Unverified50% confidenceFactExact time
This training ran on a first-generation inference-specific NPU released in 2025, using 32 chips, and is not the latest hardware.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/14/2026
First Seen
Valid until: 1/12/2027
Sources
Related Entities
Related Claims
UnverifiedAlibaba's T-Head announced the PPU (True 5V900) training and inference chip, with the company claiming performance reaches three times that of the previous generation; mass production and commercial sales are planned for Q1 2027.69% similarUnverifiedVIDRAFT's VKORE software acceleration layer, without changing the chip, both increases NPU inference performance several times over and makes Transformer LoRA training run on this kind of inference-specific hardware, with numerical precision aligned with standard PyTorch at 0.9999.67% similarUnverifiedVIDRAFT performed LoRA training on standard Transformer blocks across all 32 chips of the Korean NPU server, achieving 0.9999 consistency with standard PyTorch training.67% similarUnverifiedThe 16-series Turing architecture introduced first-generation Tensor Cores supporting FP16 mixed-precision training, while the 30-series Ampere architecture upgraded to third-generation Tensor Cores and added hardware acceleration support for BF16 and TF32.66% similarUnverifiedIn a paper published by NVIDIA in 2024, the Nemotron model trained on 98% synthetic data was shown to reach a level comparable to models trained on purely human-generated data across multiple benchmarks.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1026571API
curl https://kongchang.com/api/v1/knowledge/claims/1026571MCP
get_claim(id=1026571)