Hardware
Interactive Servers
compsrv01/02/03/04
- 2x 12-Core Intel Xeon E5-2680 v3 at 2.5 GHz
- 512 GB RAM
compsrv05/12 (dellsrv04/05)
- 4x 6-Core Intel Xeon Gold 6128 at 3.4 GHz (max: 3.7 GHz)
- 3072 GB RAM
- NVidia RTX 4000 SFF Ada 20GB
compsrv06/07
- 4x 18-Core Intel Xeon E7-8867 v4 at 2.4 GHz
- 3072 GB RAM
compsrv08/09 (x3850-4/5)
- 4x 16-Core Intel Xeon E7-8867 v3 at 2.50 GHz
- 2.8 TB RAM
compsrv10/11 (dellsrv02/3)
- 2x 8-Core Intel Xeon Gold 6144 at 3.5 GHz (max: 4.2 GHz)
- 768 GB RAM
- NVidia RTX 4000 SFF Ada 20GB
compsrv13 (dellsrv01)
- 2x 32-Core AMD Epyc 7601 at 2.2 GHz (max. 3.2 GHz)
- 1024 GB RAM
- NVidia RTX 4000 SFF Ada 20GB
Batch Servers
bdw01..08
- 2x 20-Core Intel Xeon E5-2698 v4 at 2.2 GHz
- 512 GB RAM
epyc01..02
- 2x 64-Core AMD Epyc 7702 at 2.0 GHz
- 512 GB RAM
cuda02
- 2x 16-Core Intel Xeon Gold 6226R at 2.9 GHz
- 768 GB RAM
- 2x NVidia Tesla V100 16GB
cuda04
- 2x 16-Core Intel Xeon Gold 6226R at 2.9 GHz
- 768 GB RAM
- NVidia Tesla A100 80GB
CPU Performance
The following table gives a rough estimate of the (relative) single CPU core performance of each of the above systems for standard floating point arithmetics (the higher, the better). It also shows the performance of our standard desktop systems in comparison.
Machine |
CPU |
1 core |
all cores |
|---|---|---|---|
Desktop |
i5-5250U |
100 |
197 |
compsrv01..04 |
E5-2680v3 |
119 |
2394 |
compsrv06/07 |
E7-8867v4 |
126 |
6265 |
x3850-4/5 |
E7-8867v3 |
116 |
4475 |
x3850-1 |
E7-8857v2 |
110 |
3513 |
dellsrv01 |
Epyc 7601 |
98 |
4758 |
dellsrv02/03 |
Xeon 6144 |
199 |
2914 |
dellsrv04/05 |
Xeon 6128 |
142 |
3268 |
Machine |
CPU |
1 core |
all cores |
|---|---|---|---|
bdw01..08 |
E5-2698v4 |
139 |
3957 |
epyc01/02 |
Epyc 7702 |
133 |
6961 |
cuda02/04 |
Xeon 6226R |
149 |
3999 |
GPU Performance
For the GPUs a comparison of the relative performance for FP32 and FP64 computations is shown in the following table.
Machine |
GPU |
FP32 |
FP64 |
|---|---|---|---|
cuda02 |
Tesla V100 |
112.3 |
111.9 |
cuda04 |
Tesla A100 |
165.4 |
288.4 |
dellsrv01/2/3 |
RTX 4000 |
88.8 |
— |
x3850-1 |
RTX A2000 |
50.2 |
— |
Note
The above results may depend on your program, e.g., how good it was optimized for the CPU/GPU or how efficient it can use parallel CPU/GPU cores.
Max Planck Computing and Data Facility
If the above computing resources are not enough, the MPG also has a central compute cluster, the Max Planck Computing and Data Facility, which may be used by all members of the Max Planck Society.
To apply for a login, please refer to this page. Please activate MPG HPC systems.
Available HPC systems at the RZG: