Nvidia B200
PulseAugur coverage of Nvidia B200 — every cluster mentioning Nvidia B200 across labs, papers, and developer communities, ranked by signal.
- instance of graphics processing unit 90%
- instance of GB200 90%
- used by NVFP4 90%
- competes with Readonflow Team 80%
- competes with atom 80%
- competes with MI355x 80%
- competes with graphics card model series 80%
- competes with NVIDIA H100 70%
- used by vLLM 70%
- used by SGLang 70%
- competes with Kimi k3 70%
- uses CUDA 70%
17 day(s) with sentiment data
-
NVIDIA partners with finance giants to fund $500B+ AI infrastructure buildout
NVIDIA is partnering with major financial institutions including Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR to establish financing platforms. These platforms aim to mobilize over $500 billion in t…
-
NVIDIA and Wall Street partner to finance AI infrastructure with $500B+
NVIDIA CEO Jensen Huang announced a new initiative to finance AI infrastructure, positioning GPU computing power as an investable asset class. NVIDIA is partnering with major financial firms like Apollo, BlackRock, and …
-
NVIDIA releases NemotronLabs VoiceChat 11B for real-time, full-duplex AI conversations
NVIDIA has launched NemotronLabs VoiceChat 11B, an open-source, full-duplex speech-to-speech model designed for real-time conversational AI. This unified model integrates speech recognition, language understanding, and …
-
Nvidia B200 performance benchmark claimed to be surpassed by user
A user on Mastodon claims to have achieved and surpassed the LPU performance of a single Nvidia B200 chip. The user stated this was accomplished without adding any components or making significant modifications.
-
Pokee AI launches 28B model with 10M-token context for on-premise use
Pokee AI has released Pokee-Isaac 28B, a 28 billion parameter text-only foundation model designed for deployment within private customer boundaries. This model boasts a 10 million token context window, enabling it to ma…
-
AMD MI355X GPUs offer better performance per dollar for Kimi K3 model
A blog post from Wafer.ai details how they achieved better performance per dollar by running the Kimi K3 model on AMD's MI355X GPUs. Despite Kimi K3's massive 2.8T parameter size requiring significant VRAM, the MI355X, …
-
AMD MI355X kernel optimizations show 4x performance boost, still trails competitors
A recent kernel hackathon organized by AMD and the GPU_MODE community has led to a significant performance improvement for AMD's MI355X graphics card. The Readonflow Team's optimized kernels reportedly boosted end-to-en…
-
Emerging M3D Memory Tech Promises Major Energy Savings for LLM Serving
Researchers have developed LLMET, a cross-layer simulation framework to evaluate the impact of emerging monolithic 3D (M3D) memory technologies on the energy efficiency of Large Language Model (LLM) serving. The study i…
-
AMD MI355X vLLM performance beats Nvidia B200 on Kimi K2.5 model
AMD's MI355X graphics card has demonstrated superior performance over Nvidia's B200 in vLLM benchmarks for the Kimi K2.5 model, a significant achievement driven by community-developed kernels. This advancement stems fro…
-
Moonshot AI releases Kimi K3 with 2.8T open weights, but only 1.8% are active
Moonshot AI has released the Kimi K3 model with 2.8 trillion open weights, making it the largest open-weight model to date. However, only a small fraction, approximately 1.8%, of these parameters are actively used for p…
-
AMD MI355X performance boosted by community hackathon, rivals B200 on Kimi models · 6 sources tracked
AMD, in collaboration with GPU_MODE, has launched a $1.1 million kernel hackathon that has significantly improved the performance of its MI355X graphics card. The Readonflow Team's optimizations, focusing on MoE kernels…
-
Sol-Attn speeds up video generation with efficient sparse attention
Researchers have developed Sol-Attn, a new training-free sparse attention method designed to accelerate inference for video generation models. Unlike previous methods that struggle with efficiency and accuracy due to ri…
-
Marker 2 document converter achieves 5x throughput, beats competitors on benchmark
Datalab has released Marker 2, a significantly rewritten open-source document conversion pipeline. The new version boasts a 5x increase in throughput compared to MinerU, achieving 2.9 pages per second on a single Nvidia…
-
MiniMax AI nears NVIDIA B200 performance with AMD MI355X optimization
MiniMax AI has achieved near parity with NVIDIA's B200 performance on their Minimax M3 model, utilizing AMD's MI355X hardware. This advancement was facilitated by a full-stack co-optimization of AMD's ATOM and ATOMesh t…
-
LLM Fine-Tuning Frameworks: Unsloth, Axolotl, TRL, and LLaMA-Factory Compared
A comparison of four popular LLM fine-tuning frameworks—Unsloth, Axolotl, TRL, and LLaMA-Factory—highlights their differing approaches to optimizing speed, VRAM usage, and multi-GPU scaling. Unsloth focuses on kernel-le…
-
Nota AI releases 4-bit quantized Solar Open2 250B model for NVIDIA Blackwell
Nota AI has released a 4-bit quantized version of Upstage's Solar Open2 250B model, named Solar Open2 250B — Nota NVFP4. This new version utilizes Nota AI's proprietary quantization technology, specifically designed for…
-
New GLM-5.2 Vision Model Integrates MoonViT with 1M Context
A new vision-language model, baseten/GLM-5.2-Vision-NVFP4, has been released, integrating the MoonViT vision encoder with the GLM-5.2 reasoning model. This model achieves this integration by freezing the weights of both…
-
GMI Cloud unveils full-stack AI solutions at WAIC 2026
GMI Cloud showcased its comprehensive AI infrastructure solutions at WAIC 2026, highlighting its AI Cloud, MaaS, and Agentbox platforms. As a NVIDIA Cloud Partner, GMI Cloud offers GPU cloud services powered by high-per…
-
Cloud GPU rental guide for LLMs: Optimizing cost by model size
The optimal cloud GPU rental for running large language models (LLMs) in 2026 depends on the specific model size and workload, with a focus on tokens per dollar rather than hourly rates. For smaller models (7B-13B), bud…
-
VIDRAFT releases fully open-source Aether-7B-5Attn model
VIDRAFT has released Aether-7B-5Attn, a fully open-source foundation model under the Apache 2.0 license. Unlike many "open" models that only provide weights, Aether-7B-5Attn includes its architecture, training data reci…