Google Launches DiffusionGemma: Text Generation Speed Quadrupled, Targeting On-Device Real-Time Applications

Google has released DiffusionGemma, an open-source experimental model that disrupts the conventional token-by-token generation paradigm by generating 256 tokens in parallel with each forward pass. In real-world tests on H100 GPUs, it achieves over 1,000 tokens per second, delivering up to a fourfold speedup. The model activates only 3.8 billion parameters and, when quantized, can run on consumer-grade GPUs with 18GB of VRAM, offering developers a new technical pathway for low-latency, local workflow scenarios.

wallstreetcnJun 11 03:40

Taiwan Semiconductor faces capacity shortages; Google's TPU reportedly introduced Samsung as a contract manufacturer for the first time

Google is in talks with Samsung to produce I/O chips for its tenth-generation TPU, marking the first time it has outsourced part of TPU manufacturing away from Taiwan Semiconductor. Affected by tight capacity constraints at Taiwan Semiconductor's advanced nodes, Google is diversifying its supply chain, while the core compute engine will still be manufactured using Taiwan Semiconductor’s 1.4-nanometer process.

wallstreetcnJun 11 22:40

Intel's Stock Could Push Even Higher Thanks to This Subtle Dynamic, BofA Says

MarketWatch36minutes ago

Waymo Readies First National Ads as Rivals and Critics Proliferate -- WSJ

WSJ01:00

Waymo Launches Premier Subscription Tier for $29.99 a Month, Starting in Select Cities

CNBC00:57

Express News | S&P Global Ratings assigned Google an 'AA+' credit rating with a 'stable' outlook.

Breakings00:53