Processor
From Grid5000
Contents |
AMD Opteron
AMD PowerNow!
Availability over Grid'5000
PowerNow! availability depends on processor's model but also on motherboard and BIOS. If processor possesses PowerNow! technology but its motherboard does not recognize it, PowerNow! will be unavailable.
Availability can be known by loading the powernow-k8 kernel module on a 2.6-series Linux kernel. Following chart presents availability over Grid'5000 Opteron nodes:
| Sites \ Processors | AMD Opteron 246 | AMD Opteron 248 | AMD Opteron 250 | AMD Opteron 252 | AMD Opteron 275 |
|---|---|---|---|---|---|
| Bordeaux | |||||
| Lille | |||||
| Lyon | |||||
| Nancy | |||||
| Orsay | |||||
| Rennes | |||||
| Sophia | |||||
| Toulouse |
When powernow-k8 fails to load, if means that PowerNow! is not available. Many unavailability reasons exist:
- Power state transitions not supported
- BIOS error - no PSB or ACPI _PSS objects
Computing benchmark
High-Performance Linpack quantifies the number of floating operation per second (eg. flops) during the solving of a random dense linear system. This benchmark was applied to Grid'5000 Opteron nodes by building the HPL binary over the AMD Core Math Library (ACML).
Following chart represents a HPLinpack run on a single processor, or a single core, of each AMD Opteron node. If PowerNow! is available on nodes, the benchmark was applied to each available frequency step. Node quantity for a processor model and processor per node quantify are displayed in this summary chart to easily guess global computation capability. HPLinpack results are expressed in giga floating operation per second (eg. gflops).
| Proc@Site \ Freq (GHz) | CPUs per node qty | Node qty | 1.0 | 1.8 | 2.0 | 2.2 | 2.4 | 2.6 |
|---|---|---|---|---|---|---|---|---|
| 248@Bordeaux | 2 | 48 | - | - | - | 3.542 | - | - |
| 248@Lille | 2 | 53 | 1.718 | 3.022 | 3.336 | 3.647 | - | - |
| 252@Lille | 2 | 15 | - | - | - | - | - | 4.311 |
| 246@Lyon (capricorne) | 2 | 56 | - | - | 3.254 | - | - | - |
| 250@Lyon (sagittaire) | 2 | 70 | - | - | - | - | 3.865 | - |
| 246@Nancy (grillon) | 2 | 47 | 1.737 | 3.057 | 3.379 | - | - | - |
| 246@Orsay (gdx) | 2 | 216 | - | - | 3.388 | - | - | - |
| 250@Orsay (gdx) | 2 | 126 | 1.730 | 3.065 | 3.385 | 3.720 | 4.040 | - |
| 246@Rennes (paravent) | 2 | 99 | 1.737 | 3.059 | 3.364 | - | - | - |
| 248@Rennes (parasol) | 2 | 64 | - | - | - | 3.573 | - | - |
| 246@Sophia (azur) | 2 | 105 | - | - | 3.258 | - | - | - |
| 275@Sophia (helios) | 4 | 56 | - | - | - | 3.675 | - | - |
| 248@Toulouse (toulouse) | 2 | 58 | - | - | - | 3.586 | - | - |
Note: these results have been obtained by running the HPLinpack on the same deployed environment. Gflops variations between a same processor model can be explained by hardware variations, especially motherboard and memory ship variation.
Intel Xeon EM64T
Intel Xeon IA32
Intel Xeon E5607
Link : information
