NVIDIA

The one to get its new chip for AI with the Nvidia market in China

by

Until just a few months ago Nvidia monopolized some more than 90% of the Chinese chip market for artificial intelligence (AI), but after the entry into force of the last US sanctions package its leadership with all probability is being committed. The Chinese government is allocating a lot of resources to the development of Your own lithography teamswhich are those used to manufacture integrated circuits, and also to the tuning of their own Vanguard chips for artificial intelligence. As we told you last week, the US Department of Commerce It has imposed restrictions to the export to China of The H20 GPUand this in practice means that this chip presumably will not reach the Chinese clients of Nvidia. The company led by Jensen Huang is already paying it. His shares have fallen 6% and Nvidia has announced that this prohibition will cause a hole in its accounts of 5.5 billion dollars due to the commitments linked to the H20 GPU that the reserves of this chip had already acquired that it will finally not be satisfied. Huawei’s Ascend 920 GPU is ready to occupy the hole left by the H20 chip Some of the Chinese companies that have bought large amounts from the H20 Chip to NVIDIA and who presumably planned to continue doing them are Tencent, Alibaba or Bytedonce. The interesting thing is that this situation puts Huawei in a tray the opportunity to increase its market share in its own market taking advantage of the fact that US government prohibitions are weakening Nvidia’s position. However, this Chinese company is doing very well in this market because invoices annually about 7,000 million dollars Only in China. Huawei invoices about 7,000 million dollars only in China Huawei has lists its own GPU for iathe chips ascend AI, for more than five years. During this period of time it has been refining them and increasing their abilities with the purpose of matching or even overcome performance of the chips A100 and H100 of Nvidia. According to some analystslike those of the Chinese company Ifly Tek, the gross power of its GPU equals that of the Nvidia chips, but they are still one step behind if we stick to its performance in a real -use scenario. In any case, Huawei was prepared to react to the regulation that prevents Nvidia from delivering its most successful GPU (in China) to its Chinese clients. And it is that only one day after the US Department of Commerce formalized its latest sanctions He has presented his GPU Ascend 920a chip for AI that is clearly intended to occupy in the Chinese market the gaps that the NVIDIA H20 GPU is going to leave. The GPU Ascend 920 will begin to be manufactured on a large scale during the second half of 2025 using 6 Nm integration technology that presumably have developed side with huawei elbow and SMIC. The characteristics of this lithographic node have not yet been officially confirmed, but it will probably use the technique known as Multiple patterningwhich is the same one that SMIC is using 7 Nm chips. In addition, the GPU Ascend 920 will reach a 4 TB/s transfer speed for the memory subsystem thanks to the use of HBM3 chips. Image | Huawei More information | Digitimes Asia In Xataka | The Nvidia pulse and US administration becomes more virulent. The B20 GPUs for danger

