Blackwell
PulseAugur coverage of Blackwell — every cluster mentioning Blackwell across labs, papers, and developer communities, ranked by signal.
- instance of Nvidia Rubin Gpu 90%
- instance of GDDR7 SDRAM 90%
- developed Vera Rubin NVL72 90%
- instance of Nvidia RTX Pro 6000 Blackwell Workstation Edition 90%
- instance of Newegg 90%
- instance of Nvidia RTX 5060 Ti 90%
- used by Arm Holdings 90%
- uses RTX Spark 90%
- instance of GeForce 50 series 90%
- competes with jalapeño 80%
- competes with Mikko Hyppönen 80%
- competes with Debian 70%
14 day(s) with sentiment data
Blackwell performance will be benchmarked against Jalapeño within 60 days
With OpenAI's Jalapeño chip now revealed and positioned as a direct competitor to NVIDIA's Blackwell, it is highly probable that independent benchmarks comparing the two will emerge. These benchmarks will be crucial for verifying OpenAI's performance claims and assessing the true competitive landscape.
OpenAI's Jalapeño chip to achieve >1.9 exaFLOPS in specific workloads
OpenAI's Jalapeño chip is claimed to offer 1.5-1.9x more work per watt and 1.7-3.6x lower latency. If these figures hold true for specific, optimized workloads, the chip could effectively surpass the 1.7 exaFLOPS mentioned in other reports, especially when considering its inference focus.
OpenAI leverages its own AI models for chip development
The report indicates that OpenAI used its GPT-Astra and Codex models to aid in writing and optimizing the low-level kernels for the Jalapeño chip. This suggests a growing trend of AI-assisted hardware design and development within leading AI labs.
-
Report: Billions in Nvidia AI chips reach China via illicit channels · 1 source tracked
A report from the Center for Advanced Defense Studies (C4ADS) details how Chinese firms are circumventing U.S. export restrictions to acquire billions of dollars worth of advanced Nvidia AI chips. The report identifies …
-
NVIDIA RTX 50-series laptop GPUs get unofficial power boost tool
A developer has created an experimental tool called NvpwrControl that allows users to increase the power limits of NVIDIA RTX 50-series laptop GPUs. This tool, available on GitHub, can potentially boost performance by u…
-
Mod enables Nvidia's RTX 50-series DLSS Multi Frame Generation on RTX 40-series GPUs
An unofficial mod has enabled Nvidia's DLSS Multi Frame Generation (MFG) feature on GeForce RTX 40-series GPUs, a capability previously exclusive to the upcoming RTX 50-series. Modder dashdogy released the mod, which ha…
-
Nvidia launches RTX Pro 5500 with 84GB VRAM, targeting AI workloads
Nvidia has introduced a new workstation graphics card, the RTX Pro 5500, which is based on the same GB202 silicon as the GeForce RTX 5090. This new card offers a significant increase in VRAM, featuring 84GB of GDDR7 mem…
-
AI Hardware Supply Chain: Focus on Third Derivative Enablers
SemiAnalysis highlights the critical, often overlooked, "third derivative" companies in the semiconductor supply chain that enable advanced AI hardware. These include material suppliers like Ajinomoto, which dominates t…
-
AI chip supply chain strained by ABF substrate bottleneck
The semiconductor industry faces a critical bottleneck in the supply of ABF substrates, essential components for high-performance CPUs, GPUs, and AI accelerators. Demand has surged due to the AI boom, with companies lik…
-
NVIDIA's Rubin and Feynman chips to unlock massive AI scaling
NVIDIA's upcoming Blackwell and Rubin chip architectures are poised to significantly accelerate AI development, with Rubin expected to offer a 4x reduction in training GPUs and a 10x decrease in inference costs for cert…
-
Together AI optimizes ThunderKittens for NVIDIA Vera Rubin Blackwell GPUs
Together AI has gained access to NVIDIA's Vera Rubin NVL72 platform, which is based on the Blackwell architecture. Their team has updated their ThunderKittens software to leverage new features of the Vera Rubin chip, sp…
-
New tool diagnoses LLM serving issues on NVIDIA Blackwell GPUs
A new tool called `blackwell-doctor` has been developed to help diagnose and resolve issues when serving large language models on NVIDIA Blackwell GPUs. The tool systematically tests various configurations, including di…
-
German scammer sells hollow RTX 5090 GPUs to two buyers
A scammer in Germany has defrauded two individuals out of thousands of dollars by selling them hollowed-out Nvidia RTX 5090 graphics cards. These cards, which lack essential internal components like silicon and memory c…
-
Perplexity Details GPU Embedding Stack for Efficient AI Search
Perplexity has detailed its GPU embedding stack, focusing on the infrastructure that serves its pplx-embed models. The company's engineering team highlighted how they optimized for both batch and online embedding worklo…
-
Nvidia offers RTX 50-series GPUs at MSRP in person at PAX West
Nvidia is making Founder's Edition RTX 50-series GPUs available at MSRP in person at PAX West. This initiative, previously seen at QuakeCon, aims to offer cards like the RTX 5090, RTX 5080, and RTX 5070 directly to cons…
-
Modders unlock Nvidia DLSS Multi-Frame Generation for RTX 40 series GPUs
Game modders have successfully bypassed Nvidia's attempts to restrict its DLSS Multi-Frame Generation feature to newer RTX 50 series GPUs. Despite Nvidia's claims that the feature requires specialized hardware only pres…
-
New FlashAttention-4 method boosts FP4 performance on Blackwell hardware
Researchers have developed a new method called Direct-P to optimize FlashAttention-4 for Blackwell's 4-bit floating-point (FP4) tensor cores, addressing performance bottlenecks caused by softmax conversion and on-chip d…
-
New RAED method improves experimental design under uncertainty
Researchers have introduced Resolution-Aware Experimental Design (RAED), a new method for selecting experiments that accounts for persistent nuisance uncertainty. RAED aims to minimize the structural candidate set while…
-
Nvidia RTX 5090 GPUs surge to $5,000 amid AI demand
Nvidia's high-end RTX 5090 gaming GPUs are experiencing dramatic price increases, with some models now costing upwards of $5,000, a significant jump from their original MSRP of $1,999. This surge is attributed to the hi…
-
OpenAI's 'Jalapeño' Accelerator Outperforms Nvidia Blackwell in Efficiency
SemiAnalysis has released details about OpenAI's new AI accelerator, codenamed "Jalapeño." This accelerator is reportedly more power-efficient than NVIDIA's Blackwell architecture, offering superior tokens per megawatt.…
-
LM Studio runtime update breaks LLM loading; rollback advised
A user encountered an issue where LM Studio silently failed to load local LLMs after an automatic runtime update. The error presented as a cryptic, large unsigned exit code, which was identified as a Windows NTSTATUS co…
-
Tech firms tackle AI costs with mixed results for consumers
Several tech companies are addressing the rising costs and complexities associated with AI, with some offering solutions that may inadvertently increase expenses for consumers. AWS is promoting its networking technology…
-
AI firms grapple with performance claims and rising costs · 1 source tracked
Multiple tech companies are facing challenges with AI development and deployment. NVIDIA and Cerebras are promoting performance metrics that may not be fully realized by customers, while Google is enforcing stricter mem…