Recently, GPU market situation has revived, Nvidia and Intel, two leading enterprises are battling. On September 21, NVIDIA officially announced the third generation RTX architecture AdaLovelace and AdaGPU can achieve 2 times the performance improvement of traditional raster games and 4 times the performance improvement of fiber tracking games. NVIDIA has three GPUs based on AdaLovelace architecture, including GeForce RTX 4090 providing 2 4G B version, GeForce RTX 4080 is available in 16GB and 12GB versions.
Figure1: three types of GPU based on Adalovelace
On September 27th, the Intel Innovation Summit was held, and Intel CEO Kissinger announced that GPU is a growth engine of Intel. For gamers, Intel has launched a sharp GPU. Pat Kissinger announced that Intel Sharp Hyun A770 will land in the retail market on October 12th with a starting price of $329 and a variety of product designs, providing excellent content creation and 1440P game performance.
Figure2: Intel CEO Kissinger
GPU has gradually expanded from early graphics rendering to high-performance computing, scientific computing and other fields, and has surpassed CPU ,FPGA ,ASIC and so on, become the ideal carrier of general parallel computing.
In the past year, PCGPU market share has changed greatly. JPR, a market research organization, released the market report of GPU graphics cards (including single and integrated graphics cards) in the first quarter of 2022, from 2021 Q1 to 2022 Q1, NVIDIA and AMD Both have taken some market share from Intel, and Nvidia still holds the leading position. In Q1 2021, the market share of Intel GPU was as high as 68%. In Q1 2022, Intel's share dropped to 60%, NVIDIA took the second place with 21%, and AMD slightly increased to 19%. What trump card will the three parties come up with this year to compete for market share? Let's start with the specific GPU market growth trend and new products.
Figure3: PC GPU vendor market share
China's GPU market will grow rapidly in the future. US restrictions will step up the pace of shipments to China
At present, from the functional point of view, GPU can be simply divided into two markets: GPU focusing on graphics and image rendering and GPGPU focusing on general computing. The insider state that computing is the largest market of CPU, According to official statistics, the graphics rendering GPU market in 2021 is 15 billion US dollars, which is higher than the computing market of 11 billion US dollars, but in fact, many renderings are sold chip are used in calculations. Therefore, 70% of the GPU market with more than 20 billion US dollars is computing.
Market research organization JPR latest according to the published GPU graphics market forecast, the annual growth rate of GPU will be 6.3% from 2022 to 2026. JPR predicts that in the next five years, the penetration rate of independent GPU in PC market will increase to 46%. Intel has long dominated the CPU market, but in the GPU market, many market shares of the former have been divided by NVIDIA and AMD.
"Although AMD, Intel and Nvidia launched new products in the second half of the year, consumers are still cautious. Therefore, our forecast for GPU this year is a moderate 2% to 3%. According to the American Federation of Large Business's economic forecast for the American economy, the real GDP growth in the United States will increase by 2.3% year-on-year, and we expect to increase by 2.1% in 2023, "said Jon Peddie, president of JPR.
On September 16th, CITIC Securities Research Report showed that the market size of China's GPU (Graphic Rendering) is US $2.7 billion in 2021, the market size of data center GPU is US $2 billion, the market size of China's GPU (Graphic Rendering) is expected to reach US $4.7 billion in 2025, and the market size of data center GPU is expected to reach US $25 billion in 2030, accounting for about 30% of the world. GPU chips in China have great growth potential.
At the beginning of September, NVDIA, a major GPU chip manufacturer in the United States, was required by the US government to restrict the export of top-level products to China AI chip one day after, Nvidia disclosed the latest progress-it has been approved by the US government to continue to provide A100 products exported to American customers (to China) before March next year, and can continue to fulfill orders for A100 and H100 in September next year. On September 26th, Taiwan Province's supply chain reported that NVIDIA, an American chip company, had asked TSMC to step up shipments, hoping that TSMC would complete the NVIDIA chip order previously planned to be completed in five months in two months, so as to help NVIDIA sell as many GPU chips as possible to the Chinese market before the deadline before March 2023.
NVIDIA launches GeForce RTX 4090 discrete graphics card
NVIDIA is currently the largest independent GPU supplier in the world, with annual revenue of US $26.91 billion in fiscal year 2022, an increase of 61% over the previous fiscal year. 2021Q2 Nvidia's global independent GPU market share reached 83%. This company invented GPU in 1999, initially focusing on PC graphics, and later expanded into intensive computing. Nvidia used GPU to create scientific computing. artificial intelligence, data science, autopilot cars, robot technology, AR and VR platform.
Figure4: RTX 4090
On September 21, Nvidia CEO Huang Renxun officially released GeForce RTX 4090, which can contain up to 76 billion chips Transistor the theoretical energy efficiency of GDDR6X memory chip using TSMC 4N process technology and Micron Technology is doubled compared with the previous generation 8nm process, which means higher overclocking potential.