It already has an alternative to Nvidia Cud

by

Nvidia’s greatest strength is not her hardware; it is CUDA (Compute Unified Device Architecture). And is that most of the projects of artificial intelligence (AI) that are currently being developed are implemented on this tool. This technology brings together the compiler and development profits used by programmers to develop their software for NVIDIA GPUs, and replace it with another option in the projects that are already underway it is a problem. Huawei, who aspires to an important portion From this market in China, it has Cann (Compute Architecture for Neural Networks), which is its alternative to CUDA, but at the moment this last software dominates the market. And at the current situation it is a problem for the Chinese companies they want, as proposed by the Government led by Xi Jinping, stop using US hardware and software. For China to have an effective alternative to CUDA is crucial right now, and it seems that it is already ready. Moore Threads is China’s Nvidia Li Guojie, a computer scientist from the Chinese Academy of Sciences that is considered an authority in China, holds that the most advanced Chinese GPUs, such as chips Ascend 910 from HuaweiThey are comparable to the current NVIDIA solutions in terms of calculation capacity. However, this expert also defends that in the current circumstances for this Asian country it is crucial to break down all the barriers imposed by CUDA. “China must develop an alternative system to achieve self -sufficiency in AI (…) Deepseek has had an impact on the Cud that exceed Cuda“, Guojie says. Perhaps Cann manages to break down little by little and manages to impose himself finally, but while he persists to some extent the development of AI in China will continue to depend on the US. “We need to establish a set of software tool systems for controllable that exceed CUDA” However, Cann is not the only trick of the country governed by Xi Jinping. Moore Threads It is one of the Chinese companies that are dedicated to the production of hardware for which companies aligned with the interests of the US and its allies cannot sell software or advanced equipment. Although it is very young (it was founded in 2020) it has something very important in its favor: its founder is Zhang Jianzhong, former general manager of the Nvidia subsidiary in China, so it is evident that he knows well what he has in hand. Moore Threads has developed several GPU for AI applications that, on paper, rival some of the advanced solutions that have placed in the Nvidia, AMD or Huawei market. The cards MTT S4000 and MTT S3000 They are its most interesting proposals right now, although, curiously, in its porpholio the MTT S80 card also appears, a proposal for games and content creation that, according to Moore Threads itself, has a calculation capacity of 14.4 Tflops in single -precision floating coma operations. It doesn’t impress, but it’s not bad at all. However, this company has something else: a software package with which the domain of CUDA seeks to break. He calls it MUSEis compatible with the range of MTT cards that I have mentioned a few lines above and incorporates a compiler, execution libraries, specialized libraries and code purification tools. However, this is not all. On paper its most attractive capacity for China is that it allows Reuse the code written in CUDAmoving it so that it can be executed About Moore Threads’ cards. It is difficult to predict what reception the hardware and software of this company will have in your country of origin, but there is no doubt that it is worth following the track. Image | Moore Threads More information | Moore Threads In Xataka | AI is the best thing that is happening to nuclear fusion. It is already accelerating the construction of Iter

The US suspects that Nvidia chips are arriving in China through Malaysia: it has decided to take action on the matter

by

The United States and China fight an increasingly aggressive commercial war. In this pulse, both have imposed export controls to protect strategic sectors. Washington focuses on the most advanced chips, While Beijin responds with critical minerals restrictions. They seem firm measures, but everything indicates that they are not being fulfilled to the letter. Chinese is avoiding restrictions. At the beginning of last year we learned that the popular liberation army He had managed to do with the most powerful NVIDIA GPUs, among which were the GPU A100 and H100. This was particularly relevant because the export of these products is prohibited by the US Department of Commerce. {“Videid”: “X8WLH9Q”, “Autoplay”: False, “Title”: “United States vs. China: The chips war”, “Tag”: “Webedia-prod”, “Duration”: “1611”} And they were not only the Chinese armed forces: also universities and research centers controlled by the government were using prohibited products. Washington believes that this has been possible by different ways, but concluded that the main channel were intermediary countries that collaborate with the Asian giant. First Singapore, now Malaysia. As The Economist points out, Singapore was one Of the countries that raised suspicions, simply because the figures did not square. In the last quarter of 2023, Nvidia multiplied by five shipments to customers in Singapore compared to the same period of the previous year, which suggested a possible detour to Chinese users. HGX H200, one of Nvidia’s most advanced products Now the focus is in Malaysia. According to Financial TimesThe United States suspects that many of the Nvidia chips enter the country end up in Chinese hands, avoiding current commercial restrictions. Given this scenario, Washington has begun to press the Malaysian government to control the trail of these latest generation chips. Tracking shipments is not so easy. The Minister of Commerce, Tengku Zafrul Aziz, has taken note of the requirement of the US and, he explains, has formed an interministerial working group to collaborate. However, he warns, that tracing chips shipments along the supply chain is not as simple as it seems and that it is a broad effort. Malaysia has become One of the great world epicenters of data centers, which explains the massive arrival of chips for the facilities that support companies such as Microsoft or Bytedance, the Tiktok matrix. According to Aziz, Washington is also promoting internal measures to reinforce control over the supply chain. In Xataka The general director of AMD is in China with one purpose: to snatch the AI ​​market to NVIDIA Waiting for results. For now, it remains to be seen if the pressure of the North American country will take effect. Malaysia has reasons to cooperate: a commercial retaliation could put its flourishing data centers at risk. Fulfill could simply be a way to protect your strategic position on the global technological map. The US does not want to give the arm to twist. The United States is doing everything possible to limit China’s access to avant -garde chips, mainly because of the Dual use risk: civil technologies that can also be applied in the military field. The concern is that these advances end up reinforcing the defensive and offensive abilities of the Chinese army. Images | Nvidia + Photoshop In Xataka | China and Russia are squeezing better than anyone the Nvidia GPUs thanks to a material need: they are vetoed (Function () {Window._js_modules = Window._js_modules || {}; var headelement = document.getelegsbytagname (‘head’) (0); if (_js_modules.instagram) {var instagramscript = Document.Createlement (‘script’); }}) (); – The news The US suspects that Nvidia chips are arriving in China through Malaysia: it has decided to take action on the matter It was originally posted in Xataka by Javier Marquez .

