1. ANTON D.E Shaw Research

2. Force Fields: Typical Energy Functions Bond stretches Angle bending Torsional rotation Improper torsion (sp2) Electrostatic interaction Lennard-Jones interaction

3. Molecular Dynamics Solve Newton ’ s equation for a molecular system:

4. Integrator: Verlet Algorithm Start with {r(t), v(t)}, integrate it to {r(t+  t), v(t+  t)}: {r(t), v(t)} {r(t+  t), v(t+  t)} The new position at t+  t: Similarly, the old position at t-  t: (1) (2) Add (1) and (2): Thus the velocity at t is: (3) (4)

5. Molecular Dynamics Iterate... and iterateIterate... and iterate Integrate Newton’s laws of motion

6. Two Distinct Problems Problem 1: Simulate many short trajectories Problem 2: Simulate one long trajectory

7. Simulating Many Short Trajectories  Can answer surprising number of interesting questions  Can be done using –Many slow computers –Distributed processing approach –Little inter-processor communication  E.g., Pande’s Folding at Home project

8. Simulating One Long Trajectory  Harder problem  Essential to elucidate many biologically interesting processes  Requires a single machine with –Extremely high performance –Truly massive parallelism –Lots of inter-processor communication

9. DESRES Goal  Single, millisecond-scale MD simulations (long trajectories) –Protein with 64K or more atoms –Explicit water molecules  Why? –That’s the time scale at which many biologically interesting things start to happen

10. Image: Istvan Kolossvary & Annabel Todd, D. E. Shaw Research Protein Folding

11. What Will It Take to Simulate a Millisecond?  We need an enormous increase in speed –Current (single processor): ~ 100 ms / fs –Goal will require < 10 m s / fs  Required speedup: > 10,000 x faster than current single-processor speed ~ 1,000x faster than current parallel implementations  Can’t accept >10,000x the power (~5 Megawatts)!

12. What Takes So Long?  Inner loop of force field evaluation looks at all pairs of atoms (within distance R)  On the order of 64K atoms in typical system  Repeat ~10 12 times  Current approaches too slow by several orders of magnitude  What can be done?

13. Parallelization (getting an idea of the level of computation needed) For every time step, every atom must communicate within its cutt-off radius with every other atom. 2) A lot of inter-processor communication that can be scaled well is needed. MD Simulator Requirements

14. Parallelization (getting an idea of the level of computation needed) Whole System is broken down into boxes (processing nodes) Each node handles the bonded interactions within NT method for non-bonded interactions (much more common). NT method for Atom Migration MD Simulator Requirements

15. 1) Need a huge number of arithmetic processing elements 2) A lot of inter-processor communication that can be scaled well is needed. 3) Memory is not an issue –With 25,000 atoms (64bytes each) total=1.6MB over 512 nodes =3.2KB/node which is < most L1 Why Specialized Hardware? Memory Communication Computation Needs

16. Consider Moore’s Law on 10X improvement in 5 years vs. Anton’s 1000X in 1 year. Can great discoveries wait? Can use custom pipelines with more precision, increased datapath logic speed, over less silicon area. Have Tailored ISA’s for geometric calculations+ Programmability for accommodating various force fields and integration algorithms Dedicated memory for each particle to accumulate forces Why Specialized Hardware? Memory Communication Computation Needs

17. ANTON Strategy  New architectures –Design a specialized machine –Enormously parallel architecture –Based on special-purpose ASICs –Dramatically faster for MD, but less flexible –Projected completion: 2008  New algorithms –Applicable to Conventional clusters Our own machine –Scale to very large # of processing elements

18. Interdisciplinary Lab Computational Chemists and Biologists Computer Scientists and Applied Mathematicians Computer Architects and Engineers

19. Alternative Machine Architectures  Conventional cluster of commodity processors  General-purpose scientific supercomputer  Special-purpose molecular dynamics machine

20. Conventional Cluster of Commodity Processors  Strengths: –Flexibility –Mass market economies of scale  Limitations –Doesn’t exploit special features of the problem –Communication bottlenecks Between processor and memory Among processors –Insufficient arithmetic power

