Data Center
The Cloud Infrastructure Platform (CLIP) consists of heterogeneous computing hardware. It is entirely virtualized using OpenStack, thus enabling rapid reconfiguration and seamless extensibility.
Available Resources
Compute Node Types and Hardware Resources
100 GBit/s
High Speed Network
160+
GPU Accelerators
8000+
CPU Cores
32+ TB
Memory
250 TB
BeeGFS Flash Storage
200 GB
HOME per User
40+ C1
CPU Compute Nodes
- 2 x 11 Core virtual processor (Intel(R) Xeon(R) Gold 6146 CPU @ 3.20GHz)
- 85 GB of RAM
- 100 Gbit/s Ethernet with hardware offloaded networking
136+ C2
CPU Compute Nodes
- 2 x 19 Core virtual processor (Intel(R) Xeon(R) Gold 6138 CPU @ 2.00GHz)
- 170 GB of RAM
- 100 Gbit/s Ethernet with hardware offloaded networking
2 M1
High Memory Nodes
- 8 x 11 Core virtual processor (Intel(R) Xeon(R) Platinum 8160 CPU @ 2.10GHz)
- 2000 GB of RAM
- 100 Gbit/s Ethernet with hardware offloaded MPI communication
3 M2
High Memory Nodes
- 4 x 19 Core virtual processor (Intel(R) Xeon(R) Gold 6138 CPU @ 2.00GHz)
- 1500 GB of RAM
- 100 Gbit/s Ethernet with hardware offloaded MPI communication
7 G1
GPU Compute Nodes
- 2 x 7 Core virtual processor (Intel(R) Xeon(R) Gold 6134 CPU @ 3.20GHz)
- 192 GB of RAM
- 8 x Nvidia Pascal class GPU P100 with 12 GB of GPU RAM
- 100 Gbit/s Ethernet with hardware offloaded MPI communication
4 G2
GPU Compute Nodes
- 2 x 15 Core virtual processor (Intel(R) Xeon(R) Gold 6142 CPU @ 2.60GHz)
- 192 GB of RAM
- 4 x Nvidia Volta class GPU V100 with 16 GB of GPU RAM
- 100 Gbit/s Ethernet with hardware offloaded MPI communication
10 G3
GPU Compute Nodes
- 2 x 15 Core virtual processor (Intel(R) Xeon(R) Gold 6226R CPU @ 2.90GHz)
- 192 GB of RAM
- 4 x Nvidia Turing class GPU TU102 with 24 GB of GPU RAM
- 100 Gbit/s Ethernet with hardware offloaded MPI communication
12 G4
GPU Compute Nodes
- 2 x 15 Core virtual processor (Intel(R) Xeon(R) Gold 6226R CPU @ 2.90GHz)
- 360 GB of RAM
- 4 x Nvidia Ampere class GPU A100 with 40 GB of GPU RAM
- 100 Gbit/s Ethernet with hardware offloaded MPI communication
Available Graphic Architectures
| Nodes | G1 P100 | G2 V100 | G3 TU102 | G4 A100 |
|---|---|---|---|---|
| Number of GPUs | 56 | 16 | 40 | 8 |
| CUDA Cores | 3,584 | 5,120 | 4,608 | 6,912 |
| Tensor Cores | – | 640 | 576 | 432 |
| RT Cores | – | – | 72 | 160 |
| Double-Precision Performance | 4.7 TFLOPS | 7 TFLOPS | – | 9.7 TFLOPS |
| Single-Precision Performance | 9.3 TFLOPS | 14 TFLOPS | 14.9 TFLOPS | 19.5 TFLOPS |
| Half-Precision Performance | 18.7 TFLOPS | – | – | 312 TFLOPS |
| Tensor Performance | – | 112 TFLOPS | 119.4 TFLOPS | 156 TFLOPS |
| Memory Buswidth | 3,072-bit | 4,096-bit | 384-bit | 5,120-bit |
| Memory Capacity | 12 GB HBM2 | 16 GB HBM2 | 24 GB GDDR6 | 40 GB HBM2e |
| Memory Bandwidth | 549 GB/s HBM2 | 900 GB/s HBM2 | 624 GB/s GDDR6 | 1.6 TB/s HBM2 |
| Max. Power Consumption | 250 Watt | 250 Watt | 250 Watt | 250 Watt |
Contact us
Please provide affiliated Organisation, and your use-case or resource use estimation (CPU-hours, GPU-hours, Memory)