After triumphing with its chips for AI, Nvidia has set another disruptive technology: quantum computers

by

Nvidia’s bet for Quantum computers It is less and less shy. Jensen Huang, the co -founder and general director of this company, has announced A few hours ago at its annual developer conference that will open a laboratory expressly dedicated to Quantum computing research. It will be housed in Boston (Massachusetts) and will allow NVIDIA engineers to work side by side with the researchers at Harvard University and the Massachusetts Technology Institute (MIT). It will begin operating at the end of 2025. This strategic movement puts on the table with absolute clarity that Huang does not want to stay out of technology that will presumably cause a medium -term disruption. The most curious thing is that before formalizing the implementation of its new quantum technologies development laboratory, this executive has not let out the opportunity to retract. At the beginning of last January A few statements They caused a very abrupt fall of the actions of some of the companies that are dedicated to the development of quantum computers. “If you said 15 years you would probably be optimistic. And if you said 30 you would be pessimistic. But if you opt for 20 years I think many of us would believe it,” Jensen Huang argued At that time. With this reflection I tried to predict when the really useful quantum machines will be ready, and, therefore, capable of dealing with a very wide range of problems. But he has changed his mind. Just two and a half months later seems to be convinced that fully functional quantum computers will be ready much earlier. With the correction of errors of quantum computers in the spotlight Nvidia flirting with quantum computers is not really new. And is that He has been collaborating for more than two years With the Israeli company Quantum Machines. This company specialized in the development of hardware and software for quantum machines, and has been ready with NVIDIA a low -performance and high performance architecture that seeks to promote the progress of quantum computing. DGX quantum seeks to help researchers who work in the field of quantum computing to develop new quantum algorithms NVIDIA has contributed its CPU/GPU grace hopper system, a beast that is designed to execute applications of artificial intelligence and offer productivity at high performance computer scenarios, and also its open source programming model CUDA QUANTUM. His partner in this project, Quantum Machines, has been in charge of the integration and set -up of a quantum platform that, according to these two companies, is specifically designed to work in hybrid systems in which classical hardware and quantum coexist in harmony. The purpose of the DGX Quantum platform, which is what is called the hardware that these two companies have developed, is to help researchers who work in the field of quantum computing to develop new quantum algorithms. It may seem surprising that it is possible to use classic hardware to develop quantum algorithms, but it is something perfectly viable. In fact, this strategy helps to put quantum computing within the reach of many more researchers who can implement and test their ideas without having access to a quantum computer prototype. However, the DGX quantum platform also serves, according to NVIDIA, to calibrate quantum systems, control them, and even aspires to have a prominent role in the tuning of a correction system that allows quantum computers amend your own mistakes. Jensen Huang emphasized this idea during his GTC 2023 conference, and there is no doubt that It is a very attractive possibility. Extraordinarily attractive. And is that, As Ignacio Cirac explained to us In the conversation we had with him, the correction of errors will give us the opportunity to solve with quantum computers really significant problems. Image | Nvidia More information | Reuters | SCMP In Xataka | Quantum computers find it impossible to do nothing. It is a mystery that has scientists on alert

