NVIDIA

After achieving what seemed impossible, Nvidia sees his future in China threatened by something terrifying: the bureaucracy

by

What is happening to Nvidia with the GPU to artificial intelligence (AI) H20 It is a real odyssey. Currently this chip is its best asset to protect its position in the Chinese market, but at the current situation it is not clear that the company led by Jensen Huang go survive In this gigantic Asian country. Interestingly, the beginnings of this GPU in China were extraordinarily promising because Their sales grew by 50% quarter to quarter since it arrived in this market in mid -2024. However, everything was complicated for Nvidia in the middle of last April. And is that the US Department of Commerce imposed new restrictions To the export to China of the H20 GPU, which in practice caused this chip to stop reaching the Chinese clients of this company. This news Nvidia’s shares sank 6% in the bag because I could no longer attend the commitments linked to the H20 GPU that it had acquired. At the beginning of July there was another unexpected turn of events. Jensen Huang met with Donald Trump and got something that seemed impossible: the trade department would allow him Sell again in China the H20 chip. Since then four weeks have passed and Nvidia continues to wait. He has not yet received the export license you need to sell this GPU in China, and, According to ReutersThe problem is that the Commerce Department is mired in the bureaucracy, which has originated a delay in the concession of export licenses that has not occurred for more than 30 years. The future of Nvidia in China is in the hands of the Chinese government This delay comes at the worst time for Nvidia. Among Chinese clients who have bought great amounts of this GPU, and that presumably plan to continue doing itare Tencent, Alibaba or Bytedance. But if the Department of Commerce takes much more the delivery of the export license to NVIDIA these commercial operations could be canceled. And it is that Jensen Huang’s company has another very important open front. The CAC is responsible for the censorship and control of the contents published in the network As We explain to you last weekthe administration of the cyberspace of China, usually known as CAC for its English denomination (Cyberspace Administration of China), he has decided Thoroughly investigate the H20 GPU. This institution is the main Internet regulatory body in China and is responsible for the censorship and control of the contents published in the Network, the supervision of technology companies and compliance with the Data Security Law and the Personal Information Protection Law. The problem that Nvidia faces now is that The CAC has decided to investigate it Because he suspects that the H20 chip could incorporate a rear door of difficult location by Chinese experts. If so, the possibility of China to use this GPU could be possible. At the moment the CAC has limited himself to questioning those responsible for NVIDIA in China and ask them to demonstrate that the H20 Chip does not represent a threat to the interests and security of the country led by Xi Jinping. As expected, Nvidia has immediately responded to the Chinese authorities and is collaborating to dissipate as soon as possible the doubts that loom about the H20 chip. According to SCMPthose responsible for the company in China have assured CAC researchers that the GPUs for the develops They do not incorporate any “back door” implemented to facilitate espionage by the US government. “Cybersecurity is of vital importance for us”, has declared A NVIDIA executive. “We have no rear doors in our chips that can give someone remote access or the ability to control them.” Probably during the next few days we will know how this conflict ends. Image | Nvidia More information | Reuters In Xataka | The US gives Huawei a great opportunity: to get its new chip for AI with the Nvidia market in China

The US hardened their restrictions for Nvidia chips not to reach China. So they are sweeping the black market

by

A few days ago Nvidia got what It seemed impossible: That the US government allowed you to sell your H20 GPU very soon to Chinese clients. It was an important turn in the Trump administration policy, which since April had raised his restrictions. Despite the hardness of the sanctions, the plan has had lagoons, according to an investigation of the Financial Times. What happened. The Financial Times has accessed sales contracts, to presentations of companies already involved in the industry and has drawn a conclusion: three months after Trump harden the export control, chips worth $ 1 billion were sent to China. In context, it is a huge figure, if one takes into account that Nvidia entered 17,000 million last year. The process began in May according to the Financial Times, when Chinese distributors began marketing GPU on which restrictions on data centers that work with Chinese Laboratories of AI weigh. The investigation points to something that the US already suspected: Many chips enter from Southeast Asia They are not any chips. The investigation reveals that Chinese AI companies are managing to acquire the GPU B200 of Nvidia, a beast that It already has a successorbut that is a candy to train models. And it promised multiply by four The performance of the desired H100 in MLPERF 4.1. The investigation also mentions the sale of other chips on which export controls weigh, such as the aforementioned GPU H100 and its successor, the H200. The price of a Rack Of eight B200 ready to use is approximately $ 489,000, and has dropped in price since they arrived in China in May. The difference with the sale that is authorized? An extra 50%. Publication in social networks announcing the sale of ASUS H200 racks. Image: Financial Times Nvidia’s position. The company has maintained a defense position to remain in China these months, and it is that this market supposed 13% of its global income. Jensen Huang has been praising the Chinese models of AI At a complicated moment for the company, in need of strengthening its complicity among Chinese companies while just selling chips. In full Dilemma for the US to sell or not sellHuang has placed the power of Huawei’s solutions at the height of the H200. A path full of difficulties. As Nvidia told Financial Times, buying chips clandestine is not something that allows expected yields. Mounting a data center with the GPUs is not just a process of installing parts, but about giving them service and support, something that the company does not supply to chips sold outside authorized channels. According to an operator of a data center, the export control does not prevent NVIDIA chips from reaching China, and what it does is create inefficiencies and “huge profits for intermediaries that assume risks.” Even so, a distributor recognized that “there is no shortage.” The effect of relaxation with H20. With the announcement of the permission to the marketing of the H20 to China, the sales of the B200 and other chips marketed in the black market have fallen, according to several distributors. The reasons are not clear, but buying Nvidia guarantees the aforementioned support as well as a more competitive price. Despite this, there are Chinese distributors announcing stock of the B300, which is not yet manufactured in mass. Image | Nvidia and Flickr In Xataka | The US machinery to win the “war” of AI to China is already underway. And it goes faster than expected

Nvidia cannot sell her most powerful chips to China for sanctions. So you have found a plan B: Risc-V

by

Nvidia has announced that Its CUDA platform will be compatible with RISC-V processors. He has done so during the Risc-V Summit in China and the chosen place is not accidental: this announcement clearly points to the Chinese market. For the first time, the technology that allows applications to communicate with the NVIDIA GPUs will be extended beyond ARM and X86, towards an open source architecture. Why is it important. CUDA It is the software that operates the Nvidia’s ecosystem. Without a CUDA, the GPU would lose much of their parallel calculation apacities. That Nvidia opens this technology to RISC-V It means that processors based on this open architecture can now serve as the main CPU in NVIDIA GPU systems. The background. The announcement, in addition to making in China, comes while China is accelerating its efforts to reduce its dependence of western processors. Nvidia can’t sell your most powerful models GB200 and GB300 to China for US sanctions, so in this way finds a way to maintain relevant Cuda in the Chinese market. Between the lines. There is a lot of geopolitical strategy in this decision: NVIDIA has been integrating RISC-V nuclei for years into its own GPU for low-level control tasks. Now it makes the jump to support RISC-V as the main processor. And that responds to a reality: if China is going to develop its own processors using open architectures, Nvidia wants to be there from the beginning. In detail. The configuration shown by NVIDIA shows a heterogeneous system: The GPU handles parallel loads. The RISC-V processor executes the CUDA controllers and the application logic. And a DPU manages network tasks. This architecture allows you to orchestrate GPU computations completely within the CUDA environment, something impossible so far with RISC-V. Deepen. Historically, Nvidia has behaved Cuda to each important architecture: X86, ARM, PowerPC and even Sparc de Sun. The company understands that it must be present from the first day on any platform that can take off in the business sector. With a value already exceeding 4 billion dollarsNvidia can afford to bet on all promising architectures. And now the movement positions RISC-V as a viable alternative for future designs of AI processors and high performance computing. If the stars align, other manufacturers could follow the example of Nvidia. And that would accelerate the adoption of RISC-V in data centers beyond China. Outstanding image | Wikimedia Commons In Xataka | On his way to the authentic quantum supremacy, China has set an objective: a “real” quantum computer before 2030

