[HN Gopher] Jensen Huang keynote at CES 2025 [video]
       ___________________________________________________________________
        
       Jensen Huang keynote at CES 2025 [video]
        
       Author : subharmonicon
       Score  : 100 points
       Date   : 2025-01-07 02:42 UTC (4 days ago)
        
 (HTM) web link (www.youtube.com)
 (TXT) w3m dump (www.youtube.com)
        
       | behnamoh wrote:
       | I liked this part:                   "one small step at a time,
       | and one giant leap, together."
       | 
       | I didn't like this part:                   5090 for $2000, about
       | $500 more than 4090 when it was announced.
       | 
       | They didn't mention VRAM amount though, and I doubt it's more
       | than 24GB. If Apple M4 Ultra gets close to 1.8 TB/s bandwidth of
       | 5090, it'll crush GeForce once and for all (and for good).
       | 
       | Also nitpick: the opening video said tokens are responsible for
       | all AI, but that only applies to a subset of AI models...
        
         | jsheard wrote:
         | It's 32GB, that was leaked well before the event.
         | 
         | > If Apple M4 Ultra gets close to 1.8 TB/s bandwidth of 5090
         | 
         | If past trends hold (Ultra = 2x Max) it'll be around 1.1 TB/s,
         | so closer to the 4090.
        
           | behnamoh wrote:
           | Jensen didn't talk about it. Maybe he knows it's
           | embarrassingly low. Everyone knows nvidia won't give us more
           | VRAM to avoid cannibalizing their enterprise products.
        
             | aseipp wrote:
             | The official specs for the 5090 have been out for days on
             | nvidia.com, and they explicitly state it's 32GB of GDDR7
             | with a 512-bit bus, for a total of 1.8TB/s of bandwidth.
        
               | behnamoh wrote:
               | 32GB is still not a lot.
        
               | BryanLegend wrote:
               | "640k should be enough for anybody"
        
         | toshinoriyagi wrote:
         | edit: 32GB is confirmed here https://www.nvidia.com/en-
         | us/geforce/graphics-cards/50-serie...
         | 
         | Supposedly, this image of a Inno3D 5090 box leaked, revealing
         | 32GB of VRAM. It seems like the 5090 will be more of a true
         | halo product, given the pricing of the other cards.
         | 
         | https://www.techpowerup.com/330538/first-nvidia-geforce-rtx-...
        
         | ksec wrote:
         | For what 5090 is offering with 32GB RAM I thought it is a
         | pretty decent price comparatively to 4090. I thought the whole
         | lineup is really well priced.
        
         | headcanon wrote:
         | Market price for a 4090 is ~$2700 today:
         | https://stockx.com/nvidia-founders-geforce-rtx-4090-24gb-gra...
         | 
         | When you have a retail price so far below "street" price, it
         | just makes it harder to obtain and scalpers take a bigger cut.
         | Raising the price to something more normal at least gives you
         | more of a chance at the big-box store.
        
           | heyoni wrote:
           | Or scalpers won't be dissuaded and street price for a 5090
           | will be $3200 or more. $1500 was already an insane price tag
           | for scalpers to pay but they did it anyways.
        
       | magicalhippo wrote:
       | AMD barely mentioned their next-gen GPUs, NVIDIA came out
       | swinging right from the start. AMD announced two new models which
       | by their own cryptic slide wouldn't even compete with their
       | current top-end. Then NVIDIA came and announced a
       | 4090-performance GPU for $549...
       | 
       | If that's not just hot hair from NVIDIA, I totally get the
       | business decision from AMD but man, would love some more
       | competition in the higher end.
        
         | behnamoh wrote:
         | AMD has strategically decided to abandon the enthusiasts market
         | [0]. Really sad that nvidia's monopoly has got solidified even
         | further. That's one of the reasons I contribute to Apple's MLX
         | instead of CUDA.
         | 
         | 0: https://www.techradar.com/computing/gpu/amd-announces-new-
         | ra...
         | 
         | Update: typical HN behavior. someone is downvoting all my
         | comments one by one...
        
           | jiggawatts wrote:
           | Someone with a lot of NVIDIA shares, no doubt...
        
             | blitzar wrote:
             | More likely someone holding the AMD bag
             | 
             | edit: confirmed.
        
           | ksec wrote:
           | >Update: typical HN behavior. someone is downvoting all my
           | comments one by one...
           | 
           | You cant downvote older comments or comments older than 1 / 2
           | days.
        
