Download presentation
Presentation is loading. Please wait.
1
ALICE TPC Online Tracking on GPU David Rohr for the ALICE Corporation 25.5.2010 Lisbon
2
ALICE TPC Online Tracking on GPU Introduction Tracking Algorithm NVIDIA CUDA Tracking on GPU Results Summary Large Hadron Collider
3
ALICE TPC Online Tracking on GPU Introduction Tracking Algorithm NVIDIA CUDA Tracking on GPU Results Summary The ALICE Experiment
4
ALICE TPC Online Tracking on GPU Introduction Tracking Algorithm NVIDIA CUDA Tracking on GPU Results Summary Tracking ClustersTracks
5
ALICE TPC Online Tracking on GPU Introduction Tracking Algorithm NVIDIA CUDA Tracking on GPU Results Summary Real TPC pp-Event in Online Event Display
6
ALICE TPC Online Tracking on GPU Introduction Tracking Algorithm NVIDIA CUDA Tracking on GPU Results Summary Simulated Heavy Ion Event
7
ALICE TPC Online Tracking on GPU Introduction Tracking Algorithm NVIDIA CUDA Tracking on GPU Results Summary Tracks found in simulated Heavy Ion Event
8
ALICE TPC Online Tracking on GPU Introduction Tracking Algorithm NVIDIA CUDA Tracking on GPU Results Summary Tracks in simulated Central Heavy Ion Event
9
ALICE TPC Online Tracking on GPU Introduction Tracking Algorithm NVIDIA CUDA Tracking on GPU Results Summary TPC Clusters divited into Slices
10
ALICE TPC Online Tracking on GPU Introduction Tracking Algorithm NVIDIA CUDA Tracking on GPU Results Summary The ALICE TPC
11
ALICE TPC Online Tracking on GPU Introduction Tracking Algorithm NVIDIA CUDA Tracking on GPU Results Summary TPC Tracker and Merger
12
ALICE TPC Online Tracking on GPU Introduction Tracking Algorithm NVIDIA CUDA Tracking on GPU Results Summary One ALICE TPC Sector (Traking is performed row by row)
13
ALICE TPC Online Tracking on GPU Introduction Tracking Algorithm NVIDIA CUDA Tracking on GPU Results Summary Tracking Algorithm Category of TaskName of TaskDescription on Task (Initialization) Combinatorial Part (Cellular Automation) I: Neighbors Finding Construct Seeds (Track Candidates) II: Evolution Kalman Filter Part III: Tracklet Construction Fit Seed Extrapolate Tracklet Find New Clusters IV: Tracklet SelectionSelect good Tracklets (Tracklet Output)
14
ALICE TPC Online Tracking on GPU Introduction Tracking Algorithm NVIDIA CUDA Tracking on GPU Results Summary Step I: Neighbors Finder (Fit best straight lines) dx +
15
ALICE TPC Online Tracking on GPU Introduction Tracking Algorithm NVIDIA CUDA Tracking on GPU Results Summary Step II: Evolution (Keep coinciding links)
16
ALICE TPC Online Tracking on GPU Introduction Tracking Algorithm NVIDIA CUDA Tracking on GPU Results Summary Step III: Tracklet Construction Green: Seed Red: Extrapolation Clusters close to the extraplation point are searched
17
ALICE TPC Online Tracking on GPU Introduction Tracking Algorithm NVIDIA CUDA Tracking on GPU Results Summary Neighbours Finder on Real Data
18
ALICE TPC Online Tracking on GPU Introduction Tracking Algorithm NVIDIA CUDA Tracking on GPU Results Summary Evolution Step on Real Data
19
ALICE TPC Online Tracking on GPU Introduction Tracking Algorithm NVIDIA CUDA Tracking on GPU Results Summary Tracklet Construction on Real Data
20
ALICE TPC Online Tracking on GPU Introduction Tracking Algorithm NVIDIA CUDA Tracking on GPU Results Summary Tracklet Selection on Real Data
21
ALICE TPC Online Tracking on GPU Introduction Tracking Algorithm NVIDIA CUDA Tracking on GPU Results Summary NVIDIA CUDA GPU
22
ALICE TPC Online Tracking on GPU Introduction Tracking Algorithm NVIDIA CUDA Tracking on GPU Results Summary Parallel Tracklet Construction Current Row Tracklets are independent and can be processed simultaneously Because of data locality the tracklets are processed for a common Row Single Instruction Decoder Either parallel fitting or extrapolation Red: Initial Seed Green: Extrapolation
23
ALICE TPC Online Tracking on GPU Introduction Tracking Algorithm NVIDIA CUDA Tracking on GPU Results Summary Initial GPU Tracker Performance (Peripheral Pb-PB) (8 threads on Nehalem CPU with 8 virtual / 4 physical cores) Focus on Tracklet Construction when optimizing
24
ALICE TPC Online Tracking on GPU Introduction Tracking Algorithm NVIDIA CUDA Tracking on GPU Results Summary Active GPU Threads for the First Implementation Active GPU Threads: 19% Colors represent Tracklet Constructor Steps: Black:Idle Blue:Track Fit Green:Track Extrapolation x-axis: threads y-axis: time
25
ALICE TPC Online Tracking on GPU Introduction Tracking Algorithm NVIDIA CUDA Tracking on GPU Results Summary Active GPU Threads using Dynamic Scheduling Active GPU Threads: 62% Colors represent Tracklet Constructor Steps: Black:Idle Blue:Track Fit Green:Track Extrapolation x-axis: threads y-axis: time
26
ALICE TPC Online Tracking on GPU Introduction Tracking Algorithm NVIDIA CUDA Tracking on GPU Results Summary Pipelining Initialization / Output on CPU, Tracking on GPU and DMA Transfer can overlap Time
27
ALICE TPC Online Tracking on GPU Introduction Tracking Algorithm NVIDIA CUDA Tracking on GPU Results Summary Final Speedup (Central Heavy Ion) (CPU performance was doubled as a side effect)
28
ALICE TPC Online Tracking on GPU Introduction Tracking Algorithm NVIDIA CUDA Tracking on GPU Results Summary Final Speedup in Contrast to Event Size (PP Mode: Special variant optimized for small scale events)
29
ALICE TPC Online Tracking on GPU Introduction Tracking Algorithm NVIDIA CUDA Tracking on GPU Results Summary Tracking Efficiency CPU GPU
30
ALICE TPC Online Tracking on GPU Introduction Tracking Algorithm NVIDIA CUDA Tracking on GPU Results Summary Twofold performance increase on CPU 2.5-fold performance increase compared to CPU Tracking efficiency matches the CPU tracker‘s efficiency A common source code ensures maintainability CPU still available during the GPU tracking Perspective for the Future Next NVIDIA GPU generation Fermi may result in another boost Track Merger may be adapted to run on the GPU
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.