It’s been quiet a while since Rambus released RDRAM.
However, adoption of this new technology has been somewhat slow, despite support
from Intel in its 820 chipset-based motherboards. One of the reasons for this
has been the much higher cost of RDRAM compared to SDRAM.
The
major difference between RDRAM and other kinds of memory is the clock speed.
Whereas the memory used in most computers today runs at 100 MHz or 133 MHz,
RDRAM has the capability to scale up to 800 MHz. It’s available in three
different speeds–600 MHz, 700 MHz, and the fastest and costliest version at
800 MHz. It may seem like 800 MHz RDRAM would be six times faster than 133 MHz
SDRAM, but this is not true, since the true measure of the capabilities of RAM
is not just its frequency, but also factors like bandwidth, latency, etc. These
are some vital areas where RDRAM doesn’t outperform SDRAM as much as one would
expect.
Memory bandwidth
To understand how data is transferred from the memory using a
bus, imagine a highway with multiple lanes. Each lane has the capability to
accommodate one car. Depending on the number of lanes, a highway will have the
capacity to handle more and more traffic. This is exactly how a bus works as
well. In our case, these "cars" are nothing but single bits coming
from memory. The frequency of the memory decides how quickly traffic exits the
highway. The number of "lanes" differs between different memories.
SDRAM has a 64-bit bus width. This means that at 133 MHz, its bandwidth stands
at 1.064 GB/sec (64x133/8). In contrast, RDRAM has a 16-bit bus width but runs
at 800 MHz giving it a theoretical bandwidth of 1.6 GB/sec (16x800/8). Thus,
these figures show that RDRAM will not work all that fast compared to SDRAM.
To say that RDRAM has 800 MHz clock is not entirely accurate
as well. Actually, it’s fed with a 400 MHz clock, but transfers data on the
rising as well as falling edge of the clock pulse, thus effectively making it
work at 800 MHz. Though this works out well in the case of PC800 RAM, it creates
problems in slower versions. For example, in the case of PC600 RDRAM, the
external clock speed is actually a measly 266 MHz, with effective clock speed at
532 MHz. A little mathematics, and the bandwidth for PC600 RDRAM comes out to be
1.064 GB/sec, which happens to be exactly the same as PC133 SDRAM.
Memory organization
SDRAM
is usually mounted on a small PCB (printed circuit board) that’s called a DIMM
(Dual Inline Memory Module). RDRAM, on the other hand, is mounted on a different
slot called RIMM (Rambus Inline Memory Module). The number of SDRAM or RDRAM
chips in a module differ, depending on the bus width supported by individual RAM
chips. For example, one SDRAM module –which has a 64-bit data bus–could have
sixteen 4-bit wide or eight 8-bit wide SDRAM chips.
SDRAM is parallel in nature. This means that all memory chips
in a module of SDRAM are connected in parallel to the data bus that’s used to
read and write data. On the other hand, RDRAM has a serial nature in that
individual chips are connected in serial to the data bus. So, it’s arranged
logically as a single strip of RDRAMs through which the common 16-bit bus runs.
As a result of this, no RIMM slot on a motherboard can be left empty and must be
filled up by a special kind of PCB, called a continuity module.
Latency
All memories take some time to process a request for data and
transfer the same. This is called latency. Due to the serial nature of RDRAM,
the chips closest to the memory controller take much less time to respond to the
controller, compared to those that are located further away. This difference in
time can be quite a lot, since the farthest RDRAM chip can be about a foot away
from the memory controller. Hence, the controller must find a way to manage all
these different latencies. To do this, the controller finds out the highest
latency value in all the RDRAM chips during the boot phase, and then programs
the rest to work at that latency. Thus, even though the actual latency for RDRAM
may be very low, more often that not, the RIMM ends up working at a much higher
latency value.
Both the latest SDRAM as well as RDRAM have 20 nanoseconds
latency. But because of the reasons given above, RDRAM always has latency
greater than published figures.
To the benchmarks
We decided to see for ourselves how the two technologies
performed in comparison to each other. Our test setup included 128 MB of PC800
RDRAM on an Intel VC820 motherboard and 128 MB of PC100 SDRAM on an Asus CUBX
motherboard. All testing was done using a GeForce display adapter, PIII/700 MHz
CPU and 5,400 rpm Seagate hard drives.
First up, Business Winstone 99. This benchmark is a popular
suite of tests used for evaluating performance in business applications like MS
Office, Lotus SmartSuite, etc. This is an area where the primary advantage of
RDRAM–its bandwidth–is not stressed very much. The results favored SDRAM
slightly, as it managed to score 29.6 in comparison to RDRAM’s 28.8.
Next came Quake III Arena. Games are known to be very heavy
applications and quickly saturate whatever bandwidth and processing power you
throw at them. However, with the continued development of more and more advanced
AGP cards, the focus has been shifting away from the RAM and processor to the
display card. Our results showed as much, as both our test systems consistently
gave us the same frame rates. So, we thought we’d use CAD applications for
testing, thinking that we might be able to find some difference. For this, we
used the Indy3D benchmark to stress the PC. The benchmark has highly
customizable tests to stress your PC to the limit. No luck here either.
Everywhere we would get the same scores, no matter what we’d try.
The verdict
RDRAM has certainly been a step in the right direction. All
indicators in the current-day scenario point towards a future where more and
more bandwidth will be required. SDRAM simply cannot satisfy this insatiable
appetite. However, one has to think whether RDRAM is the right solution. The
most prohibitive factor in the whole RDRAM affair is its exorbitant cost. One
simply cannot be expected to pay the same amount for RAM, as for a whole new
computer. Then there are performance issues as well. A new memory called DDR
SDRAM is just around the corner. Expected to cost a little more than the current
SDRAM, DDR (Double Data Rate) memory will have even higher bandwidth than the
best RDRAM available today. Hence, for now, RDRAM seems to be a waste of money.