STPC ATLAS
76/109
Issue 1.1 - March 23, 2004
6.2. STPC CONFIGURATION
The STPC is a very flexible product thanks to
decoupled clock domains and to strap options
enabling a user-optimized configuration.
As some trade off are often necessary, it is
important to do an analysis of the application
needs prior to design a system based on this
product. The applicative constraints are usually
the following:
- CPU performance
- graphics / video performances
- power consumption
- PCI bandwidth
- booting time
- EMC
Some other elements can help to tune the choice:
- Code size of CPU Consuming tasks
- Data size and location
On the STPC side, the configurable parameters
are the following:
- Synchronous / asynchronous mode
- HCLK speed
- MCLK speed
- Local Bus / ISA bus
6.2.1. LOCAL BUS / ISA BUS
The selection between the ISA bus and the Local
Bus is relatively simple. The first one is a standard
bus but slow. The Local Bus is fast and
programmable but doesn't support any DMA nor
below summarize the selection:
Before implementing a function requiring DMA
capability on the ISA bus, it is recommended to
check if it exists on PCI, or if it can be
implemented differently, in order to use the local
bus mode.
6.2.2. CLOCK CONFIGURATION
The CPU clock and the memory clock are
independent unless the "synchronous mode"
strap option is set (see the STRAP OPTIONS
chapter). The potential clock configurations are
The
advantage
of
the
synchronous
mode
compared to the asynchronous mode is a lower
latency when accessing SDRAM from the CPU or
the PCI (saves 4 MCLK cycles for the first access
of the burst). For the same CPU to Memory
transfer performance, MCLK has to be roughly
higher by 20MHz between SYNC and ASYNC
modes to get the same system performance level
(example: 66MHz SYNC = 86MHz ASYNC) .
In all cases, use SDRAM with CAS Latency
equals to 2 (CL2) for the best performances.
The advantage of the asynchronous mode is the
capability to reprogram the MCLK speed on the
fly. This could help for applications where power
consumption must be optimized.
The last, and more complex, information to
consider is the behaviour of the software. In case
high CPU or FPU computation is needed, it is
sometime better to be in DX2-133/MCLK=66
synchronous
mode
than
DX2-133/MCLK=90
asynchronous mode. This depends on the locality
of the number crunching code and the amount of
data manipulated.
right column correspond to the configuration
Obviously, the values for HCLK or MCLK can be
reduced compared to
Table 6-2 in case there is no
need to push the device at its limits, or when
avoiding to use specific frequency ranges (FM
radio band for example).
6.3. ARCHITECTURE RECOMMENDATIONS
This
section
describes
the
recommend
implementations for the STPC interfaces. For
more
details,
download
the
Reference
Schematics from the STPC web site.
Table 6-1. Bus mode selection
Need
Selection
Legacy I/O device (Floppy, ...), Super I/O
ISA Bus
DMA capability (Soundblaster)
ISA Bus
Flash, SRAM, basic I/O device
Local Bus
Fast boot
Local Bus
Boot flash of 4MB or more
Local Bus
Programmable Chip Select
Local Bus
Table 6-2. Main STPC modes
CMode
HCLK
MHz
CPU clock
clock ratio
MCLK
MHz
1
Synchronous
66
133 (x2)
66
2
Asynchronous
66
133 (x2)
90
Table 6-3. Clock mode selection
Constraints
C
Need CPU power
Critical code fits into L1 cache
1
Need CPU power
Code or data does not fit into L1 cache
3
Need flexible SDRAM speed
2