21. General-Purpose Scientific Supercomputer  E.g., IBM Blue Gene  More demanding goal than ours –General-purpose scientific supercomputing –Fast for wide range of applications  Strengths: –Flexibility –Ease of programmability  Limitations for MD simulations –Expensive –Still not fast enough for our purposes

22. Anton: Special-Purpose MD Machine  Strengths: –Several orders of magnitude faster for MD –Excellent cost/performance characteristics  Limitations: –Not designed for other scientific applications They’d be difficult to program Still wouldn’t be especially fast –Limited flexibility

23. Anton System-Level Organization  Multiple segments (probably 8 in first machine)  512 nodes (each consists of one ASIC plus DRAM) per segment –Organized in an 8 x 8 x 8 toroidal mesh  Each ASIC equivalent performance to roughly 500 general purpose microprocessors –ASIC power similar to a single microprocessor

24. 3D Torus Network

25. Why a 3D Torus?  Topology reflects physical space being simulated: –Three-dimensional nearest neighbor connections –Periodic boundary conditions  Bulk of communications is to near neighbors –No switching to reach immediate neighbors

26. Source of Speedup on Our Machine  Judicious use of arithmetic specialization –Flexibility, programmability only where needed –Elsewhere, hardware tailored for speed Tables and parameters, but not programmable  Carefully choreographed communication –Data flows to just where it’s needed –Almost never need to access off-chip memory

27. Two Subsystems on Each ASIC Specialized Subsystem Flexible Subsystem  Programmable, general-purpose  Efficient geometric operations  Modest clock rate  Pairwise point interactions  Enormously parallel  Aggressive clock rate

28.  28 Anton 33M gate ASIC Two computational subsystems connected by communication ring Hardware datapaths compute over 25 billion interactions/s Full machine has 512 ASICs in a 3D torus 13 embedded processors

29. Where We Use Specialized Hardware Specialized hardware (with tables, parameters) where: Inner loop Simple, regular algorithmic structure Unlikely to change Examples: Electrostatic forces Van der Waals interactions

30. Example: Particle Interaction Pipeline (one of 32)

31. –Executes Non-bonded MD interaction calculations (Charge Spreading & Force Interpolation) –Accumulates forces on each particle as data streams through. –ICB Controls flow of data through the HTIS, programmable ISA extensions, acts as a buffering, pre-fetching, synchronization, and write back controller High-Throughput Interaction Subsystem

32. Array of 32 Particle Interaction Pipelines

33. Advantages of Particle Interaction Pipelines  Save area that would have been allocated to –Cache –Control logic –Wires  Achieve extremely high arithmetic density  Save time that would have been spent on –Cache misses, –Load/store instructions –Misc. data shuffling

34. Where We Use Flexible Hardware –Use programmable hardware where: Algorithm less regular Smaller % of total computation -E.g., local interactions (fewer of them) More likely to change –Examples: Bonded interactions Bond length constraints Experimentation with -New, short-range force field terms -Alternative integration techniques

35. Forms of Parallelism in Flexible Subsystem  The Flexible Subsystem exploits three forms of parallelism: –Multi-core parallelism (4 Tensilicas, 8 Geometry Cores) –Instruction-level parallelism –SIMD parallelism – calculate on 3D and 4D vectors as single operation

36. Overview of the Flexible Subsystem GC = Geometry Core (each a VLIW processor)

37. Geometry Core (one of 8; 64 pipelined lanes/chip)

38. But Communication is Still a Bottleneck  Scalability limited by inter-chip communication  To execute a single millisecond-scale simulation, –Need a huge number of processing elements –Must dramatically reduce amount of data transferred between these processing elements  Can’t do this without fundamentally new algorithms: –A family of Neutral Territory (NT) methods that reduce pair interaction communication load significantly –A new variant of Ewald distant method, Gaussian Split Ewald (GSE) which simplifies calculation and communication for distant interactions –These are the subject of a different talk.

39.  39 Anton in Action

40. 500X NAMD80-100X Desmond 100X Blue Matter Simulation Evaluations

41. GPU+FPGA ??? GPU 6*GDDR5 FPGA HIGH SPEED SERIAL I/O UP TO 2 Tbit/S LVDS FFT and LJ 16*PCIe