H20 GPUs

Extracted Attributes

• Memory

  96GB HBM3

• L2 Cache

  60 MB per chip

• Developer

  Nvidia

• Interface

  PCIe 5.0

• CUDA Cores

  14,592

• Interconnect

  NVLink

• Market Focus

  China (designed to comply with US export restrictions)

• Primary Purpose

  AI inference, large-scale model computation, cloud applications

• FP16 Performance

  148 TFLOPS

• FP32 Performance

  44 TFLOPS

• FP64 Performance

  1 TFLOPS

• INT8 Performance

  296 TFLOPS

• Memory Bandwidth

  4.0 TB/s

• NVLink Bandwidth

  900 GB/s

• TF32 Performance

  74 TFLOPS

• Microarchitecture

  Hopper

• HGX Configurations

  8-Way

• Performance vs H100

  41% fewer cores, 28% lower performance compared to top H100 configurations

• Estimated Price (Chip)

  $12,000 - $15,000 USD

• Multi-Instance GPUs (MIG)

  Up to 7

• Thermal Design Power (TDP)

  400W

• Estimated Price (8-chip Server)

  1.4 million yuan

Timeline

• Nvidia's Hopper microarchitecture, which H20 GPUs are based on, was leaked. (Source: Wikipedia)

  2019-11

• The Hopper microarchitecture was officially revealed by Nvidia. (Source: Wikipedia)

  2022-03

• US export restrictions were imposed, leading Nvidia to develop H20 GPUs specifically for the Chinese market. (Source: CNN)

  2023-Q4

• Small batches of H20 products were expected to reach important clients in China. (Source: TechPowerUp)

  2024-Q1

• The H20 chip was effectively banned by Washington, impacting its availability. (Source: CNN)

  2024-04

• A wider release of H20 products in China was anticipated. (Source: TechPowerUp)

  2024-Q2

• Mass production of H20 GPUs was expected to begin around Spring. (Source: TechPowerUp)

  2024-03

• Nvidia announced plans to resume sales of H20 GPUs to China, expecting to receive necessary licenses soon. (Source: CNN, Reuters)

  2025-07-15

Hopper (microarchitecture)

Hopper is a graphics processing unit (GPU) microarchitecture developed by Nvidia. It is designed for datacenters and is used alongside the Lovelace microarchitecture. It is the latest generation of the line of products formerly branded as Nvidia Tesla, now Nvidia Data Centre GPUs. Named for computer scientist and United States Navy rear admiral Grace Hopper, the Hopper architecture was leaked in November 2019 and officially revealed in March 2022. It improves upon its predecessors, the Turing and Ampere microarchitectures, featuring a new streaming multiprocessor, a faster memory subsystem, and a transformer acceleration engine.

Web Search Results

• NVIDIA H20: The Next-Generation AI Inference GPU

  The NVIDIA H20 is a data center GPU based on the Hopper architecture, specifically designed for AI inference, large-scale model computation, and cloud applications. It inherits key technologies from the H100 but is optimized for power efficiency and cost-effectiveness, making it an ideal choice for enterprise AI applications, especially for deploying large language models (LLMs) and AI inference tasks in cloud environments. 🔍 Detailed H20 Specifications [...] The H20 is built on the Hopper architecture and features 14,592 CUDA cores. It integrates Tensor Cores optimized for AI workloads and supports the Transformer Engine, enabling highly efficient deep learning acceleration. For memory, the H20 is equipped with 96GB of HBM3 memory with an ultra-high bandwidth of 4.0TB/s, significantly improving data transfer speeds. It supports NVLink for multi-GPU interconnect and uses the PCIe 5.0 interface. [...] • H200 is designed for ultra-large-scale AI training, such as next-generation large models like GPT-5. 💡 Conclusion: H20 is the Best Choice for AI Inference & Cloud Computing The NVIDIA H20 delivers H100-class AI performance with lower power consumption and a more cost-effective approach, making it perfect for cloud-based AI inference, virtual assistants, medical AI, autonomous driving, and financial AI applications. 📩 Looking to purchase AI GPUs or solutions? Contact us now!

• NVIDIA's China-Compliant H20 GPU Has 41% Fewer Cores & 28 ...

  Officially, the NVIDIA Hopper H20 GPU was designed for the HGX platform with 96 GB of HBM3 memory, offering 4.0 TB/s of bandwidth versus the 8 TB/s bandwidth offered by the top solution. This chip featured 296 TFLOPS of INT8, 148 TFLOPS of FP16, 74 TFLOPS of TF32, 44 TFLOPS of FP32, and just 1 TFLOPS of FP64 performance. Other specs include support for up to 7 multi-instance GPUs, 60 MB L2 cache per chip, 8-Way HGX configurations, NVLINK bandwidth of 900 GB/s, and a 400W TDP. [...] NVIDIA never disclosed the GPU core count but based on a recent Geekbench 6 entry, it looks like we finally know what the chip has to offer. According to the data, NVIDIA's Hopper H20 AI GPU for China features just 78 SMs out of the total 144 SMs for the H100 GPU. The fastest H100 GPUs are equipped with 114 and 132 SMs so that's already a 41% drop in core count versus the top dies that are accessible to everyone else besides China. Furthermore, the GPU comes with 96 GB of HBM3 memory which is [...] Published Time: 2024-07-09T08:40:03+00:00 NVIDIA's China-Compliant H20 GPU Has 41% Fewer Cores & 28% Lower Performance Versus Top Hopper H100 Config =============== Menu) Home News - [x] Hardware All Hardware Intel AMD NVIDIA TSMC - [x] Gaming All Gaming Sony Microsoft Electronic Arts Epic Games Ubisoft - [x] Mobile All Mobile Apple Samsung Android iOS

• NVIDIA Readying H20 AI GPU for Chinese Market | TechPowerUp

  Reuters report suggests that Team Green is taking pre-orders for a refreshed "Hopper" GPU—the latest China-specific flagship is called "HGX H20." NVIDIA web presences have not been updated with this new model, as well as Ada Lovelace-based L20 PCIe and L2 PCIe GPUs. Huawei's competing Ascend 910B is said to be slightly more performant in "some areas"—when compared to the H20—according to insiders within the distribution network. [...] "distributors are offering H20 servers, which are pre-configured with eight of the AI chips, for 1.4 million yuan. By comparison, servers that used eight of the H800 chips were sold at around 2 million yuan when they were launched a year ago." Small batches of H20 products are expected to reach important clients within the first quarter of 2024, followed by a wider release in Q2. It is believed that mass production will begin around Spring time. [...] The leakers reckon that NVIDIA's mainland distributors will be selling H20 models within a price range of $12,000 - $15,000—Huawei's locally developed Ascend 910B is priced at 120,000 RMB (~$16,900). One Reuters source stated that: "some distributors have started advertising the (NVIDIA H20) chips with a significant markup to the lower end of that range at about 110,000 yuan ($15,320). The report suggests that NVIDIA refused to comment on this situation. Another insider claimed that:

• Nvidia's resumption of AI chips to China is part of rare earths talks ...

  "The H20 is a powerful chip that, according to our bipartisan investigation, played a significant role in the rise of PRC AI companies like DeepSeek," Moolenaar said, referring to a Chinese startup that claims to have built AI models at a fraction of the cost paid by U.S. firms such as OpenAI. "It is crucial that the U.S. maintain its lead and keep advanced AI out of the hands of the CCP." [...] But critically, H20 chips work with Nvidia's software tools, which have become a de facto standard in the global AI industry. CEO Jensen Huang, who is visiting Beijing and set to speak at an event on Wednesday, has argued that Nvidia's leadership position could slip away if the company cannot sell to Chinese developers being courted by Huawei Technologies with chips produced in China. [...] Nvidia's plan to resume sales has set off a scramble at Chinese firms to buy H20 chips, two sources told Reuters. The chips that Nvidia will resume selling are the best it can legally offer in China but lack much of the computing power of the versions for sale outside of China because of previous restrictions put in place by Trump's first administration and then President Joe Biden's administration.

• Nvidia says it will restart sales of a key AI chip to China, in a ... - CNN

  The world’s most valuable firm is filing applications with the US government to resume sales to China of the H20 graphics processing unit (GPU), and expects to get the licenses soon, Nvidia said in a statement. Ad Feedback “The US government has assured NVIDIA that licenses will be granted, and NVIDIA hopes to start deliveries soon,” it said. [...] Nvidia (NVDA) said it plans to resume sales of its H20 artificial intelligence chip to China, days after its CEO, who is visiting Beijing, met US President Donald Trump. Nvidia’s AI chips have been a key focus of US export controls designed to keep the most advanced chips out of Chinese hands amid national security concerns, restrictions that the US-listed company said would cut its revenue by $15 billion. [...] The H20 chip was developed specifically for the Chinese market after US export restrictions were imposed on national security grounds in late 2023. The AI chip was Nvidia’s most powerful legally available product in China until it was effectively banned by Washington in April. What do you think?Leave a comment View Comments