Hardware Reference
In-Depth Information
While waiting for the results, the CPU may well be able to continue with other
work. For example, a cache miss while prefetching an instruction does not inhibit
the execution of one or more instructions already fetched, each of which may refer
to data not in any cache. Thus, multiple transactions on the two LPDDR2 inter-
faces may be outstanding at once, even for the same processor. It is up to the
memory controller to keep track of all this and to make actual memory requests in
the most efficient order.
When data finally arrives from the memory, it can come in 4 bytes at a time. A
memory operation may utilize a burst mode read or write, which will allow multi-
ple contiguous addresses within the same DRAM row to be read or written. This
mode is particularly efficient for reading or writing cache blocks. Just for the
record, the description of the OMAP4430 given above, like that of the Core i7 be-
fore it, has been highly simplified, but the essence of its operation has been de-
scribed.
The OMAP4430 comes in a 547-pin
ball grid array
(PBGA), as shown in
Fig. 3-48. A ball grid array is similar to a land grid array except that the con-
nections on the chip are small metal balls, rather than the square pads used in the
LGA. The two packages are not compatible, providing further evidence that you
cannot stick a square peg into a round hole. The OMAP4430's package consists of
a rectangular array of 28
26 balls, with two inner rings of balls missing, plus two
asymmetric half rows and columns of balls missing to prevent the chip from being
inserted incorrectly in the BGA socket.
×
Figure 3-48.
The OMAP4430 system-on-a-chip pinout.
It is difficult to compare a CISC chip (like the Core i7) and a RISC chip (like
the OMAP4430) based on clock speed alone. For example, the two ARM A9 cores
in the OMAP4430 have a peak execution speed of four instructions per clock
cycle, giving it almost the same execution rate as that of the Core i7's 4-wide
superscalar processors. The Core i7 achieves faster program execution, however,
since it has up to six processors running with a clock speed 3.5 times faster (3.5
GHz) than the OMAP4430. The OMAP4430 may seem like a turtle running next
