Nvidia RTX Spark: per què el nou xip ARM amb GPU Blackwell entusiasma i encara genera dubtes

I can't remember a more polarizing tech launch than any video RTX Spark. There are people who absolutely loved it that uh just so happen to get a big bag of money from a new video and others who hated it because of a knee jerk reaction to Jensen mentioning AI. And like I want to see Saml, let me get kicked in the marriage just as much as anyone else. But if you're going to hate on this launch maybe it should be because it uses a two year old media tech CPU. That's from a company who I don't trust with my Wi-Fi cards, let alone the whole system. So here my totally unsponsored thoughts on the Nvidia RTX Spark and if you want to support content like this, maybe check out our Patreon down below. So what is RTX Spark? Well, but simply it combines an ARM CPU and an Nvidia GPU onto a single chip. We get up to 6,144 blackwell GPU cores. That's a tiny bit more than what you'll find in RTX 5070 Ti mobile, up to 128 gigabytes of unified memory, and a 20 core ultra-efficient CPU. From the keynote, it wasn't entirely clear, but you will not need to buy 128 gigabytes of RAM to get an Nvidia spark. There are only models starting with 16 gigabytes of RAM, meaning that we will only have to sell this Civic to buy one and not this Civic. It seems like the launch of RTX Spark has been pretty rough. Based off of leaks it was supposed to be announced around this time of last year but it got delayed significantly. Given my experiences using Qualcomm laptops I have a sneaky suspicion that Windows on ARM just was not good enough for Nvidia but has this year delay resulted in proofing and becoming some beautiful bread or has the industry kind of moved on.

From what I've been able to gather based on leaks online, the CPU will have 10 P cores, specifically the ARM Core Tech's X925 and 10 E cores, the ARM Core Tech's A725. The X925 is quite a stout chip. It offered a 15% IPC increase, but the only problem is that was in 2024. So, two years later, that performance is not quite as impressive. We can get a basic idea of the performance to expect based off of benchmarks of the X925 in mobile CPUs, specifically the DEMONST-9400. In a phone we'll running at 3.63 gigahertz. It was able to record a single core score of about 2900 in Geekbench 6. So once it's overclocked a bit and a laptop, this is going to have a much higher TDP. We can expect single-core performance that's in the mid-3000s.

This means that because of the delays, Nvidia won't be able to offer the single core performance that their other ARM competitors have. Like Qualcomm is in the high 3000s with their new X2 chips, and Apple dominates everyone with more than 4,300 points. The thing is, I suspect Nvidia doesn't even care about that. X86 is so comically inefficient that AMD and Intel can't hit. 3000 and deep bench in most of their laptops. So even with emulation overhead, there's a good chance this thing's gonna be faster. Man, I would hate to be in teller right now, put an all that effort, make a process, no, that's now your angstroms and just get completely coked by a two year old mobile CPU. And even though the CPU performance probably isn't quite there, I don't think Nvidia is going to be too worried about Qualcomm or Apple either because of one big thing, gaming. As for gaming, unsurprisingly, this is in video and it works great. Like, right here is Pragmata. And, yep, it's running awesome. This right here is being fully emulated and you cannot tell because the Nvidia driver stack is being run natively on the GPU itself. So you're going to be getting the same GPU performance that you would if it's emulated or not. And you are going to take a hit and see few bound games.

Most of them are GPU bound anyway, so you're not gonna see much of a difference. You also have an interesting rate reconstruction demo here where in the past when you have your raised bounce and around, there's going to be a lot of noise generated like internally in the engine. And to get rid of that in video would just sort of like smooth it all out, I'm really simplifying this. But that can cause problems where We have these TVs right here and the models like, okay, I need to smooth out that static in the TV and you end up just completely losing that entire thing. But if we enable the new model, wow, the static actually works again. Given we have an ARM-based CPU though, the biggest concern is anti-cheat-like. I have had a lot of ARM laptops that will run games pretty well until you know you try and boot up Fortnite and it just... doesn't work in the slightest because easy anti-cheat is like, ah what the heck is this? But here we can see Fortnite. It's just running perfectly and that's because easy anti-cheat actually now has an arm version of it. So even though Fortnite itself is being emulated, you can get into Fortnite and it runs pretty darn well.

What is particularly wild though is that even though we don't know the TDP yet like literally no one does still is getting tuned I was told that all of these are running at under 100 watts total system power so that is wild though like an RTX 5070 mobile is more than that right? Yeah for creative workflows there's some stuff that could be game-changing like for us personally right now and use shooting 10 bit 4k and 4221 is FX3 and even though we're shooting a 30 and this footage here is a 60 wow is this a is this a powerpoint slide joy here yeah so that's trying to play back we have not gotten a single frame we do shoot a 30 fps this is at 60 but it doesn't really matter when this is a sort of performance that you're getting fortunately Adobe has rewritten oh whole bunch of their software. So all of the black VLGP is that already have 422 encoding built in. Can now do this sort of thing. This is 4K streams running at 60 FPS all at the same time. They said they can do 6 if they really want to. But like the performance difference is just staggering. Now people on DaVinci already have access to this. This is more of a Adobe catching up sort of thing.

