure-Command, which is a PowerShell commandlet that lets me time the operation. So, I'm going to say Python main, and then pass in that parameter that they recommend on the site. Let's do the same thing here on the Mac to see how that multicore will really affect it. Boom. And I like this test because it uses up all the cores at the same time. It really spins them up real hard. MacBook Air, you really can't hear anything because there's no fans there at all. This one, it's actually surprisingly calm. The fan is pretty chill. I mean, you can hear it all the time. It's pretty consistent, but it's not like spiking or anything.
Well, the MacBook is done. 14 seconds total for that one. This one is still working on it. Since this is a pretty intense operation while it's running, I'm seeing that we are hitting 52, 53 degrees in certain spots like the hot spots right over there. It looks like there's two exhaust areas. So, it does get pretty toasty up there. Woah. 54 right over there. The fan is not going super crazy, but the machine is pretty toasty. Look at that CPU curve. And it's done. 71 seconds. Um, I am having a hard time believing that that did not throttle because that should not take that long. And I will believe that it is throttling because if we take a look at the CPU, it's going at 1.97 GHz right now, which is quite a bit lower than expected. And I have a feeling that's due to the high temps. I mean, you're putting an Ultra 9 inside a really, really thin laptop. And I'm demanding a lot from it right now.
Now, just a little bit behind the scenes here. I did run this test before recording the video and I got about 30 seconds on this test. So, it's still about at least two times slower than the MacBook Air. 74 this time around. So, yeah, definitely something going on with the temperatures, but that just goes to show you if you're really going to be pushing this thing really hard, you should expect a little bit of a slowdown.
Since we're nice and toasty, let's see how quickly Visual Studio opens up because that tool is sometimes takes a little while to open up. So, let's go. Boom. Mhm. Mhm. We're thinking about it. We're going to. Hey, did it great. It's not terrible, right? Let's create a new project here. Blazer web app. Next. I just want to get a sense of how quickly we can get through this and launch the web application. And let's go create it. .NET Net 8 long-term support is what we'll use. Not sure why that's the default. Should be Net 9 at this point. Maybe even 10. And we'll go and launch that. Boom. Oh, don't ask me again. It's going to ask me again. It's going to ask me again. Here we go. Do you want to install the certificates? Yes. It is asking me again. See? Yes. And yes. Four times I got to say yes. It's always happening for years. Many, many years. And there it is. So, it's not bad. It's not terrible. Let's see what happens if I stop this and relaunch it. And boom. There we go. Pretty good. Once we get going with the project, it's not bad.
Now, despite being warm, uh, the general operation, switching between windows is actually pretty spiffy. Everything is moving nice and fast. I'm not seeing any issues with this.
Now, you've seen that the MacBook does pretty well against this machine, but you know, the MacBook pricing is quite a bit different, too. The Geekbook is trying to be in a pro kind of market, and MacBook Pros are pretty untouchable as far as pricing goes. MacBook Air is what I have here. If you try to get a, let's say a 13-inch, which is a smaller screen than this, and you go with a lower 8-core GPU, for example, and you get the same amount of memory as this machine, 32 gigs, actually 32 gigs requires that you go with a 10-core GPU. And the default SSD is 512. We're going to need to update that to 2TB. And we've ballooned our price to $2,200. So, this is a pretty decent alternative.
Also, what can you do with 32GB? Well, we can run larger large language models, larger than would fit on this base model MacBook Air, for example. Let's go to LM Studio. I can basically run any models that are smaller than 18 gigs because that's what's available for the GPU. I can do larger models than that, but then things will be offloaded to the CPU and we don't really want to do that. So, GPT4 OSS 20B, pretty popular model, 11.28 GB on disk, can have 100% GPU offload. And there goes the memory. Yep. With the context and everything, we're we're using quite a lot of it. 30GB being used right now, but let's see what happens if we give it a nice little prompt here. "Design a scalable web application architecture for an e-commerce platform." And so on. Nice little web application architecture prompt there. Boom. Let's see what's happening here. Is this being processed by the GPU? And indeed, it is. Let's just switch over to compute here. Yep. You can see down here we are using 13.2 GB out of the 18, the 18 that's available to allocate for the GPU on this SOC, resulting in about 96, 97% utilization for this model on this machine, which is just just about where you want it to be. So this is a pretty decent model to run on this machine. It's not going to be the fastest thing in the world, but it runs and it fits and it's going to give you a decent output. So, there it goes. Drawing a high-level overview for me. Thanks. I'm going to stop that. 12.5 tokens per second. Let me know what you think. It's very light, portable, can do things, very capable machine. Is it the best thing out there? No. But it's also not priced to be the best.
Now, here GeekCom is entering the laptop market for the first time. Last year, they made a pretty indestructible little mini PC, and I kind of want them to do something like that for a laptop. Make something that's going to be not dainty. Not that this is dainty, but something that's going to be more rugged and indestructible for a laptop. That'd be kind of cool.
All right, thanks for watching the video. Let me know what you think in the comments down below. Would you buy this thing? And I'll see you next time.