Huang Renxun particularly emphasized that the new generation RTX4090 flagship graphics card has 24GBGDDR6X memory, which is higher than the 3090 of the previous generation architecture on different game platforms Ti the performance increased by 2-4 times, and the price was US $1,599 (RMB 12,999), which was officially released on October 12th. Two versions of the RTX 4080 start at $899 for 12GB of memory, while the 16GB version starts at $199.
For the data center, Nvidia released its latest flagship product H100 in April this year TensorCoreCPU can mass production.
Figure5: product comparison
Intelrelease Arc A770 to provide more cost-effective graphics cards to consumers
In June this year, Intel released its first desktop-level discrete graphics card, positioning the entry-level A380. On September 28th, Intel CEO Kissinger announced that Intel Ruixuan A770 will land in the retail market on October 12th with a starting price of US $329 and a variety of product designs, providing excellent content creation and 1440p game performance.
A770 is equipped with 16GB of GDDR6 memory, with 2100MHz frequency and 32 Xe cores. Compared with the entry-level A380 released in June this year, 6GB of memory and 8 Xe kernels are greatly improved. At present, Intel has not released the complete test results of A770 graphics card. The A770 graphics card also supports XeSS supersampling technology similar to DLSS technology on NVIDIA RTX graphics card. Intel said that thanks to this technology, the ray tracing performance of A770 graphics card is significantly better than that of NVIDIA RTX 3060 at the same price.
Compared with NVDIARTX3060, the newly listed IntelA770 is more cost-effective, with a price of only $329. "For a long time, the price of GPU has remained between 200 and 300 US dollars, but recently GPU has become very expensive. People are tired of paying too high a price for high-performance graphics cards. We don't think it is necessary, and we are solving this problem and providing more choices for players." Pat said in his speech.
Pat Kissinger said that the blade server with built-in Intel data center GPU codenamed Ponte Vecchio has been shipped to Argonne National Laboratory, which will provide driving force for Aurora Supercomputer.
Figure6: ARC A770
AMD uses this product to Hold 1044P games
Liu Wenzhuo, technical director of AMD China Enterprise and Business Division, told the media that, first of all, AMD is the first in the industry to apply 7nm process technology to x86 CPU and GPU products semiconductor enterprises.
In 2022, the Radeon RX 6000 series graphics card family has been growing, and AMD Radeon RX 6750 XT, one of the new members, is mainly for 1440P, that is, 2K resolution game people. It can provide high-performance experience at the highest setting, and can hold both AAA masterpieces and e-sports games.
Radeon RX 6750 XT graphics card adopts AMD RDNA 2 architecture, with process optimization and firmware and software enhancement. It is equipped with 96MB AMD Infinity Cache with high bandwidth and low latency, 40 computing units and light accelerator, and ultra-fast AMD Infinity Cache, with 12GB GDDR6 maximum memory, 2150 MHz base frequency and 2600MHz acceleration frequency, which brings excellent performance to players.
It supports Microsoft Windows 11 and Microsoft DirectX 12 Ultimate, and can improve image quality and performance for games supporting FSR super-resolution sharp painting technology; Bring super-native image quality to games supporting FSR 2, and unlock new performance levels at the same time. Not long ago, AMD announced a new iteration of the popular time domain sampling amplification technology-FidelityFX Super Resolution (# FSR #) 2.1.
FSR 2.1 includes a pair of Algorithm the update can improve the image quality in the game and reduce artifacts such as ghosting and splash screen. At the same time, it also announced 11 games that have joined or are expected to join FSR 2 support, and the total number of games that can support FSR 2 has reached 45.
For a long time, the global discrete graphics GPU market has been monopolized by NVDIA and AMD. Although Intel occupies the highest share in the integrated graphics market, its presence in the profitable discrete graphics market is not strong. This time, Intel launched Arc770 series, which is an attempt to break this monopoly.
At present, although the complete test results have not been disclosed, Intel revealed that Arc A770 outperforms RTX 3060 in ray tracing games with 1080P resolution. In addition to facing the consumer market, Intel's launch of unique products may also be a defense against NVIDIA's high-performance computing products and aim at the future market demand.
For domestic enterprises, NVIDIA's A100 and H100 for data centers may be cut off in March 2023, which also stimulated the upsurge of domestic GPU enterprises to quickly launch alternative products. It is reported that the first 7nm general GPU "Tiangai 100" launched by DeepSpark has been mass-produced. As of the beginning of September, the accumulated order amount exceeded 230 million yuan. Biren Science and technology compared with NVIDIA's flagship "A100", the launched GPU chip "BR100" leads in some indicators. We look forward to more news of domestic GPU chip mass production.
GPU Market Competition is Fiery, Which has a Leg up on Market? -China.exportsemi.com
