...
| Model | User | Pro SMT  | Contra SMT  | ||||||||||||||||||||||||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| idle | admin | - | (∑ Esocket[0-7] according to lm_sensors: | ||||||||||||||||||||||||||||||||||||||||||
| stress-ng stream | admin | -- | ~13% slower | ||||||||||||||||||||||||||||||||||||||||||
| FESOM | NEC | Using 128 Threads per node: 3% faster (probably because the (buggy) GXFS daemon can use a virtual core) | Using 256 Thread per node: 10% slower | ||||||||||||||||||||||||||||||||||||||||||
| Python AI | vhelm | no impact/difference | no impact/difference | ||||||||||||||||||||||||||||||||||||||||||
| matlab #SBATCH --cpus-per-task=16 | vhelm | Runtime: 1440s instead of 1366s → ~5% slower | |||||||||||||||||||||||||||||||||||||||||||
| unzip 262 about ~50 MB files in parallel: S=$(date +%s); parallel -P$P gunzip -c > /dev/null ::: /tmp/input/* ; echo "$(( $(date +%s) - $S )) sec" salloc -psmp --qos=12h --time=12:00:00 --ntasks-per-node=128 salloc -psmpht --qos=12h --time=12:00:00 --ntasks-per-node=256 --mem=249G | lkalesch mthoma | If a user does only request --ntasks-per-node=1 (the default) and uses P=0 (use all available cores) then you get
But please note: This is a improper use of slurm/HPC, so this "advantage" does not justify SMT | no advantage (if slurm is used properly)
|
GPU nodes (A40 vs. A100)
| Model | User | A40 vs. A100 | |
|---|---|---|---|
| tensorflow-gpu AI application | vhelm | no difference | |
python3, matrix operations with with numpy (fat) vs cupy (gpu) | sviquera |
| |
...
| albedo | ollie | ||||||
|---|---|---|---|---|---|---|---|
| Application | user | node internal /tmp (NVMe) | 100 Gb Infiniband /albedo (GPFS) | 10 Gb Ethernet /isibhv (NVMe) | node internal | 100 Gb Omnipath /work (BeeGFS) | 10 Gb Ethernet |
| idmidl: reading 244 data files | vhelm | ~9 sec | 10~13 sec | 8~11 sec spikes up to 181 sec | 27~29 sec | 27~37 sec | 29~60 sec spikes up to 98 sec |
| ls -f ls # default with stat/color ls -f | directory with 30000 entires | 0.08 sec 0.19 sec | 0.04 sec 6~15 0.3 sec | 0.03 sec 0.2 sec | 0.1 sec 0.4 sec | 0.2 sec 1.6 sec | 0.08 sec 0.3~0.7 sec |
- ...
Runtime compared with ollie
| Model | User | albedo GPFS | albedo local NVMe | ollie BeeGFS | |
|---|---|---|---|---|---|
| idl vhelm | vhelm | Cumulative time for loop and if conditions: | 0.32 s ( 3.11 %) | 0.13 s ( 1.40 %) | 3.48 s (12.73 %) 0.05 s ( 0.19 %) 23.77 s (87.07 %) 27.30 s 34269 ( 1441 MB/s) |
| tensorflow | vhelm | AI for trained modell | 4 times slower | ~10 minutes | |