Memory, Evolution or Revolution?

Memory
Evolution or Revolution?

Like most things, that was then and this is now..

Given the abundance of new motherboard chipsets and new processors on the horizon, there had to be new breakthroughs in memory technology. Let’s look at what has already arrived, and what is on the horizon.

HSDRAM

On August 8, 2000, Ramtron International Corporation, through two of its subsidiaries, Enhanced Memory Systems, Inc. and Mushkin Inc., announced its next generation of the enhanced high-speed SDRAM, naming it HSDRAM. As with most SDRAM module types, it is in the form of DIMM modules. These new modules operate at CAS2 latency and enable the boosting of the memory bus speed of the Intel 815E chipset to 150 MHz and greater. Obviously, the primary market for this new SDRAM type are those computer users who over-clock their computers. Quoting David Bondurant, vice president of marketing and applications for Enhanced Memory Systems, “Enhanced Memory Systems continues to offer the fastest and lowest-latency DIMM modules for over-clocked PC systems…”

HSDRAM is a high performance version of the industry standard PC133 SDRAM. This form of  SDRAM specifies a clock rate of 150 MHz clock rate with 2:3:2 latency and an improved clock access time of 4.5 ns. This is one of the first memory modules to be specified at 150 MHz with a CAS2 latency. Tested on an ASUS CUSL2 motherboard using Intel’s 815E chipset, memory bus speeds can be pushed as high as 166 MHz with a 2:3:2 latency. 

You ask, why not push those memory chips a bit higher, given that processors have already passed the 1 GHz range, right? Well its not quite that easy. The DRAM chips are on a DIMM of about 12 cm (½”) apart, end to end. Signals travel with a speed much lower than the speed of light, somewhere around 120,000 km per second or 12 cm per ns. If the SDRAMs were clocked at around 500 MHz (cycle time = 2 ns), the clock cycle signal of the first chip would be about half a clock cycle (about 1 ns) out of phase with the previous one, thereby losing the synchronous nature of SDRAM. Hmm, so what’s the point..

High Speed SDRAM (HSDRAM) provides higher speeds than normal SDRAM, somewhere around 166 MHz as compared to the normal 133 MHz, largely because it uses higher quality memory chips. This reflects on our earlier comment about two modules that are identical according to the standard, but actually may be miles apart in actual performance. The frequency “headroom” of HSDRAM is quite small, and the advertised 25% increase in peak bandwidth is not really something to rave about.

Enhanced SDRAM (ESDRAM)

ESDRAM, is a different look at SDRAM technology. In order to overcome some of the inherent latency problems that accompany standard DRAM memory modules, chip manufacturers have inserted a small amount of SRAM directly into the chip, effectively creating an on-chip cache. One such design that is getting immediate attention is ESDRAM from Ramtron International Corporation. Ramtron announced this new SDRAM development in 1997 and proposed that it be added as an enhanced standard to current SDRAM technology. Increasing the clock speed and the width of the bus have been the most obvious target solutions, and Ramtron apparently believes that it has found one.

ESDRAM is essentially SDRAM, plus a small amount of SRAM cache which allows for lower latency times and burst operations up to 200MHz. Just as with external cache memory, the goal of a cache DRAM is to hold the most frequently used data in the SRAM cache to minimize accesses to the slower DRAM. One advantage to the on-chip SRAM is that a wider bus can be used between the SRAM and DRAM, effectively increasing the bandwidth and increasing the speed of the DRAM even when there is a cache miss. SRAM buffers can eliminate latencies like the precharge latency (Precharge time, T_rp), and they also have lower CAS, and RAS to CAS, latencies, If the buffers contain the right information, the critical data can be sent in 5 to 6 cycles instead of the typical 7 cycles of normal SDRAM. ESDRAM is based on HSDRAM chips, which were also developed by Ramtron. Unfortunately, ESDRAM offers little or no real benefit for streaming applications, as there is only a tiny bit more bandwidth. In most cases though, ESDRAM can offer allot lower critical data latencies. One nice feature of ESDRAM is that it is totally compatible with SDRAM. You can install ESDRAM into any DRAM slot. Okay, these are the good features, but what are the bad ones?

First, the BIOS has to be able to recognize that the CAS, and CAS to RAS, latencies are much lower. If it can’t, the faster SRAM (cache) of the ESDRAM will be accessed at the same rate as that of a normal SDRAM. In addition, if the SRAM does not contain the correct data, the result will be a few extra cycles compared to regular SDRAM since you will have to reload the right data. Lastly, ESDRAM is a bit expensive since it uses the highest possible quality memory chips and is not manufactured in large quantities. Ramtron has been offering ESDRAM for quite some time now, however like HSDRAM, it has be relegated to the high-end gaming or over-clocking markets. In spite of its higher price, it can provide an easy upgrade for many SDRAM systems where the motherboards BIOS is able recognize lower latencies.

VC SDRAM

Virtual Channel SDRAM, designed by NEC, also contains SRAM caches. These caches, however, are not really buffers for the sense amps like ESDRAM. VC SDRAM contain 16 virtual channels, or 16 1 KB SRAM caches. While the ESDRAM module handles caching internally, the VC SDRAM cache is managed by the chipset. This has two important consequences, in that the VC SDRAM will only work properly when paired with a chipset that supports it, and the performance of VC SDRAM depends largely on the quality of the chipset implementation.

VIA has included support for VCM in their Apollo 133 and KX/KT 133 chipsets. If you have a need for faster SDRAM, you may want to consider VC SDRAM. VC SDRAM does not have a high speed core like ESDRAM, but the NEC VC SDRAMs are high quality DIMMs offering a small, but significant boost (2-5%) to system performance. This is especially important for computers that has a CPU with a smaller cache, such as Celeron and Duron.