NVIDIA Quadro RTX 6000 (Turing, 24 GB)
nvidia turing (TU102) 2018
Fields
id
accelerators/nvidia-quadro-rtx-6000Vendor
nvidia
Architecture
turing (TU102)
Release year
2018
Process (nm)
12
Form factors
pcieworkstation
VRAM (GB)
24
Memory type
gddr6
Memory bus (bit)
384
Memory bandwidth (GB/s)
672
Bandwidth basis
Datasheet prints 'Memory Bandwidth 672 GB/s' next to 'Memory Interface 384-bit' and '24 GB GDDR6'. Arithmetic checks: 14 Gbps x 384 / 8 = 672 GB/s.
Peak FLOPS
precisionfp32tflops16.3denseyesvendor_claimyessource_idnv-quadro-rtx-6000-datasheetbasisvendor-spec-sheetbasis_detailDatasheet 'Single-Precision Performance 16.3 TFLOPS', footnote-free peak based on GPU boost. FP32 has no 2:4 sparse path, so dense:true.unitTFLOPS
TDP (W)
260
Unified memory
no
Consumer part
no
notes
USED-MARKET WORKSTATION CARD. Predecessor of the Ampere RTX A6000 and the card people mean when a used listing says 'RTX 6000' without 'Ada'. 4608 CUDA cores, 576 Tensor cores, 72 RT cores, ECC, NVLink (two cards, 48 GB combined, only if the application supports it). Total board power 260 W, active dual-slot, PCIe 3.0 x16. Do not confuse with nvidia-rtx-6000-ada (48 GB, Ada, 960 GB/s) or nvidia-rtx-pro-6000-blackwell (96 GB).
Notes
USED-MARKET WORKSTATION CARD. Predecessor of the Ampere RTX A6000 and the card people mean when a used listing says 'RTX 6000' without 'Ada'. 4608 CUDA cores, 576 Tensor cores, 72 RT cores, ECC, NVLink (two cards, 48 GB combined, only if the application supports it). Total board power 260 W, active dual-slot, PCIe 3.0 x16. Do not confuse with nvidia-rtx-6000-ada (48 GB, Ada, 960 GB/s) or nvidia-rtx-pro-6000-blackwell (96 GB).