S THE MAKING OF DGX SATURNV: BREAKING THE BARRIERS TO AI SCALE. Presenter: Louis Capps, Solution Architect, NVIDIA,

Transcription

1 S THE MAKING OF DGX SATURNV: BREAKING THE BARRIERS TO AI SCALE Presenter: Louis Capps, Solution Architect, NVIDIA, lcapps@nvidia.com

2 A TALE OF ENLIGHTENMENT Basic OK List 10 for x = 1 to 3 20 print x 30 next x 1 FPS Run OK 2

3 DEEP LEARNING A NEW COMPUTING PLATFORM Assembly LDX #$00 dec: INX JSR printx CPX #$03 BNE dec BRK 30 FPS DL --> 3

4 SATURNV PURPOSE 124 Node Supercomputing Cluster Innovation is fueled by the right engine! Deep Learning scalability; move outside the box Drive research and Deep Learning application Partner with university research, government and industry collaborations Enable data science in HPC 4

5 NVIDIA DGX SATURNV ARCHITECTURE 124 node Cluster nvidia.com/dgx1 124 NVIDIA DGX-1 Nodes 992 P100 GPUs 8x NVIDIA Tesla P100 SXM GPUs NVLINK CubeMesh 2x Intel Xeon 20 core GPUs 512TB DDR4 System Memory SSD 7 TB scratch TB OS Mellanox 36 port EDR L1 and L2 switches 4 ports per system Fat tree topology Ubuntu 14.04, CUDA 8, OpenMPI a1, Docker, DL Frameworks NVIDIA GPU BLAS + Intel MKL (NVIDIA GPU HPL) Deep Learning applied research Many users, frameworks, algorithms, networks, new approaches Embedded, robotic, auto, hyperscale, HPC 5

6 SATURNV STACK 6

7 DGX-1 MULTI-SYSTEM 7

8 NVIDIA DGX SATURNV Greenest Supercomputer 8

9 NVIDIA DGX-1 SATURNV HPL RUN 124 node Supercomputing cluster HPL Setup Problem contained mainly in GPU memory (~16GB / GPU) 124 nodes * 8 GPU/node * 16 GB mem/gpu = 15,872 GB mem --- N = Measurement PDU input power time-stamped during full run All cluster hardware nodes, switches, storage SATURNV produced groundbreaking 9.4 GF/W at full scale --> Sets the stage for future Exascale class computing Performance HPL Rpeak 4,896 TF HPL Rmax 3,307 TF Pwr Full run avg KW Pwr Core avg KW ~15KW sustained per rack 9.4 GF / Watt 40% better than nearest competing technology 9

10 NOV2016 TOP GREEN500 SYSTEM Green500.org Top500.org SATURNV produced groundbreaking 9.4 GF/W at full scale --> Sets the stage for future Exascale class computing 10

11 WHAT IS HPL, TOP500, GREEN500? HPL High Performance Linpack Multi-system benchmark - measures optimized double-precision floating performance Solves system of dense linear equations One system or many connected in a cluster - usually Ethernet or InfiniBand Single problem split across many systems single final performance number Well designed to scale across large clusters and push limits Top500 (top500.org) List of the fastest HPL clusters in the world Updated twice a year June and Sept Published at ISC and SC conferences Green500 (green500.org) Same HPL clusters, but rank by power used during the HPL run Published at same time as Top500 11

12 DGX-1 SUPERCOMPUTER CHALLENGES Giant Leap Towards Exascale AI Compute Significant math performance FP32, FP16, INT8 Highly optimized frameworks Training, Inference Interconnect Multiple compute units inside node Multiple systems Storage Low latency, high bandwidth Equal perf to all systems Local caching for DL workloads Facilities Sufficient for bursts Maintain inlet air temp always High power density 12

13 NVIDIA DGX-1 COMPUTE NCCL Collective Library 13

14 DGX-1 COMPUTE AND MULTI-SYSTEM DGX-1 single system considerations Higher performance per system 27x to 58x faster Ingest data faster, provides faster results Also more power and heat High data ingest for DL workloads More storage and I/O into single system Cache data locally NFS cache on local SSD for training data Higher power/thermal density Example: KW vs 1,000 KW Ambient temperatures very important Silicon uses more higher temps Clocks will gate at thermal and power limits Variability lowers overall performance of multi- GPU and multi-system runs 14

15 DGX-1 COMPUTE CONSIDERATIONS #1 Recommendation - Using containers improves performance - Access to latest NVIDIA tuned codes - Latest NCCL libraries Clocking - CPUs set to performance mode to improve memory/i/o bandwidth Leave GPU clocks at default if you do set them, use base or slightly higher - Running set at max can cause extreme variation and reduced performance depending on workload - Monitor with nvidia-smi dmon 1320 Time Effects of Clocking 15

16 DGX-1 COMPUTE CONSIDERATIONS Affinity - Best performance when CPU/GPU/mem/IB affinity are aligned - E.g. cpu socket 0<->gpu0/1<->mlx5_0 Interrupt traffic can be high - Keep core 0 and core 20 free for interrupts 16

