How can you have a memory of the future? Not that kind of memory silly! The future holds some interesting changes in Random Access Memory for computers, shoes lets us know what to expect.

 

The nature of most computer games is such that responsiveness from the machines that run them is a very important part of the player’s experience with the game. The key to this responsiveness is usually a good amount of memory. Random Access Memory has come a long way since the days of the 4MB SIMM modules used along side the first 486 processors. There are many different kinds and sizes of RAM, and for a while now it has been pretty much widely accepted that 1 gigabyte of decent speed memory would suffice for even the meatiest of systems. This idea was recently challenged when, with the release of Battlefield 2, EA demonstrated that 1GB of memory is in fact not always enough. With the future holding even more demanding games and computer components that are sure to gobble up bandwidth like the cookie monster out of rehab, we will be seeing the arrival of several new memory interfaces in the next two years. The following will attempt to explain the basic characteristics of these different kinds of memory and the advantages they should provide to computers of the future.

XDR
When Rambus, in a partnership with Intel, introduced their RDRAM to answer the demand for higher memory clock speeds from bandwidth-hungry components, it was met with stiff competition from a new DDR SDRAM. The race to the market was very close, and it came down to availability, price, and chipset support. Despite it’s superior performance with updated Intel chipsets for the Pentium 4, and although it’s potential for the future was vast, Rambus’ RDRAM eventually failed as a result of ridiculously expensive development costs along with Intel’s inability to persuade manufacturers, through benchmarks, that RDRAM was the way to go.

That was six years ago, and now Rambus is at it again.

Their new product is called eXtreme Data Rate (XDR) DRAM and it is already making waves in the PC industry for its implementation as the standard memory solution for cell processors, like the one that will be used in the Playstation 3. It is quite simply the fastest memory technology in existence; Samsung proved this point in January 2005 with the release of its first 256MB module with a transfer speed of 8 gigabytes per second. Rambus has already taken advantage of a few of XDR’s features to gain a foothold in the computing market.

One such aspect of XDR that has proven to be very good for Rambus is their FlexPhase technology. Have you ever noticed the little gold strips that cover motherboards and other PCBs in intricate patterns? Those are just the wires, called traces, which carry signals from component to component. The reason they are so bungled and complicated is because a signal must arrive at a certain time to a component, and the length of the trace determines how long it takes for the signal to get there. As computers get more and more complicated and bus widths increase as often as they do, it gets increasingly difficult to manipulate the traces in such a way that everything is timed out perfectly. FlexPhase technology solves this problem by aligning the data on the chip with each clock in a way that will allow for synchronization with the associated clocks, meaning precise data transfers. This not only simplifies the manufacturing process for systems that utilize XDR, but also allows the overall size of systems to decrease. This quality alone makes XDR very attractive for adoption into next generation equipment.

 

{mospagebreak heading=Intro, XDR&title=XDR (cont.), DDR3}
As mentioned before, XDR also provides some enormous bandwidth potential. This is the result of its Dynamic Point-to-Point Technology. DPP basically allows XDR to operate automatically in multiple channel modes. When another XDIMM (the term used for a stick of XDR) is added to an existing setup, the whole thing is dynamically reconfigured to make it a dual channel setup. Rambus has said this can be done with up to 8 XDIMMS for now, and they expect to be able to have it work with more in the future. The aforementioned 8GB/s is the bandwidth of one module on its single channel. Expectedly, an 8 channel setup would produce 64GB/s of bandwidth, which is insane. To break down this number, take the memory clock of 500Mhz and multiply it by the amount of bytes per transfer, 2, the amount of transfers per clock, 8, and the amount of channels, 8. Rambus also has plans to design XDR modules capable of much higher clock speeds, which would increase bandwidth even more. However, chipset and processor manufacturers really are not able to utilize this kind of bandwidth quite yet, and for this reason XDR, in the form of XDR2 (potentially 2 times the bandwidth of XDR), is expected to be used as memory on graphics cards before it sees any kind of mainstream PC action.

