https://www.youtube.com/watch?v=wSp6AiNIrsY
>>108953015arm vvon
ai music still sound like shit
GET THE FUCK IN HERE NOW
>>108953177Nah Enjoy your slop
I HAVE NVIDIA SHORTS QUEUED FOR OPENING TOMORROW AFTER THIS DISASTERCLASS
I wonder how good the the N1X's performance will be for rendering in Blender.
where is the Y axis label
nvidia invented usecases for ai
NINE TRILLION DOLLARS
why is he talking about this base level shit
>compilation of random clips>epic music >this will happen>that will happen>agents>agents>agents>everything will go well>we will all make lots of moneyfatigue
AI is gonna run everywhere, every region.I love AI so much it is unreal.
>Please clapThis nigga channeling the ghost of Jeb! Busch
>tune in>first words are literally AI cloudyep. Nvidia presentation
maybe china should invade
>>108953345JEB IS A MESS
>tune in>see this>close stream
they don't care about gamers anymore, right?
>>108953398Gamers didn't make Nvidia the world's richest company.
>>108953418well.Kind of though
A CPU built for agents.
its over for amd
CPU shortage is next, I presume
>there only 1 billion of humansspoilers
There's only 1 billion humans?
X86 SIS, WE LOST
https://nvidianews.nvidia.com/news/nvidia-unveils-vera-the-cpu-for-agentsNVIDIA WONx86 IS KILL
Agentic CPU is not a thing, is it?
I am extremely sad to be poor.
And the mini Vera is N1XIntel/AMD is unironically finishedIt's Nvidia vs Apple now
>>108953486Marketing. The memory bandwidth is higher on Vera vs the other platforms on Phoronix's paid article so it's not surprising it won there. But Michael, Phoronix's editor, was only allowed to run what Nvidia allowed for that article.https://www.phoronix.com/review/nvidia-vera-benchmarksNot saying it isn't impressive being that competitive but it is cherry picked as fuck.
>>108953494N1X/N1 are using standard ARM cores, not Olympus cores in VeraSuper excited for future Nvidia Nxx SoCs when Nvidia brings Olympus successors cores and tech to it
>We are gonna hire thousands of agentsinteresting word choice
I just wanted to play videogames
running on human blood
N1X N1 ANNOUNCEMENT NOWGET THE FUCK IN HERE NOW
Here we go, he showed Windows.
Here we go!
>pc 2
OH NO NO NO NO NO
>personal agentno one asked for this
>agents will replace appsOh no.
Nvidia RTX Spark
>128 GB of unified memoryGo on...
NVIDIA WONx86 IS KILL
Say hello to the Nvidia N1X.
I don't really understandAre they trying to turn away even the most hardcore fan of AI or what?
>Mediatek
good news everyone, we are making architects redundant
>>108953592
>>1089534501.2 tb/s lpddr5x eccsounds very affordable
>claude, create me giant futa vr dating simulator
>>108953589<no price<intro prices are never get-able anywaylulz
https://nvidianews.nvidia.com/news/nvidia-microsoft-windows-pcs-agents-rtx-sparkNVIDIA WONTHIS KILLS THE x86
>>108953602my sister works in a company that does architecting and interior decorationI have to fear for her employement
BUTDoes it run Crysis?
>>108953590when nvidia starts the world religion, will you visit china to pay homage to jensen's jacket and wig?
NVIDIA JUST INVENTED PERSONAL COMPUTERS
>>108953612well? Let's see some swimsuit pics, come on.
No Asians thankyou.
20 pedoflops on your desk
How does it compare to whatever Apple is doing? It took a while before the MLX ecosystem developed, no? Adding a third one will be something. Is there out of the box compatibility with Cuda?
brainlet here, how did they fit 128gb on a chip that size?
Bravo George!
>>108953628You can run any model of Pytorch or Jax.
>>108953634unified memory doesnt mean the memory is in the chipit refers to ram chips, but this soc is designed to have really wide memory interface that is uncommon in consumer systems of the past outside of apple who does it because of mobile focus on energy efficiency
This is going to be out of the price range of the average person.
NVIDIA ROSA FENYMAN SPARK NEXT GEN SHOWN IN THE ROADMAPLET'S FUCKING GO
>>108953642Nothing will matter, because the price will suck.
>DONT YOU GUYS WANT STAR WARS SLOP IN YOYR HOME???
>>108953628more expensive but it plays games real gudmlx is irrelevant to this discussion on hardwarewhats the second one?it runs cuda duh hell you can run cuda host code on ryzen cpus too
whats the privacy policy of an Nvidia agent running locally (but still conneceted to the internet)?
>>108953653How about a jewish actress gf who is literally dressed like a clown?
>>108953656>gamesnobody cares.
>>108953665Its something a mac cant do.
>>108953679Like I said, nobody cares. Everyone now knows that gpu <> ai. the overlap is quickly vanishing.
>>108953465underrated post
>>108953656>mlx is irrelevant to this discussion on hardwareI mean, I would think their target market is machine learning practitioners, no?
>SIX FEET HUNDRED FIFTY POUNDS>JUST LIKE MEjensen
WE ARE SO BACK
>>108953717So nvidya being jews is what finally kills X86? is this really it?
jensen, if you wanna disrupt industries you should make a fucking game engine because unreal is so fucking bad
WHAT THE FUCK IS THIS LMAO
Kill all robots
LMAO what the fuck
>>108953725its pressing the X button time
these boomers are so out of touch
>>108953718Hell intel wants to kill X86 at this point.
LAUNCH THE NUKES I DON'T WANT TO LIVE IN THIS SOCIETY
>>1089537172028 Vera Ruby Spark
HUMANS ARE DONE FOR
>>108953730On the contrary, that whole room is full of boomers and Chinese ones at that. You are not the audience.
i cannot believe this is fucking rweal
Fuck clankers
Holy shit, Windows is back.
AI generated music sounds so low-bandwidth, rough and grainy.
Look Jensen, I get that every Nvidia presentation has to have some genAI slop graphics, but you gotta put *some* kind of effort into it
>>108953763Is that hideo kojima?
>>108953773I dont think nvidia owns the IP yet
https://www.youtube.com/watch?v=BR8Qs31FrxwHe finally killed his high school bully
>>108953778vera rubins literally use intel xeon cpus
>>108953778windows arm cpus wouldnt be possible without the snapdragon push right
https://www.nvidia.com/en-us/products/rtx-spark/WE ARE SO BACK, PC GODS
oh boy AI slop in an overpriced microsoft laptop!
zero (0) mention of a 6090 or new gpus.>lol datacenter shit only>u buy AI ok?
>>108953836>Welcome to the PC where agents work alongside you—running tasks, generating assets, writing code, on demand.LITERAL MICROJEET SLOPyou will rely on cloud AI and own nothing.
I still use my 1660 and it’s great, no „agent“ bullshit
>Get RTX Spark in a small, ultra-efficient desktop. Built to run personal AI agents 24/7 right at your desk plus game and create with the full power of RTX graphics.fucking garbage. you use virtual AI shit instead of relying on hardware you own.
I couldn’t watch this because of the time difference but was there something interesting? Are they winning over Apple‘s M-chips?
>>108953895We already knew this was the case.
>>108953917Nothing anyone didn't know a year ago.
>>108953926yeah no one wants nvidia to even release anything and have a 5070 tier gpu cost 1000€ because half of the cost is in VRAM
>>108953628Apple is better.
>5060 equivalent>AI slop focus>thousands of dollarslmao, fuck Nvidia, if only Lisa Su wasn't sucking Jensen's dick (her cousin) we might have some alternatives.
