저자: Alex · 외국인
요약 보기
CES 2025에서 엔비디아의 Project DIGITS(개인용 AI 슈퍼컴퓨터, 약 $3,300), RTX 50 시리즈(5070이 4090급 성능을 $549에), DLSS 4 및 Omniverse+Cosmos 등 발표를 소개. 개인 AI 워크스테이션 수요 확대로 AI 시장 성장과 엔비디아 수혜를 강조. 특히 가격 인하로 AMD 게이밍 GPU 마진 압박 가능성을 언급하며 AMD 게이밍 사업 축소 가능성도 제기. 중간에 데이터 브로커 삭제 서비스 유료 광고 포함.
다듬은 글
the best way to find great long-term Investment huge growth happens when a company has the perfect product for a hot new market well I went to CES in Las Vegas to see exactly what Nvidia and AMD have been cooking up for 2025 and boy is there a ton to talk about sorry this video is a little late I got back from CES last week and I immediately got super sick I'm actually still a little under the weather but I'm too excited to share what I learned to keep waiting I'm also not here to waste your time this video is going to cover three big announcements that Nvidia made during their CES keynote and what those announcements could mean for AI stocks going forward specifically I'm going to talk about project digits and personal supercomputing I'll talk about the new RTX 50 Series gpus and affordable neuro rendering as well as generative AI That's grounded in reality thanks to Technologies like L lss4 there's a lot of technical jargon there so let me break down all three announcements for you real quick project digits is a tiny super computer that Jensen Wong showed off during nvidia's keynote at CES it costs around $33,000 and it can handle AI models of up to 200 billion parameters for comparison qualcomm's current aipc costs around $11,000 but can only handle models up to 13 billion parameters so nvidia's project digits performs about five times better per dollar which is a big leap in performance for developers and data scientists the new Blackwell GeForce RTX 50 Series Desktop gpus also offer insane performance per dollar nvidia's keynote focused more on performance but I think the real star of the show was the price since the RTX 5070 5070 TI and 580 are all cheap enough to put serious pricing pressure on amds upcoming Radeon RX 9000 series gpus at this point it almost doesn't matter whether Nvidia picks up any market share in PC gpus because these aggressive prices are going to squeeze the margins out of the market altogether on the flip side Nvidia talked about two sets of AI technologies that could lead to much higher margins overall dss4 and Nvidia Omniverse plus Cosmos in traditional Computing you get better Answers by spending more resources whether that means allocating more cores or more time but now these tools open up a choice of how much of the final answer to compute in the first place versus how much to generate with AI which changes the economics behind Computing in a big way all right now that you understand the basics we can get into a little more detail including what I learned from getting into the exclusive press Q&A session with Jensen after his keynote let's start with project digits none of this would be possible if not for uh this incredible project that we started uh about a decade ago inside the company was called project project digits deep learning GPU intelligence training system digits well before we launched it uh I shrunk at the dgx the reason why we built it was because we wanted to uh make it possible for researchers and startups to have an out-of-the-box AI supercomputer but now artificial intelligence is everywhere it's not just in researchers and and and startup Labs you know we want artificial intelligence is mentioned in the beginning of our talk this is now the new way of doing Computing this is the new way of doing software every software engineer every engineer every creative artist everybody who uses computers today as a tool will need a AI super computer everybody who uses a computer today will need an AI supercomputer at first I wrote this off as marketing speak but I'm actually interacting with AI more and more every day myself for example almost every tool I use to edit videos like descript Adobe Premiere Pro and Photoshop are adding new AI features to their workflows and on the research side Google Sheets and notion are baking in more AI as well so knowledge workers and content creators needing an AI supercomputer or at least a dedicated AI chip isn't as crazy as it might first sound anyway back to project digits everybody who uses computers today as a tool will need a AI supercomputer I just wish that djx1 was smaller and um you know so so um you know imagine ladies and gentlemen our this is nvidia's latest AI supercomputer it runs the entire Nvidia AI back all of Nvidia software runs on this dgx Cloud runs on this it's even a workstation if you like it to be and um it's based on a a super secret chip that we've been working on called GB 110 the smallest Grace Blackwell that we make this top secret chip uh we did in collaboration the CPU the grace CPU was a uh is built for NVIDIA in collaboration with mediatech uh they're the world's leading s so company and they worked with us to build this CPU this CPU so and connect it with chipto chip mvy link to the Blackwell GPU this little thing here is in full production uh we're expecting this computer to be available uh around May time frame all right so here's why project digits is a much bigger deal than I first thought at the CES press Q&A with Jensen he mentioned that some of the biggest unmet needs in Computing are in areas like software development and data