snatch the AI ​​market to Nvidia

by

Nvidia leads the GPU market to artificial intelligence (AI) with a shocking forcefulness, especially if we stick to training the great language models. According to the consultant Jon Pedie Research The company led by Jensen Huang controlled 90% of the market During the third quarter of 2024, and presumably their numbers will not have changed much since then. AMD reached a 10%quota during that period, so the presence of Intel in this sector is anecdotal. The sanctions that it has deployed The US government During the last two and a half years they prevent Nvidia from selling its GPU for the most advanced, such as chips H100, A100, H800 or A800. However, the GPU that at the time the US Department of Commerce does allow you to sell its Chinese clients is the H20 cut chip. And he is having a lot of success. In fact, its sales in the country led by Xi Jinping grow 50% quarter to quarter despite the delicate current situation. AMD wants a cake portion in China The Chinese government wants its technology companies to stop buying integrated circuits to NVIDIA. The step he took at the beginning of last October was merely informative, but, in any case, It was expressly addressed to companies that acquire GPU for AI with the purpose of recommending that they buy chips produced in China, and not the solutions of the Jensen Huang company. The current situation does not seem ideal for another American company to try to grow in the Chinese market Nvidia is losing competitiveness in China despite the good sales of her H20 chip. It is inevitable. And this GPU on paper is much less capable than The most sophisticated chips for the that Jensen Huang’s firm is currently selling. At the moment the Ministry of Industry and Information Technology of China has not officialized anything, but the administration led by Xi Jinping has restricted since last August the purchase of NVIDIA H20 GPUs by Chinese companies. On the other hand, Huawei has its own GPU for ia, The chips ascend tofor more than five years. During this period of time it has been refining them and increasing its capacities with the purpose of matching or even overcoming the performance of the A100 and H100 chips of Nvidia. This situation does not seem ideal for another American company to try to grow in the Chinese market, but this panorama does not seem to have discouraged Lisa her, the general director of AMD. And this executive is now in China. In fact, according to SCMP He has begun his visit by highlighting the role that his company is playing in the development of AI models of the country’s main actors, among which are Deepseek and Alibaba. Smooth his has remarked that THE INSTICT GPU of AMD are compatible with the models of AI of these last two companies in a clear attempt to Defend your position in the Chinese market. Besides, has pointed out which has observed a perceptible increase in the performance of Depseek models thanks to AMD hardware optimization. This statement is quite surprising if we take into account that Depseek uses Nvidia chips in training and huawei in inference. Whatever it will be interesting to verify how AMD hardware evolves for the Chinese market during the next months. Image | AMD More information | SCMP In Xataka | The Nvidia pulse and US administration becomes more virulent. The B20 GPUs for danger

The B300 GPU is the new Nvidia beast for Ia. And we already know what prepares for 2026 and 2027

by

Jensen Huang, the co -founder and general director of Nvidia, has not let out the opportunity to publicize the next GPU for artificial intelligence (AI) that have put their engineers ready in the framework of the GTC 2025 (GPU Technology Conference). The spectacular thing this electrical engineer has presented is The DGX B300 platform. This hardware is the most powerful Nvidia for generative, although according to this company it is also its most efficient proposal from an energy point of view. The Blackwell Ultra GPUs work on the B300 platform elbow with a 2.3 TB Map of HBM3E memory, delivering according to NVIDIA 72 Pflops in training processes with precision FP8 and nothing less than 144 pflops in inference tasks with precision FP4. These figures are a real monstrosity. In fact, the B300 platform is 11 times faster in inference and 4 times in training than its predecessor, the B200. This is the hardware with which Nvidia wants to maintain her leadership If we look at the consumer figures announced by NVIDIA we will see that apparently the energy efficiency of B200 and B300 platforms is similar. The first consumes approximately 14.3 KW maximumand the second one 14 kW. However, there is something that we should not overlook: the GPUs of both solutions have been implemented on the Blackwell microarchitecture, but they are not the same. The Blackwell Ultra chips of the B300 platform are more powerful than the Blackwell to dry infrastructure B200. The B300 platform integrates 50% more memory, allowing you to deal with larger AI models In addition, the B300 platform integrates 50% more memory, which in theory allows this hardware to deal with larger and more parameters. This proposal will reach the first data centers During the second semester of 2025. In any case, Nvidia has not only spoken of her current hardware in this edition of her conference dedicated to AI; He has also anticipated what his engineers are working for 2026 and 2027. The microarchitecture that will replace Blackwell is known as Rubin, and, as expected, it will be even more powerful than his predecessor. An interesting detail is that Rubin will be compatible with Blackwell at the infrastructure level, which will allow Nvidia customers to combine both solutions. In any case, Rubin will deliver 1.2 EXAFLOPS in training processes with precision FP8 compared to 0.36 EXAFLOPS of the B300 platform. It will arrive during the second half of 2026. And during the second semester of 2027 Nvidia will launch Rubin Ultra, a review that according to this company will reach 5 exaflops In training tasks with FP8 precision, so your performance in this scenario will be almost four times greater than Rubin’s. A last interesting note: Rubin will use HBM4 memory, while Rubin Ultra will have HBM4E. Image | Nvidia More information | Nvidia In Xataka | AI is already our best ally to solve the mathematical problems that seem impossible

