3:14
Arm yourselves. There's a new mini PC format in town. Arm yourselves because the ASUS sent Q and 10 is not an x86 based. It's arm. It's an arm mini PC. Arm promising high performance, low power, quiet, affordable performance cuz it's based on the Snapdragon Elite X2, which is, you know, the newer processor. It's not the this is the second edition. We got to take a look at There's There's benchmarks. There's other benchmarks. This runs Windows.
Windows on arm? And Windows on arm. We'll take a little bit of a Linux detour, don't worry, but this is mostly about Windows for reasons I'll explain. So, you get the mini PC. Internally, you've got a 2280 M.2, and that's pretty much your only option for upgradeability. We have at the rear 2 and 1/2 gig HDMI, two 10 gig type A, and a 40 gig USB. We'll try that in a minute. At the front, we've got two more type A and two more USB 40 gig. Now, the power brick is twice as big by volume and weight as the mini PC itself, and it is 240 W.
Now, the configuration that I'm looking at here, that X Elite X2, 18 cores. Well, 18 threads. 32 gigs of LPDDR5, and our peak clock speed in Windows as reported by Windows is about 4.15 GHz. The big thing they're putting in all the marketing material is it's 80 tops, 80 trillion AI tops. So, it has an NPU, and 80 is greater than the 50 that is more typical of x86, but you know, things that use the built-in AI NPU, it's really int 8 performance. It's a different story. I don't want to crap on it. It's good for a mini PC, but at the low end for NPU, the amazing things you can do with AI is not quite as compelling as having actually more AI horsepower. Most of the
work and optimization of things like that are happening at the low end. If you need a little bit of help like organizing a document or running a small model, it's cool having the extra capability, but remember Windows wants a certain amount of that for Copilot, and the Copilot has not yet managed to convince everybody that it's a compelling feature of Windows. So, there are some aspects of this that you have to keep in mind are a little bit benchmarked, or maybe it's just a bug in AIDA64. It says the LPDDR5 memory latency is 6.9 nanoseconds. Nice, but also incorrect. And the overall benchmarks, not just the leading AIDA64 screenshot benchmarks, 7.8 nanoseconds of latency. That's a testing
artifact. But, you know, just for comparison, we got the AIDA64 Mini Forum MS-03 and the Expert Center PN55. That's the one that's based on an AMD CPU. And I have a feeling that under the hood our Snapdragon really probably is on the order of 100 nanoseconds. It's just that prefetch and some of the underlying machinery here is uh doing a good job hiding the latency of the memory. There's really not a lot of comparable benchmarks, but if we do Cinebench, Cinebench 2023, our single and multi-core score for the ASUS Ascent QM1 looks pretty good. You know, our single-core score is a little worse, and our multi-core score for 18 cores is a
little worse, but overall this is pretty good performance. Now, if we set up to 2024, uh you know, again, single-core score is a little worse, the multi-core score is, you know, roughly the same. Our Panther Lake CPU is doing pretty good. But, Geekbench. If we look at Geekbench, it's is oh, Geekbench is substantially faster single core and our multi-core score is phenomenal." Geekbench, you know, it's it really is an apples-to-oranges comparison with Geekbench on x86 versus Geekbench on Linux on x86 versus Geekbench on Macs and this is more comparable to Mac performance or at least the numbers that Geekbench
produces for Macs on the CPU side. So, again, just which synthetic benchmark do you want? It's it's really down to real-world application usage. GPU, again, ASUS QN Ascent QN10, the GPU looks really strong here. It's like, "Let's load up Steam and let's play some games." And it's like, "Well, not so fast." Thunderbolt GPU, well, not so fast. It's not quite there in terms of video driver and some of the other machinery that you need to do this. This is a productivity machine. Vulcan? Okay, Vulcan things narrow a little bit, you know, our AMD iGPU is pretty solid, pretty mature drivers, but again, the Vulcan score here looks phenomenal. 46,862.
It's a little bit Part of this is a is into our local AI story, being able to run Vulcan with this sort of stuff, but 32 gigs of LPDDR5 for local AI, probably not. Prob- not so much. Like, maybe the 64-gig configuration, but not the 32-gig configuration. It's really just meant for that INT8 super local, you know, 80 tops model, which is a different thing than trying to run AI through Vulcan. But overall, this is pretty respectable numbers. It's just that because it's the ARM platform with Windows, I don't have quite as much of surface area to give you for overall benchmarking. And I don't have I don't really have any surface area for the gaming side of it, either. This is really productivity. And if you're a developer,
this could be pretty good, but this is a different a different set of tests. Thunderbolt, you say? I'll just add an external Thunderbolt enclosure with my GeForce RTX card. Not so fast. Nvidia doesn't publish ARM drivers for Windows yet. They're going to have to soon. But, just going to the download page uh there's some bootleg drivers you can get. It's not fabulous. But, our Thunderbolt enclosure does seem to work, but there's not really a way for me to fully test it on Windows, which is maybe part of the plan because there's a lot of things you have to get right with PCIe tunneling over USB. The BIOS is sort of interesting. It's pretty minimal. There's not a lot of options there. If you're looking for
HDMI CEC, you're not going to find any HDMI CEC options. It does have some secure boot options, but it's limited because it's the ARM platform. So, you know, you get your TPM module and it's basically seems like it's always on. The fast boot option also disables your ability to boot from USB. So, you'll have to go into BIOS and disable fast boot in order to try booting from USB if you are going to go the ARM route for that. You can also control the fan speed. This is the only way that I found to reliably set the cooling fan speed to 100%. It doesn't need to run at 100% all the time. Like I say, even the AIDA64 torture test doesn't run it at 100%.
