➀ The NVIDIA GeForce RTX 5090 will use the GB202 'Blackwell' GPU, 22% larger than the AD102 'Ada Lovelace' GPU in the RTX 4090; ➁ The GB202 GPU has a die size of 744mm², making it 22% larger than the AD102 GPU in the RTX 4090; ➂ The RTX 5090 is expected to have 20,000+ CUDA cores, 32GB of next-gen GDDR7 memory, and up to 2.0TB/sec memory bandwidth.
Recent #Blackwell news in the semiconductor industry
➀ NVIDIA has confirmed that Blackwell GPUs are in full production despite cooling challenges; ➁ The company's AI chips have significantly contributed to its success; ➂ NVIDIA's AI data center business accounted for a staggering $30.7 billion of its quarterly earnings.
➀ NVIDIA's Blackwell AI GPUs are 'sold out' for the next 12 months due to high demand; ➁ The GPUs are in high demand among major tech companies like Google, Meta, and Microsoft; ➂ Despite the delay in production, Blackwell is expected to deliver significant performance improvements for AI training and inference.
➀ OpenAI receives one of the first engineering builds of NVIDIA's new Blackwell DGX B200 AI system; ➁ The system features 8 x B200 AI GPUs with up to 1.4TB of HBM3 memory and 72 petaFLOPS of training performance; ➂ OpenAI expresses excitement about the new system and its potential to accelerate AI compute.
➀ SK hynix has begun mass production of the world's first 12-layer HBM3E memory with a capacity of up to 36GB and a bandwidth of 9.6Gbps; ➁ The new memory is designed for AI GPUs and is set to be supplied to NVIDIA within 12 months; ➂ SK hynix aims to maintain its leadership in AI memory with the introduction of this new technology.
➀ NVIDIA's Blackwell AI GPUs are facing significant design issues requiring a major redesign. ➁ The new B200A AI GPU is being developed to address these issues and is scheduled for release in the second half of 2025. ➂ The B200A will feature a B102 die, 4 stacks of HBM, and will be available in 700W and 1000W HGX form factors with up to 144GB of HBM3E memory.
➀ NVIDIA recently unveiled its RTX Pro 6000 Blackwell graphics cards with impressive specifications for modern workstations and servers.
➁ The RTX Pro 6000 Blackwell Workstation Edition and RTX 6000 Blackwell Max-Q Workstation Edition models are priced over $8,000.
➂ The RTX Pro 6000 Blackwell features 24,064 CUDA cores, 96GB of GDDR7 memory, and a 600W TDP, while the Max-Q version has a 300W TDP.
➀ The Nvidia GeForce RTX 5070 Founders Edition offers a 20% performance increase over its predecessor and a $549 MSRP.
➁ Concerns over retail availability and pricing, with potential for significantly higher prices in the short term.
➂ AMD's RX 9070 launch scheduled for the same day, posing competition and availability challenges.
➀ Asus US web store and Best Buy US have announced the pricing for their Blackwell laptops, starting from $1,899;
➁ The ROG Strix Scar 16 with an RTX 5080 GPU and an Intel Core Ultra 9 275HX CPU is priced at $3,199.99;
➂ The ROG Strix Scar 18 with an RTX 5090 GPU and the same processor is priced at $4,299.99.
➀ NVIDIA unveiled the GeForce RTX 50 series, including the RTX 5090, RTX 5080, RTX 5070 Ti, and RTX 5070, with the RTX 5090 offering 170 Streaming Multiprocessors and 32GB of GDDR7 memory. The series boasts significant performance improvements over the RTX 4090. ➁ NVIDIA introduced DLSS 4, RTX Neural Shaders, and RTX Neural Face, enhancing image quality and performance. ➂ The RTX 50 series also features RTX Mega Geometry for improved ray-tracing capabilities. The GeForce RTX 5090 is set to be released this month, with laptops following in March.
➀ Nvidia's RTX 5080 is rumored to launch on January 21; ➁ Expected to debut at CES with sales embargo lifting two weeks later; ➂ Features 84 SMs or 10,752 CUDA cores, 16GB of GDDR7 memory.
➀ The upcoming GeForce RTX 5090 is rumored to have a 600W TGP, but recent leaks suggest a slight decrease; ➁ The power rating of the current GeForce RTX 40 series is provided for comparison; ➂ The GeForce RTX 5090 is expected to have 21,760 CUDA cores and 32GB of GDDR7 memory.
➀ Nvidia reported a 94% year-over-year revenue increase to $35.1 billion, with a profit of $19.3 billion; ➁ Revenue growth for the current quarter is forecasted at 69.5% year-over-year; ➂ Datacentre sales grew by 112% year-over-year, accounting for $30.77 billion of total revenue.
➀ Leaked RTX 50 mobile GPU lineup revealed; ➁ Flagship RTX 5090M rumored to use GB203 GPU; ➂ No increase in memory capacity over last generation.
➀ The Wall Street is focusing on NVIDIA's Blackwell production increase due to the influx of capital into AI-related investments. Morgan Stanley has released another report on NVIDIA's latest GPU architecture, revealing more information about the exact reasons for the production bottleneck in the summer. ➁ NVIDIA confirmed the rumors about minor defects in the Blackwell design during the recent earnings call, but assured investors that some minor modifications to the photomask can solve the low yield problem caused by it. ➂ Morgan Stanley believes that the low yield problem was only discovered in the encapsulation stage, resulting in the scrapping of不合格 products, leading to losses in CoWoS and HBM3e. These products were already in short supply, which further exacerbated NVIDIA's supply issues.
➀ NVIDIA is expected to launch the GeForce RTX 5090 soon; ➁ Jensen Huang's CES keynote is likely to unveil the new GPU; ➂ The GeForce RTX 5090 may feature a GB202-300-A1 Blackwell GPU with 21,760 CUDA cores and 32GB GDDR7 memory.
➀ Supermicro introduces its datacenter-scale liquid cooling solution for Nvidia Blackwell servers; ➁ The solution supports up to 96 B200 GPUs per rack and includes advanced cooling components; ➂ The technology aims at ultra-dense AI servers and features rapid deployment capability.
➀ NVIDIA's next-generation Blackwell AI chips are delayed due to design flaws, affecting major clients like Meta, Google, and Microsoft. ➁ The production was initially planned for Q3 2024 but was halted by TSMC engineers discovering issues. ➂ Despite initial optimism, the delay is more significant, impacting the deployment plans of companies like Supermicro.
1. NVIDIA's next-generation GeForce RTX 50 series may be announced at CES in January, according to a prominent leaker. 2. The GeForce RTX 5090 and 5080 are expected to be based on the Blackwell architecture and could feature GDDR7 memory. 3. The launch timing deviates from previous patterns, where NVIDIA typically unveiled new GPU series in September.