The future will be full of robots, says Nvidia. And just launch a tools to win that race

by

The event GTC 2025 Nvidia began yesterday with an inaugural talk in which Jensen Huang machine -gunned us with a real Building burst. Among them, especially those focused on an area in which the company is especially optimistic: robotics. Foundational model. In March 2024 NVIDIA He already told us about his Project Grootand now said project has crystallized in the call Isaac Gr00t N1which qualify as the first open foundational model for humanoid robots. That is to say: the model is pretended, but it can be customized for all types of scenarios, both “domestic” and industrial. “The era of generalist robotics is here”. Huang declared how this new era began with this launch, and pointed out that thanks to Isaac Gr00t N1 and new data generation systems and learning for robots, developers can create much more capable robots. Robotic labor. The CEO of Nvidia highlighted how “it is very likely that this is the largest industry of all. At the end of this decade, at least 50 million workers will miss in the world. We would be more delighted to pay $ 50,000 to each one to come to work; we will probably have to pay the robots 50,000 dollars a year (instead).” Humanoid robots in sight. In The last part of your presentation Two and a half hours, the Nvidia CEO focused precisely on robotics, and for example showed videos of humanoid robots Neo gamma of the company 1x, which are based on Isaac Gr00t N1 and that already show their Potential as domestic robots. Other companies like Boston Dynamics –Creators of Atlas-, Quesoe Robotics and Neura Robotics are also developing this type of robots and have enjoyed preliminary access to the Nvidia robotic platform. Think quickly, think slowly. This foundational model uses a double architecture inspired by human cognition. Thus, system 1 is “a fast thinking action model” similar to what humans have with reflexes and intuition. System 2, enhanced by a vision model, is “a model of slow thought” that reasons about its environment and the instructions he receives and then raised his actions. Learning to move. This model therefore raises a pillar to develop robots (humanoids or not) that can operate in the real world. Thus, GR00T N1 can generalize and adapt to common tasks such as taking objects, moving them with one or two arms or carrying them from one place to another. Basic and limited operations are at the moment, but they are an important first step for the robots of the future. Basically these robots are learning to walk. A customizable model. Nvidia offers large information so that developers can start working with this model. Thus, we have both the Technical information Like project files available in Github. And data to train robots. Robots need huge amounts of data to be trained and learn to interact in the physical world. Nvidia ACBA to offer A huge and available data set in Hugging Face for this area. They are 15 TB of data that represent “more than 320,000 trajectories for robotic training”, in addition to 1,000 “scenarios” Scene Description (Openusd) to contribute to that training. Collaboration with Disney and Deepmind. Another of the robotic projects is the alliance with these companies to develop Newtonan Open Source physics engine that allows robots to learn how to manage tasks of all kinds with greater precision. Disney will be one of the first to use these robots, and in fact Huang finished the demo with a unique co -presenter: a robot that he called Blue and that was inspired by the BDX droids of Star Wars. It is something that Disney already pointed out years agoand it is likely that this type of entertainment robots will end in their attraction parks. Image | Nvidia In Xataka | We believed that Atlas was the pinnacle of modern robotics. Boston Dynamics has just killed the famous hydraulic robot