Nvidia says that China has the best open source AI in the world. These praises have a very clear intention

by

Jensen Huang, the co -founder and general director of Nvidia, is in China. At the end of last week He met with Donald Trump with the purpose of defending the interests of his company before an administration that has drastically restricted your business In the country led by Xi Jinping. During the last fiscal year, which expired on January 26, 2025, China represented approximately 13% of total income of Nvidia with a figure of about 17,000 million dollars. In practice, this Asian nation is the third best client of this company only behind the US and Taiwan. The decisions made by Jensen Huang during the last months clearly reflect that he is determined to defend your presence in this market. And he is doing it. A few hours ago he has made a statement during his visit to the International Exhibition of the China Supply Chain, in Beijing, which is worth not overlooking. Jensen Huang does not give stitch without thread “The models of artificial intelligence (AI) Open source of Deepseek, Qwen and Kimi are the best reasoning models in the world today. They are very advanced. ” Huang has pronounced These words during a talk with Wang Jian, the founder of Alibaba Cloud, the AI and cloud computing unit of this gigantic Chinese company. And this statement has not arrived at any time. Whether or not, objectively, the three models mentioned by this Executive are very competitive, as the reports of Demodazzle either Mediumamong others. The really interesting thing is that Jensen Huang has praised these Chinese models at a very delicate time for Nvidia. At a time when You need to strengthen your complicity with Chinese companies to which they have been selling their GPU for several years. Last week Jensen Huang got a promise from Donald Trump: Nvidia can sell again Your H20 GPU To his Chinese clients very soon. The reception that this chip initially received was very good even though its capacities are clearly lower than those of the other proposals for this company. Jensen Huang has praised these Chinese models at a very delicate time for Nvidia. At a time when you need to strengthen your complicity In fact, initially the Department of Commerce allowed its sale in China because this integrated circuit met the restrictions it had imposed. And despite its limitations Their sales grew by 50% quarter to quarter since it arrived in this market in mid -2024. Everything was complicated for Nvidia in the middle of last April. And is that the US Department of Commerce imposed new restrictions To the export of the H20 GPU, which in practice caused this chip to stop reaching the Chinese clients of this company. During the next few weeks this GPU will come back to the facilities of Tencent, Alibaba or Bytedance, among other Chinese companies, but Nvidia will have it more difficult than ever. And is that Huawei, his main Chinese competitor, is being reinforced with great intensity. Only one day after the US Department of Commerce formalized its latest sanctions He presented his GPU Ascend 920a chip for AI that is clearly intended to occupy in the Chinese market the gaps left by the NVIDIA H20 GPU. 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 have presumably 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 | Nvidia More information | SCMP In Xataka | The US gives Huawei a great opportunity: to get its new chip for AI with the Nvidia market in China

The United States is convinced that the Chinese army already uses its chips for ia. Nvidia has a different opinion

