Under pressure from US restrictions, Intel is reportedly preparing to follow in NVIDIA’s footsteps by developing “special edition” versions of its AI acceleration chips, Gaudi 3, for the Chinese market. These two related products are rumored to be launched at the end of June and the end of September.

According to reports from The Register, Intel recently unveiled its new generation AI acceleration chip, Gaudi 3. Intel stated in the Gaudi 3 white paper that it is preparing to launch a special edition Gaudi 3 tailored for the Chinese market. This would include two hardware variants: the HL-328 OAM-compatible Mezzanine Card and the HL-388 PCIe Accelerator Card. The HL-328 is said to be scheduled for release on June 24, while the HL-388 follow suit on September 24.

In regard of the specifications, the made-for-China edition and the original version share the same features, including 96MB of on-chip SRAM memory, 128GB of HBM2e high-bandwidth memory with a bandwidth of 3.7TB per second, PCIe 5.0X16 interface, and decoding standards.

However, due to US export restrictions on AI chips, the comprehensive computing performance (TPP) of high-performance AI needs to be below 4,800 to export to China. This means the Chinese special edition’s 16-bit performance cannot exceed 150 TFLOPS (trillion floating-point operations per second).

For comparison, the original Gaudi 3 achieves 1,835 TFLOPS in FP16/BF16. This contrasts with NVIDIA’s H100, which is approximately 40% faster in large model training and 50% more efficient in inference tasks.

Therefore, the made-for-China edition will need to significantly reduce the number of cores (the original version has 8 Matrix Multiplication Engines [MME] and 64 Tensor Processor Core [TPC] engines) and operating frequency. Ultimately, this could result in reducing its AI performance by approximately 92% to comply with US export control requirements.

Analyses cited in the same report further suggest that Intel’s launch of the made-for-China edition for AI performance will be comparable to NVIDIA’s AI accelerator card H20 tailored for the Chinese market.

The made-for-China edition of Intel’s Gaudi 3 boasts a performance of 148 TFLOPS in FP16/BF16, slightly below the 150 TFLOPS limit. However, in terms of high-bandwidth memory (HBM) capacity and bandwidth, the Chinese special edition Gaudi 3 will be lower than NVIDIA’s H20, potentially putting it at a competitive disadvantage against the H20. Still, pricing will also be a key factor in determining whether it holds any competitive advantage.

As per a previous report from Reuters, the prices of the chips were said to be comparable to those of its competitor Huawei’s products. Reportedly, NVIDIA priced orders from Chinese H20 distributors between USD 12,000 and 15,000 per unit.

TrendForce believes Chinese companies will continue to buy existing AI chips in the short term. NVIDIA’s GPU AI accelerator chips remain a top priority—including H20, L20, and L2—designed specifically for the Chinese market following the ban.

Read more

• [News] Expert Insights on NVIDIA’s AI Chip Strategy – Downgraded Version Targeted for China, High-End Versions Aimed Overseas
• NVIDIA Experiences Strong Cloud AI Demand but Faces Challenges in China, with High-End AI Server Shipments Expected to Be Below 4% in 2024, Says TrendForce

(Photo credit: NVIDIA)

Meta Platform, parent company of Facebook, has announced its latest generation AI chip of its Training and Inference Accelerator (MTIA) on April 10th, fabricated using TSMC’s 5nm process. According to a report from Commercial Times, this move is expected to reduce Meta’s reliance on NVIDIA’s chips and enhance computational power for AI services.

In its shift towards AI services, Meta requires greater computational capabilities. Thus, last year, Meta introduced its AI models to compete with OpenAI’s ChatGPT. The latest AI chip, Artemis, is an upgraded version of MTIA introduced last year, assisting platforms like Facebook and Instagram with content ranking and recommendations.

Meta’s new generation AI chip will be produced by TSMC using the 5nm process. Meta reveals that Artemis offers triple the performance of the first-generation MTIA.

In October last year, Meta announced plans to invest USD 35 billion to establish infrastructure supporting AI, including data centers and hardware. CEO Mark Zuckerberg told investors, “In terms of investment priorities, AI will be our biggest investment area in 2024 for both engineering and compute resources.”

