Tensor Cores

نوى Tensor Cores

وحدات حوسبية متخصصة في معالجات NVIDIA (بدءاً من Volta) تنفّذ عمليات ضرب المصفوفات بصيغة FP16 مع تجميع النتائج بصيغة FP32 بسرعة فائقة.

Specialized compute units in NVIDIA GPUs (starting with Volta) that perform FP16 matrix multiplications with FP32 accumulation at very high throughput.

Also translated asنوى التنسور

First appears in this corpus in: Mixed Precision Training (2018)

Appears in these papers

FlashAttention-2: Faster Attention with Better Parallelism and Work Partitioning2023 in the sky ✦
FlashAttention-2: Faster Attention with Better Parallelism and Work Partitioning2023 in the sky ✦
Mixed Precision Training2018 in the sky ✦
Mixed Precision Training2018 in the sky ✦