BTC $64,679.21 -0.31%
ETH $1,813.24 -1.27%
BNB $592.44 -0.27%
XRP $0.9871 +2.50%
SOL $75.20 +3.26%
TRX $0.3078 +0.61%
DOGE $0.0595 +0.80%
ADA $0.1540 -0.12%
BCH $209.16 -0.04%
LINK $9.25 +0.72%
HYPE $68.95 -1.67%
AAVE $111.93 -0.52%
SUI $0.7307 +0.37%
XLM $0.1414 +2.11%
ZEC $963.08 +4.41%
AAPL $350.35 -0.71%
AMZN $258.64 +0.12%
GOOGL $282.29 -0.78%
MSFT $508.58 -1.65%
META $620.63 -0.55%
NVDA $158.13 +0.44%
TSLA $383.22 +1.02%
SNDK $822.19 +0.65%
INTC $56.01 +2.28%
SPCX $142.28 +3.11%
MU $731.53 +1.07%
AMD $515.04 +0.89%
BTC $64,679.21 -0.31%
ETH $1,813.24 -1.27%
BNB $592.44 -0.27%
XRP $0.9871 +2.50%
SOL $75.20 +3.26%
TRX $0.3078 +0.61%
DOGE $0.0595 +0.80%
ADA $0.1540 -0.12%
BCH $209.16 -0.04%
LINK $9.25 +0.72%
HYPE $68.95 -1.67%
AAVE $111.93 -0.52%
SUI $0.7307 +0.37%
XLM $0.1414 +2.11%
ZEC $963.08 +4.41%
AAPL $350.35 -0.71%
AMZN $258.64 +0.12%
GOOGL $282.29 -0.78%
MSFT $508.58 -1.65%
META $620.63 -0.55%
NVDA $158.13 +0.44%
TSLA $383.22 +1.02%
SNDK $822.19 +0.65%
INTC $56.01 +2.28%
SPCX $142.28 +3.11%
MU $731.53 +1.07%
AMD $515.04 +0.89%

for

All
Article
Flash

first_img Online materials: Ascend 950DT's inference performance reaches 86% of H200

On November 18, the AI content account Max For AI stated that a document titled "Ascend 950DT Performance Testing and Business Benefit Calculation Report" was circulating in computing power-related communities. The document is dated October of this year and shows content from an internal testing report of China Mobile. According to the document, on October 10, a joint team from multiple departments of China Mobile went to Huawei's Lianqiu Lake R&D Center to conduct tests on the Zhiyuan GLM 5.2 and DeepSeek V4 Pro using a 64-card Ascend 950DT super node.The performance data provided in the document is based on the GLM 5.2, 16K input / 2K output scenario. In terms of total throughput per card, the Ascend 910B4 achieved 169 tokens/s, the 910C achieved 483 tokens/s, and the 950DT achieved 1951 tokens/s, which is approximately 11.5 times that of the 910B4 and 4 times that of the 910C. The document cites vLLM data, estimating the NVIDIA H200 in the corresponding scenario to be about 2200 tokens/s, which corresponds to the 950DT being about 89% of the H200; another set of 7K/2K data is about 82%. The final conclusion of the document states that the inference performance of the 950DT reaches 86% of the H200.In terms of cost estimation, the document lists the prices of 8-card servers based on bulk purchasing: the 950DT is approximately 2.97 million yuan, the 910C is approximately 2.15 million yuan, and the 910B4 is approximately 650,000 yuan. Including equipment, maintenance, supporting facilities, and IDC, and amortizing over five years, the monthly costs are approximately 63,600 yuan, 48,700 yuan, and 16,100 yuan, respectively. Based on this, the document calculates that the unit cost inference performance of the 950DT is about 3.1 times that of the 910C and 2.9 times that of the 910B4. The bare metal rental is estimated at 84,300 yuan per month for the 950DT and about 130,000 yuan for the H200, stating that the cost-performance ratio of the 950DT is about 30% higher than that of the H200.