>>108953926there will be no 60 series. Jensen wants you to spend $200/month for a virtual AI 6090. (Geforce now)
>MediaTekLMAOLÖLhave fun with your trash, pathetic that not even Nvidia doesn't make the chips
>>108954025Nvidia has never made its own chips.
>no pricing info>no gaming info>no mention of geforce5000 supers never6000 series never
>>108954032they could have started now instead of going with MediaTek
>>108954040>no pricing infobecause the regular person won't be able to afford it.
some mediatek shit wont beat apple's mobile silicon.
>>108954041They have been fab-less since 1993, why start now?
>>1089539465060 ti is under 5000 cuda cores.
>>108953917>>108954061The GPU definitely will win watt for watt, Nvidia put in the same amount of cores as a 5070 inside the SOC so M5 Max will trade blows with it and anything else loses. However, the CPU will not since it is using base ARM core designs. Better multicore than M series but better than x86 which is all that is needed to compete with x86 SOCs. Nvidia will probably bring their custom CPU design next generation to the stack since as far as Snapdragon is concerned, they need to compete with them at some point and having better graphics will only get you part of the market.
>>108954105*Better multicore than M5 base series with worst single core but better than x86
>>108954105The N1x is rumored to have 16 memory channels, which is equivalent to only the M5 Pro. The Max chips have 24 and 32 channels, so up to double the memory bandwidth
waitchuds HOLD
use case?
>>108954040>5000 supers never>6000 series neverwhy bother?the culling is near anyhow
only two questions1. how much $2. how many points in passmark 1T
>you are HERE
>>108954120Memory bandwidth isn't everything and I wish a caveman would bash your stupid Apple fanboyism skulls in each time for even remotely suggesting that is all that matters online here and elsewhere. Apple M GPU has less compute than Nvidia's GPU, and not to mention it doesn't support anything below 16 bit data types for machine learning which is why it loses on prefill and long context tasks and other sorts of tasks that are NOT LLMs like video/image/audio gen which is more important in most AI systems today and where it is most useful. You can see it in benchmarks comparing what Apple has vs the DGX Spark today and when the N1x comes out, it won't be any different. The whole advertising angle with this machine is that it can do everything competently rather than what Apple has which was more of a happy accident that it was good for CERTAIN kinds of AI tasks.
>>108954205The highest spec N1X still has has <300GB/s memory bandwidth vs the max M4 with 550GB/sAll those CUDA cores in the sparks but they're starved for memory bandwidth. Maybe 3x faster prefill for LLMs but no faster token output than a rtx3090. And this only matters for AI. For gaming you'll have to contend with whatever fucking compatibility layer they use with Windows for ARM.
>>108954234and your currynigger silicon still gets destroyed by literally everything in every test recorded in history
>>108953733Well, yes, we can train it on any kind of content... Can I estimate your models for a week, Mr Huang?
>>108953634$3000-4000The various required subscriptions to properly use your "local" agents not includedGeforce NOW sub for gaming also not included
>>108953015is it on chip memory ?
>>108954202Jesus fucking Christ. This timeline went straight into the shitter.Take me back to the 90's, please.PLEASE!
>>108953015someone tldr this thx
>>108954202THANKYOUBASEDAPPLE
>slop AI>HuangI swear chinks are tasteless creatures. Only cultures based on ancient europe and Christianity can produce beauty.
>>108954411Nvidia pronounced the death of diy pc. Now you will either buy their all in one shit or have to pay $$$$ for the components (ram, GPU, HDD)
isnt it a bit odd to go "we made vera the bestest cpu ever that beats x86 like no tomorrow" and then follow up with a pc cpu made by mediatek
Why does LINE MUST GO UP when the line has already slammed into the ceiling?