by

The CEO of Nvidia, Jensen Huang, recently granted An interview to the CNN medium, where he could speak without capujos about the current context of his company, about Nvidia’s position in the AI racethe benefits of this technology, and other relevant topics of current affairs. In addition, he has also provided his opinion about The commercial war with China and restrictions, as well as the possible use of American technology in Chinese military terrain. Chips for the Chinese army. The United States has shown concerns in the past about whether advanced Nvidia chips would be used by the Chinese army. Those concerns gave rise to The restrictions of chips A100 and H100. Despite this, There are indications that the chips would also be reaching the Chinese army. Nvidia is not worried. Huang It has subtracted importance To Washington’s concerns about the military use of its artificial intelligence chips by China, arguing that the Chinese army cannot depend on American technology that can end up being restricted at any time. Their statements come while the company promises to resume sales of Its H20 processors to the Chinese market and days after meeting with President Donald Trump in the White House. Huang’s position against Washington. In the interview, the manager defended that “we do not have to worry” about the military use of American technology in China, since “they simply cannot trust it.” His argument is that Beijing will avoid depending on US components due to the risk of future limitations. Huang added that China already has “a lot of computing capacity” and does not need Nvidia chips to develop its military capabilities. The restriccions have hit Nvidia. US administrations have maintained export controls on the most advanced semiconductors for fear of strengthening Chinese military capacities. These measures, applied in a bipartisan manner, have cost Nvidia billions of dollars in potential revenues. According to the company, restrictions made them lose approximately about 15,000 million dollars in sales After the latest restrictions imposed by the Trump administration to prohibit their chips to China. The company affirms which also had to amortize 5,500 million in inventory. Between Washington and Beijing. It is clear that the company is making extraordinary efforts to maintain a balance between Washington and Beijing. His CEO openly criticizes control policies, arguing that They are counterproductive For American technological leadership. His thesis is that for the United States to maintain its dominant position in AI, American technology must be available globally, included in China, where half of the world’s developers are located. American senators They have warned him specifically that avoids meeting with companies linked to the Chinese army or intelligence organizations during your visit to Beijing. Meanwhile, Nvidia faces Huawei’s growing competition and other Chinese chip manufacturers, although the country’s technology companies continue to demand their processors due to their platform CUDA. And now what. Nvidia has announced that you will request licenses for Resume Sales of the H20 Chip to Chinawith the US government ensuring that it will grant them soon. The company has also developed a new processor specifically designed to comply with current regulations: the RTX Pro GPU, which is part of the architecture Blackwell and is completely adapted to export controls. China represents 13% of the total income of Nvidiaabout 17,000 million dollars, which explains why Huang maintains its favorable speech towards the Chinese market while sailing among the geopolitical pressures of both countries. The H20 chip had been developed specifically for the Chinese market after the restrictions of the late 2023, becoming the most powerful product legally available until its effective prohibition in April. Cover image | Nvidia In Xataka | The Nvidia’s supercomputer costs millions of dollars. And to work we use a switch with three kilometers of cable

Nvidia is the first company to reach 4 billion dollars of capitalization

by

Nvidia has achieved something historical: becoming the first company quoted to reach 4 billion dollars of stock market. During Wednesday’s session, the company’s shares They reached this symbolic milestonesurpassing both Apple and Microsoft. An ephemeral but significant achievement. Although Nvidia’s actions closed the day with A 1.8% riseplacing its valuation slightly below the 4 billion, the fact of having touched this figure during the session marks a historical moment in the markets. The company has managed to overcome giants like Apple, which began the year as The most valuable company in the world With 3.9 billion dollars, and Microsoft, which for months has exchanged positions with Nvidia as the ranking leader. The AI ​​revolution drives Nvidia. The meteoric rise of Nvidia It is explained by its central role in the boom of artificial intelligence. Their specialized chips feed on data centers that companies such as Microsoft, Amazon and Google need for their AI models and cloud services. This strategic positioning has catapulted its actions 22% so far this year, despite geopolitical turbulence and commercial restrictions with China. The numbers speak for themselves. In its last fiscal quarter, Nvidia generated $ 44,100 million in revenue, an increase of 69% compared to the same period of the previous year. The forecasts of the sector are even more optimistic: the global expenditure on the infrastructure of the 200,000 million dollars in 2028 is expected to exceed 2028, according to The International Data Corporation research firm. Obstacles on the road. Not everything has been a rose path for the company. Export restrictions of Your H20 chips China has cost about 8,000 million dollars in lost sales. In addition, the irruption of Deepseekthe Chinese startup that was planted out of nowhere with it developed from a powerful model and promising to have reached its milestone With ridiculous costs compared, caused doubts On whether the expensive Nvidia chips were really essential for the advance of AI, temporarily sinking your actions in January. Overflowing expectations. Despite these setbacks, Wall Street analysts maintain overflowing optimism. The Loop Capital firm esteem That Nvidia could reach a capitalization of 6 billion dollars in 2028, arguing that the company essentially maintains a monopoly in critical technology for the AI ​​sector. For its part, CEO Jensen Huang, who has become The tenth richest person in the worldaccording to the Bloomberg index, with a equity of 140,000 million dollars, See a promising future Ahead: “Of course, we know that AI is this incredible technology that will transform all industries, from the way we do the software to medical care and financial services to retail trade until, I suppose, all industries, transport, manufacturing … and we are at the beginning of that.” Cover image | Nvidia In Xataka | Depseek marked a turning point in OpenAi: now reinforces its safety while GPT -5 appears on the horizon, according to FT

