엔비디아가 직접 딥시크 최적화 시켜본 결과 ㅎㄷㄷㄷㄷㄷ

이미지 텍스트 확인

NVIDIA AI
Developer
“Iola
@NVIDIAAIDev
Introducing DeepSeek-R1 optimizations for Blackwell, delivering 25x
more revenueat2Oxlower cost pertoken, compared with NVIDIA H1OO
justfour weeks ago
Fueled by TensorRT DeepSeek optimizations for our Blackwell
architecture, including FP4 performance with state-of-the-art
production accuracyit scored 99.89 of FP8 on MMLU general
intelligence benchmark
FP4-optimized DeepSeek checkpoint now available on @huggingface:
huggingface colnvidia/DeepSee.
25X Higher DeepSeek-R1 Inference Throughput
20X Lower Cost
Output TokenslSecond
25,000
1.OOx
21,088
20.000
통
5,000
올
통
O,50x
돌
1OOOO
통
훌
5,899
5.000
14gg
844
OOOx
HIOO
HOO
H2oO
B200
January 2025
January 2025
February 2025
February 2025
8.49 AM
Feb 25,2025
197.8K Views