>>108953015why does this guy still have an accent? speak American, idiot
>pull request is when they download software and modify it>commit is when they push it back upyeah this guy has never used git. embarrassing presentation
Say hello to the same old rake to step on.PeeCees in the 90s became just good enough for their tasks. Secretaries and school students were not using even 50% of functions of Word 95, and were totally fine. And if your computer printed letters just fine, why would you need a new one.The solution was quickly found: entertainment, games and multimedia. Both hardware and software makers knew that their businesses depend on constant growth of system requirements, and obsolescence of existing systems. Intel, for example, was really involved in gaming and multimedia projects, both when everything was software-rendered, and in the 3D accelerator era. It also was the main developer of HDCP and other DRM schemes, so you pay them a tiny amount each time you play a (legally licenced) video. Microsoft also ensured its dominant position through Windows-only gayming.This is exactly the same thing. They know no one is going to be “productive”, everyone will fap to their waifus, and auto-generate LinkedIn level crap for “official” tasks. They will ride on public escapism once again. Slop is not an accident. You will have to buy DirectAI® 2 compatible system to have a functioning Windows™ slopware, then DirectAI® 3 upgrade next year, then DirectAI® 4 will become obsolete before the release, and they will jump to version 5, and so on. The obsolescence rate like in the '80s and '90s is just perfect (and it's already working even on the giant corporate customers).Another old trick is talking about open this and open that. Models are open — if you buy the hardware and/or use services of companies that lease it. They are open and useless on your Commodore 64. This is just vendor lock-in.Oh, and one more thing. Please provide your official ID when logging in to make sure you don't infringe on our partners' copy rights in your creative process. Thanks. Your content will be uniquely watermarked in secure and anonymised fashion.
>>108954490You can't cut 1 cylinder from a race car engine, and put into a cheap car, it won't work. Their datacenter CPUs are too specialised for high performance to scale to single user system, so they get conventional solutions from a maker of a non-power-hungry chips.
>>108954630I assume the server designs overlap a bit with consumer oriented designs and they can share both for gpus and cpusJensen talked a lot how the cpu is designed for AI agents, and the laptop is sold as agent running laptop so its a bit, well wheres my agent running cpu
>and humans, there's only one billion of usseriously what the fuck is wrong with this guy. I guess he's too old so his brain is completely fried?
>>108954670people who dont an nvidia gpu dont count
>>108953015I dont get it, how will that slop replace my trash PC? I won't pay a trillion dollars for my next PC and I'm sure it's the same for most people. Is he so rich that he is completely detached from reality?
>>108954687there are people who own a 128gb ram system before AI boom and they'll be happy to own spark laptop for work, and most of the price is in the ddr5 memoryobviously this is not a gamer laptop, but for pros who need high ram for their app of choicegamers might get these things when they get cheaper in 5-10 years
have cooling solutions for laptops kept up?
>>108954663You won't micro-optimise your code for the specific hardware, like they do on the space heaters, you, as a general user, will use software someone provides.“AI agents” is a buzzword, it means nothing. You need to figure out what people actually imply when they say it, as it can be different each time.
I cannot fucking believe that regular old Joes are the ones bringing us the Singularity..... Like this guy started out cleaning fucking bathrooms, let that sink in, he was cleaning bathrooms and now he is elevating all of humanity, to literal fucking Godlike status.....The TV series Pantheon is a documentary -- Charles Stross's Accelerando is a documentary -- Star Trek is a documentary. THIS IS THE WORLD WE LIVE IN NOW. And it's fucking insane...
>AI software slop in commercial uses replacing workers, customer services and programmers I sleep as Nvidia literally invents make belief use cases to sell their chipsets to retarded investors when AI hallucinations cannot construct anything scalable without human supervision. >Nvidia hardware giving us a PC alternative to Apple's M4 with unified 128 GBs of memory and 6144 cuda coresThat's the real game changer, Nvidia should just stick to creating hardware instead of breaking their drivers with AI written slop. Spark will be a major deal for any coomer who wants to locally host their degen AI porn. Running models at their native precision points without quantization or by hosting larger (up to 200b parameters) models without delay.
So this the Windows macbook equivalent. Goodbye freedom.
