RTX 4090 Tensor Core throughput increased from 165.6 to 331.6 TFLOPS FP16 after hidden AI compute limits were removed. Modifications remove software-level limits on certain Tensor Core operations without hardware changes or a new VBIOS. Gains apply to AI and compute workloads, not gaming.
Key points
RTX 4090 FP16 Tensor Core throughput: 165.6 → 331.6 TFLOPS
RTX 3090 Ti BF16 performance: 87.5 → 157.4 TFLOPS
Sources
MicroDuck shows waddle, sit and somersault at GTC Berlin
MicroDuck demonstrated waddle, sit and somersault motions at NVIDIA GTC Berlin. The tiny robots learned these moves in simulation before performing them in the real world.
Key points
Demonstration: MicroDuck waddles, sits and does somersaults.
Training: Moves were learned in simulation prior to real-world execution.
Sources
Prized Sandboxes – agent runs 100 subagents, today
Prized Sandboxes are available today and let an agent run 100 subagents, each on its own copy of your computer via the Prized CLI. The offering provides separate computers for each subagent.
Key points
Agent can run 100 subagents.
Prized Sandboxes are available today.
