Site sponsored by: Idera Try Idera’s new SQL admin toolset
SQL Server Performance

  • Home
  • Articles
  • Forums
  • Tips
  • Quiz
  • FAQ's
  • Blogs
  • Software
  • Books
  • About Us
RSS Feeds
Sign in | Join


Article Topics

All Articles
Performance Tuning
Audit
Business Intelligence
Clustering
Reporting Services
Developer
General DBA
ASP.NET / ADO.NET

Write for Us

Share you SQL Server knowledge with others and raise your profile in the community More...
Latest Articles

Filtered Indexes in SQL Server 2008
Importance of Database Backups and Recovery Plan
Data Compression in SQL Server 2008
SQL Server 2008 MERGE Statement

More     
 
Latest FAQ's

ALTER TABLE SWITCH statement failed because the object '%.*ls' is not ...
ALTER TABLE SWITCH statement failed because column '%.*ls' at ordinal %d ...
ALTER TABLE SWITCH statement failed because table '%.*ls' has %d columns ...
SQL Server Reporting Server (SSRS) service is failing to start ...

More     
   
Latest Software Reviews

Spotlight on ApexSQL Doc 2008
ApexSQL Enforce
Embarcadero Change Manager
SQL Server DBA Dashboard

More     

articles >> performance tuning >> Processor Performance, 2002

Processor Performance, 2002

By : Joe Chang
Jul 31, 2004

Page 2 / 5



Processor Frequency

Performance generally does not increase exactly in step with frequency. That is, a 10% frequency increase should not lead to a 10% performance gain. In order for performance to scale in step with frequency, the program must either run entirely within on-die cache, or non-cached memory accesses must be pre-fetched with sufficient lead time to completely hide the memory access latency. Memory access times, including signal delays, are on the order of 100ns. A 1GHz processor has a clock cycle time of 1ns, so completely hiding memory access latency is not a simple matter. Performance will not scale well with frequency if either memory bandwidth or access latency are limiting. A program that performs simple operations on a large memory set can be constrained by memory bandwidth. A program that depends on serialized memory accesses, where one memory access must be completed before the next memory access can be issued, will be performance bottlenecked by memory latency.

Figure 5 shows the Pentium III performance from 800MHz to 1000MHz. All processors have 256K L2 cache, 133MHz FSB and use the Intel D815EEA2 motherboard. Note the unusual GZIP results. The 866MHz and 1GHz results are better than expected based on the 800MHz and 933MHz results.

Figure 5. Pentium III performance from 800MHz to 1GHz, 256K, 133MHz FSB.

Figure 6 shows the Pentium 4 performance from 2.26GHz to 2.8GHz. All processors have 512K L2 cache, 533MHz FSB and use the Intel D850EMV2 motherboard, and the compiler is version 6.0.

Figure 6. Pentium 4 performance from 2.26GHz to 2.80GHz, 533MHz FSB.

Figure 7 below shows the Pentium III percentage performance gain from 866MHz to 1GHz, a frequency increase of 15.5%. EON, CRAFTY and PERLBMK performance gains are nearly instep with frequency increase. MCF has negligible performance gain with frequency. Overall performance gain is 0.63% per 1% frequency increase.

Figure 7. Pentium III performance gain from 866MHz to 1GHz.

Figure 8 shows the Pentium 4 percentage performance gain from 2.26GHz to 2.80GHz, a frequency increase of 23.5%. The Eon performance gain is 25.4%, which is more than what should be possible, unless there is some thing else happening. Overall performance gain is 0.67% per 1% frequency increase.

Figure 8. Pentium 4 performance gain from 2.26 to 2.80GHz, 533MHz FSB.

Some applications scale well with frequency (GZIP, CRAFTY, EON and PERLBMK). Others less so and MCF scales poorly.


<< Prev Page     Next Page>>    








Home | Peformance Articles | Audit Articles | Business Intelligence Articles | Clustering Articles | Developer Articles | Reporting Services Articles | DBA Articles | ASP.NET / ADO.NET Articles | DBA FAQ's | Developer Peformance FAQ's | DBA Peformance FAQ's | Developer FAQ's | Clustering FAQ's | Error Messages | Audit Tool Reviews | Backup Tool Reviews | Coding Tool Reviews | Compare Tool Reviews | Documentation Tool Reviews | Design Tool Reviews | Monitoring Tool Reviews | Log Tool Reviews | Reporting Tool Reviews | Clustering Tool Reviews | Security Tool Reviews | Change Management Tool Reviews | Remote Access Tool Reviews | Book Reviews | Security Tool Reviews | QDPMA Performance Tuning | ADO.NET / ASP.NET | Administration | Analysis/OLAP Services | Application Development | Configuration | Components | ETL | Hardware | High Availability | Hints | Index | Misc | Operating Systems | Performance Tuning | Replication | T-SQL | Views


              © 1999-2008 by T10 Media. All rights reserved