Re: complaints about SPEC and a request for FORTRAN validation codes

jwreilly@mipos2.intel.com (Jeffrey Reilly) writes:
>- These are benchmarks aimed at measuring CPU performance (not compiler
> compliance).


grover@brahmand.Eng.Sun.COM (Vinod Grover) writes:
>That is surprising. I thought that SPEC benchmarks were for measuring
>*system performance*. For many benchmarks runtimes are dependent on system
>features such as malloc (espresso), alloca (gcc), and multiprocessing
>(compress), curses libraries (sc). A system having faster versions of
>these can skew SPECmarks significantly.


Relevant points... however I would also be careful about using the term
"system performance"; the term "system" having different meanings to
different people. Networking, video, audio and I/O capability would be
things I would look for in a "system" benchmark. CINT92 and CFP92 do not
cover the performance of these "system" aspects as well (if at all).


Perhaps the following would be more accurate...


These benchmarks (CINT92 and CFP92) are computationally (or CPU)
intensive. As such, they are significantly dependent on the processor (or
processors/ coprocessors), memory subsystem and compiler (both generated
code and libraries) technology.


Jeff
SPEC CINT92/CFP92 Release Manager


Jeff Reilly
Intel Corporation
jwreilly@mipos2.intel.com
(408) 765 - 5909
--

Return to the comp.compilers page.
Search the comp.compilers archives again.