Nvidia tries to placate the criticism of the distribution of the GeForce RTX 50. Your answer is not entirely convincing

by

Several weeks after its launch it is still very difficult to find in the stores a family graphics card GeForce RTX 50 with a price aligned with its official cost. The easiest model to get with a price Similar to the marked by Nvidia It is the GeForce RTX 5070but the others or have an important extra cost or are not available. In this scenario it is understandable that users who have decided to get one of these graphics cards are disappointed. Many of you have clearly expressed it in The comments of our articlesand forums dedicated to PC hardware are full of users They complain due to the low availability of the new NVIDIA graphics cards. Given the circumstances, the company led by Jensen Huang has been forced to face with the purpose of appeasing players and trying to explain what is happening. This is Nvidia’s response As we can see in the graphics that we publish on the cover of this article, Nvidia says that during the first five weeks after the launch has distributed twice the graphics cards of the GeForce RTX 50 family that on its day of the range GeForce RTX 40. However, this explanation deserves a calm analysis. And, as they point out in Tom’s hardware and Redditthis comparison does not scrupulously reflect what is happening. During the first five weeks of the launch of the GeForce RTX 40, only the RTX 4090 was available Jarred Walton, one of Tom’s hardware editors, explains very well: During the first five weeks after the launch of the GeForce RTX 40, only the RTX 4090. This graphics card arrived on October 12, 2022, and the RTX 4080 landed on November 16, just five weeks later. All users know that RTX 4090 is not an adequate graphic solution for a wide range of players or for its price or for their benefits. However, during the first five weeks after the launch of the GeForce RTX 50 family, the four graphics cards that are currently available: the RTX 5090, the RTX 5080, the RTX 5070 TI and the RTX 5070 arrived. The first two landed together on January 30, 2025, while the GeForce RTX 5070 TI arrived on February 20. Finally, the GeForce RTX 5070 appeared almost at the end of this period: on March 5. If we have all this present it is reasonable to conclude that the comparison made by Nvidia It is not balanced. Not even accepting the possibility that this company has added the RTX 4080 to the figure of the GeForce RTX 40 family. Hopefully both availability and prices are normalized as soon as possible for the good of users. Image | Nvidia More information | Tom’s hardware In Xataka | If Nvidia lived only from PC GPUs, she would be about to die of success. And US tariffs don’t help

AMD and NVIDIA are promoting unreal prices for their new graphics cards. They end up costing much more

by

The same thing happens to each new generation of graphics cards. We excite ourselves thinking that we will finally have an opportunity to update our PC for a reasonable amount, and then the tortazo arrives. Because what manufacturers promise to which a world actually happens. On the one hand, Nvidia. Nvidia recently launched its RTX 5070 at a recommended price of 659 euros. Nvidia’s own edition, which had that price, It has been exhausted from the zero minute And it doesn’t even seem to be available in stores. The RTX 5070 TI, which has a recommended price of 899 euros, is possible find it at that pricebut is exhausted right now. Those that are available are more expensive versions. On the other, AMD. With the new Radeon RX 9070 / XT, three quarters of the same. The cards have gone to the market with recommended prices of 549 and 599 dollars respectively. There are not even PVPS in euros, but it doesn’t matter, because just being available in stores, its price is much higher. The most affordable we have found, a Gigabyte RX 9070, is 759.99 euros. If we want an RX 9070 XT, the most affordable model we have seen, 835 eurosIt is not available. Yes we have found one from Sapphire to 1,087.99 euros. It is a lot of difference with the prices recommended by AMD. But are they improved versions, right? The manufacturers take the reference designs of AMD and NVIDIA and try to raise additional improvements. For example, they preconfigure them with some overclocking to be something better in performance, they place additional cooling systems or add other elements, but the difference seems too high with respect to the recommended prices. Only a few at recommended prices. In Videocardz They have managed to talk to INET.se, a chain of stores and stores that sell these cards. Those responsible indicate that the recommended prices “will be applied to a limited number of cards.” In fact, they stand out, “they will only apply to the first shipment of each model.” Thus, it is not only that a few cards come out at those announced prices – which are exhausted right away – but it is very likely that such prices will not be repeated again, and go up depending on who sells them. AMD tries to soften the situation. In The Verge They have managed to contact AMD responsible, who neither confirmed or denied those allegations. What they said is the following: “It is inaccit that the recommended PVPs of 549 /599 dollars are launch prices. We hope that the cards are available in several suppliers to 549/599 dollars (excluded the tariffs and / or specific taxes of each region) based on the work we have done with our AIB partners (ADD-IN-IN-IN BOARD AIBs have different premium configurations at higher prices, which will also continue to exist. “ Deceptive advertising? The technique used by AMD and Nvidia is really worrying. Although there are users who in social networks claim to have found cards at recommended prices, many others point out how these models are exhausted and the only options are theoretically improved versions that manufacturers put at much higher prices. There is also a lot of resale and speculation, something that has been lived other times in the past. A lot of availability … if you buy an entire PC. Many of these stores offer these graphics without problems, but only if you buy them as part of a new gaming equipment. There it is more difficult to assess whether the price of the graph is really closer to the recommended or not. What is happening? The truth is that the situation is terrible for consumers, who can hardly access those graphs at the prices promised by AMD and NVIDIA. In Xataka we have contacted pccomponent, one of the main chains of sale of these products, to know if they can clarify the situation, but for now we have not obtained an answer. We will expand this article if we receive it. In Xataka | Nvidia has beat its income record again. Paradoxically, Wall Street does not seem to import much