science where professional users want to have their own local compute resources without worrying about the cloud or usage fees this has a few key benefits first doing everything on a local machine is way more secure when everything runs on one machine access can be controlled and there's much less risk of compromising company data that's what companies care about most and it's especially important when you're talking about proprietary company data used to train a custom AI model or customer data which always requires additional security and speaking of data I found out that dozens of online data Brokers were selling my personal data and if you've been getting more spam phone calls texts or emails lately they might be selling yours too that's why I joined delete me the sponsor of this video delete me is a hands-free subscription service that will remove your personal information from hundreds of online data Brokers you just sign up enter your information and let their experts get to work they reviewed over 44,000 listings for me so far after 7 days you'll get a detailed privacy report showing everything they've done I just got my eighth report and delet me found 400 New pieces of my personal information across 34 different data Brokers not just my name and address but my familes and my fiance's too and here's the best part they'll keep scanning these websites even after they remove my information I even get a privacy adviser if I want to talk to a real person or make a custom removal request so if you care about your data and your family's privacy you can get 20% off any consumer plan with my code symbol2 by going to join delet me.com sls20 or use my link in the description and a big thank you to delete me for supporting the channel and keeping my family's data safe all right security is just the start having a personal AI supercomputer also helps reduce latency and energy costs compared to building testing and running generative AI models on the cloud inside the box which is roughly the size of your hand is a Blackwell GPU and a 20 core Grace CPU that are connected together by Envy link and share a 128 GB pool of unified memory that means these chips don't need to copy data back and forth which is especially nice for AI workloads involving big data sets you can also connect two units together to double the compute capacity and handle models with up to 400 billion parameters ERS which helps with future proofing as AI models continue to grow in size and complexity on top of that project digits can connect to cloud and data center infrastructures running on nvidia's ecosystem so developers can build and fine-tune AI models locally and then deploy them to bigger environments without worrying about compatibility issues as a result I think project digits will pick up some real market share with AI researchers data scientists and developers but what I think most investors don't realize is what this could mean for the rest of the artificial intelligence Market which that what happens to that growth if AI researchers and developers are getting workstations five times more capable than what they had a year ago that can interface directly with nvidia's AI Cloud infrastructure I think that growth goes up across the board with Nvidia as the biggest AI beneficiary I'll follow up on this once project digits launches this may but while project digits opens up new markets for nvidia's Blackwell gpus the gForce RTX 50 Series is locking one down so let's talk about that next here it is our brand new GeForce RTX 50 Series Blackwell architecture the GPU is just a beast 92 billion transistors 4 4,000 tops four pedop flops of AI three times higher than the last generation Ada 380 Ray tracing teraflops so that we could for the pixels that we have to compute compute the most beautiful image you possibly can and of course 125 Shader teraflops there's actually two dual shaders one is for floating point one is for integer G7 memory from Micron 1.8 terabytes Per Second Twice the performance of our last generation and we now have the ability to intermix AI workloads with computer graphics workloads and one of the amazing things about this generation is the programmable Shader is also able to now process neuron networks right let me explain why this is important for investors at a high level programmable shaders being able to process neural networks means Developers can describe complex materials textures and lighting interactions as small neural networks instead of many individual layes which means they can essentially train and fine-tune certain parts of their Graphics data instead of having to manually code it all this also has a lot of benefits for developers ranging from much better and faster texture compression material processing and Ray tracing while using a lot less memory and for the end user that means more frames per second at higher resolutions you using lower power and don't forget there's much more to Graphics processing than just video games faster texture compression material processing and rate tracing also benefits physics simulations robotics training generating data to train self-driving cars and so on so this really is an upgrade to everything gpus get used for across the board but here's the crazy part so how does it compare well this is rtx4 490 well now with the Blackwell family RTX 570 4090 performance at $549 5070 490 performance $549 