3. SECTIONS:
- W HAT IS TILE64 ?
- APPLICATIONS
- BASIC ARCHITECTURE.
- W HAT IS TILE ?
- IMISH
- TAPERED FAT TREE TOPOLOGY
- MEMORY MODEL
- POW ERMODEL
- HARDWALL TECHNOLOGY
- REFERANCES
4. What is TILERA TILE64 ?
• The name for the first processor in the family of
Tile Processor chips from Tilera Corporation.
• The TILE64 processor is based on an architecture
that can scale to hundreds, or even thousands of
cores.
• The processor contains 64 full-featured,
programmable cores, each capable of running its
own operating system.
5. Cont.
• Tilera's architecture eliminates the on-chip bus
interconnect by :
• placing a communications switch on each processor
core
• arranging Cores in a grid fashion.
• homogeneous cores.
• Each of the 64 cores is a general-purpose processor
that includes L1 5MB and L2 caches, as well as an
innovative distributed L3 cache.
6. TILE64 applications and uses.
• Advanced networking:
• Unified Threat Management (UTM).
• Network Security Appliances.
• Deep Packet Inspection (DPI) is a networking
technology that Internet Service Providers use
to monitor customers' data traffic
• Network Monitoring.
7. Cont.
• Digital Video:
• Video Conferencing.
• Video-on-Demand (VoD) Servers, IPTV technology
• Video Surveillance. is the monitoring of the
behavior.
• Media ‘Head-End’ Services.
8. Cont.
• Cloud Computing applications such as web
indexing, search engine and cache acceleration
servers
12. Cont.
• Using multiple processors require a system to
allow communication among them.
• Old Solution: bus interconnection.
• Problem: more cores added to chips bus
creates data congestion, limiting performance
scalability with the increased number of cores.
• Tilera’s solution: iMesh.
14. Cont.
• Five physical mesh networks
• UDN, IDN, SDN, TDN, MDN
• TDN and MDN are used for handling memory traffic.
• Memory requests transit TDN
• Large store requests, small load requests
15. Cont.
• Memory responses transit MDN
• Large load responses, small store responses
• Includes cache-to-cache transfers and off-chip
transfers.
• MIMD processor.
16. TAPERED FAT-TREE
Good for many-to-few connectivity
• Fewer hops Shorter latency
• Fewer routers Less power, less area
17. TILE64 WITH TAPERED FAT TREE
Legend
- Level 3 Routers
- Level 2 Routers
- Level 1 Routers
(Connect to memory controllers)
18. Tapered fat-tree topology (TFT)
• Physical design of the tapered fat-tree is more
difficult.
• The TFT topology can reduce memory latency
and power dissipation for many-core systems
19. MEMORY MODEL
• Directory-based cache coherence.
• Directory cache at every node.
• Off-chip directory controller.
• Tile-to-tile requests and responses transit the TDN.
• Off-chip memory requests and responses transit the
MDN.
20. POWER MODEL
• Like the CELL processor, unused tiles (cores) can
be put into a sleep mode to. further decrease
power consumption
• 500MHz – 866MHz operating frequency.
• ClearSpeed MTAP Co-processor.
• 15 – 22W @ 700MHz all cores active.
• Lower operating cost.
21. Multicore coherent cache
• Cache subsystem high performance, two-level,
nonblocking ,cache hierarchy.
• Each tile's cache can be shared with other tiles
each tile can access the aggregate multi-megabyte
cache.
22. Cont.
• Each tile can view the collection of on-chip caches of all
tiles, serving as an L3 cache.
• Neighborhood caching to provide an on-chip distributed
shared cache.
24. Multicore Hardwall Technology
• Enables the user to define one or many cores as
a processing island, eliminating communication
between it and other cores unless specified.
• If a packet attempts to cross the established
boundary, an interrupt is signaled and control is
passed on to the hypervisor. the established
boundary, an interrupt is signaled and control is
passed on to the hypervisor.