# AMD Related Articles

HTX News Center provides the latest articles and in-depth analysis on "AMD", covering market trends, project updates, tech developments, and regulatory policies in the crypto industry.

CPU Makes a Comeback to the Table, A $170 Billion "Power Seizure" Drama Begins

A new era is dawning for the server CPU (Central Processing Unit), driven by the shift from AI model training to large-scale reasoning and the rise of Agentic AI. This article explores how the CPU is reclaiming a central role in the AI data center. For years, the focus has been on the GPU (Graphics Processing Unit) for AI training. However, as AI moves to the inference and Agent phase—where tasks involve complex, multi-step reasoning, tool calls, and data management—the workload balance is flipping. Studies show CPUs now handle over 70% of the workload in Agentic AI, up from 10-30% in training. This is because Agent tasks generate massive intermediate data (KV Cache) that exceeds GPU memory, forcing it to be offloaded to the CPU's larger, more scalable memory pools. This increased importance is translating into market changes. Major players are taking note: NVIDIA launched its first standalone CPU line, Vera, based on ARM architecture and optimized for Agent performance. AMD doubled its server CPU market forecast to over $1200 billion by 2030. Analyst reports project the total server CPU market could reach $1700 billion by 2030, with AI-driven demand being a primary driver. Furthermore, the classic ratio of CPUs to GPUs in AI servers is rapidly changing, converging from 1:8 toward 1:1 for Agent deployments. This surge in demand has led to a rare industry-wide price increase of 10-15% for server CPUs from Intel and AMD, breaking a decade-long trend of "more performance for the same price." Demand is bifurcating into high-core-count CPUs for in-rack GPU support and moderate-core CPUs for standalone Agent task orchestration. In China, this global trend presents an opportunity for domestic CPU manufacturers like Hygon (海光信息) and Huawei Kunpeng, who are bolstered by both growing AI infrastructure needs and national policies promoting technological self-reliance ("xin chuang"). The maturity of their software ecosystems is also accelerating, evidenced by faster adaptation to new AI models. In conclusion, the narrative is shifting from a GPU-centric view to one where CPU-GPU synergy is critical. The CPU is no longer a peripheral component but a performance-defining bottleneck and a key growth driver in the AI hardware stack, opening a massive new market estimated in the hundreds of billions of dollars.

marsbit5h ago

CPU Makes a Comeback to the Table, A $170 Billion "Power Seizure" Drama Begins

marsbit5h ago

Dylan Patel: Founder of SemiAnalysis, Praised by Jensen Huang, is a 'Beekeeper' and 'Forum Enthusiast'

Dylan Patel, founder of the independent research firm SemiAnalysis, has an unconventional background. A former beekeeper from rural Georgia, he entered the semiconductor world as a self-taught "forum warrior," discussing chip technology anonymously online from a young age. He launched the SemiAnalysis blog in May 2020, which later transitioned to a paid subscription model. The firm has grown from a one-person operation to a global team of around 60, with a dedicated teardown lab. Its detailed, technically-focused analysis on semiconductor supply chains, AI infrastructure, and products has earned significant industry recognition. Notably, NVIDIA founder Jensen Huang has publicly cited their reports. In a landmark case, a critical 2024 report on AMD's MI300X GPU software stack led to a 90-minute call with AMD CEO Lisa Su, who thanked him for the constructive feedback. SemiAnalysis later acknowledged AMD's improvements. The firm's influence on markets was seen when a report on NVIDIA's Rubin memory configuration was partially shared, affecting memory stock prices. Dylan Patel emphasized the importance of context, contrasting the shared excerpt with the report's actual title. SemiAnalysis, now a multi-faceted consultancy with revenue projected to reach $100 million, is known for its deep technical insights that influence major industry players and investment decisions.

marsbit18h ago

Dylan Patel: Founder of SemiAnalysis, Praised by Jensen Huang, is a 'Beekeeper' and 'Forum Enthusiast'

marsbit18h ago

Dylan Patel: SemiAnalysis, Praised by Jensen Huang, is Founded by a 'Beekeeper and Forum Warrior'

Dylan Patel, founder of the independent research firm SemiAnalysis, has an unconventional background. Growing up in rural Georgia, he later worked as a beekeeper in Minnesota. His entry into semiconductors began as a self-taught "forum warrior," engaging anonymously in online tech communities from a young age. In May 2020, he started the SemiAnalysis blog on WordPress, later moving it to Substack as a paid subscription service. The firm has since evolved from a one-person operation into a global company with around 60 employees, featuring a dedicated chip teardown lab. Its revenue, reaching $20 million last year, is projected to surpass $100 million this year. SemiAnalysis is highly regarded in the AI and semiconductor industry for its deep technical analysis. NVIDIA founder Jensen Huang has publicly praised its reports. In a notable instance, a critical report on AMD's MI300X GPU software shortcomings prompted a 90-minute call with CEO Lisa Su, who thanked Patel for the "constructive feedback." A later report acknowledged AMD's subsequent improvements. The firm's analyses have significant market impact. For example, a June report discussing potential memory configuration changes in NVIDIA's next-generation servers was cited as a factor in pressure on memory-related stocks. Patel plans to establish a venture capital firm, having already made personal investments in about 20 startups. SemiAnalysis combines roles as a consultancy, model platform, and tech lab, focusing on the practical bottlenecks in AI infrastructure.

