Recent #GPU news in the semiconductor industry

about 1 year ago
➀ NVIDIA's GeForce RTX 5080 is expected to feature the fastest GDDR7 memory at 32Gbps, with 16GB GDDR7 at 1TB/sec memory bandwidth; ➁ The GPU will have 10,752 CUDA cores and is expected to match RTX 4090 performance; ➂ The new RTX 5080 will offer significant performance improvements over the RTX 4080 and RTX 4080 SUPER.
Blackwell GPU architectureCUDA CoresDLSS 4GDDR7GPUGeForce RTX 5080Memory BandwidthNVIDIA
about 1 year ago
➀ The RTX5090 boasts 21760 CUDA Cores, a 33% increase over the RTX4090; ➁ It features 32 GB GDDR7 memory with a 512-bit bus, offering up to 2.00 TB/s bandwidth; ➂ The GPU uses the new Blackwell architecture and N4P process technology; ➃ It has a PCIe 5.0 x16 interface with double the theoretical bandwidth of the RTX4090; ➄ The TDP is a significant 600W, a 33% increase from the RTX4090.
GPUNVIDIAperformancespecs
about 1 year ago
➀ AMD showcased its latest products in CPU, GPU, and UA interconnection at Computex 2024; ➁ The new Zen 5 core is called the most powerful and efficient processor core so far; ➂ The second generation XDNA NPU architecture introduces new Block FP16 (BF16) floating-point precision, with AI engine performance three times that of the second generation AMD Ryzen AI; ➃ AMD launched the Ryzen 9000 CPU, the fastest consumer-grade PC processor globally, featuring Zen5 core and AM5 platform; ➄ AMD's next-generation ultra-thin and high-end laptop processor 'Strix Point' is designed for the next generation AI PC/Copilot+PC; ➅ AMD's new Versal AI Edge Gen 2 series provides the first single-chip adaptive solution for preprocessing, inference, and post-processing; ➆ AMD plans to launch a faster, larger memory MI325X AI GPU this year, followed by the MI350 series with the new cDNA4 architecture in 2025, and the MI400 series with the new cDNA architecture in 2026; ➇ AMD is making significant progress in promoting the development of high-performance AI network infrastructure systems with the upcoming UA-Link 1.0 standard.
AMDGPUcpu
about 1 year ago
➀ A 30 billion parameter LLM is demonstrated with a prototype inference device equipped with 16 IBM AIU NorthPole processors, achieving a system throughput of 28,356 tokens/second and a latency below 1 ms/token; ➁ NorthPole offers 72.7 times better energy efficiency and lower latency compared to GPUs at the lowest GPU delay; ➂ NorthPole architecture is inspired by the brain, optimized for AI inference, and demonstrates superior performance in LLM推理.
GPULLMenergy efficiency
about 1 year ago
➀ The Chinese Internet Investment Fund has invested in瀚博半导体,a high-tech semiconductor company specializing in high-end GPU chips and full-stack solutions; ➁ The investment signifies a new phase for瀚博半导体 in technology development, market expansion, and industrial upgrading; ➂ 钱军,the founder of瀚博半导体, brings over 25 years of experience in high-end chip design and has held positions in companies like LSI Logic, Cisco, and AMD.
GPUinvestmentsemiconductor