Developers: | Nvidia |
Date of the premiere of the system: | 2017/05/11 |
Technology: | Processors |
Content |
Volta is architecture of GPU.
The processor on the Volta platform – GPU for data centers of Tesla V100 provides the high speed and scalability of training and data output of neural networks, accelerates productive and graphic calculations.
Each GPU Nvidia V100 includes 21 billion transistors (providing performance tasks in deep learning, equivalent 100 CPU), 640 Tensor-cores, HBM2 NVLink and DRAM 900GB/with technology that provides a 50% gain of performance in comparison with GPU of the previous generation. Peak capacity of Volta is five times higher than architecture of Pascal - the operating graphic architecture of NVIDIA, and 15 times above Maxwell.
GPU V100 go complete with the software optimized under Volta, including CUDA 9.0 and SDK for deep learning which includes TensorRT 3, DeepStream SDK and cuDNN 7 and also all main AI frameworks. According to Nvidia, hundreds of thousands of GPU accelerated applications are available to different hard tasks, including training of neuronets and inferens, high-performance computing, graphics and difficult data analysis.
The table of results of performance Tesla V100 in comparison with accelerators of Tesla of the previous generation.
Tesla Product | Tesla K40 | Tesla M40 | Tesla P100 | Tesla V100 |
---|---|---|---|---|
GPU | GK110 (Kepler) | GM200 (Maxwell) | GP100 (Pascal) | GV100 (Volta) |
SMs | 15 | 24 | 56 | 80 |
TPCs | 15 | 24 | 28 | 40 |
FP32 Cores / SM | 192 | 128 | 64 | 64 |
FP32 Cores / GPU | 2880 | 3072 | 3584 | 5120 |
FP64 Cores / SM | 64 | 4 | 32 | 32 |
FP64 Cores / GPU | 960 | 96 | 1792 | 2560 |
Tensor Cores / SM | NA | NA | NA | 8 |
Tensor Cores / GPU | NA | NA | NA | 640 |
GPU Boost Clock | 810/875 MHz | 1114 MHz | 1480 MHz | 1455 MHz |
Peak FP32 TFLOP/s* | 5.04 | 6.8 | 10.6 | 15 |
Peak FP64 TFLOP/s* | 1.68 | 2.1 | 5.3 | 7.5 |
Peak Tensor Core TFLOP/s* | NA | NA | NA | 120 |
Texture Units | 240 | 192 | 224 | 320 |
Memory Interface | 384-bit GDDR5 | 384-bit GDDR5 | 4096-bit HBM2 | 4096-bit HBM2 |
Memory Size | Up to 12 GB | Up to 24 GB | 16 GB | 16 GB |
L2 Cache Size | 1536 KB | 3072 KB | 4096 KB | 6144 KB |
Shared Memory Size / SM | 16 KB/32 KB/48 KB | 96 KB | 64 KB | Configurable up to 96 KB |
Register File Size / SM | 256 KB | 256 KB | 256 KB | 256KB |
Register File Size / GPU | 3840 KB | 6144 KB | 14336 KB | 20480 KB |
TDP | 235 Watts | 250 Watts | 300 Watts | 300 Watts |
Transistors | 7.1 billion | 8 billion | 15.3 billion | 21.1 billion |
GPU Die Size | 551 mm² | 601 mm² | 610 mm² | 815 mm² |
Manufacturing Process | 28 nm | 28 nm | 16 nm FinFET+ | 12 nm FFN |
2017
Solutions of partners based on Nvidia Volta for AI
Nvidia and partners of the company Dell EMC Hewlett Packard Enterprise, IBM and Supermicro provided on September 28, 2017 more than ten servers based on Tesla V100 GPU accelerators with architecture of Nvidia Volta. The multiprocessor systems of vendors based on V100 will open for users ample opportunities of GPU Nvidia for acceleration of researches to areas AI and creations of products and services in this area.
As specified in Nvidia, the graphic processors Nvidia V100 which capacity in problems of deep learning exceeds 120 TFLOPS, accelerations of analytics and other resource-intensive computing tasks are created especially for deep learning of neuronets and an inferens, high-performance computing. One GPU Volta provides performance equivalent 100 CPU, allowing scientists, researchers and engineers to solve problems which solution was presented too difficult or impossible earlier.
The following systems based on V100 are announced:
- Dell EMC — PowerEdge R740 with support to three GPU V100 for PCIe, PowerEdge R740XD with support to three GPU V100 for PCIe and PowerEdge C4130 with support to four V100 for PCIe or four GPU V100 for Nvidia NVLink in SXM2 form factor;
- HPE — HPE Apollo 6500 with support to eight GPU V100 for PCIe and HPE ProLiant DL380 with support to three GPU V100 for PCIe;
- IBM — servers of the next generation of IBM Power Systems based on the Power9 processor with support of several GPU V100 and NVLink technology with superfast interconnect of GPU-to-GPU and CPU-to-GPU OpenPOWER for fast data transmission;
- Supermicro — the product line with support of new GPU Volta includes workstation 7048GR-TR for high-performance GPU calculations, servers 4028GR-TXRT, 4028GR-TRT and 4028GR-TR2 for the most exacting applications of deep learning and servers 1028GQ-TRT for, for example, difficult analytical tasks.
Besides, a number of the partner systems was replenished with solutions of the Chinese producers, including Inspur, Lenovo and Huawei which announced systems based on architecture of Volta for data centers of the Internet companies.
Announcement
On May 10, 2017 the Nvidia company provided architecture of the graphic processors Volta and a series of the hardware-software developments focused on acceleration of work with the systems of artificial intelligence.
According to Jensen Huanga's statement of CEO Nvidia, Volta will become the standard of high-performance computing. Consolidation of cores of CUDA and Volta Tensor in the unified architecture, one server on GPU Tesla V100 platform will be able to replace hundreds of central processors in the field of high-performance computing.