Contents | < Browse | Browse >

                              MOTOROLA 68040

The primary feature of the G-Force is the Motorola MC68040RC-33
microprocessor.  Some features of the 68040:

* 6 stage pipeline, MC68030 compatible integer unit

* MC68881/882 compatible FPU

* Independant instruction and data MMUs

* Simultaneously accessible, 4K instruction and data caches

* Multimaster/multiprocessor support via bus snooping

* Concurrent IU, FPU, MMU, and Bus Controller operation

* 4 Gigabyte address space

I am very impressed with the performance, and it is quite a step up from my
28MHz 68030/68882.  The acceleration suffers a very noticable loss of speed
without the 68040's special modes (cache, burst, copyback) but still stays
leaps and bounds faster than a comparable 68030.  The 68040 has a fan and
heatsink mounted on the chip.

As long as there is available 32-bit RAM, you can expect full speed from
your system.  If there is any 16-bit fast RAM, you can expect RAM access to
be SLOWER (it will run at ~98% of the speed of an unaccelerated A2000)
*when* you run out of 32-bit RAM.  This is to be expected.  16-bit RAM is
not a good idea in an accelerated system.  If present, it should be clearly
understood that the system will perform very poorly when the 16-bit RAM is
used.  I eventually got rid of all my 16-bit RAM.  The RAM is placed
outside the 24-bit address space of Zorro II, so using cards that need that
space such as 24-bit video cards, bridgeboards, and RAM expansions are not

There is also a software selectable fallback mode which disables the 68040,
but you lose both the SCSI host adapter and all RAM expansion.

Before you view the "benchmarks" remember that benchmarks can be
misleading.  Ultimately, the only way to realize the value of any product
you need to try using the system yourself.  Note that the RAM seems
somewhat slow compared to an A3000.  This is due to the copyback cache and
the fact than my particular G-Force is not in "burst" mode.  AIBB reports
that the RAM runs at 0.9 times the speed of an A3000 with copyback, 1.52
times faster without copyback.

View benchmark information