Fable 5
Jul 7, 2026

Fable 5 Tops KernelBench-Mega With 18.71x Speedup and the First Single-Launch Megakernel

Fable 5 has taken the top spot in the latest KernelBench-Mega benchmark by delivering an 18.71x speedup on RTX PRO 6000 using a fully hand-written CUDA kernel. The reported result places it well ahead of Claude Opus 4.8 at 14.4x, GPT-5.5 at 4.34x, and Sonnet 5 at 4.0x. The tested workload was 02_kimi_linear_decode, a Kimi-Linear W4A16 mixed decoding task with 4-bit weights and bf16 activations, under a strict setup that allowed only one autonomous session and a three-hour wall-clock limit. What makes the result stand out is not only the score but the implementation: according to the report, Fable 5 is the first model in KernelBench-Mega to produce a true end-to-end megakernel, compressing the inference path into a single GPU kernel launch per decoded token. Anthropic co-founder Jack Clark said the development could mark the beginning of a recursive self-improvement loop, arguing that once models can optimize the low-level systems used to train and run future models, the feedback cycle may accelerate substantially.

1020
Fable 5 Tops KernelBench-Mega With 18.71x Speedup and the First Single-Launch Megakernel
NVIDIA
Jul 1, 2026

SemiAnalysis's Dual Assessment: NVIDIA Rubin Revenue May Beat by 20%, But Flagship Chip Shrinks by Half

Semiconductor research firm SemiAnalysis published two contrasting reports on NVIDIA. On the positive side, the resolution of HBM4 supply shortages and front-end wafer preparations could boost NVIDIA's data center computing revenue in H2 FY2027 by about 20% above consensus. On the negative side, the original 4-chip Rubin Ultra, unveiled at GTC 2026, was canceled just three months later; the replacement design is half the size, halving performance. Additionally, the research noted that Anthropic and other clients are adopting multi-platform architectures (Google TPU, Amazon Trainium), gradually eroding NVIDIA's CUDA moat.

1910
SemiAnalysis's Dual Assessment: NVIDIA Rubin Revenue May Beat by 20%, But Flagship Chip Shrinks by Half
Nvidia
Jun 30, 2026

SemiAnalysis: Nvidia's Rubin Platform Faces Both Optimistic Revenue Outlook and Flagship Downsizing

Semiconductor research firm SemiAnalysis has released two contrasting assessments of Nvidia's Rubin platform. On the bullish side, its Accelerator Model predicts Nvidia's data center computing revenue for the second half of fiscal 2027 will exceed Wall Street consensus by approximately 20%, citing resolution of HBM4 supply issues and sufficient front-end wafer capacity. On the bearish side, the original 4-chip Rubin Ultra design was canceled about three months after its GTC 2026 unveiling, replaced by a half-sized version with halved performance due to advanced packaging difficulties. SemiAnalysis also highlights that its supply-chain-based model differs from traditional analysts' conservative forecasts, and notes that Anthropic's multi-platform architecture (TPU, Trainium, GPU) is slowly eroding Nvidia's CUDA moat.

2010
SemiAnalysis: Nvidia's Rubin Platform Faces Both Optimistic Revenue Outlook and Flagship Downsizing
NVIDIA
Jun 30, 2026

SemiAnalysis: NVIDIA Rubin Platform Revenue Estimates Revised Up 20%, but Flagship Product Shrinkage and CUDA Moat Under Pressure

Semiconductor research firm SemiAnalysis published two contrasting assessments on NVIDIA's Rubin platform. On the bullish side, HBM4 supply constraints have been resolved and front-end wafer capacity is secured, leading to a projection that data center computing revenue in H2 FY2027 could exceed Wall Street consensus by ~20%. On the bearish side, the original 4-chip Rubin Ultra was canceled about three months after its GTC 2026 unveiling, replaced by a halved version with reduced performance. Furthermore, SemiAnalysis warns that the CUDA moat is slowly eroding, citing Anthropic's multi-platform architecture comprising Google TPU, Amazon Trainium, and NVIDIA GPUs. These conflicting signals paint a dual narrative of execution strength versus technological vulnerability.

1980
SemiAnalysis: NVIDIA Rubin Platform Revenue Estimates Revised Up 20%, but Flagship Product Shrinkage and CUDA Moat Under Pressure
Total 23 items
Go topage