DDR3
The third and fastest generation of the memory standard that overcame RDRAM has been having a large impact on the PC scene for a while now. Even though its only implementation thus far has been as GDDR3 (Graphics Double Data Rate), the advantages it holds over other RAM types is very apparent. DDR3 maintains the same basic properties as DDR2, but also uses less power and can scale quite a bit higher. Voltage requirements went from 2.6 to 1.8 and now to 1.5 with DDR, DDR2, and DDR3 respectively. This low voltage allows the clock speed of the DDR3 to scale higher without the heat issues that would occur with DDR and DDR2. This quality makes DDR3 RAM ideal for laptops and other low-power consuming products like HTPCs. The high potential clock speeds also make it quite attractive for high-end desktops - less heat and more speed, what every enthusiast wants.

There really is not much to be said about DDR3 other than its speed, as basically all of the fundamental characteristics are the same as DDR2. The fastest proposed DDR3 offers a peak bandwidth of 12.8GBps for a single channel configuration. This number is obtained by multiplying the maximum clock rate of 800Hz with the number of bytes per transfer, 8, and the number of transfers per clock, 2. As is the case with its predecessors, DDR3 can also be used in dual-channel, which brings the maximum theoretical bandwidth up to 25.6GBps. Again, this amount of bandwidth is quite a bit higher than what can be utilized by processor and chipset manufacturers, which again is why we see GDDR3 memory on graphics cards.

 

{mospagebreak title=FB DIMM}
FB DIMM
For a long time now, servers have been using Error Checking and Correction (ECC) registered memory for making sure the data transferred through the RAM is accurate and free of errors. This RAM has always been slower than the normal, unbuffered memory found in most desktops because data transfers are basically delayed so that the data can be checked. The introduction of Fully Buffered Dual In-line Memory Modules (FBDIMM) changes all that by being the first memory interface to really use the serial interface to its advantage. The serial interface basically allows less data to flow at a time, but at much higher rates than the parallel interface. FB DIMM memory is a variant of DDR2 RAM that incorporates something called an Advanced Memory Buffer (AMB) chip on each module. These modules are stringed together through a bi-directional serial bus, and data is coordinated through each module until it reaches the first module that serves as the host to the memory controller. The role the AMB plays is sort of like a supervisor for the little pieces of information. It repeats, checks, and corrects the information on each module so that it is precisely coordinated with the memory controller. This means that all modules stay in sync in the event of an error, because all of the AMBs are checking data at the same time.

The reason FBDIMM is faster than traditional ECC Registered memory is due to the inherent speed advantages of the serial interface. FBDIMMs also hold an advantage in potential capacity over normal ECC memory. Thanks to 64-bit technology in processors that can handle much larger amounts memory than their 32-bit counterparts, FBDIMM is able to run in as many as 6 channels. Maxing out each channel with 8 modules, and each dual layer module to its extents, the potential capacity of an FBDIMM memory system is 192GB per controller.

Granted, with the other memory options available for gaming systems, FBDIMM RAM might not be the best choice. However, the advantages of checking memory are pretty clear: no errors. It has been considered that the next best type of application to use ECC capable memory on, after workstation/server tasks, is gaming. Although not as fast as XDR or DDR3, FBDIMM should not be overlooked as a memory solution for a future gaming computer.

Both XDR and DDR3 have some amazing features that are sure to make computers of the future a dream for gaming enthusiasts looking for maximum responsiveness. They both offer ridiculous amounts of bandwidth, which makes me wonder what kind of beastly processors will eventually utilize it all. It should be very interesting to see how the future PC market ends up after such a hard choice between the two. Round two of the memory face-off is well underway; can Rambus’ technology succeed this time around? Nobody knows for sure quite yet, and nobody will know until the first systems utilizing these technologies show their stuff, but for now we can look forward to the what the future memory scene has in store for us.

Readers interested in learning more about the technical aspect of RAM, refer to ZeGermans’ article, here.

Credits:
Information for this article was gathered from these manufacturer’s websites:
-SAMSUNG Semiconductor
-Rambus
-Infineon Technologies

Popularity: 4% [?]

Support FPSLabs! Share and Enjoy:
  • Digg
  • del.icio.us
  • Facebook
  • Mixx
  • Google
  • Reddit
  • Slashdot
  • StumbleUpon

You Should Also Check Out These Post:

More Active Posts: