Blackwell
PulseAugur coverage of Blackwell — every cluster mentioning Blackwell across labs, papers, and developer communities, ranked by signal.
15 day(s) with sentiment data
-
vLLM releases GLM-5.2 for NVIDIA Blackwell; Mixture of Agents 2.0 unveiled
The vLLM project has announced the availability of GLM-5.2 in NVFP4 format, optimized for NVIDIA's Blackwell architecture. This release enables efficient deployment of the GLM-5.2 model. Separately, Teknium introduced M…
-
Nvidia DGX Spark brings data center AI GPU to local desktops for $4,699
Nvidia has released the DGX Spark, a compact desktop computer featuring a full data center GPU, specifically the Blackwell architecture. This device is priced at $4,699 and is designed to bring powerful AI processing ca…
-
SK Hynix eyes fab expansion; diamond heat sinks to boost AI server cooling
SK Hynix is reportedly considering an expansion of its NAND wafer fab investment in Cheongju, South Korea, with plans to be announced by June 29th. Meanwhile, a report from CICC suggests that diamond heat sinks, due to …
-
AI Servers to Feature Diamond Heat Sinks and Liquid Cooling
A new thermal management solution for high-end AI servers is emerging, combining diamond heat sinks with full liquid cooling. This approach addresses the increasing power consumption and heat generation of GPUs like NVI…
-
ComfyUI users get custom Blackwell GPU wheels for SageAttention
A user has created and shared custom Linux wheels for SageAttention and FlashAttention, specifically tailored for Nvidia's Blackwell architecture GPUs like the RTX 5090 and RTX Pro 6000. These wheels are necessary becau…
-
New book details modern GPU programming for AI workloads
A new book titled "Modern GPU Programming for MLSys" aims to demystify high-performance GPU kernel development for machine learning systems. The book, originating from Carnegie Mellon University's Machine Learning Syste…
-
NVIDIA Blackwell platform dominates MLPerf Training 6.0 benchmarks
NVIDIA's Blackwell platform has set new records in the MLPerf Training 6.0 benchmarks, achieving the fastest times across all seven tests. The platform demonstrated strong scaling, with clusters of up to 8,192 GPUs show…
-
NVIDIA Blackwell platform dominates MLPerf Training 6.0 benchmarks · 4 sources tracked
NVIDIA's Blackwell platform has achieved top performance across all seven benchmarks in the MLPerf Training 6.0 industry standard tests. The platform demonstrated the fastest training times and enabled the largest-scale…
-
Nvidia Rubin GPU promises 10x cheaper tokens but with significant caveats
Nvidia has announced its Vera Rubin NVL72 GPU, promising up to a 10x reduction in cost per token compared to its Blackwell architecture. However, this significant cost saving is contingent on several factors, including …
-
Together AI benchmarks Blackwell hardware for agent infrastructure
Together AI has released benchmarks demonstrating the performance of their inference stack on NVIDIA's Blackwell hardware, showing a 31% increase in transactions per second compared to other open-source engines. This pe…
-
MiniMax AI releases open-weight M3 model with 1M context
MiniMax AI has released its new open-weight model, MiniMax M3, featuring a 1 million token context window and advanced capabilities. The model utilizes a novel sparse attention architecture called MSA, which includes de…
-
NVIDIA Blackwell Systems Lead New Agentic AI Benchmarks
NVIDIA has set new performance records on the first agentic AI benchmarks, AgentPerf and Agentic AI Benchmark. The company's GB300 NVL72 system, powered by Blackwell architecture, demonstrated up to a 20x performance le…
-
E-scooter founder raises $5M for space data centers
Orbital, a new company founded by e-scooter entrepreneur Euwyn Poon, has secured $5 million in seed funding to develop data centers in space. The company aims to leverage the growing demand for AI compute power by deplo…
-
Nvidia, Microsoft push AI agents to dedicated edge hardware
Nvidia and Microsoft are both investing in dedicated hardware for personal AI agents, aiming to shift computation from the cloud to edge devices. Nvidia's RTX Spark integrates CPU and GPU for local model execution, whil…
-
Tokens per Watt to Dictate 2026 GPU and Cooling Decisions
The primary constraint for AI compute in 2026 will shift from raw processing power to efficiency, specifically tokens per watt. This is because inference, which now accounts for the majority of AI compute spend, is fund…
-
Nvidia reportedly plans 2026 RTX 50 Super GPU refresh
Nvidia is reportedly planning to release its RTX 50 Super series of GPUs in 2026, according to recent leaks. The lineup is expected to include refreshed versions of existing cards like the RTX 5070 Super, RTX 5070 Ti Su…
-
US chip export loophole may grant China access to NVIDIA Blackwell
A potential loophole in US export restrictions could allow China access to NVIDIA's Blackwell chips, raising concerns about technology and geopolitics. Meanwhile, InnoGrit is developing high-capacity PCIe Gen6 SSDs desi…
-
New tool optimizes llama.cpp models with advanced NVFP4/MXFP6 quantization
A developer has released an advanced quantizer tool for llama.cpp, designed to create NVFP4 and MXFP6 GGUF models. This tool goes beyond basic quantization by evaluating various methods and incorporating custom techniqu…
-
AI compute verification may rely on board/software changes by 2028
Achieving compute verification for advanced AI systems within the next few years will likely require focusing on methods beyond die-level hardware modifications. The author suggests that board-level or software/firmware…
-
NVIDIA launches Vera Rubin AI platform and Nemotron 3 Ultra model
NVIDIA CEO Jensen Huang announced the new Vera Rubin AI platform, which is now in full production and utilizes high-bandwidth memory from Micron, SK Hynix, and Samsung. This platform is designed as a cabinet-level AI fa…