         | Permik wrote:
         | It depends totally how you define "hot air" in this instance.
         | Later in the keynote there's a mention how the $549 card
         | achieves the 4090-like rendering performance with DLSS, i.e.
         | not with raw graphics, number crunching horsepower.
         | 
         |  _Personally?_ It 's a no for me dawg, DLSS unfortunately
         | doesn't actually replace the need for the raw GPGPU crunch.
         | 
         |  _For the average layman and a consumer?_ Nvidia will be
         | selling those GPU 's like hotcakes.
        
           | y-c-o-m-b wrote:
           | I kind of agree with this. On the one hand when DLSS works
           | well, it's amazing. I didn't notice any issues at all playing
           | Senua's Sacrifice and Senua's Saga. They were both as good as
           | playing on everything maxed without DLSS. On the other hand
           | when using it on Jedi Survivor, it's a very janky experience.
           | You can clearly see artifacts and it trying to "self-correct"
           | when using DLSS vs without DLSS.
        
             | kcb wrote:
             | DLSS upscaling is good. On a 4k display I can't see any
             | visual reason not to run the Balanced preset. DLSS frame
             | gen on the other hand I'm much more skeptical of. Every
             | time I've tried it something just feels off.
        
               | heyoni wrote:
               | DLSS in cyberpunk has a lot of issues. Faces look weird,
               | tons of ghosting, hair and particle effects looking very
               | jagged.
        
               | josephg wrote:
               | They're claiming DLSS4 (only available on their new
               | cards) fixes a lot of these issues. But we'll have to
               | wait for the cards to get in the hands of reviewers
               | before we'll know for sure. I'm still pretty skeptical.
               | 
               | That said, if you read between the lines, it looks like
               | the new 5090 is about ~40% faster than the 4090 at
               | rasterisation. That's a solid inter generational
               | improvement.
        
         | nomel wrote:
         | > Then NVIDIA came and announced a 4090-performance GPU for
         | $549
         | 
         | Never trust vendor performance claims (these specifically rely
         | on 3x frame generation), and never assume cards will be
         | available at MSRP.
        
           | belter wrote:
           | Correct: "Debunking the CUDA Myth Towards GPU-Based AI
           | Systems" - https://arxiv.org/abs/2501.00210
        
         | ekianjo wrote:
         | > Then NVIDIA came and announced a 4090-performance GPU for
         | $549...
         | 
         | That's obviously a marketing hoax. Spray DLSS and other tricks
         | and you can make this kind of claims while the raw power is
         | clearly on the side of the 4090
        
         | uluyol wrote:
         | AMD is not ready to have a full announcement for their new
         | consumer GPUs: https://morethanmoore.substack.com/p/where-was-
         | rdna4-at-amds...
        
         | caseyy wrote:
         | 4090 (no frame generation) performance with a $549 card (4x
         | frame generation), isn't it?
         | 
         | In other words, the same performance at somewhere in the range
         | of 2x-3.5x lighter workload. In other words, vastly different
         | performance under same workload. It is veeery on the edge of
         | false advertisement for them to omit the information in
         | parentheses until later in the presentation. :)
         | 
         | That line should have been "4090 frame rates on budget hardware
         | thanks to DLSS 4" or something like that.
        
       | ksec wrote:
       | I actually got a lot more upbeat about the potential of AI after
       | watching this keynote more so than any other demonstration or
       | presentation.
       | 
       | Blackwell is still based on N4, a derivative of 5nm. We know we
       | will have N3 GPU next year, and they should be working with TSMC
       | on capacity planning for N2. Currently Blackwell is pretty much
       | limited by capacity of either TSMC, HBM or packaging. And unless
       | the AI hype dies off soon ( which I doubt will happen ), we
       | should still have at least another 5 years of these GPU
       | improvements. We will have another 5 - 10x performance increase.
       | 
       | And they have foreshadowed their PC entry with MediaTek
       | Partnership. ( I wonder why they dont just acquired Mediatek )
       | and may be even Smartphone or Tablet with Geforce GPU IP.
       | 
       | The future is exciting.
        
         | AlotOfReading wrote:
         | The same industry opposition that killed the ARM acquisition
         | would kill a mediatek acquisition.
        
           | satellite2 wrote:
           | The acquisition of ARM would have put Nvidia in the position
           | of being able to impose to competitors uncompetitive
           | restrictions on a quasi monopolistic market. Is it the case
           | with MediaTek?
        
             | AlotOfReading wrote:
             | Mediatek is the largest smartphone SoC manufacturer, so
             | yes. They're a supplier for Apple and the biggest
             | competitor to Qualcomm, the two companies that were the
             | most vocal in opposing the ARM acquisition.
        
         | brookst wrote:
         | My guess is mediatek's margins are much lower so an acquisition
         | would tank Nvidia's stock price by leading to lower returns.
         | That and/or not wanting the distraction of operating mediatek's
         | many businesses that aren't aligned to Nvidia core competence.
        
