PRO6000PCLE Server Edition
-
Detail
Architecture: Blackwell, fifth-generation tensor core, fourth-generation optical tracking core
CUDA core: 24064
Tensor core: 752 (5th generation)
Light chasing core: 188 (4th generation)
Display memory: 96GB GDDR7 (with ECC error correction), 512bit width, bandwidth 1597GB/s
Commetic power (peak)
FP4: 4 PFLOPS
FP8: 2 PFLOPS
BF16/FP16: 1 PFLOPS
TF32: 234 TFLOPS
FP32: 120 TFLOPS
Optical tracking performance: 355 TFLOPS
Video encoding and decoding: 4×NVENC, 4×NVDEC, 4×NVJPEG
Interconnection: PCIe 5.0 x16, two-way 128GB/s, no NVLink
Power consumption: 400~600W configurable, upper limit 600W
Form: double-slot full-length FHFL, passive heat dissipation (server air cooling / liquid cooling), PCIe board, no HGX carrier board required
MIG: Support, up to 4 instances, single instance 24GB
Display output: 4×DP2.1 (server version is disabled by default)
Others: Support confidential computing, secure startup; CUDA ecological compatibility
-
Customer ReviewsNo comments