Ace's Guide to Memory Technology: Part 3
By
Johan De Gelas
Saturday, August 10, 2002 1:57 AM EDT
|
|
DDR333, DDR400, DDR-II, and 32-bit PC1066 RDRAM
What is the memory standard of the future? RDRAM, DDR400 or DDR-II? It is a
question that continues to fuel
heated discussions on many internet hardware
forums, including our own message board. In this article, we'd like to offer you
some insight in how these different technologies compare. This article's
difficulty level is a bit steep, but we are convinced
that with a bit of technical background you will be able to make a wiser
choice between the different DRAM types out there.
|
First, I'd like to immediately address a question that some of our readers
may have: "why still include RDRAM?" Indeed, over the past months there
have been many reports that RDRAM is dead. Intel has indeed stopped RDRAM
chipset development efforts and it is clear that DDR-II has a much brighter
future ahead. Nevertheless, RDRAM is far from dead.
Despite Intel's half-hearted launch of the i850E, the good performance
of PC1066 has sparked some interest and quite a few manufactures have launched
or will launch boards with official PC1066 support. The largest motherboard
manufacturer, ASUS has launched two motherboards officially supporting PC1066
RDRAM. DFI, EPOX, BIOSTAR and IWill have followed in ASUS footsteps.
| Advertisement:
|
On top of that, Silicon Integrated Systems (SiS), the manufacturer
of the first DDR400 chipset for the P4, has launched the PC1066 RDRAM-
based R658. At the same time, Intel is still validating PC1066 support
for the i850E. After erasing RDRAM from their roadmap a few months ago,
Elpida
(NEC + Hitachi) has decided to go ahead with the PC1066 RDRAM production
anyway. So while DDR is by far the most popular DRAM in today's market,
RDRAM can still be found in quite a few high-end desktops and workstations.
As many of our readers are power-users, we decided to take a look at the
different chipsets and DRAM technologies available for the Pentium 4 platform
once more.
But before we start talking about benchmarks in our next article,
this article will provide some technical background about the differences
between RDRAM and DDR SDRAM. The objective is to understand which technology
has the best chances in the long term. What the strong and weak point are
of DDR SDRAM and RDRAM? Why RDRAM can reach so much higher clockspeeds?
Can DDR-II adopt some of the techniques that make RDRAM a faster clocked
technology?
This article will be pretty technical, but even if you are not so technically
inclined, don't worry. Every interested reader will be able to understand
the discussion. However, to understand this article well, you should understand
the basics of DRAM memory, which are explained in our previous articles
here.
High Clockspeeds for DRAM
Many articles point out that RDRAM uses a 16-bit bus for the data signals
and that this narrow 16-bit path is the main reason why RDRAM is able to
run at speeds up to 533 MHz (DDR, effective clockrate: 1066 MHz). Indeed, one of
the problems that parallel
wired systems face is skew. The longer and the faster
a synchronous bus (with
a clocksignal) gets, the more likely it is that some
data signals will arrive
"too soon" or "too late" and are not within the limits
of the clocksignal.
That is why we see that many I/O interfaces go from typical
slow parallel
busses to fast serial busses. Examples are the move from the
parallel port
to USB or the current IDE ATA to serial ATA.
But while a 16-bit datapath can be clocked a bit higher than a 64-bit
one, it does not explain the differences in clockspeed between RDRAM and
DDR SDRAM. The very best DDR SDRAM overclocks to 200 MHz or slightly more, while the best RDRAM has been able to run at 640 MHz (stable)
and even more when we tested. Interestingly, overclocking from 166 MHz (PC2700) and 533 MHz (PC1066), both memory technologies achieve a 17% overclock. There'll be a full report on this later on.
Packaging is another reason why RDRAM is able to run faster than DDR
SDRAM. Most DDR SDRAM uses TSOP (Thin Small Outline Package), while RDRAM uses
FBGA or Fine pitch Ball Grid array. TSOP chips have rather long contact
pins on each side, while FBGA chips have tiny balls at the underside to
make contact with the outside world. Those very small soldered balls have
a much lower capacitive load than the TSOP pins. In fact, industry sources
told us that the current FBGA DDR SDRAM chips are able to run at 200-266
MHz, while the same chips with TSOP packaging are limited to 150-180 MHz.
Therefore a decent DDR400 module should come FBGA and not TSOP packaged
chips. Of course, this doesn't mean that FBGA modules with mediocre memory
chips are superior to TSOP modules with quality memory chips.
Nevertheless, it is clear that the manufacturers who produce DDR-400
(Samsung and Micron) in TSOP packaging are somewhat stretching the possibilities
of their chips. Don't expect high overclocking margins and be careful with
highly loaded (= all sockets filled) memory systems. Already we have noticed
that quite a few unofficial (=non JEDEC) DDR2700 CAS 2.5 won't work at
166 MHz DDR on many boards when you install two or more DIMMs of 512 MB.
While we have yet to perform a comprehensive test of DDR400 modules, it
is likely that we will encounter the same problems. It is no coincidence
that JEDEC has decided that all DDR-II modules, which will work at 200
and 266 MHz, will be FBGA chips.
All Content is Copyright (C) 1998-2003 Ace's Hardware. All Rights Reserved.
|