and here's the whole family starting from 5070 all the way up to 5090 5090 twice the performance of a 4090 of course we're producing a very large scale availability starting January I'm not so sure the RTX 70 will actually have the same performance as the 4090 across the board but it almost doesn't matter at $549 well it might matter to AMD since nvidia's new RTX 5070 5070 TI and 80 all fall in the same price range as amd's current Radeon RX 7000 series gpus and I assume their upcoming RX 9000 Series gpus as well amd's gaming segment has been shrinking steadily for years now going from one of their biggest Revenue generators to their smallest segment by far like I mentioned at the start of the video I don't think it even matters whether Nvidia adds to their market share with their new 5000 series gpus what does matter is squeezing any remaining margins out of one of their only competitors at least in this market Nvidia also has a big Advantage by using the same Blackwell architecture for PCs and data centers while amd's radon gpus for PCs and Instinct gpus for data centers are actually very separate kinds of products if things keep trending the way they have been there could come a time where AMD has to give up their Radeon product line alt together and I'll follow up on this once AMD releases more information on their 9,000 Series gpus which was notably absent from CES as well as their earnings call in a couple of weeks and that's actually a really nice lead in to grounded generative AI with Technologies like dss4 and Nvidia Omniverse plus Cosmos let me explain them both and then call out the bigger Trend here dlss stands for deep learning super sampling and it's used to improve frame rates in video games by rendering a low resolution image and then upscaling it using AI the AI takes in information from the low resolution image as well as previous frames to generate new pixels effectively creating a higher resolution image for the cost of that lower resolution render D lss4 takes this to the next level by being able to predict up to three frames into the future so now for every one lower resolution frame that's actually rendered dlss 4 can display up to four high resolution frames by the way this is how the RTX 5070 can achieve the same performance as the 4090 but at onethird the price the RTX 5070 has dlss 4 which means it can generate additional frames that the 4090 has to actually render so it's not exactly an apple to Apple's comparison and we should wait for thirdparty benchmarks my point here here is that dlss 4 allows this computer to generate around 15 pixels for every one that it calculates which is a massive increase in output per watt or output per dollar another example is NVIDIA Omniverse plus Cosmos Omniverse is a gigantic simulator grounded in the laws of physics I cover it a lot on this channel because it's used to power platforms like Nvidia Drive Sim which simulates traffic patterns road conditions and the inputs and outputs of different car sensors to build scenarios for training and testing self-driving AIS or Nvidia Isaac Sim which is a similar idea but for robot manipulation and navigation but Jensen just announced something called Nvidia Cosmos at CES 2025 Cosmos is a world Foundation model kind of like how GPT 4 and llama are Foundation text and image models at a high level the world Foundation model just means that Cosmos speaks in physics you supply it with a simulation or a scene from Omniverse and it will output tons of photorealistic videos based on that scene so now developers can build a model of one busy intersection in Omniverse and generate a million videos of cars flowing through it using Cosmos or they can build a model of their Factory and generate a million videos of robots navigating through it game designers can build a 3D asset in Omniverse and then quickly generate millions of different skins or variations for it all grounded by the physical properties of the scene and objects that they built in Omniverse both dlss 4 and Cosmos are examples of a bigger Trend I think we're going to start seeing everywhere which is Computing only a small amount of ground truth data whether that's grounding words or pixels or frames or objects in an environment or proteins or motions and then generating the rest the more variance you're willing to allow the less truth points you have to compute the more strict you need to be with the final output the more more truth points and constraints you need to give the model upfront imagine being able to give an AI 10 key moments in a story and allowing it to generate the rest or telling a robot that it has to be in these five places at these five times every day but otherwise letting it generate its own path and now we've come full circle because that's a pretty big change from how we think about Computing today a change that might actually require graphics cards to handle more kinds of AI calculations in the future a change that slowly moves us from PCS and smartphones to AI PCS and smartphones that has nothing to do with chatbots or co-pilots a change that just maybe does justify every researcher data scientist and content creator getting a personal AI supercomputer or at least a dedicated AI chip in their existing one that's a future I'm excited to invest in and this thanks for hanging out until the end even though I summarized everything for you up front
Comments 0