Laughing Hyena
  • Home
  • Hyena Games
  • Esports
  • NFT Gaming
  • Crypto Trends
  • Game Reviews
  • Game Updates
  • GameFi Guides
  • Shop
Tag:

Blackwell

Nvidia CEO Jensen Huang Next to a B200 Node
Product Reviews

Nvidia RTX PRO 6000D (B40) Blackwell GPUs reportedly set to supersede banned H20 accelerators in China

by admin May 26, 2025



Following the ban of its Hopper H20 accelerations in China, Nvidia is reportedly planning on launching new Blackwell-based solutions at a lower price this year, per Reuters. With mass production anticipated by June, we can expect these solutions to be widely available in the Chinese market by Q3 or Q4. While technical details are still emerging, we can already discern some important details and specifications.

Due to stringent U.S. export policies targeting China, the Hopper family has largely been a cat-and-mouse chase between U.S. regulators and Nvidia. Even before their official debut, the flagship H100 and H200 accelerators were already subject to export bans. Nvidia introduced the H800 to circumvent these regulations, which eventually faced a similar fate in October 2023. The cut-down H20 served as Nvidia’s primary AI solution for the Chinese market in the interim until its recent ban under the current administration last month, which forced Nvidia to write off $5.5 billion in GPU supply.

Reuters reports that Nvidia’s follow-up to the H20 will be based on the Blackwell architecture, more specifically, the RTX Pro 6000D. Further clarification by tipster Jukanlosreve at X, citing a report from China’s GF Securities, suggests the RTX Pro 6000D will be dubbed B40 (likely a successor to the Ada Lovelace L40). Reuters classifies this as a server-class GPU which uses traditional GDDR7 memory instead of HBM, and notably avoids the use of TSMC’s CoWoS packaging technology, likely signaling at its monolithic nature.


You may like

Silicon possibilities and B40 pricing

There are two possibilities based on the available data. This GPU can either be based on datacenter grade GB1XX Blackwell or consumer-grade GB2XX Blackwell silicon. The former is unlikely, as it only features HBM controllers at the silicon level. If the B40 utilizes GB2XX dies, it would be a derivative of the GB202 chip (found in the RTX 5090 and RTX Pro Blackwell 6000) and would lack NVLink support. The report estimates the price of the B40 between $6,500 and $8,000, which is less than the H20 and comparable to Nvidia’s global RTX Pro 6000 workstation models.

The HGX H20 could be configured in an 8-GPU configuration, but without NVLink, the B40 would likely face challenges in multi-GPU setups. Nvidia’s latest RTX Pro Blackwell servers employ up to eight RTX Pro 6000 GPUs, connected via ConnectX-8 SuperNICs with integrated PCIe 6.0 switches, for GPU-to-GPU communication. This setup is likely what we’ll see for the B40, with scaling beyond eight GPUs expected to be handled by Nvidia’s Spectrum-X networking platform. Since details are scarce, this is just speculation on our part, so please don’t read it as gospel.

Follow Tom’s Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.

Get Tom’s Hardware’s best news and in-depth reviews, straight to your inbox.



Source link

May 26, 2025 0 comments
0 FacebookTwitterPinterestEmail
GB200 Grace Blackwell Superchip
Gaming Gear

DGX B200 Blackwell node sets world record, breaking over 1,000 TPS/user

by admin May 24, 2025



Nvidia has reportedly broken another AI world record, breaking the 1,000 tokens per second (TPS) barrier per user with Meta’s Llama 4 Maverick large language model, according to Artificial Analysis in a post on LinkedIn. This breakthrough was achieved with Nvidia’s latest DGX B200 node, which features eight Blackwell GPUs.

Nvidia outperformed the previous record holder, SambaNova, by 31%, achieving 1,038 TPS/user compared to AI chipmaker SambaNova’s prior record of 792 TPS/user. According to Artificial Analysis’s benchmark report, Nvidia and SambaNova are well ahead of everyone in this performance metric. Amazon and Groq achieved scores just shy of 300 TPS/user — the rest, Fireworks, Lambda Labs, Kluster.ai, CentML, Google Vertex, Together.ai, Deepinfra, Novita, and Azure, all achieved scores below 200 TPS/user.

Blackwell’s record-breaking result was achieved using a plethora of performance optimizations tailor-made to the Llama 4 Maverick architecture. Nvidia allegedly made extensive software optimizations using TensorRT and trained a speculative decoding draft model using Eagle-3 techniques, which are designed to accelerate inference in LLMs by predicting tokens ahead of time. These two optimizations alone achieved a 4x performance uplift compared to Blackwell’s best prior results.


You may like

Accuracy was also improved using FP8 data types (rather than BF16), Attention operations, and the Mixture of Experts AI technique that took the world by storm when it was first introduced with the DeepSeek R1 model. Nvidia also shared a variety of other optimizations its software engineers made to the CUDA kernel to optimize performance further, including techniques such as spatial partitioning and GEMM weight shuffling.

TPS/user is an AI performance metric that stands for tokens per second per user. Tokens are the foundation of LLM-powered software such as Copilot and ChatGPT; when you type a question into ChatGPT or Copilot, your individual words and characters are tokens. The LLM takes these tokens and outputs an answer based on those tokens according to the LLM’s programming.

The user part (of TPS/user) is aimed at single-user-focused benchmarking, rather than batching. This method of benchmarking is important for AI chatbot developers to create a better experience for people. The faster a GPU cluster can process tokens per second per user, the faster an AI chatbot will respond to you.

Follow Tom’s Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.

Get Tom’s Hardware’s best news and in-depth reviews, straight to your inbox.



Source link

May 24, 2025 0 comments
0 FacebookTwitterPinterestEmail

Categories

  • Crypto Trends (952)
  • Esports (722)
  • Game Reviews (674)
  • Game Updates (841)
  • GameFi Guides (944)
  • Gaming Gear (904)
  • NFT Gaming (927)
  • Product Reviews (894)
  • Uncategorized (1)

Recent Posts

  • KPop Demon Hunters Gets Perfectly Remade As A Lego Movie
  • Head of IRS Crypto Work Exits as U.S. Tax Changes Loom For Digital Assets
  • Bitcoin Miners Hit by $100M Tariffs
  • Get an Extremely Close-Up Look at Tobey Maguire’s Spider-Man Costume
  • Jerome Powell rate-cut hint jolts Ethereum price toward 2021 ATH territory

Recent Posts

  • KPop Demon Hunters Gets Perfectly Remade As A Lego Movie

    August 22, 2025
  • Head of IRS Crypto Work Exits as U.S. Tax Changes Loom For Digital Assets

    August 22, 2025
  • Bitcoin Miners Hit by $100M Tariffs

    August 22, 2025
  • Get an Extremely Close-Up Look at Tobey Maguire’s Spider-Man Costume

    August 22, 2025
  • Jerome Powell rate-cut hint jolts Ethereum price toward 2021 ATH territory

    August 22, 2025

Newsletter

Subscribe my Newsletter for new blog posts, tips & new photos. Let's stay updated!

About me

Welcome to Laughinghyena.io, your ultimate destination for the latest in blockchain gaming and gaming products. We’re passionate about the future of gaming, where decentralized technology empowers players to own, trade, and thrive in virtual worlds.

Recent Posts

  • KPop Demon Hunters Gets Perfectly Remade As A Lego Movie

    August 22, 2025
  • Head of IRS Crypto Work Exits as U.S. Tax Changes Loom For Digital Assets

    August 22, 2025

Newsletter

Subscribe my Newsletter for new blog posts, tips & new photos. Let's stay updated!

@2025 laughinghyena- All Right Reserved. Designed and Developed by Pro


Back To Top
Laughing Hyena
  • Home
  • Hyena Games
  • Esports
  • NFT Gaming
  • Crypto Trends
  • Game Reviews
  • Game Updates
  • GameFi Guides
  • Shop

Shopping Cart

Close

No products in the cart.

Close