         | bloomingkales wrote:
         | There was something eerily epic about his assertions. To
         | suggest that they are _all_ of those companies is pretty wild.
         | 
         | AI is bewitching.
        
         | dehrmann wrote:
         | I was taking these sort of improvements for granted. They'll
         | certainly make ai cheaper, but most of the cost is up-front for
         | generative models, and the models don't seem to be getting
         | linearly better.
        
         | kcb wrote:
         | A major chip acquisition by Nvidia will probably never get
         | global regulatory approval. I guess the question is what does
         | Mediatek provide in the partnership. Nvidia is well versed in
         | developing ARM SoCs already.
        
           | bgnn wrote:
           | 5g, wifi, bluetooth etc. what makes a mobile processor monile
           | processor other than the CPU
        
             | knowitnone wrote:
             | yeah, those are just modules that anybody can just buy.
             | NVIDIA already makes jetsons which are basically SBCs with
             | decent GPU. so again, what does Mediatek provide?
        
         | VHRanger wrote:
         | How would shrinking the processor size help if the current GPU
         | generation is already completely bottlenecked by VRAM bandwidth
         | saturation?
         | 
         | We aren't seeing nearly the same gains on VRAM bandwidth as we
         | are on compute bandwidth
        
           | ffsm8 wrote:
           | Well, at least this generation almost doubles that bandwidth,
           | right?
        
       | contingencies wrote:
       | As someone who has gone pretty deep in to robotics over the last
       | 9 years I skipped right to the physical AI portion, and wasn't
       | impressed.
       | 
       | This has been stated on HN in most robotics threads, but the core
       | of what they show, once again, is content generation, a feature
       | largely looking for an application. The main application
       | discussed is training data synthesis. While there is value in
       | this for very specific use cases it's still lipstick ("look it
       | works! wow AI!") on a pig (ie. non-deterministic system being
       | placed in a critical operations process). This embodies one of
       | the most fallacious, generally unspoken assumptions in AI and
       | robotics today - that it is desirable to deal with the real world
       | in an unstructured manner using fuzzy, vendor-linked,
       | unauditable, shifting sand AI building blocks. This assumption
       | _can make sense_ for driving and other relatively uncontrolled
       | environments with immovable infrastructure and vast cultural,
       | capital and paradigm investments demanding complex multi-sensor
       | synthesis and rapid decision making based on environmental
       | context based on prior training, but it makes _very little sense_
       | for industrial, construction, agricultural, rural, etc.
       | Industrial is traditionally all about understanding the problem
       | or breaking it in to unit operations, design, fabricate and
       | _control the environment to optimize the process_ for each of
       | those in sequence, and thus lowering the cost and increasing the
       | throughput.
       | 
       | NVidia further wants us to believe we should buy three products
       | from them: an embedded system ("nano"), a general purpose robotic
       | system ("super") and something more computationally expensive for
       | simulation-type applications ("ultra"). They claim (with
       | apparently no need to proffer evidence whatsoever) that "all
       | robotics" companies need these "three computers". I've got news
       | for you: we don't, this is a fantasy, and limited if any value
       | add will result from what amounts to yet another amorphous
       | simulation, integration and modeling platform based on broken
       | vendor assumptions. Ask anyone experienced in industrial, they'll
       | agree. The industrial vendor space _is_ somewhat broken and rife
       | with all sorts of dodgy things that wouldn 't fly in other
       | sectors, but NVidia simply ain't gonna fix it with their current
       | take, which for me lands somewhere between wishful thinking and
       | downright duplicitous.
       | 
       | As for "digital twins", most industrial systems are much like
       | software systems: emergent, cobbled together from multiple poor
       | and broken individual implementations, sharing state across
       | disparate models, each based on poorly or undocumented design
       | assumptions. This means their view of self-state, or "digital
       | twin", is usually functionally fallacious. Where "digital twins"
       | can _truly_ add value is in areas like functional safety, where
       | if you design things correctly you avoid being mired in
       | potentially lethally disastrous emergent states from
       | interdependent subsystems that were not considered at subsystem
       | design, maintenance or upgrade time because a non-exhaustive,
       | insufficiently formal and deterministic approach was used in
       | system design and specification. This very real value however
       | hinges on the value being delivered _at design time, before
       | implementation_ , which means you're not going to be buying
       | 10,000 NVidia chips, but most likely zero.
       | 
       | So my 2c is the Physical AI portion is basically a poorly founded
       | forward-looking application sketch from what amounts to a
       | professional salesman in a shiny black crocodile jacket at a
       | purchased high-viz keynote. Perhaps the other segments had more
       | weight.
        
       ___________________________________________________________________
       (page generated 2025-01-11 23:00 UTC)