wiki home

compute at leeds. note that Leeds' uses cluster to mean "lab" and research compute when they mean "cluster". (I mean technically, I guess we have clusters of urninals in many gent's bathrooms, I still wouldn't write cluster on the door)

Main compute resources are the local ARC, regional >Bede (why Power9 N8, why - what did IBM do to achieve this?!), and national Jade 2 (not yet online as of March 2021).

Resources for using ARC are here + Rebecca's cheat sheet.

Comparing different clusters is non-trivial. This table just counts the number of GPUs - this is normally what ML-ey graphics-ey people use. We note that server-side GPU is expensive pared to (gamer grade) local (under desk GPUs). Local is also much easier to debug (trying to debug something run with a scheduler is a nightmare). (a list of national HPC facilities).

institutionclusterGPU hardwareGPU count
YorkViking8 x v1008
Ireland/ICHECKay32 x v10032
LeedsARC 3/44 x k80, 24 x P100, 12 x V10040
Bede (Tier 2)NICE-1980 x V100 (IBM Power9)80
Cirrus (Tier 2)Cirrus152 x v100152
BristolBlueCrystal 3/476 x k20, 65 x p100141
Jadev1176 x V100176
EdinburghEddie + James and Charleseddie ( 44 x k80, 80 x titan X) + james (59 mixed) 183
BirminghamBaskerville (July 21)184 x a100184
Cambridge (Tier 2)Wilkes2360 x p100360
KAUSTIbext32 x 2080ti, 32 x 1080ti, 20 x P100, 4 x p6000, 272xv100340
Jade 2 (Tier 2)v2504 x V100504
UMassMGHPCCunspecified mix1092