Meta’s proprietary AI chips are deployed in data centers to power AI applications. Meta has several ongoing projects aimed at expanding MTIA’s application scope, including supporting generative AI workloads.

The trend of tech giants developing their own AI chips is evident, with Meta joining competitors like Amazon, Microsoft, and Google in internal AI chip development to reduce reliance on NVIDIA. Google recently unveiled its latest data center AI chip, TPU v5p, on the 9th. Meanwhile, Intel is targeting NVIDIA’s H100 with its new AI chip, Gaudi 3.

Read more

• [News] NVIDIA’s H100 AI Chip No Longer Out of Reach, Inventory Pressure Reportedly Forces Customers to Resell

(Photo credit: Meta)

The global laptop and PC market is experiencing a gradual recovery, driven by the growing trend of AI-powered PCs (AIPC). Consequently, as per a report from TechNews, the competition to enhance AI chip computing power has emerged as a key global focus. 

One of the competitors, Intel, during its Vision 2024 event, showcased its next-generation laptop chip, Lunar Lake. Intel CEO Pat Gelsinger stated that this chip will deliver over 100 TOPS (trillion operations per second) of AI performance, with the NPU alone contributing 45 TOPS. This marks a threefold increase in AI performance compared to Intel’s current generation of chips and meets the 45 TOPS NPU performance threshold previously set by Intel for the next generation of AI PCs.

Currently, Intel’s Meteor Lake processor NPU can only deliver 10 TOPS, which falls short of the standard required for the next generation of AI PCs. However, the NPU performance of Lunar Lake precisely meets the 45 TOPS standard.

Pat Gelsinger did not provide detailed breakdowns of the remaining 55+ TOPS performance between the CPU and GPU, but it can be reasonably speculated that the GPU contributes around 50 TOPS, while the CPU cores contribute 5-10 TOPS.

As for Intel’s competitors, AMD’s current-generation Ryzen Hawk Point platform offers NPU performance of 16 TOPS, which is also below Intel’s envisioned standard for the next generation of AI PCs.

However, AMD has recently indicated that their next-generation products will make significant breakthroughs to meet the demands of AI computing, incorporating a robust architecture with powerful CPU, GPU, and NPU components. This design philosophy has been consistent for AMD from the Ryzen 7040 series to the current 8040 series.

At an AI event in December last year, AMD unveiled the next-generation Ryzen Strix Point mobile processor featuring the XDNA 2 architecture, boasting a threefold increase in AI performance compared to the previous generation.

Yet, AMD has not provided detailed performance allocations for each component. Nonetheless, a simple calculation suggests that if the NPU performance triples, then the NPU performance of Ryzen Strix Point would reach 48 TOPS.

Qualcomm’s Snapdragon X Elite platform represents another competitor in the escalating competition, with chips based on ARM-based architecture scheduled to launch in mid-2024. Qualcomm has stated that its NPU performance will reach 45 TOPS, further heightening the competition among Intel, AMD, and Qualcomm for dominance in the next generation of AI computing.

Read more

• [News] AMD Hosts Innovation Summit, CEO Lisa Su Highlights the Beginning of AIPC Era
• [News] PC Market on the Rise, Tremendous Potential for AIPC Next Year

(Photo credit: Intel)

Intel revealed on the 2nd that its manufacturing business’s operational losses deepened, dealing a significant blow to Intel’s attempts to surpass TSMC and regain its leading position.

According to Reuters, Intel stated that its manufacturing division incurred a $7 billion loss in 2023, a more severe downturn than the $5.2 billion loss in the previous year, with revenues reaching $18.9 billion, a 31% decrease from the previous year.

In the same report, Intel CEO Pat Gelsinger admitted that 2024 would be the year of worst operating losses for the company’s wafer fabrication business, with the break even expected by 2027.

The primary causes for the losses were attributed to bad decisions, including opposition to using ASML EUV equipment a year ago. The cost of this equipment could exceed $150 million but is more cost-effective than earlier chip manufacturing equipment.

Gelsinger believes that due to these mistakes, Intel has outsourced approximately 30% of its wafer production to external contract manufacturers like TSMC, with a goal to reduce this share to about 20%.