The AI ​​has no future without nuclear energy when even Nvidia has begun to pray to Bill Gates reactors

by

Data centers will be responsible for 10% of the increase in energy demand until 2030, according to the International Energy Agency (IE). The rise of artificial intelligence (AI) What we are living has triggered the proliferation of these facilities In the US, China, Japan, Singapore, India, Germany, Netherlands or Ireland, among other nations. And for the moment there is no indication that invites us to anticipate that this trend will be exhausted in the medium term. A data center dedicated to large AI can exceed 150 MWand, precisely, these are the facilities that are proliferating the most. In fact, in 2024 its global consumption amounted to about 415 TWH, a figure that represents around the 1.5% of global electricity consumption. To solve this challenge and guarantee to data centers the delivery of energy that more and more companies need nuclear. The last one who has done is Nvidia. And is that the company led by Jensen Huang has participated in a financing round Of 650 million dollars to support Terrapower projects, the nuclear energy company founded by Bill Gates in 2006. With this decision NVIDIA adds to the strategy that defends the use of Compact modular reactors (known as SMR for its denomination in English) with the purpose of delivering to the data centers the electricity they need. And, incidentally, put one more leg in a sector with an indisputable growth potential. Terrapower is already building the first Natrium nuclear reactor The nuclear fission reactor that this company has designed is a modular and compact design refrigerated by sodium that uses a molten salts storage system. Because of its characteristics, it is about A fourth generation machine That, according to those responsible, it will be able to generate electricity in half of the cost that a conventional nuclear fission reactor. Whatever the interesting thing is that the first Natrium nuclear reactor in Terrapower is being built in a Wyoming Mining town (USA), and, according to Bill Gates, will be completed in 2030. Nvidia has participated in a financing round of 650 million dollars to support Terrapower projects It sounds good, but we must not overlook that it is a new generation design, so a priori the five years that Terrapower manages seem too optimistic. However, this reactor has an important asset in your favor: on paper Its tuning should be faster and cheaper than that of conventional reactors. In addition, a Spanish public company is participating in the construction of this machine. It is called Ensa (Nuclear Teams, SA), is Cantabrian and has more than five decades of experience in the field of design and manufacturing large components for the nuclear industry. There is no doubt that the fact that Terrapower has decided to ally with it is a boost that will surely reinforce its international image. And, perhaps, he opens the door of other latest generation nuclear energy projects. “This is the first reactor of these characteristics that is manufactured following the highest standards of safety and quality in accordance with the most demanding nuclear regulations,” has declared A Enso spokesman. Interestingly, this Spanish company will participate in The manufacture of the Natrium reactor lid. A last interesting note: currently also intervenes in the construction of ITER (International Thermonuclear Experctor reactor), The experimental reactor of nuclear fusion that an international consortium led by Europe is pointing in the French town of Cadarache. Image | Terrapower More information | The Register In Xataka | “We are already on the last step”: how Spain has done with the key to realize nuclear fusion