first_img Some NVIDIA board members are pushing for Jensen Huang's return, but the chairman said it is not being considered

According to Semafor citing informed sources, despite Nvidia stating on Tuesday that it does not plan to have former CEO Jensen Huang return, a portion of the board is still pushing for his return. Huang has met with several executives from tech companies and discussed the possibility of returning with at least one board member. This communication with the board occurred before the company's earnings report was released and prior to Huang's public statement on Tuesday. In his statement, Huang expressed that he is willing to return if the board determines that it is in the best interests of shareholders, employees, the company, and the United States.Nvidia Chairman Harvey Jones stated during the earnings call a few hours later that the company is not considering Huang's return, and the board has unanimously voted to accept his resignation. Both Huang's side and Nvidia's spokesperson declined to comment. Informed sources say that while a few board members are still advocating for his return, other board members have begun to narrow the search for a successor. ARM CEO Rene Haas and Marvell CEO Matt Murphy have both been considered. One informed source stated that Haas has discussed taking on the role with at least some Nvidia board members, but he is concerned that taking over amid calls for Huang's return could create perception issues. Spokespeople for ARM and Marvell did not immediately respond to requests for comment. It is currently unclear whether Murphy can take on the role of Nvidia CEO, as Marvell is advancing a large merger with competitor Lumentum.

Xiaomi was reported to be discussing independent financing for MiMo, with Lei Jun and Luo Fuli having differing opinions

According to Top Chinese Technology Innovation Society, Xiaomi is discussing whether to spin off the MiMo large model business and seek independent financing. Several well-known venture capital and corporate investment institutions have made preliminary contacts, but Xiaomi Group has not yet presented a clear plan.Luo Fuli and the core research team of MiMo hope to separate the large model business and introduce external financing, but Lei Jun has different considerations, and the two sides have not reached an agreement. Luo Fuli has begun to contact a few investors and financing consultants, and does not rule out the possibility of starting her own venture. Luo Fuli officially joined Xiaomi in November last year and was promoted to the highest level 22 in September this year. The MiMo-V2.6-Pro she led to launch once topped the Artificial Analysis global open-source model ranking.In March this year, Lei Jun announced that Xiaomi would invest at least 60 billion yuan in the AI field over the next three years. In July, Xiaomi adjusted its organizational structure, clearly stating that the MiMo team would provide basic model capabilities for Xiao Ai, serving mobile phones, cars, and other terminal businesses. Currently, the news about the spin-off financing and the existing differences between the two parties has not been confirmed by Xiaomi or the parties involved.

Tencent launched the WeFund mini-game open platform, integrating distribution middle platform capabilities

According to "Reading Plus," Tencent has quietly launched WeFund (Youkeying) mini-game open platform, positioned as a one-stop publishing empowerment platform focused on the mini-game sector under Tencent Interactive Entertainment's Traffic Ecology Department, integrating core capabilities and resources across five dimensions: product, user acquisition, operations, materials, and community.The platform's core relies on the OSDK open interface, allowing developers to integrate once and adapt to both WeChat and Douyin ecosystems simultaneously, significantly reducing cross-platform adaptation costs. Additionally, the platform provides features such as data collection, event tracking verification, A/B testing tools, and marketing dashboards, helping small and medium teams gain large enterprise-level data diagnostic capabilities. Currently, several mini-games on the platform are in the internal testing phase, and a player assistant has been launched on WeChat.In terms of cooperation models, WeFund mainly offers two methods: domestic full-channel agency publishing and custom development. The former is fully handled by Tencent, covering all aspects of publishing, including advertising, operations, and materials, while the latter involves Tencent funding and joint development. This move aims to lower the barriers to mini-game development and publishing, further strengthening Tencent's voice in the publishing ecosystem of mini-games on WeChat and Douyin.
app_icon
ChainCatcher Building the Web3 world with innovations.