home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.sys.amiga.hardware
- Path: sparky!uunet!usc!sol.ctr.columbia.edu!eff!ssd.intel.com!ichips!tjehl
- From: tjehl@sedona.intel.com (Timothy Jehl)
- Subject: Pentium vs 68060 (was: Intel vs Motorola)
- Message-ID: <1992Dec11.215853.25996@ichips.intel.com>
- Originator: tjehl@sedona
- Sender: news@ichips.intel.com (News Account)
- Reply-To: tjehl@sedona.intel.com
- Organization: Intel Corporation.
- References: <andy.00a7@onkyo.incubus.sub.org>
- Date: Fri, 11 Dec 1992 21:58:53 GMT
- Lines: 142
-
-
- andy@onkyo.incubus.sub.org (Andreas Goehler) writes:
- >
- > Who knows ANYTHING about Intel's new 586 alias P5 alias Pentium ???
- >
- > Will it be 64-bit Data and Address Bus. Will it be 8-bit internal as all
- > other x86 ??? Does it include a 386 unit as planed earlier ??? Is it any
- > competition to the new MC68060 as soon as both of these processors are out.
- > Another thing, what power usage will the 586 have ???
- >
-
- PART I - the Pentium
- ====================
-
- Well, let me tell you a bit about what is public knowledge about the
- Pentium processor.
- 1) It is superscaler (two integer execution units, as well as a
- pipelined floating point unit)
- 2) Expected performance of 100+ MIPS
- 3) Floating point with a 5-10x improvement over a 33Mhz i486(tm)
- 4) 64 bit external bus
- 5) 32 bit internal integer path
- 6) 64 bit floating point path
- 7) more total cache, with separate code and data caches
- 8) 100% compatible with previous generation software
- 8) Systems in production in 1993
-
- Editorial comment: "8-bit internal bus"?!?!?! Where'd you get that?
-
-
- PART II - the 68060
- ===================
-
- Now, a reprint of a previous article which I posted to comp.sys.amiga.misc:
-
-
- 1) Design targets
-
- >3x performace over 040-25 with existing compilers
-
- ** editorial comment - there is an implication here that the architecture
- will be such that code ordering will affect performance (no surprise on
- a modern processor), and that new compiler technology will be part of the
- performance increase (i.e. the software that you already own may not get
- the full benefit of the performance increase)
-
- >2 million transitors, .5 micron triple metal, 3.3v static
-
- '040 style package with similar bus
-
- '040 code compatible for user-mode code
-
- 2) Architecture
-
- Superscaler pipeline
- 4 stage inst. fetch pipeline
- parallel 4 stage operand execution pipelines
-
- ** editorial comment - looking at the block diagram, there appears to
- to be a single floating point unit, which is attached to one of these
- pipelines, implying that you can execute two integer instructions in
- parallel, or one integer and one floating point if the code is ordered
- properly.
-
- Branch cache for instruction fetch prediction (the block diagram
- implies that it is a brance target cache, rather than a branch
- cache)
-
- Internal Harvard architecture
- Execution pipelines are in lock-step
- Simultaneous but in-order execution
- Supports precise exception model
-
- '040 compatible bus
- 32 bit address and data buses
- support for cache line bursts (16 bytes = 4 words)
-
- 3) Assorted block diagrams which I am not going to attempt to duplicate
- in an ASCII file.
-
- 4) Dispatch algorithm
-
- A bunch of verbage describing the dispatch algorithm.
-
- On integer code, analysis shows 50-60% of instructions from existing
- compilers would be executed as pairs.
-
- ** editorial comment - this is a low number, but is a result of no
- optimization on existing code. New compiler technology should be
- able to drive this number up near the 75-80% range.
-
- 5) Branch cache
-
- yep, it's a branch target cache
-
- branch history information is kept for branch prediction
-
- one clock penalty for incorrect branch prediction
-
- 6) FPU
-
- compatible with 68040 FPU programming model
- extended precision operation
- IEEE 754 compliant
-
- Implemented in one of the integer pipelines
-
- 3 units - add, multiply, divide
-
- 3 cycle adds, 4 cycle multiplies, 24 cycle divides
-
- 7) Pipeline diagram which I am not going to attempt to duplicate
- in an ASCII file.
-
- 8) Summary
-
- 060-50 w/ new compiler ~3.5x 040-25 with existing compiler
-
- design targeted for 50/66 MHz, with 3.3 V operation
-
- ** editorial comment - lower voltage operation also normally
- implies a lower frequency specification on a CMOS product
-
- production volumes in the first half of 1994
-
- PART III - the comparison
- =========================
-
- I have no intention of getting into the processor flame wars. Both
- companies have given reasonable comparisons of their next generation
- products with their present products. However, the Intel part will be
- in production systems in 1993, while the Motorola part will not be out
- before 1994.
- I hope this is at least a partial answer to some of your questions.
-
- TJ
-
-
- --
-
- "It really only matters to the people who care."
-
-