home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.sys.sgi
- Path: sparky!uunet!cis.ohio-state.edu!magnus.acs.ohio-state.edu!usenet.ins.cwru.edu!wsu-cs!hal!atems
- From: atems@hal.physics.wayne.edu (Dale Atems)
- Subject: Memory interface and performance of R4K Indigo
- Message-ID: <1992Jul28.021405.27412@cs.wayne.edu>
- Followup-To: Re: Memory interface in R4000 Indigo
- Originator: atems@hal
- Sender: usenet@cs.wayne.edu (Usenet News)
- Organization: Dept. of Physics, Wayne State University
- Date: Tue, 28 Jul 1992 02:14:05 GMT
- Lines: 18
-
- Thanks to all who responded to my question about the CPU-main memory interface
- on the R4000 Indigo. All are agreed that the penalty for a secondary cache miss
- is about 100 CPU cycles. This brings up two questions:
-
- 1). Why 100 cycles? If the CPU is clocked at 50 MHz, a single clock cycle takes
- 20 ns. If main memory consists of 80 ns DRAM, I would expect a penalty on the
- order of 4-5 CPU cycles, not 100. (I assume my understanding of these issues
- is flawed somewhere. Please correct me here, I do not claim any expertise.)
-
- 2). Standard CPU/FPU benchmarks should fit easily into a 1MB cache, so I don't
- expect to see the effects of cache misses reflected there. Has SGI/Mips done
- any internal studies on how significant this effect will be on the performance
- of "real" code?
-
- Thanks for any insights/information,
-
- Dale Atems
- E-Mail: atems@hal.physics.wayne.edu
-