We knew that humanoid robots would reach factories. Nvidia has already chosen where and when to start, according to Reuters

by

When did humanoid robots stop being a spectacle to become a tool? Maybe that’s right there. Sources consulted by Reuters They assure that Nvidia and Foxconn are in conversations to display them in a server manufacturing plant of artificial intelligence in Houston. Nvidia has trusted the Taiwanese giant to lift a new server manufacturing plant in Houston, Texas. The objective: produce the GB300its new AI servers based on architecture Blackwellwithin the ambitious plan for relocate part of its production in US territory. As Reuters has advanced, both companies are in conversations to display humanoid robots in this factory. The intention would be that they begin to operate in the First quarter of 2026. If concrete, it will mark a double milestone: it would be the first time that a NVIDIA product is manufactured with the help of these tools, and also the first use of this technology by Foxconn in a production line of AI servers. Houston is not any factory: something new is prepared here For now, the details are scarce. It is not known how many robots will be used, how will they look or what exact functions they will perform. But there are indications. In an internal presentation of May, Foxconn showed how he was training humanoid robots for tasks such as manipulating objects, inserting cables or making basic assemblies, usual activities in the manufacture of servers. Houston’s choice is not accidental. Being a new plant, spaces are being designed with margin to integrate these technologies From the beginningsomething much more complex to achieve in already operational facilities. According to one of the sources consulted, that design would facilitate the incorporation of humanoid robots in the line. NVIDIA GB300 has a rack scale design That Nvidia bet on humanoid robots in its production chain is not just a logistics movement. It is also a declaration of intentions. Until now, no company product had been manufactured with the help of this type of robots. And Foxconn, the largest manufacturer on the commission of the world, had not used them in a production line dedicated to AI servers. The decision, according to what the sources have told Reuters, would mark the beginning of a new stage for both companies. In the case of Foxconn, it would also serve to show the world the advances in robotics who has been developing with Nvidia, although third -party models such as those of China Ubtech have also been tested. For Nvidia, the movement fits with its broader strategy. The company not only designs chips for AI models training: it also offers A development platform Specific for humanoid robots, with visual, motor and cognitive abilities based on their own architectures. In March, Jensen Huang himself He predicted that The generalized use of humanoid robots in industrial environments would come “in less than five years.” They are not alone: ​​Tesla, Mercedes, BMW, China The idea of ​​incorporating humanoid robots into the assembly lines is no longer a rarity. Although its deployment is still limited and experimental, several manufacturers have been testing this technology for some time in controlled environments or in very specific tasks. Among them BMW stands out, that has made trials in American plants. And it is known that Teslawhich has developed its own humanoid robot called OptimusHe has put at least two units to work in a production line. But interest is not limited to the great western brands. China has converted humanoid robotics into a national strategic priority Within its Made in China 2025 plan. Companies like Ubtech – whose model has also been evaluated by Foxconn – are being driven directly by the government with a view to transforming the country’s industrial fabric. Strategic alliances are part of this mission Like Huawei and Ubtech Specified this year. This possible deployment of humanoid robots in Houston does not occur in a vacuum. Is part of a broader movement, driven by political pressure and the strategic need of Relocate production Technological on American soil. In April, Nvidia announced its intention To manufacture AI infrastructure of up to 500,000 million dollars in the US in the next four years, with partners such as TSMC, Wistron and Foxconn itself. For many companies, automating is a matter of survival. The Houston factory, still under construction, is part of that strategy. But producing locally implies facing at least one new problem: the shortage of labor. And that is where automation would come into play. Perhaps not essentially for these factories, but as a test field for possible future expansions. For many companies, automating is no longer a matter of improvement. It is a matter of survival. Thus, more and more local actors are developing humanoid robots designed specifically for the industry. Tesla, Figure, Apptronik or Agility Robotics They are among the companies that have opted for this new generation of machines. Jeff Burnstein, president of the Association for Advancing Automation, summed up axios The new industrial reality: “This is how it competes today”, so “you have to take advantage of the best available tools.” Humanoid robots lived for years with skepticism: beautiful exhibitions, Little useful in practice. Now, that perception is turning. We are faced with a change that aims to be important, but whose real range we will know only over time. Images | Nvidia | Boliviainteligent In Xataka | The US is willing to do anything for advanced chips not to reach China. And Malaysia is an obstacle