>>108954697Gamers are not the majority of PC users. Most people use it for internet browsing, communication (4chan hello) or watching content like movies. I just can't see how his offer can appeal to most people and reinvent PC forever.
>>108954773dont take a talk hyping up a product with speculation about the future that seriouslypeople will buy it for the hardware +cuda here and now, or not
>>108954687Yes, the entire thing was basically doubling and tripling down on the massive failure that is the concept of "agents." Nothing interesting was revealed. Imagine paying $5-10k in hardware just to be able to produce nondeterministic, unreliable, extremely low-quality slop locally. There is no use case for this.If SOTA AI resulting from multibillion dollar investments still produces unreliable slop, your $5k AI hardware has no chance, it's just ewaste.Bubble is popping soon enough. We're approaching the peak.
>>108954581Nice hypothesis but no, that's not what's happening.>obsolescence rate like in the '80s and '90s is just perfectYou probably mean 90s and 00s as 80s was actually kind of slow. But if they were to speed up the obsolescence rate again, they would have to actually offer better price/perf every year. Something they didn't want to do even before they had datacenters paying 10x more in nvidias case.Intel actually demonstrated the business model 2012-2017. You only give 10% per generation incremental improvements to mainstream consumer/client, offer money-sink workstation products, and only offer the best to datacenters.When you're in a near-monopoly position there's enough automatic sales coming in every year from worn down laptops, mindless consooomers, businesses and institutions have fixed cycles anyway... The savings from giving almost nothing to the mainstream class are bigger than the lost sales, even if it means the 2700k guy gets to brag that his 5 year old build is still top-tier (Now the 3090 guy, I guess)
Topping out the keynote with some horrible sounding AI song.
>>108953015The cores in that new spark aren't even vram, fucking worthless. In the trash it goes.Enjoy your 5 thousand dollar subpar third world tier chinktech level lazy overpriced dogshit hardware.
>>108954865some people might want to run a 100gb llm locally.
>>108954748Windows on ARM already has very specific restrictions that Windows on x86 traditionally doesn't have.
>>108954873Aren't local llms all retard tier level intellect dogshit?Even frontier models constantly fuck up when your paying for them.I tried claud and grok today to get a modmade for a set of nich functions I wanted for the neoforge server and I had to baby Claude and it gave up after 3 tries... Grok 4.3 not only made the mod but fully automated downloading of portable non-globalized prereqs, code compilation, and deployment but even wrote me a powershell script to paste run in terminal any folder to build it. it took 27 tries but it did work... still.. 27.. fuck sake 30 a day is the soft cap.. retarded.
>Hey the stuff you do on your MacBook Pro? you can do it here too! With the help of AI... And maybe a little gaming with select titles here and there but idk about that bwo :3I was enthusiastic at first but now I'm struggling to find a usecase.
Also btw the existence of these chips will inadvertently enshittify Windows.
>>108954920Here you go my good retard.
>>108954865This.Did everyone forget that they already did this in desktop form?The GPU is too slow, the ram most importantly is way too slow, it can't handle 128gb models.>>108954873And it's been available for cheap forever. I had 128GB ram before chatGPT came out.Running large models on this thing is very slow, just like running it on CPU. You will be waiting minutes for your chatbot to write a response.
>>108954936The only practice use case I see with this shit would by like 4 dgx sparks working in tandem with a harness on a ... 90 bil param model loaded into each maybe? 1 mil token space for each, a collaborative whitespace for them with a project dump output folder, 1 worker monitors the other 3 retards while 2 code and a third checks their work.. when space nears full capacity it opens a new session, links back to where they left off and they slowly plod through coding, deploying / testing watching crashlogs and looping while you play vidya on a deck or thor or touch grass or sleep or some shit.. just leave rust desk or ssh on your collab rig and.. remote in with your phone periodically to do next thing when they... agonizingly slowly succeed eventually?
