Recently, Reuters reported that SoftBank Group acquired Graphcore, a company often referred to as the “UK’s NVIDIA,” though the amount of the deal was not disclosed. Graphcore is a startup in the field of artificial intelligence (AI) that has designed a new type of Intelligent Processing Unit (IPU). In certain model tests, its performance has surpassed that of NVIDIA’s GPU systems, and thus, the industry is optimistic about its potential to compete with NVIDIA’s GPU.

• The Differences between IPU and GPU

As a processor specifically designed for AI computation, also known as an AI processor, IPU is excels in fields such as deep learning, machine learning, and natural language processing, boasting the capability of accelerating various AI-related tasks.

GPU, on the other hand, was initially designed to meet the demands of graphics rendering and image processing. With the rapid proliferation of AI and big data technologies, high-performance GPU, known for their powerful parallel processing capabilities, can handle multiple data points and tasks simultaneously, thereby speeding up training and inference processes, which together enable GPU to be gradually applied in the AI field, particularly in deep learning and machine learning.

Although both IPU and GPU can be used in the AI domain, they differ a lot in several aspects, such as computational architecture and memory architecture.

Previously, Lu Tao, the President and General Manager of Greater China at Graphcore, explained that the Graphcore C600 has 1,472 processing cores per IPU, capable of running 8,832 independent program threads in parallel. In comparison, NVIDIA’s GPU SM Core (Stream Multiprocessor Core) has around 100 cores, varying with different product configurations.

In terms of memory architecture, NVIDIA’s GPUs have a two-level memory structure. The first level consists of around 40-50 MB of memory within the chip, with external HBM or VRAM attached. Graphcore’s IPU, however, contains 900 MB of on-chip SRAM storage, which is distributed.

Lu summarized that the IPU’s architecture shows greater advantages for tasks with high requirements for sparsity and high dimensions relative to to GPU. For matrix operations, its performance might be similar to GPU or slightly less competitive.

Read more

• [News] SoftBank Founder Masayoshi Son Plans to Raise USD 100 Billion to Establish AI Chip Company

(Photo credit: Graphcore)

According to the industry sources cited in a report from Economic Daily News, TSMC is gearing up to start production of NVIDIA’s latest Blackwell platform architecture graphics processors (GPU) on the 4nm process. In response to the strong customer demand, NVIDIA has reportedly increased its orders to TSMC by 25%.

This surge not only underscores the unprecedented boom in the AI market but also provides substantial momentum for TSMC’s performance in the second half of the year, setting the stage for an optimistic annual outlook adjustment, the report notes.

TSMC is set to hold an earnings conference call on July 18, in which it is expected to release the financial results of the second quarter as well as the guidance for the third quarter.

As TSMC will reportedly commence the production of NVIDIA’s Blackwell platform architecture GPU, which may be regarded as one of the most powerful AI chips, it is anticipated to be a focal point of discussion at TSMC’s upcoming earnings call.

Packed with 208 billion transistors, NVIDIA’s Blackwell-architecture GPUs are manufactured using a custom-built 4NP TSMC process with two-reticle limit GPU dies connected by 10 TB/second chip-to-chip link into a single, unified GPU.

The report further cited sources, revealing that international giants such as Amazon, Dell, Google, Meta, and Microsoft will adopt the NVIDIA Blackwell architecture GPU for AI servers. As demand exceeds expectations,NVIDIA is prompted to increase its orders with TSMC by approximately 25%.

As NVIDIA ramps up production of its Blackwell architecture GPUs, shipments of terminal server cabinets, including the GB200 NVL72 and GB200 NVL36 models, have seen a simultaneous significant increase. Initially expected to ship a combined total of 40,000 units, this figure has surged to 60,000 units, marking a 50% increase. Among them, the GB200 NVL36 accounts for the majority with 50,000 units.

