GPUs

Please click on the respective model to get further information.

Please click here for a NVIDIA® TESLA® QUICK REFERENCE GUIDE.

Peak double precision
7.50 TeraFLOPS
 

 

 

Peak single precision
15.00 TeraFLOPS
   
Deep Learning
120.00 TeraFLOPS
   
Number of GPUs
1 x VP100
   
Number of CUDA cores
5120
   
Number of Tensor cores
640
   
Memory size per board (HBM2)
16GB at 900 GB/s
   
Features

Volta Architecture pairing NVIDIA® CUDA® and Tensor Cores to deliver the performance of an AI supercomputer in a single GPU

System
Servers NVLINK / PCIe Gen3
   
Peak double precision
4.70 teraflops
   
Peak single precision
9.30 teraflops
   
Number of GPUs
1 x GP100
   
Number of CUDA cores
3584
   
Memory size per board (HBM2)
16GB at 720 GB/s or 12GB at 540 GB/s
   
Features

Pascal Architecture, 16nm FinFET, HBM2 stacked memory, Unified Memory, Compute Preemption, and New AI Algorithms

System
Servers PCIe Gen3
   
Peak double precision
5.30 teraflops
   
Peak single precision
10.6 teraflops
   
Peak half precision
21.2 teraflops
   
Number of GPUs
1 x GP100
   
Number of CUDA cores
3584
   
Memory size per board
16GB CoWoS HBM2 720 GB/s
   
Features

Pascal Architecture, 16nm FinFET, HBM2 stacked memory, Unified Memory, Compute Preemption and New AI Algorithms

System
Servers SXM-2
   
Peak single precision
12 teraflops
   

 

Integer Operations (INT8)
47 TOPS* (Tera-Operations per Second)
   
Number of GPUs
1 x GP102
   
Number of CUDA cores
3840
   
Memory size per board (GDDR5)
24GB at 346GB/s
   
Max Power
250W
   
Features

Enhanced Programmability with Page Migration Engine, ECC Protection, Server-Optimized for Data                            Center Deployment, Hardware-Accelerated Video                            Engine 1x Decode Engine, 2x Encode Engine

System Interface
PCI Express 3.0 x16
   
Peak single precision
5.5 teraflops
   
Integer Operations (INT8)
22 TOPS* (Tera-Operations per Second)
   
Number of GPUs
1 x GP104
   
Number of CUDA cores
2560
   
Memory size per board (GDDR5)
8GB at 192GB/s
   
Max Power
50W/75W
   
Features

Enhanced Programmability with Page Migration Engine, ECC Protection, Server-Optimized for Data                            Center Deployment, Hardware-Accelerated Video                            Engine 1x Decode Engine, 2x Encode Engine

System Interface
Low-Profile PCI Express Form Factor
   
Peak double precision
1.87 teraflops
   
Peak single precision
5,60 teraflops
   

 

Number of GPUs
2 x GK210
   
Number of CUDA cores
4992
   
Memory size per board (GDDR5)
24GB
   
Memory bandwidth for board (ECC off)
480 GBytes/sec
   
Architecture features
SMX, Dynamic Parallelism, Hyper-Q
   
System
Servers
   
Data Sheet

Peak double precision
0.3 teraflops
   
Peak single precision
9.7 teraflops
   
Number of GPUs
2 x GM205
   
Number of CUDA cores
4096 (2x 2048)
   
Memory size per board (GDDR5)
16 GB (8 GB per GPU)
   
Memory bandwidth for board (ECC off)
320 GBytes/sec (2x 160 Gb/s)
   
Architecture features
SMX, Dynamic Parallelism, Hyper-Q
   
System
Server
   
Peak single precision
7 teraflops
   
Number of CUDA cores
3072
   
Memory size per board (GDDR5)
12 GB
   
Memory bandwidth for board (ECC off)
317 GBytes/sec
   
Architecture features
SMX, Dynamic Parallelism, Hyper-Q
   
System
Server
   
Peak single precision
7 teraflops
   
Number of CUDA cores
3072
   
Memory size per board (GDDR5)
24 GB
   
Memory bandwidth for board (ECC off)
288 GBytes/sec
   
GPU computing applications
Artificial Intelligence / Deep Learning
   
Architecture features
sMX, Dynamic Parallelism, Hyper-Q
   
System
Server only
   
Data Sheet

Number of CUDA cores
1536
   
Memory size per board (GDDR5)
8 GB
   
Max Power
100W
   
Form Factor
MXM
   
Cooling
Solution Bare Board
   

Peak double precision
0,7 TeraFLOPS mit NVIDIA GPU Boost
   
Peak single precision
2.2 teraflops
   
Number of CUDA cores
1024
   
Memory size per board (GDDR5)
4 GB
   
Memory bandwidth for board (ECC off)
88 GBytes/sec
   
Architecture features
SMX, Dynamic Parallelism, Hyper-Q
   
Max. Energieverbrauch
50–75 W
   
Kühlung
passiv
   
Formfaktor
Low-Profile
   
Berechnungs-APIs
NVIDIA CUDA, DirectCompute, OpenCL, OpenACC
   
System
Hyperscale-Server
   

Peak double precision
1.43 teraflops
   
Peak single precision
4.29 teraflops
   
Number of GPUs
1 x GK110B
   
Number of CUDA cores
2880
   
Memory size per board (GDDR5)
12GB
   
Memory bandwidth for board (ECC off)
288 GBytes/sec
   
GPU computing applications
CFD, CAE, financial computing, computational chemistry and physics, data analytics, satellite imaging, weather modeling
   
Architecture features
SMX, Dynamic Parallelism, Hyper-Q
   
System
Servers and Workstations
   
Data Sheet

Personal contact:

Michael Rechenmacher
Sales Manager HPC
Phone: +49 (0)89-558 909 6-14
Michael.Rechenmacher@fluidyna.de
Get more information about this technology
Use our configurator: