NVIDIA TITAN V
nvidia volta (GV100) 2017
Fields
id
accelerators/nvidia-titan-vVendor
nvidia
Architecture
volta (GV100)
Release year
2017
Process (nm)
12
Form factors
pcie
VRAM (GB)
12
Memory type
hbm2
Peak FLOPS
precisionothertflops110denseyesvendor_claimyessource_idnvidia-titan-v-product-pagebasisvendor-spec-sheetbasis_detailProduct page states more than 110 TeraFLOPS of deep-learning performance from 640 Tensor Cores, and the launch press release says 110 teraflops. Neither names the precision, so this is filed as other rather than guessed as fp16. It is a tensor-core peak, not a CUDA-core FP32 figure.
Unified memory
no
Consumer part
yes
notes
First-party spec table: Volta, 12 GB HBM2, 1455 MHz boost clock, 640 Tensor Cores, 5120 CUDA cores, 21.1 billion transistors, TSMC 12 nm FFN, launched December 2017 at $2,999. Bandwidth and board power are null because neither the product page nor the launch press release prints them, and an aggregator figure is not substituted. 12 GB of HBM2 holds a 7B Q4 with modest KV headroom and does not hold a 13B Q4. No NVLink is stated on the product page, so the host PCIe link is the only interconnect recorded.
Notes
First-party spec table: Volta, 12 GB HBM2, 1455 MHz boost clock, 640 Tensor Cores, 5120 CUDA cores, 21.1 billion transistors, TSMC 12 nm FFN, launched December 2017 at $2,999. Bandwidth and board power are null because neither the product page nor the launch press release prints them, and an aggregator figure is not substituted. 12 GB of HBM2 holds a 7B Q4 with modest KV headroom and does not hold a 13B Q4. No NVLink is stated on the product page, so the host PCIe link is the only interconnect recorded.