Nvidia announces Tesla P100 accelerator with Pascal GPU and memory to 16GB hbm2

Apr

6

2016

Nvidia has announced a new accelerator card for high performance computing. The Tesla P100 has an GP100-GPU that is based on Pascal-architecture and is intended for applications such as deep-GPGPU learning and development of artificial intelligence.

Jen-Hsun Huang, CEO of Nvidia, announced the Tesla P100 during the GPU Technology Conference. The accelerator features the new GP100 GPU consisting of 15.3 billion transistors. That is almost double the GM200 GPU of the Maxwell-generation, which consists of up to eight billion transistors. According to Huang, there is been working for three years on the chip and there was research & development two to three billion dollars in costs incurred.

Nvidia has not released during the keynote on new graphics cards for consumers. Interestingly, the Tesla P100 uses hbm2 16GB memory, probably from Samsung . Recent rumors just seemed to indicate it would get the new GeForce cards gddr5x memory. It is certain that future GeForce graphics cards will use the same Pascal architecture as the Tesla P100.

The Tesla P100 has further 4MB and 14MB L2 cache sm RF, which can communicate with a speed of 80TB / s with the chip. The GP100 GPU is made on a 16nm FinFET process-and has a processing power of 5.3 teraflops at fp64-double precision calculations. FP32-in single precision is 10.6 teraflops and fp16 take it increased to 21.2 teraflops.

Nvidia Tesla P100
Nvidia Tesla P100Nvidia Tesla P100Nvidia Tesla P100
The GP100 is an assembly of graphics processing clusters, streaming multi-processors and memory controllers. The chip has six gpc, up to 60 sm and eight 512bit memory controllers, which represents a memory of 4096bit wide. Every Streaming Multiprocessor on the GPU has 64 cudacores and 4 texture units. For a total of 3840 cudacores and 240 texture units. In his 3584 Tesla P100 cores enabled. The GPU has a clock speed of 1328MHz with boost clock of 1480MHz.

Nvidia GP100Nvidia GP100
The Tesla accelerators are intended for business applications, and P100 is located on the top of that segment. The chip is currently in mass production and Nvidia says the first units to be delivered as soon as possible to large companies for use in their Hyperscale -datacenters. Later OEMs like Dell, HP and IBM gain access to the accelerators so they can build it into servers. Those servers with Tesla P100 come in the first quarter of 2017 on the market, according to Nvidia.

Nvidia itself brings the DGX-1 break, describes itself as a “supercomputer” which includes eight Tesla P100 accelerators. The cards communicate with the Nvidia nvlink interface, which is five times the speed of PCI-E-3.0 offers. A single node is good for 170 teraflops of computing power with fp16 precision. With rack full of these servers is 2 petaflops possible. The Nvidia DGX-1, which cost $ 129,000, down two Intel Xeon E5-2698 v3 processors, 512GB DDR4 RAM and four 1,92TB-SSDs in RAID 0 set up. The first copies will be delivered to research departments of universities.

Nvidia DGX-1

Tesla Accelerators Tesla K40 Tesla M40 Tesla P100
GPU GK110 (Kepler) GM200 (Maxwell) GP100 (Pascal)
sm’s 15 24 56
TPC’s 15 24 28
FP32 cudacores / sm 192 128 64
FP32 cudacores / gpu 2880 3072 3584
FP64 cudacores / sm 64 4 32
FP64 cudacores / gpu 960 96 1792
Base Clock 745 MHz 948 MHz 1328 MHz
GPU Boost clock 810/875 MHz 1114 MHz 1480 MHz
FP64 GFLOPS 1680 213 5304
texture Units 240 192 224
memory Interface 384-bit GDDR5 384-bit GDDR5 4096-bit hbm2
memory Size up to 12GB up to 24GB 16GB
L2 cache
1536Kb 3072KB 4096KB
Register file size / sm 256KB 256KB 256KB
Register file size / gpu 3840KB 6144KB 14336KB
TDP 235 Watts 250 Watts 300 Watts
transistors 7.1 billion 8 billion 15.3 billion
GPU which format 551mm² 601mm² 610mm²
design 28nm 28nm 16nm

Viewing:-129

In: Technology & Gadgets Asked By: [15764 Red Star Level]

Answer this Question

You must be Logged In to post an Answer.

Not a member yet? Sign Up Now »