00:00 / 00:00
Sleep timer ended
⏪ 10s
⏩ 10s
39.8K views
1 week ago
Keep your emails and files private with Proton Mail and Proton Drive. Protect your data with end-to-end encryption and get started for free: https://proton.me/l…
...more
100 Comments
Never trusting a company that dictates what benchmarks reviewers can and can't use.
At some point, I came to the conclusion that it's cheaper for me to run a cheap laptop with Moonlight over 5G 24/7 and run the actual computer system at home, so I made myself a Linux system that just boots into a VM at home. If I did things like Blender, that would also mean 100Gbps connection to the render farm, which no laptop can ever beat. I mean, my laptop is fairly nice, it's a Lenovo Yoga 7 14AKP10 and it was north of a thousand dollars, but that is probably the highest I will go for a laptop. I own a couple of DGX Spark's and for me, the laptop version would be a compute unit with a built-in console.
"considerable" being max 30 seconds is just whatever to me. The difference is made in features like DLSS. Also factoring some of these tests, especially the games aren't arm native it seems? It's a good showing to me, and imo better than a mac on plenty of measures. But it is true, you have to look beyond the benchmark.
Apple silicon remainds goated, just wish they'd make a more concentrated effort on gaming
6k... holy crap.
I think the idea is good, the execution is held back by oldish ARM cores and being a first generation product. I need a new laptop and this isn't it but a second generation might be if I can postpone the purchase long enough.
give us a NUC with RTX Spark instead of laptops...
Did the 27B model get mentioned... but then the result not shared? Edit mistake or am I blind? Like the only thing these laptops <i>should</i> excel in is being able to run larger models whilst still having access to CUDA (I assume). That combination is... very interesting to a lot of people who actually care about local models. The mentioned 27B model is... absolutely fantastic, and actually good enough for a lot real usecases. Good luck running that (well) with 16GB of VRAM.
NO matter what they offer, windows on ARM is Bullshit
Soooooo I should get a 5070 ti then.
I want to hear Linus say Sally sells sea shells by the sea shore 😁
Doesn't outperform my px13 much if any while costing literally 4000 more is wild
"We sold all 5090s to data centers for 10k a pop. Now let's sell the 5070 to consumers at 5090 prices." Nvidia - we are passionate... about making money.
Why no core ultra X7 in the comparisons?
I couldn't afford the price of one of these, but I was really hoping it would be a better performer than this. I think the best way forward for AI is to be as local as possible. Fewer data centers is better for everyone. <br><br>But the price to performance just isn't there. The offerings from AMD and Apple are just so much further ahead.
Put it in an handheld
I genuinely don’t understand the point of this laptop. <br><br>Gaming? 5060-5070 mobile level gaming. <br><br>Local AI? Ram so slow that unless you are getting the $6k version there is no real advtange over a 5090.<br><br>CPU? It’s ok…<br><br>Battery life? It’s alright…<br><br>Super dissapointed, but I think the next gen version of the software gets ironed out and the memory becomes a ton faster that it could be a real competitor.
I want to know what laptops we can get for gaming, I recently got a job and since I don't have any expenses I already have a bit of a large $1000 budget. But I dont know what to spend it on trying to compare all these "AI Laptops" in their gaming usecase. I am torn between intel/amd SOC's or waiting for new lineups from either of them. I do not want nvidia dealing with nvidia and linux is annoying, and I have tons of projects that I still work on which have assembly code so I dont want to bother looking at arm based chips
I wanna see benchmarking for the 5070 ti vs the SPARK using a text generation model that has to be split across VRAM and RAM. I wanna know how it fares using all 128GB of that unified memory when the discrete GPU has to deal with the layer split.
It can’t keep up with a binned down M5 Pro and is priced like a M5 Max, aight
arm architecture is new industry standard right?