The report estimates suggest that the average selling price of the GB200 NVL36 server cabinet is USD 1.8 million, while the GB200 NVL72 server cabinet commands an even higher price of USD 3 million. The GB200 NVL36 features 36 GB200 super chips, 18 Grace CPUs, and 36 enhanced B200 GPUs, whereas the GB200 NVL72 boasts 72 GB200 super chips, 36 Grace CPUs, and 72 B200 GPUs, which all contribute to TSMC’s momentum.

TSMC former Chairman Mark Liu, before handing over the reins in June, had already predicted that the demand for AI applications looks more optimistic compared to a year ago. Current Chairman C.C. Wei has also indicated that AI applications are just beginning, and he is optimistic like everyone else.

Read more

• [News] End of an Era under Mark Liu’s Leadership, TSMC Discusses Future Development Plans
• [News] TSMC’s 4nm Process Powers NVIDIA’s Blackwell Architecture GPU, AI Performance Surpasses Previous Generations by Multiples

(Photo credit: TSMC)

With the widespread adoption of AI, the demand for GPUs and high-bandwidth memory (HBM) is soaring. The Semiconductor Equipment Association of Japan (SEAJ) has revised its forecast for sales of Japanese-made semiconductor equipment, predicting that for the first time in history, sales will exceed JPY 4 trillion in the 2024 fiscal year. It is also estimated that sales will further exceed JPY 5 trillion in the 2026 fiscal year.

Per a report by Nikkei on July 5th, SEAJ’s forecast report indicates that for the 2024 fiscal year (April 2024 to March 2025), the sales of Japanese-made chip equipment (including sales by Japanese companies both domestically and overseas) have been revised from the previous estimate of JPY 4.0348 trillion (in January 2024) to JPY 4.2522 trillion, marking a significant increase of 15.0% compared to the 2023 fiscal year.

This will be the first time in history that annual sales exceed JPY 4 trillion, setting a new record. The main reason for this growth is the widespread adoption of AI, which has led to a strong demand for GPUs used in AI servers and a continuous surge in demand for HBM used in conjunction with them.

Reportedly, SEAJ stated that, due to anticipated steady investments in logic/foundry and memory, the sales forecast for Japanese chip equipment in the 2025 fiscal year (April 2025 to March 2026) has been revised upward from JPY 4.4383 trillion to JPY 4.6774 trillion, representing a 10.0% year-on-year increase.

Additionally, the demand for chip equipment is expected to be driven by AI-related semiconductors, resulting in a projected 10.0% year-on-year increase in sales for the 2026 fiscal year, reaching JPY 5.1452 trillion. This will mark the first time annual sales exceed JPY 5 trillion.

The compound annual growth rate (CAGR) for Japanese chip equipment sales during the 2024-2026 period is estimated to be 11.6%. Japan’s global market share for chip equipment (in terms of sales) is approximately 30%, making it the second-largest in the world, following the United States.

SEAJ has indicated that in addition to servers, AI will be integrated into into PCs and smartphones at a faster pace in the near future. SEAJ President Toshiki Kawai mentioned that by 2027, 30-40% of PCs and smartphones are expected to incorporate AI, which is anticipated to have a more significant impact on increasing the demand for chip equipment compared to servers.

Toshiki Kawai further mentioned that as the self-sufficiency in manufacturing equipment remains insufficient in the Chinese market, the demand remains consistent and robust for Japanese-made equipment.

On June 25th, SEAJ released statistics indicating that in May 2024, Japan’s semiconductor equipment sales (3-month moving average, including exports) reached JPY 400.954 billion. This marked a significant increase of 27.0% compared to the same month last year, marking the fifth consecutive month of growth and the largest increase in 19 months (since October 2022, with a growth of 27.6%).

Monthly sales exceeded JPY 300 billion for the seventh consecutive month and surpassed JPY 400 billion for the first time in history, setting a new monthly sales record (previously, the highest record was JPY 389.106 billion in April 2024).

