1 / 21

EV7

EV7. Peter Bannon Staff Fellow HP pbannon@hp.com. Alpha Microprocessor Roadmap. EV7 Features. Alpha 21264 core with enhancements Integrated L2 Cache Integrated memory controller Integrated network interface. M. IO. M. M. IO. IO. M. IO. IO. M. IO. M. M. M. IO. M. IO. M.

lise
Télécharger la présentation

EV7

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. EV7 Peter Bannon Staff Fellow HP pbannon@hp.com

  2. Alpha Microprocessor Roadmap BARC 2003

  3. EV7 Features • Alpha 21264 core with enhancements • Integrated L2 Cache • Integrated memory controller • Integrated network interface BARC 2003

  4. M IO M M IO IO M IO IO M IO M M M IO M IO M IO IO M IO M IO EV7 System Block Diagram 364 364 364 364 364 364 364 364 364 364 364 364 BARC 2003

  5. I/O Router Mem0 Mem1 L2 Tag N E S W L2 Data L2 Data P 0 1 2 3 P 4 5 6 7 EV68 Core BARC 2003

  6. Integrated L2 Cache • 1.75 MB, 7-way set associative, with ECC • 20 GB/s total read/write bandwidth • 16 Victim buffers for L1 -> L2 • 16 Victim buffers for L2 -> Memory • 9.6ns load to use latency • Tag access start every cycle • Data access in 4 cycle blocks • Couple Tag/Data access to minimize latency • Decoupled Tag access to minimize resource use. BARC 2003

  7. Two Integrated Memory Controllers • RDRAM memory • Directly connect to the processor • High data capacity per pin • 800 Mb/s operation • 75ns load to use latency • 12.8 GB/sec peak bandwidth • 6 GB/sec read or write bandwidth • 2048 open pages • 64 entry directory based cache coherence engine • ECC SECDED • Optional 4+1 parity in memory BARC 2003

  8. Cache Coherence Engine DRAM Scheduling Data Path QUE ROW ROW OUT 32 MEMORY REFERENCES 4 PRQ ROW COL COL OUT COL 4 RSQ PA DATA OUT SLOT MAP Data R DATA IN D 8 RCAS CC STATE MACHINE CHK COR REMAP R 8 WCAS A M Directory out Directoryin Data to Core ZBox Block Diagram BARC 2003

  9. Integrated Network Interface • Direct processor-to-processor interconnect • 4 links 6.4 GB/second per link • 32 bits + ECC at 800 Mb/s each direction • 18ns processor-to-processor latency • ECC, single error correct, double error detect, per hop • Out-of-order network with adaptive routing • IO, Request, Forward, Special, and Response channels • VO, V1, and Adaptive virtual networks • Asynchronous clocking between processors • 3 GB/second I/O interface per processor BARC 2003

  10. Bytes Adap. V0 V1 Request 12 8 1 1 Forward 12 8 1 1 Block Resp. 76 3 1 1 Resp. 12 8 1 1 Write I/O 76 1 2 2 Read I/O 12 1 2 2 Special 12 8 0 0 ~1KB per port E W N S IO C Z0 Z1 I I I I I I I I Q Q Q Q Q Q Q Q OF O O O O O O O E W N S IO L0 L1 Rbox Block Diagram BARC 2003

  11. while (p) p=*p; BARC 2003

  12. BARC 2003

  13. BARC 2003

  14. CPU INT 2000 BARC 2003

  15. CPU FP 2000 BARC 2003

  16. Database Performance BARC 2003

  17. BARC 2003

  18. BARC 2003

  19. 64P Running TTOY, 32 memory controllers BARC 2003

  20. 64P Running TTOY, 64 memory controllers BARC 2003

  21. BARC 2003

More Related