RTX5090 Turbo Edition
-
Detail
Architecture: Blackwell, GB202-300 core, TSMC 4N technology; fifth-generation tensor core, fourth-generation optical tracking core
CUDA core: 21760
Tensor core: 680 (5th generation)
Light chasing core: 170 (4th generation)
Graphics memory: 32GB GDDR7, 512bit bit width, graphics memory bandwidth 1792GB/s, no ECC
Boost frequency: 2407MHz
Computing power
FP4: 13.2 PFLOPS
FP8: 6.6 PFLOPS
BF16/FP16: 3.3 PFLOPS
TF32: 770 TFLOPS
FP32: 132 TFLOPS
Optical tracking power: 1100 TFLOPS
Interconnection: PCIe5.0 x16, bandwidth 128GB/s; no NVLink
Power consumption TDP: 575W, 16Pin PCIe5.0 power supply
Form: Double-slot turbine blows heat, and heat is directly discharged from the chassis, which is suitable for multi-card deployment of rack servers.
MIG: MIG multi-instance cutting is not supported
Video output: 3×DP2.1a + 1×HDMI2.1
Media engine: NVENC/NVDEC, supports video encoding and decoding
Software: CUDA Ecology, supports DLSS4
-
Customer ReviewsNo comments