I'll bet it's trash for daily use just like snapdragon if you use any even slightly non standard software.
this is what I call a 'slop chip'
Having to ask "what runs on everything?" also means this thing is dead. Every game you want to run will run on x86. 😂
try crossover, it is just proton for macOS
<a href="https://www.youtube.com/watch?v=aV81SqB6HQg&t=335">5:35</a> i do wonder how the DLSS will have changed things (even if a bit unfair i do think due to how good dlss 4.5 is, i will have liked to see it has to see a "real life application"
I would only consider this type of system for running large MoE architecture LLM's on it (Qwen 3.8 flash next for example). That is the main thing so much unified memory allows you to do. MoE models also mean that it will be much faster or similar speeds as the small models you tested. Running such models with 32GB of RAM would be maybe an order of magnitude or two slower.
Am I tripping or did they forget to put Qwen's performance numbers in?
wow so they're hot ass got it
If these things sell well and replace true laptops, we will have the fundamental example of "That's why we can't have nice things".
Regardless of performance, this thing is a flop because it can't compete on price/performance.
Marketing: 💪🏾<br>Reality: 🤏
Linus quietly going full "F u Nvidia imma do it my way" is kinda based and hilarious "it will perform like this plus a dlss" crack me up ngl
What a waste of 128GB of ram.
I mean, to be fair everyone should hate macbooks
I expect a better product from Nvidia
I really could not care any less for this product.
You really can tell which videos Linus LOVES making and which videos he feels obligated to make as a tech influencer.
No use dllss in benchmarks…. Why ?
Jesus, for that price i can build maxed out setup at home and get a thinkpad for outside use.......... Nvidia's next best option is to replace every laptop with RTX spark so it becomes the norm and then normalize abnormal prices but even then why would you pay 7k for a laptop....... 😭
Can it run Linux?
Sad I don't even get excited for hardware anymore. I always was pumped about the new hardware releases but its pointless now.
First off, thank you for keeping the ai crap in its own segment so I could skip it easily. Second, wow is this ever unimpressive. I wasn't very hopeful for this given the dated CPU designs, but wow, this was worse than expected. Guess I'm happy I'm not in the market for a $6000 laptop - though if I was I'd probably just go with Apple anyway.
Terrible Ai test choices!
This will be big fail like snapdragon laptop, even ai performance sucks and gaming performance is just like rtx 4060 laptop $6000 is not the price for this good job nvidia 🤣
Kind of seems like it's another Copilot+ situation. Good in theory, but vastly underwhelming in practice.
<a href="https://www.youtube.com/watch?v=aV81SqB6HQg&t=747">12:27</a> even then if you where flying and needed ai there still seems to be better options overall 🤷🏻♂️
btw 2700 single geek 7 and snapdragon score 4000 in single core
Nvidia is the most expensive tech company that even puts apple to shame
You're holdin... benchmarking it wrong
Yooo
M5 Max would blow all of them while at the same price as these RTX Spark Laptops
......why not compare it against the m5 max? It would've been more fair from a price standpoint
A 3090 generates more tokens per seconds than the DGX Sparks and the like. The only advantages of those devices is the large amount of unified ram so you can run large models. Otherwise the bandwith is 1/4 of a 3090. you can still run good models on 16 GB or 24 GB of VRAM.
When the Macbook pro is just plain better value at your pricepoint you really have to think hard about wtf are you doing...
Considering it's still early, it being just kinda mid and not outright bad is fairly promising, things could get interesting in a couple more years.
Summary: A solution for a problem we haven't found yet.
<a href="https://www.youtube.com/watch?v=aV81SqB6HQg&t=640">10:40</a> এটা আবার কি 😅
Linus should have been in bed it seems
Airplanes have starlink now lmao
<a href="https://www.youtube.com/watch?v=aV81SqB6HQg&t=545">9:05</a> yay
Weren't you supposed to get rid of those obnoxious braces a while ago?
why does AI only mean LLM to these guys? Colmap? TTS? STT? Music and Video gen (LTX 2.5)? such lazy work.
As far as I get it, it's a PC dedicated for being an AI workstation, with gaming capabilities as a side bonus.
