mboeller
2016-06-20, 15:11:58
beim MIT haben sie eine neue(?) Multicore Architecture entwickelt; SWARM.
Links:
http://people.csail.mit.edu/emer/
edit: http://people.csail.mit.edu/mcj/ (Links zu 1-2 weiteren PDF)
edit: http://techxplore.com/news/2016-06-world-processor-chip.html gehört nicht zu der News
http://news.mit.edu/2016/parallel-programming-easy-0620
https://www.computer.org/csdl/mags/mi/preprint/07436649.pdf
Der Speedup für ein 64-Core System schaut im Vergleich zu einer Software-Lösung IMHO gut bis sehr gut aus (im PDF auf Seite 9 sind ein paar Charts).
ABSTRACT
The authors present Swarm, a parallel architecture that exploits ordered parallelism, which is abundant but hard to mine with current software and hardware techniques. Swarm programs consist of short tasks, as small as tens of instructions each, with programmer-specified order constraints. Swarm executes tasks speculatively and out of order and efficiently speculates thousands of tasks ahead of the earliest active task to uncover enough parallelism. Several techniques allow Swarm to scale to large core counts and speculation windows. The authors evaluate Swarm on graph analytics, simulation, and database benchmarks. At 64 cores, Swarm outperforms sequential implementations of these algorithms by 43 to 117 times and state-of-the-art software-only parallel algorithms by 3 to 18 times. Besides achieving near-linear scalability, Swarm programs are almost as simple as their sequential counterparts, because they do not use explicit synchronization.
edit:
IMHO auch witzig:
Cores: 64 cores in 16 tiles (4 cores/tile), 2GHz, x86-64 ISA, IPC-1 except misses and Swarm instructions
L1 caches 16 KB, per-core, split D/I, 8-way, 2-cycle latency
L2 caches 256KB, per-tile, 8-way, inclusive, 7-cycle latency
L3 cache 16MB, shared, static NUCA [40] (1MB bank/tile),
16-way, inclusive, 9-cycle bank latency
Links:
http://people.csail.mit.edu/emer/
edit: http://people.csail.mit.edu/mcj/ (Links zu 1-2 weiteren PDF)
edit: http://techxplore.com/news/2016-06-world-processor-chip.html gehört nicht zu der News
http://news.mit.edu/2016/parallel-programming-easy-0620
https://www.computer.org/csdl/mags/mi/preprint/07436649.pdf
Der Speedup für ein 64-Core System schaut im Vergleich zu einer Software-Lösung IMHO gut bis sehr gut aus (im PDF auf Seite 9 sind ein paar Charts).
ABSTRACT
The authors present Swarm, a parallel architecture that exploits ordered parallelism, which is abundant but hard to mine with current software and hardware techniques. Swarm programs consist of short tasks, as small as tens of instructions each, with programmer-specified order constraints. Swarm executes tasks speculatively and out of order and efficiently speculates thousands of tasks ahead of the earliest active task to uncover enough parallelism. Several techniques allow Swarm to scale to large core counts and speculation windows. The authors evaluate Swarm on graph analytics, simulation, and database benchmarks. At 64 cores, Swarm outperforms sequential implementations of these algorithms by 43 to 117 times and state-of-the-art software-only parallel algorithms by 3 to 18 times. Besides achieving near-linear scalability, Swarm programs are almost as simple as their sequential counterparts, because they do not use explicit synchronization.
edit:
IMHO auch witzig:
Cores: 64 cores in 16 tiles (4 cores/tile), 2GHz, x86-64 ISA, IPC-1 except misses and Swarm instructions
L1 caches 16 KB, per-core, split D/I, 8-way, 2-cycle latency
L2 caches 256KB, per-tile, 8-way, inclusive, 7-cycle latency
L3 cache 16MB, shared, static NUCA [40] (1MB bank/tile),
16-way, inclusive, 9-cycle bank latency