17 DGX-1 MULTI-SYSTEM CONSIDERATIONS IB Leaf Switch GPU0 GPU1 MLX0 GPU3 GPU4 MLX1 GPU5 GPU6 MLX1 GPU7 GPU8 MLX1 PCIe PCIe PCIe PCIe CPU0 CPU1 Example affinity with numactl: MEM0 MEM1 mpirun \ -np 4 bind-to node -mca btl openib,sm,self --mca btl_openib_if_include mlx5_0 -x CUDA_VISIBLE_DEVICES=0 numactl --physcpubind=1-4./mycode : \ -np 4 bind-to node -mca btl openib,sm,self --mca btl_openib_if_include mlx5_0 -x CUDA_VISIBLE_DEVICES=1 numactl --physcpubind=6-9./mycode : \ -np 4 bind-to node -mca btl openib,sm,self --mca btl_openib_if_include mlx5_1 -x CUDA_VISIBLE_DEVICES=2 numactl --physcpubind=10-13./mycode : \ -np 4 bind-to node -mca btl openib,sm,self --mca btl_openib_if_include mlx5_1 -x CUDA_VISIBLE_DEVICES=3 numactl --physcpubind=15-18./mycode : \ -np 4 bind-to node -mca btl openib,sm,self --mca btl_openib_if_include mlx5_2 -x CUDA_VISIBLE_DEVICES=4 numactl --physcpubind=21-24./mycode : \ -np 4 bind-to node -mca btl openib,sm,self --mca btl_openib_if_include mlx5_2 -x CUDA_VISIBLE_DEVICES=5 numactl --physcpubind=25-28./mycode : \ -np 4 bind-to node -mca btl openib,sm,self --mca btl_openib_if_include mlx5_3 -x CUDA_VISIBLE_DEVICES=6 numactl --physcpubind=30-33./mycode : \ -np 4 bind-to node -mca btl openib,sm,self --mca btl_openib_if_include mlx5_3 -x CUDA_VISIBLE_DEVICES=7 numactl --physcpubind=35-38./mycode 17

18 DGX-1 MULTI-NODE INTERCONNECT DESIGN 6,012 GB/s Design topologies that reduce latency and improve total bandwidth Fat-tree topologies for instance Equal bandwidth from a system all the way up to top level switch Ensure GPUDirect RDMA enablement DL and many computation workloads rely on fast synchronization Collectives Consistent iteration times System hierarchy - CPU0 <-> GPU0/1 <-> mlx5_0 - CPU0 <-> GPU2/3 <-> mlx5_1 - CPU1 <-> GPU4/5 <-> mlx5_2 - CPU2 <-> GPU6/7 <-> mlx5_3 If designing with only two IB ports, hook up mlx5_0, mlx5_2 18

19 DGX-1 MULTI-SYSTEM INTERCONNECT DGX-1 multi-system considerations High node to node communications DL and HPC workloads 4 IB ports à 2 ports DL: up to 5% loss Compute: up to 18% loss 1 IB port per system low performance Significant contention for many workloads Can t GPU Direct RDMA across full system Switch hierarchy critical Low bandwidth on second level Same issues as lowering ports per system Contention, lower bandwidth, variability 19

20 DGX-1 STORAGE CONSIDERATIONS Storage needs HPC needs well known Parallel FS like Lustre and Spectrum Scale well suited DL workloads just being understood Read dominated Input data rarely changes Can be raw or formatted in a DB (like LMDB) Large group of random read, then reread same data later Approaches Local caching helps significantly Can be many GB (>16GB for instance) Another approach is keep full datasets local (>100GB for ImageNet) Local SSD RAID Alternately, copy all data to nodes at beginning of job Reference designs 10Gb attached Central NFS with local caching Spectrum Scale IB attached (still evaluating) Lustre IB attached (still evaluating) 20

21 AI GRAND CHALLENGES CANDLE - Accelerate cancer research Energy / Fusion Future of low cost energy Weather and Climate Disaster Preparedness Astrophysics Our future? Autonomous Cars 21

22 Summary DGX-1 DL SCALABILITY SUMMARY DGX-1 crafted for AI and Computational workloads High compute density, but also high power and thermal density Watch ambient can cause large variability Single system has large demands in data ingest and GPU to GPU communication Multi DGX-1 systems have large demands on inter-node communication for most workloads Need at least two IB rails per system (1 EDR IB for every 2 GPU) DL Storage needs are very high But read dominated (vs writes with HPC) Many codes benefit significantly when watching affinity Align CPU/memory with GPUs and IB cards Avoid cores handling interrupts NVIDIA pre made containers significantly reduce user work Affinity is already handled Provides technologies like NCCL and the latest, tuned code and frameworks 22

23 DGX-1 DL SCALABILITY SUMMARY Thanks!!! More info at NVIDIA DGX-1 System Architecture: CANDLE sessions ( S CANDLE: PREDICTING TUMOR CELL RESPONSE TO DRUG TREATMENTS S THE DOE AND NCI PARTNERSHIP ON PRECISION ONCOLOGY AND THE CANCER MOONSHOT S BUILDLING EXASCALE DEEP LEARNING TOOLS TO HELP UNDERSTAND CANCER BIOLOGY AT THE MOLECULAR SCALE S BUILDING EXASCALE DEEP TEXT COMPREHENSION TOOLS FOR EFFECTIVE CANCER SURVEILLANCE S WHAT'S NEXT IN DGX SERVER SOLUTIONS FOR DEEP LEARNING Thursday, May 11, 10:00 AM - 10:50 AM Room 210B 23

24