Nvidia announces Titan V with Volta GPU
Nvidia has announced a new card in its Titan series. Like the Tesla V100 accelerator, the Titan V is equipped with a GV100 GPU, which is based on the Volta architecture. The card has a suggested retail price of $3,000.
Nvidia CEO Jen-Hsun Huang presented the Titan V at the NIPS 2017 conference. He showed off the gold-colored card, which otherwise features the same design and vapor chamber cooler as its predecessor, the Titan Xp.
It is the first card for PCs with a GPU based on the Volta architecture, the GV100. Nvidia also uses this chip with a size of 815mm² and 21.1 billion transistors for its Tesla V100 accelerator. As with that card, Nvidia emphasizes deep learning performance. Thanks to the presence of 640 so-called tensor cores, the performance for training deep learning networks would be 110 tflops.
Compared to the Tesla V100, the Titan V has less memory: 12GB hbm2 versus 16GB hbm2. Also, the 3072-bit memory bus is less wide than that of the Tesla, which has a 4096-bit interface.
The Titan V is a dual slot card with three displayport and hdmi. The tdp is 250W and the card is powered via the 8+6-pin connector, with Nvidia recommending the use of at least a 600W power supply. The card will receive a suggested retail price of USD 3000 from Nvidia, converted and with VAT that is EUR 3,078. The company targets the professional consumer market with the Titan cards.
Nvidia Titan vs Tesla Specs | ||||||
Titan V | Tesla V100 (pci-e) |
Tesla P100 (pci-e) |
Titan XP | |||
Cuda cores | 5120 | 5120 | 3584 | 3840 | ||
Tensor cores | 640 | 640 | – | – | ||
Core clock sn. | 1200MHz | ? | ? | 1485MHz | ||
Boost clock speed | 1455MHz | 1370MHz | 1300MHz | 1582MHz | ||
Memory | 1.7Gbit/s hbm2 | 1.75Gbit/s hbm2 | 1.4Gbit/s hbm2 | 11.4Gbit/s gddr5x | ||
Memory interface | 3072-bit | 4096-bit | 4096-bit | 384-bit | ||
Mem Bandwidth | 653GB/s | 900GB/s | 720GB/s | 547GB/s | ||
Mem Quantity | 12GB | 16GB | 16GB | 12GB | ||
L2 cache | 4.5MB | 6MB | 4MB | 3MB | ||
single precision | 15 tflops | 14 tflops | 9.3 tflops | 12.1 tflops | ||
Double precision | 7.5 tflops? (1/2 rate) |
7 tflops (1/2 rate) |
4.7 tflops (1/2 rate) |
0.38 tflops (1/32 rate) |
||
Tensor Performance (Deep Learning) |
110 tflops | 112 tflops | AFTER | AFTER | ||
GPU | GV100 (815mm²) |
GV100 (815mm²) |
GP100 (610mm²) |
GP102 (471mm²) |
||
Transistors | 21.1 billion | 21.1 billion | 15.3 billion | 12 billion | ||
tdp | 250W | 250W | 250W | 250W | ||
form factor | pci-e | pci-e | pci-e | pci-e | ||
Cooling | Active | passive | passive | Active | ||
Production process | TSMC 12nm FFN | TSMC 12nm FFN | TSMC 16nm FinFET | TSMC 16nm FinFET | ||
Architecture | Volta | Volta | Pascal | Pascal | ||
Launch date | 07/12/2017 | Q3’17 | Q4’16 | 07/04/2017 | ||
MSRP | $2999 | ~$10000 | ~$6000 | $1299 |
Table compiled by AnandTech