Intel plans to invest $100 billion in constructing or expanding chip factories across four states in the United States and convincing other companies to use Intel’s manufacturing services. Consequently, Intel has begun segregating its wafer foundry business into an independent division and financials, heavily investing to catch up with TSMC and Samsung.

According to TrendForce’s previous report on the fourth quarter of 2023, global semiconductor foundry revenue rankings showed that Intel Foundry Services (IFS), which ranked ninth globally in the third quarter of 2023, was pushed out of the top ten by PSMC and Nexchip due to factors such as the transition between old and new CPU generations and lackluster inventory momentum. At the same time, the top three semiconductor foundries globally were TSMC, Samsung, and GlobalFoundries.

(Photo credit: Intel)

Amid the overwhelming wave of Artificial Intelligence, the importance of advanced process chips is becoming increasingly prominent. Currently, the 3nm process is the most advanced node in the industry. Meanwhile, manufacturers such as TSMC, Samsung, Intel, and Rapidus are actively promoting the construction of 2nm wafer fabs. TSMC and Samsung previously planned to mass-produce 2nm chips in 2025, while Rapidus to begin trial production in 2025.

2nm Wafer Fabs to Complete Construction as Soon as this Year?

Recently, the Semiconductor Equipment and Materials International (SEMI) announced that it is expected that both TSMC and Intel will potentially complete the construction of 2nm wafer fabs by the end of this year.

Intel is anticipated to be the first to achieve commercialization of 2nm chips. The Intel PC CPU Arrow Lake product will utilize the 2nm process node. TSMC’s 2nm process is expected to be applied in Apple’s iPhone AP chips. Subsequently, TSMC’s 2nm capacity will soar up.

According to a report from Commercial Times the installation of equipment for TSMC’s 2nm process is accelerating. TSMC’s Fab20 P1 plant in Hsinchu, Baoshan is scheduled to install equipment in April this year, with pilot production expected to commence in 2H24 and small-scale production in 2Q25.

As for Intel, ASML already delivered the world’s first High Numerical Aperture (NA) EUV EXE:5200 to Intel in late 2023, supporting the latter in producing 2nm chips. Later, Intel kicked-start the calibration of lithography machine, which has been well on track.

Samsung and Rapidus all Set to Move

In terms of Samsung, its previously announced technology roadmap indicates that it will first mass-produce 2nm process chips for mobile terminals starting in 2025, followed by high-performance computing (HPC) products in 2026. It plans to expand to automotive chips by 2027.

Rapidus is setting up a 2nm chip fab in Chitose City, Hokkaido, Japan. Its pilot production line is scheduled to start operation in April 2025, aiming to commence mass production in 2027.

Recently, it’s reported that in order to promote the development of advanced wafer fabs in Japan, several Japanese manufacturers will supply products to Rapidus. Among them, Dai Nippon Printing (DNP) will begin mass production of masks for 2nm chips at its Fukuoka plant and other operations in Japan in 2027, which will be provided to Rapidus.

In addition to DNP, Japanese company TOPPAN Holdings is also collaborating with IBM to develop masks for 2nm chips and achieve mass production by 2026, and Rapidus is reportedly the purchaser. Moreover, companies like Tokyo Ohka Kogyo (TOK), JSR, Shin-Etsu Chemical are also expected to be the suppliers of Rapidus.

1nm Chip Plans Brought to Light

Following 2nm, 1nm chip will be the next target for wafer fabs. In light of manufacturers’ plans, the industry is expected to see mass production of 1nm-level chips from 2027 to 2030.

TSMC plans to reach the A14 node (1.4nm) in 2027 and the A10 node (1nm) in 2030. Recent reports from Economic Daily News indicated that TSMC intends to establish a factory in the Science Park of Taibao City, Chiayi County in central Taiwan to produce 1nm chips.

Samsung anticipates to launch the 1.4nm process by late 2027. It is reported that Samsung’s SF1.4 (1.4 nm) process can increase the number of nanosheets from 3 to 4, which is expected to significantly improve performance and power consumption.

Intel’s latest foundry roadmap shows that the Intel 14A (1.4nm-level) node will put into production in 2026, and Intel 10A (1nm-level) will start development or production in late 2027.