AMD’s problem is not that it does not make good gpus for ia. Is that it is not even close to Nvidia

by

AMD is doing things well, but even doing them still unable to compete with Nvidia. The company has just raised its renewed road map with promising models, but that is not a guarantee of anything to a NVIDIA that will not let its absolute leadership position escape. The problem for AMD is not to be, but get others to take note. IDC consultancy data indicate that Nvidia dominates the AI ​​chips market with 85.2% market dick, for 14.3% AMD. Other analysts like Jon Pedie Research go beyond and According to your data The NVIDIA quota in this segment is 92%. AMD instinct mi350 are just the beginning. The GPUS for IA, which AMD calls “accelerators”, follow its evolution. During the event they presented their family or Instinct Mi350 series with two variants, MI350X and MI355X. According to the manufacturer, these chips are four times higher in general performance with respect to the previous generation, but are up to 35 times more powerful in the field of inference AI (that is, in the practical use of models such as Chatgpt, which “infers” “their responses from our prompts). They have 288 GB of HBM3E memory and a memory bandwidth of 8 TB/s. Its yield is 18.45 pflops in FP4 precision and 9.2 pflops in precision FP8. Instinct Mi400 in 2026. Next year the new family of AMD’s accelerators will arrive. It’s about future MI400 instinctwhich will arrive with up to 432 GB of HBM4 memory, 19.6 TB/s of bandwidth of that memory, and a performance of 40 pflops in precision FP4 and 20 Pflops in precision FP8. These monsters will be sold in future racks with infrastructure “Helios“, that You can house Up to 72 Mi400 with up to 260 TB/s total bandwidth thanks to its interconnection technology, Ultra Accelerator Link. EPYC VENICE. AMD not only talked about GPUS: it also has its future processors for servers in data centers in full development. The Epyc Venice will arrive in 2026 and will be based on Zen 6 architecture. Among the variants, an especially spectacular with 256 cores that will offer up to 70% more performance compared to the previous generation. These processors will be built with future MI400 instinct. They are expected to be manufactured with the N2P (2 Nm) node of TSMC. Helios against Oberon. The aforementioned Rack Helios will compete with not already with the current Nvidia AI server, the GB200 NVL72 which connects 36 CPUS Grace and 72 Gpus Blackwell. He is destined to compete with his successor, which has Oberon’s code name and will use IA B300 GPUS with Vera Rubin architecture. The yields and benefits of these future racks are absolutely dizzy, and for example their Precision Power FP8 is 1.4 Exaflops. The same in some things, better in others. AMD promises to match NVIDIA in several sections, but also ensures that it will exceed it remarkably (50% more) in memory quantity and width, something crucial for training and inference AI. Be careful, because at the end of 2027 NVIDIA prepares the Rubin Ultra architecture, which promises racks with up to 5 Exaflops in FP8 precision, three times more than Helios or Oberon. In 2027 we will have another “summer”. The AMD roadmap goes further, and they have already prepared the development of their new generation of chips for summer Epyc servers, which will replace the Epyc Venice. These CPUS will be paired with the future MI500X instinct, and it is expected – although it is not safe – that both types of chip take advantage of the one already announced TSMC A16 node (1.6 Nm), which will begin to be used at the end of 2026. There are no specifications for these developments, surely because they will depend on the manufacturing node that AMD ends up using to produce them. Frantic race. All these ads show that AMD does not want to be left behind in that race to place their solutions in data centers worldwide. The Crusoe company, which is dedicated to the construction of large AI data centers, advertisement A few days ago I would spend 400 million dollars in AMD’s chips, and even Sam Altman, CEO of OpenAi, made a surprise appearance During the inaugural talk of the Lisa Su, CEO of AMD event. Altman said they will also use AMD chips in the data centers they use, and highlighted that the new AMD ia gpus “will be somewhat amazing.” AMD presumes to be more efficient (and cheap). AMD’s message was clear during the event: its MI355 offer much more efficiency and are cheaper than NVIDIA B200 and GB200 with comparable yields. The sales prices of those GPUS are not known, but we do know that at the beginning of 2024 the MI300x of AMD They cost a maximum of $ 15,000 for the more than $ 40,000 that cost The NVIDIA H100. The biggest challenge is still CUDA. The benefits of AMD AI chips are not in fact the problem of this company. Detailed studies revealed months ago that MI300X are clearly higher than NVIDIA H100 and H200 on performance and power. However, Nvidia has a Cudathe de facto standard in the industry for services of services and applications of AI. Using AMD native software is feasible, yes, but software experience, They assured in SEMIANALYSIS“Software is full of errors that make training (AI models) with AMD it is impossible.” AMD’s hope is Rocm. In that AMD event also presented Rocm 7, the latest version from your own Open Source programming platform for your GPUS. In AMD they indicated that this version is 3.5 times more powerful than Rocm 6, and even claim that it is 30% more powerful than CUDA in the B200 when serving the model Deepseek R1. Even so, they indicate In another report of semi -health, it is still lower in some sections. Getting that component allows developers to take advantage of all the potential of AMD’s chips is precisely key to the future of those efforts. Even … Read more