There's really not a lot of options under advanced. You can disable the USB ports and that kind of thing, which is useful in like business use case type scenarios, but usually you want remote or out-of-band management type options uh you know, Qualcomm is not really there yet. It's going to take a partner like ASUS uh and probably a buckets of money for them working together to really build a first-class business management type thing. Also, boot from network, it seems like it should work, but I could not actually get it to boot from the network. And I'm not sure if that's because it's a you know, legacy free PXE boot or whatever.
It might be a BIOS bug. Um it should work. It might totally be my network, but it didn't the network booting and configuration for MAC address did not work for me for whatever reason. Now, the Snapdragon X2 Elite Linux story is a little more complicated and it's a little more that I can get into in this video. Right now, sad to say, well, I mean, just cuz it's me, that Windows is a more first-class experience on the QNAP 10 than Linux. Uh, but at the same time, ASUS is a first-class citizen in trying to add Linux support, get Qualcomm and the appropriate parties together to add Linux support because ASUS not only has this mini PC, but they also have the X X2 Elite Snapdragon
laptop and that has had some action. So, Canonical has their experimental distro and there are patches that have been submitted to the Linux kernel for drivers and some other things, but it is not cohesive. The grain of salt I would also point you toward is that with the not X2, the older one, you know, we're we're coming up on 2 years, I think, now. Um, and that platform has not been super well supported in Linux. It's been up to a lot of individuals to sort of carry the weight there and get it toward a more first-class Linux experience. It's been at least as much, if not dramatically more, of a community effort than something commercially supported. At the same time, Canonical and ASUS have put in a lot of work,
which is nice to see in terms of Linux support. So, if you're interested in Linux support on this, especially after you look at those benchmark numbers, like those benchmark numbers seem like this should be just an amazing device for KDE and a very low-power desktop machine that just does what you need to and you can remote into a powerful server and do your powerful server stuff there, which is kind of the workflow that I have with Linux. And so, yeah, engage below because, uh, this could be some follow-up content or something like that when the Linux experience is a little more cohesive. So, what's the bottom line after using this thing like actually earnestly using this
thing? It's surprisingly snappy. Um my expectations were really not super high because I had taken some of the earlier Elite Snapdragon laptops for a spin and it did really didn't seem like the best of all worlds because, you know, Apple sort of sets the arm expectations astronomically high and x86 historically hasn't really, you know, performed all that well. But, this is entering the world at a time where Panther Lake CPUs from Intel that family of CPUs, you know, code name Panther Lake, um it sort of prom sort of lives up to lofty promises that Intel had for a long time. And so, it's a delightful experience on the x86 side. And yet, this is still I feel like that if Qualcomm had launched with this a processor of this
class, we would be having wildly different conversations than, you know, just a couple of years later. Windows on arm is a shockingly polished experience given my experience with Windows on x86. Windows has this weird square dance when you are a software developer. They have this thing called dev drive because uh not to put too fine a point on it, but the Windows volume management stuff is so trash when combined with antivirus. The fix is to deploy a file system within a file system, dev drive, or you can run that on a separate piece of hardware, but it runs better when it's a nested file system, which should tell you all you need to know about architectural choices inside
Windows. And yet, it's pretty solid experience on this. That has not been my experience on other platforms, but this not because the drive is particularly remarkable or anything else. It's just that I guess maybe the code pathways are less crufty on arm than they are on x86, but Visual Studio Code dealing with a billion tiny files on dev drive and the Windows Subsystem for Linux, I was expecting the Windows Subsystem for Linux to be a dumpster fire, but in reality, basically everything worked. So, if you need the Linux experience locked in under Windows, that is kind of an option on this platform. And Visual Studio Code scanning a bunch of files, working with dev drive, all of that was a standout experience on this drive.
So, yeah, that's pretty much it. Uh for a productivity machine for Windows arm stuff, yeah. I'm not speaking anything to the x86 compatibility layer, that is another dimension of testing that's needed. This does run x86 binaries okay, uh but it's not the fastest. And if you depend on x86 uh deeply, then binaries running on arm, then this is probably not the platform for you. But if you are doing coding tasks and you're working with everything natively and everything's made of arm, and Windows on arm, I really Windows on arm was nowhere near this good when certainly when the Surface machines launched. And again, when the Elite launched. But the Elite X2, uh Windows has had a lot of time to improve, and it really seems that they have. So,
I don't know, that's pretty much it for this one. I hope this machine makes another appearance in the future once Linux matures, but we'll see. We'll see, maybe. Uh if you have any other questions or I missed anything, hit me up at the forum.level1techs.com. I'm Wendell up, signing out, and I'll see you there. Let's chat about all things arm, you know. I mean, we could get there. It could be good.
