It's going well! As Norm said in the talk, Ironwood features AI-based ALU circuits and floor plan optimization. Here are some slides on the topic from a talk I gave at the NSF AI4EDA workshop at NeurIPS in Dec last year: https://
drive.google.com/file/d/1Gq8baK
6aUBwNSNXG5U7V8zJligr3gvHU/view?usp=drivesdk
… Last slide points to many papers.
INNOVATION
-
Ironwood AI-Based ALU Circuits and Floor Plan Optimization
By
–
-

Google Unveils TPUv7 Ironwood: 9216 Chips Reaching Zettaflop Scale
By
–
My Google colleagues Norm Jouppi & Sridhar Lakshmanamurthy gave a talk today at Hot Chips on TPUv7 ("Ironwood"). The TPUv7 system offers 9216 chips / pod (42.5 exaflops of fp8), but we can scale across many of these pods to provide multiple zettaflops.
-
NotebookLM Expands Support to Multiple Languages
By
–
Great to see new NotebookLM features that support many languages.
-
Open-Source Models Transformed Into High-EQ Characters
By
–
Today’s Lesson: Making Open-Source Models Fun By building tech to quickly transform any OSS model into a high-EQ Character, we can now ship better, more entertaining experiences faster than ever. Since testing our updated models, we’ve seen: 22% increase in time spent 13%
-
Grok Code Fast Now Available Through Poe API
By
–
Grok Code Fast is available through the Poe API on your favorite coding apps including RooCode, Cline, and KiloCode. You can also try the bot at https://
poe.com/Grok-Code-Fast
-1
… and across all platforms. (2/2) -

xAI Launches Grok Code Fast Model on Poe Platform
By
–
New on Poe: Grok Code Fast! This high-performance, cost-efficient model from xAI is optimized for agentic coding applications. It offers visible reasoning traces, strong steerability, and supports a 256k context window. (1/2)
-
Google DeepMind Releases New Capable Image Generation Model
By
–
@GoogleDeepMind has also released a new highly capable image generation model alongside nano-banana Try it on Replicate:
-

Google Nano Banana Model Launch Announcement
By
–
Have you tried it yet? http://
replicate.com/google/nano-ba
nana
… -
Image Generation AI: The Shifting Bar of Quality
By
–
The bar on image generation keeps shifting: 2019: We can make faces, but the hairline's kinda weird.
2022: Wow, realistic people, but look at the 6 fingers, ick.
2023: This person doesn't exist, but the text on their t-shirt makes no sense.
2024: I asked it for a new viewpoint, -
Trillion Labs Cuts LLM Training Time Using NVIDIA NeMo Curator
By
–
Trillion Labs used NVIDIA NeMo Curator to cut data prep from days to hours to train highly accurate LLMs in Korea. Explore the key results in advancing sovereign AI development: Reduced training time Improved language accuracy Increased cost and energy savings