China urgently needs an alternative to Cuda de Nvidia

by

Chinese companies that are dedicated to artificial intelligence (AI) They already have at their disposal a wide range of GPUs designed and manufactured in China. Huawei is the best positioned company thanks to its Chips Ascend 910 family, but in the country led by Xi Jinping There are many other specialized companies in the tuning of hardware for which they also have great potential. Metax, Biren Technology, Moore ThreadsINNOSILICON, ZHAOXIN, ILUVATAR COREX, DEGLINAI OR VAST AI TECH ARE SOME OF THE MOST IMPORTANT. “China is dedicating mass resources to the implementation of emerging companies specialized in the development of GPU. Do not underestimate them.” This warning was probably not ignored. He was aimed at the US government and came from someone who knows well what he says: Jensen Huang. The general director of Nvidia He pronounced these words Last year, during the celebration of Computex, and it is evident that its intention was to prevent US administration about the consequences that They would have the sanctions that seek to stop the technological development of China. Hardware is no longer the problem; It is CUDA Li Guojie, a computer scientist from the Chinese Academy of Sciences that is considered an authority in China, has confirmed What we have just seen: GPUs for the most advanced Chinese, such as Huawei Ascend 910 chips that I mentioned above lines are comparable to current NVIDIA solutions in terms of calculation capacity. However, the authentic strength of the company led by Jensen Huang is not its hardware; is its ecosystem CUDA (Compute Unified Device Architecture). “Deepseek has had an impact on the CUDA ecosystem, but has not completely overcome it because barriers persist” Most of the AI ​​projects that are currently being developed are implemented on CUDA. This technology brings together the compiler and development tools used by programmers to develop their software for NVIDIA GPUs, and replace it with another option in the projects that are already underway It is a problem. Huawei, who aspires to an important portion From this market in China, it has Cann (Compute Architecture for Neural Networks), which is its alternative to CUDA, but for the moment CUDA dominates the market. The most surprising thing is that not only Chinese companies need to part with this software; The entire AI industry wants to end Cuda. This is at least what Pat Gelsinger saidthe former director general of Intel. In December 2023, this executive wet and explained what was the official position of his company in the context of the AI ​​sector. “The whole industry is determined to eliminate market CUDA (…) We see it as a shallow and small pit (…)”, Gelsiter defended in the framework of the event “Ai EveryWhere” held in New York. According to Li Guojie“China must develop an alternative system to achieve self -sufficiency in AI (…) Deepseek has had an impact on the CUDA ecosystem, but it has not completely overcome it because barriers persist. In the long term we need to establish a set of software tool systems for controllable that exceed Cuda“. This is undoubtedly one of the great challenges that China faces in this area. Perhaps Cann manages to break down gradually and manages to finally impose himself, but while he persists to some extent the development of AI in China will continue to depend on the US. Image | Moore Threads More information | SCMP In Xataka | We can forget an AI without hallucinations for now. The general director of Nvidia explains why

Log In

Forgot password?

Forgot password?

Enter your account data and we will send you a link to reset your password.

Your password reset link appears to be invalid or expired.

Log in

Privacy Policy

Add to Collection

No Collections

Here you'll find all collections you've created before.