<a href="https://www.youtube.com/watch?v=aV81SqB6HQg&t=179">2:59</a> Ouch
Theres an echo in the video
Where in the video are test of qwne3.8, you only showed the gemma
See this is why I don't care for it. Nvidia said it will run everything, that using Windows on a spark will feel as if it's a completely ordinary windows install, and the GPU gives you 5070 performance. Then when it actually comes out, they only have a vague claim for 100 games being compatible, and the GPU performs like a mobile 4050. Anytime Jensen says anything in those keynotes, it's just lies at this point.
Wow, they must have paid BIIIIG bucks for two videos on the main channel, in a row! 😨
Benchmarks have value, as they show direct COMPARABLE performance to other devices that have ran those same benchmarks, showing which devices are faster.
Did your benchmark team reply them "don't tell me what to do ?"
have to be one of the worst product from Nvidia at this price point....
I would like to have 64GB on my RTX5070.
So Jensen made a laptop for Jensen.
Id pay double to avoid apple
and the spark fizzles out into nothing…
so we finally got a mediatek cpu on real laptops?
Why not comparing with M5 Max with 40 cores? That is a direct competitor to AMD 395+...
never in my 11 years watching have I seen a video by Linus so early
huge "jus trust me bro" vibes 🥀💔
How does it perform in real time gaming? Is it similar to a 3060 TI?
Yeah, dont benchmark it, because its crap.
Nvidia lost me as a customer 2001 ending support for my 6month old GPU.
so strix halo has much better compat and better GPU performance but in turn needs extra software to run CUDA workloads<br>I guess I have my pick
they seem... fine, except that the price is a joke
The sad part is some of these designs (Surface Laptop Ultra and the XPS) are imo very nice, it's a shame they're all stuck with this chip and will probably never get an option with Panther Lake or Strix/Gorgon Halo. I reckon NVIDIA definitely used their market power to "force" these OEMs to release a fresh new design for RTX Spark.
Who else linusing their tech tip rn?
This confuses me - isn't the whole point of these laptops and the Spark chip to run AI models locally? Surely tests need to be made to take advantage of the large memory pool vs that of the 5070 ti? I cannot see anyone purchasing one of these for the purpose of gaming other than a secondary function. This review does 2 minutes on AI benchmarking when 90% of the video should of been about that. If it compares equally to other laptops and then can do AI a lot more effectively and offline using a variety of models then it is worth its value - as burning those tokens online is a significant expense. Also what power draw does it do while computing that AI vs something like a 5070 ti.
to me this is a launch on paper. it only supports win11? yeah no thanks, pass.
This is the crutch of memory bandwidth limitations. This is why APUs need HBM or some type of revolutionary architecture to mitigate those flaws of ram speed.
hope one day we can see the performance of native apps. conversion is never ideal, especially with microsoft.
Linux support?
Video suggestion: how to setup a local agent for pcs on the older side (5-10 years?) and what can an average person do with local AI agents (organize decades worth of downloaded files for health, utilities, finance, home projects?) or how to migrate files from an old machine to these new AI centric PCs.
Who’s going to buy a $6K ARM laptop today??? Why skip a full-fledged MacBook Pro, a Strix Halo laptop, or even an Intel laptop for half the price, when you could just pay for a monthly subscription to a frontier AI model? Either way, these RTX laptops still can’t run frontier models locally, so what’s all the hype about?
Although AMD was in the gaming benchmarks, why didn't you mention it once? It seemed to beat the N1X is every game, but your "findings" @<a href="https://www.youtube.com/watch?v=aV81SqB6HQg&t=270">4:30</a> is that you want something like the spark laptops over apple or Qualcomm.
That was... underwhelming. For the price I expected better. I guess that's current NVIDIA though, I wonder when we'll get out of this age of price fixing, market manipulation and corruption.<br>We really need to start eating these billionaires.
It feels like what Nvidia WANTED was the kind of middle-ground product that hits a sweet spot of broad appeal, but ended up making a middle ground product that just doesnt really appeal to anyone except people who are REALLY into local genAI and... for some reason really feel like they need to run it on a laptop?
That... Is actually extremely disappointing. I was kind of excited for these chips
Why take a chance with a new hardware platform when you can get equivalent or better performance from proven technology? Unless they compete on price, I don't think this will be very successful.
For a first gen product this is good.
Up next
From this channel