The Nvidia IA supercomputer costs three million dollars. And to function wears a switch with three km cable

by

When Nvidia presented her new AI chips, The B200 with Blackwell architecturetook the opportunity to present an AI accelerator called GB200. And by joining 36 of those accelerators created its AI server, the monstrous DGX GB200 NVL72, which also keeps some spectacular surprises. Each node is bestial. Each of those GB200 accelerators has a CPU Nvidia Grace with 72 ARM Neaven V2 nuclei and two B200 GPUS. By combining its power we end up having a kind of bestial GPU combined with a power of 1.44 Exaflops in precision FP4. A closet that weighs a quintal. The appearance of the GB200 NVL72 DGX is that of a small and narrow closet that is above all very dense: this rack weighs 1.36 tons. Inside there are 18 Bianca computing nodes in 1u format, and each of them has two GB200, or what is the same, with four B200 GPUS (hence 18 x 4 = 72). He estimated cost of this AI server is about three million. Liquid cooling is key. The heat dissipated by these components is remarkable, which makes in this case the best option to cool those elements is the liquid cooling. This system not only applies to the CPU Grace or in the B200 GPUS, but in the NVLink chips of the switches, which can also be heated a lot due to the massive transfer of data between the accelerators. Interconnections everywhere. For all these GPUS to work together, each of the 36 GB200 has specialized network cards with NVLINK support of fifth generation that allow each of the computer nodes to be connected to others. For this there are nine switches that provide that huge amount of interconnections. 3 km cable. The system allows you to enjoy a bidirectional bandwidth of 1.8 TB/s between the 72 server GPUS. But as they point out In The Registerthe really surprising thing is that in total inside that “closet” there are 3.2 kilometers of copper cable. Only the module with the switches weighs more than 30 kilograms due to both these components and the more than 5,000 cables that are used so that all Nvidia GPUS work together and in perfect synchrony. Why copper? It may be able to opt for copper cable seems strange, especially taking into account the needs in terms of bandwidth imposed by this machine. However, the solution with fiber optic cables imposed clear problems: we would have to use electronic components necessary to stabilize and convert optical signals. That would have increased not only the cost, but the consumption of the final system. Can Crysis run? The performance of each B200 chip It is already brutal on its own: Its power is the triple than that of the GeForce RTX 5090, and the entire server includes 72 of these specialized GPUSs for AI, which demonstrates the computing capacity that said machine possesses. It also has RT (Ray-Training) nuclei of the fourth generation, which would theoretically allow you to use these AI chips to play video games, although of course that is not even its purpose. In fact your performance in this area will probably be almost as poor as the Nvidia H100. Cloud consumption. Although new chips are much more efficient than H100 –25 times less, says Nvidia – this AI server has an estimated TDP of 140 kW. Since the average consumption of an average home in Spain round The 3,000 kWh per year, in an hour of use of the Nvidia server we consume the same as an average Spanish home in 17 days. Have it on and running all year raises a consumption similar to 415 middle homes throughout the year in Spain. In Xataka | AMD has a splendid roadmap for its AI chips. The problem is still in your software

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.