Odaily星球日报18h ago

Dylan Patel: SemiAnalysis, Praised by Jensen Huang, is Founded by a 'Beekeeper and Forum Warrior'

Odaily星球日报18h ago

Bernstein Report: Agentic AI Will Transform CPU from Supporting Role to Leading Role, Bullish on Hygon Information

Bernstein research report: Agentic AI will turn CPUs from supporting players to leading roles, bullish on Hygon Information. Analysts led by David Dai argue that AI is transitioning from the chatbot era to the agentic AI era. Unlike simple query-response models, agentic AI involves complex workflows including retrieval, planning, tool calling, and multi-step reasoning. This shift dramatically increases the demand for CPU compute to orchestrate these tasks, manage memory, and prevent expensive GPU idling. The report forecasts that the GPU-to-CPU ratio in inference clusters will reverse from 8:1 in 2025 to 1:1 by 2029. In agentic AI workloads, CPUs could account for 50% of the compute, on par with GPUs. Consequently, the server CPU Total Addressable Market (TAM) is projected to surge from $37 billion in 2025 to $223 billion by 2030, representing a 6x expansion. Arm is identified as a key beneficiary due to its superior performance-per-watt and a strategic shift from IP licensing to designing its own chips, targeting $15 billion in chip revenue by 2030. Bernstein raises Arm's price target to $500. For x86 vendors, the report is Overweight on AMD (target $600) and Hygon Information (target CNY 450), citing leadership and strong growth in the Chinese market respectively. Intel's target is raised to $100, reflecting upgraded earnings assumptions. The analysis acknowledges significant supply-side risks, questioning whether foundry and memory capacity can support such rapid CPU growth. The optimistic demand forecast also heavily relies on Nvidia's guidance for over $1 trillion in annual AI infrastructure spend by 2027.

marsbit2 days ago 09:46

Bernstein Report: Agentic AI Will Transform CPU from Supporting Role to Leading Role, Bullish on Hygon Information

marsbit2 days ago 09:46

AMD Launches Compact AI Host, Directly Challenging NVIDIA DGX Spark

In June 2026, AMD announced the Ryzen AI Halo, a compact AI developer desktop to rival NVIDIA's DGX Spark. Both feature 128GB unified memory for running 200B+ parameter models locally. Priced from $2,949 to $3,999, AMD undercuts NVIDIA's $3,999+ DGX Spark. The core divergence lies in architecture and philosophy. Ryzen AI Halo uses an x86-based Ryzen AI Max+ 395 APU (CPU+GPU+NPU), runs standard Windows/Linux, and emphasizes general-purpose PC flexibility. DGX Spark uses an ARM-based Grace Blackwell Superchip, runs a custom DGX OS, and includes a high-speed ConnectX-7 NIC for cluster prototyping, anchoring it to NVIDIA's full-stack CUDA ecosystem. AMD's ROCm software has improved, with simpler installation and support for major frameworks, but still lags behind CUDA's 17-year maturity in community support and cutting-edge library availability. AMD's broader strategy focuses on becoming a viable second-source supplier. Key moves include acquiring design capabilities via ZT Systems (while outsourcing manufacturing) and securing two major 6GW GPU supply deals with OpenAI and Meta in late 2025/early 2026. These contracts validate AMD's role in diversifying the AI supply chain, rather than outright beating NVIDIA. NVIDIA counters with a tightly integrated stack from desktop (DGX Spark) to data center, emphasizing seamless scalability and enterprise software subscriptions (AI Enterprise). In summary, Ryzen AI Halo represents AMD's pragmatic path: offering a cost-effective, open-ecosystem alternative for developers wary of vendor lock-in, while its large data center contracts aim to capture share from customers seeking a second GPU supplier. The choice boils down to a familiar, flexible PC environment with potential software gaps (AMD) versus a premium, optimized, but locked-in ecosystem (NVIDIA).

marsbit06/16 09:14

AMD Launches Compact AI Host, Directly Challenging NVIDIA DGX Spark

marsbit06/16 09:14

New AMD Paper Overturns Conventional Wisdom: FP4 Training Instability's Cause Is Not Insufficient Randomness

AMD's new research challenges the conventional understanding of FP4 training instability. While reducing precision from FP8 to FP4 promises doubled computational throughput and is supported by new hardware like NVIDIA Blackwell and AMD MI350 series, training large language models natively with FP4 has been notoriously unstable, often attributed to insufficient stochasticity. The paper "Pretraining large language models with MXFP4 on Native FP4 Hardware" demonstrates successful end-to-end FP4 pre-training of Llama 3.1-8B on AMD MI355X GPUs using the MXFP4 format, achieving a 9-10% overall speedup over FP8. Crucially, it identifies the root cause of instability: not randomness, but the accumulation of *structural micro-scaling errors* along the sensitive weight gradient (Wgrad) path. Through controlled experiments, researchers found that quantizing the Wgrad operation to FP4 caused significant convergence degradation. Counterintuitively, common stochasticity-based mitigation techniques like stochastic rounding and randomized Hadamard transforms worsened performance. In contrast, applying a *deterministic* Hadamard transform successfully stabilized training by ensuring consistent error patterns, reducing the extra token cost from 26-27% to just 8-9%. This work has significant implications: 1) It provides a clear diagnostic for low-precision training instability, steering focus towards structural errors. 2) It pushes FP4 from a primarily inference-focused format into the realm of viable training. 3) It leverages the open OCP Microscaling (MX) standard, promoting cross-vendor compatibility. The research marks a critical step towards more economical large model training by further pushing the boundaries of low-precision computation.

marsbit05/27 06:19

New AMD Paper Overturns Conventional Wisdom: FP4 Training Instability's Cause Is Not Insufficient Randomness

marsbit05/27 06:19

活动图片