Read more

• [News] New Regulations Reportedly Introduced by Japan Regarding Semiconductor Exports
• [News] Booming Sales of Japanese Semiconductor Equipment Continue to Break Records, Reportedly Surpassing JPY 300 Billion

(Photo credit: TEL)

Samsung Electronics’ management has made a significant decision to invest in graphics processing units (GPUs). According to a report from Business Korea reported on June 18th, while the details of Samsung’s GPU investment have not been disclosed, this decision is noteworthy as it differs from their usual focus on memory and foundry services.

Per Business Korea citing Samsung Electronics’ governance report, the management committee approved the “GPU Investment Proposal” in March. The committee includes senior executives such as Han Jong-Hee, head of the Device eXperience (DX) division, as well as the presidents of the Mobile Experience (MX) and Memory Business divisions. Reportedly, this marks the first time since the agenda items were made public in 2012 that Samsung has decided to invest in GPUs, sparking speculation that the company aims to enhance its competitiveness in the GPU sector.

Industry sources cited in the same report interpret this investment as an internal strategy for Samsung to leverage GPUs to innovate semiconductor processes, rather than to develop or manufacture GPUs. At the “GTC Conference” held in March 2024, Samsung announced its collaboration with NVIDIA to develop AI-based digital twins, aiming to achieve full automation of semiconductor plants by 2030.

Reportedly, Samsung’s newly constructed high-performance computing (HPC) center in Hwaseong was completed in April 2024. This center houses a vast array of servers and network equipment necessary for semiconductor design, indicating a significant investment in GPUs.

According to the report, Samsung’s newly constructed high-performance computing (HPC) center in Hwaseong was completed in April 2024. This center houses a vast array of servers and network equipment necessary for semiconductor design, indicating a significant investment in GPUs.

Per another report from Bloomberg on June 4th, NVIDIA CEO Jensen Huang, during a briefing at the COMPUTEX, told reporters that NVIDIA is evaluating HBM provided by both Samsung and Micron Technology. Huang mentioned that there is still some engineering work needed to be completed, expressing the desire for it to have been finished already.

As per Huang, though Samsung hasn’t failed any qualification tests, its HBM product required additional engineering work. When asked about Reuter’s previous report concerning overheating and power consumption issues with Samsung’s HBM, Huang simply remarked, “there’s no story there.”

Read more

• [News] Reasons for Samsung’s HBM Chips Failing Nvidia Tests Revealed, Reportedly Due to Heat and Power Consumption Issues
• [News] Samsung Reportedly Establishes New HBM Team, Looking to Improve AI Chip Yield

(Photo credit: Samsung)

In response to US export bans, NVIDIA, the global leader in AI chips, has commenced to sell H20, its AI chip tailored for the Chinese market earlier this year. However, an oversupply caused the chip to be priced lower than its rival, Huawei, in some cases even at an over 10% discount, according to the latest report by Reuters.

The US Department of Commerce restricted the export of NVIDIA AI chips to China due to concerns about their potential military use in late 2022. In response, NVIDIA has repeatedly reduced product performance to comply with US regulations. The H20 chip, derived from the H800, is specifically designed as a ‘special edition’ for the Chinese market.

However, due to the abundant supply in the market, citing sources familiar with the matter, Reuters noted that H20 chips are being sold at a discount of over 10% compared to Huawei’s Ascend 910B, the most powerful AI chip from the Chinese tech giant.

The chip is reportedly to be sold at approximately 100,000 yuan per unit, while Huawei 910B sold at over 120,000 yuan per unit.

The decreasing prices underscore the difficulties NVIDIA encounters in its China operations amid U.S. sanctions on AI chip exports and rising competition from local rivals.

According to a previous report by The Information, major tech companies such as Alibaba, Baidu, ByteDance, and Tencent have been instructed to reduce their spending on foreign-made chips like NVIDIA’s, according to sources cited by the media outlet.

(Photo credit: Huawei)