Man, this is massive. Look at how well this is running. Even though this is on the spark rate now, any RTX GPU is going to benefit from this particularly the 5,000 series. Another thing that really impressed me is this Unreal Engine demo. You can see right in here we have this very detailed model. We can go outside of it and look around everything is great. And then we can also just go straight up into the air. And it turns out that this entire massive city. is just loaded in and it is running incredibly smoothly like just look at how much detail that is. Now, this isn't the bit that really impressed me. Like this is simply if you remove your liver and sell it and buy yourself 128 gigabytes of RAM, what really impressed me here is that this is totally emulated. Like not just Unreal Engine itself but like the DLSS drivers, all of the Nvidia stack. Every single bit of it is x86 running in emulation and this performance is what we are seeing. Between running this natively on arm and the X86 emulation, the performance should be basically identical in this. And that's because it's very GPU limited. Now, you are going to be seeing about a 10% CPU overhead when you're doing the emulation. So even though the actual raw FPS isn't going to change, the amount of battery life that you get when you unplug it is going to be a little bit better when you are running our native apps. And also,

This does still run really well on battery, but we will see what sort of battery life you can get when these things are cranked, you know, huge models like this. Another benefit that's going to come to everyone with an RTX card is in blender here. So we have this scene loaded up. It is rather massive and on the left is running through optics, which already is quite a big improvement over nothing. And on the right we have DLSS and the improvement in smoothness moving the scene just... cannot be understated. Like on the left, it's struggling hard and on the right, it's very nice and smooth. Like the usability of being able to just simply move the camera around your scene, it's night and day. The thing that really scares me about Windows on arm now though is emulation. Two years ago, a compu text I went to Qualcomm's booth and they showed off how awesome the emulator is and how well it works. And then

I got home and sometimes it was terrible. I'm really concerned I might get fooled twice being super bullish on windows on arm only to find it later that it just isn't what I expected but I think this time will be different for one reason. Invidious ability to swing around their big old dick. Discord for instance couldn't have cared less about making an arm version until Nvidia asked them to. Seventh zip is now made of on arm and here we can see solid works. running absolutely perfectly. I love to see this. In talking to the people at Windows, they said that first of all, they have been improving stuff the whole time, but also the Snapdragon laptops were aimed squarely at thinn-in lights, and they just simply didn't have the power to discover where the big gaps are in the emulator. Once you have that sort of juice and can run things like games and professional programs, you can start to find out where it's falling apart and fixed those and now hopefully it is good. Of course there will be gaps, particularly with niche programs, but I asked the most the Halatech NSP and now some emails have been sent off from Nvidia and Microsoft, so hopefully my personal niche problem will be solved. Now one thing that made everyone incredibly mad is how much Genson leading to AI during the Gemmin Stration. Basically saying that you won't need to use your computer anymore, AI will just do it for you. Frankly, I think this will be really great for...

people that don't work hard and don't have unique ideas, but for the rest of us, it might make them not in this task a little bit easier, but nothing will replace using your hands and your brain. And like having a computer that's able to do AI workloads locally isn't really a bad thing. You can do stuff like audio, denoising, blurring backgrounds and videos. There's loads of stuff in Photoshop, like all just selection that is awesome with AI acceleration and I think that as long as people are just using their own local hardware and not doing sewer farms that are ruining the environment like go for it dude. I don't care. As for the devices themselves, the only one that I could look at turned on was the Surface Lab Top Ultra, which is a shame because I have a sneaking suspicion. That's gonna be the worst one. Like, well, gaming, I know the pixel response times on it were awful and the entire device seemed to be just...

keeps soaking itself. But there's a very simple solution to that. Go get something else. Like this MSI Prestige 16 gets a tandem OLED panel that's 120 Hertz, thousand hits peak brightness, and there's the display issue just solves. Also, by the looks of this vapor chamber, I suspect that MSI won't have problems with the cooling either. Another bit of a set looks awesome is the ProArg P14. This thing's going to have a 1600-lit OLED, of course, the Nvidia Spark inside of it, but I'm also really stoked to put a full-size SD card reader in USB Type A. Like, once the last time you saw something that's thin and light, like it's less than a MacBook, but get some proper I. Overall, I have to say I'm excited about Nvidia's RTX Spark. I came into this video, expecting you to put them on absolutely blast. And then after using the devices and talking to the people, I'm excited again. I want the tech to be good. I want laptops to be awesome. And it... It might be, I really hope that I don't get burned twice, like holy fruit that would suck. But what doesn't suck is that you made it to the end of this video. Please hit like, get subscribed, maybe check out our Patreon because we didn't put any sponsors on this video. And of course, have a great all day.