I am sure there are many of you that have heard of, but aren’t sure about what DualDDR is, so let’s spend a moment to give you a brief introduction to the technology.
Your typical memory bus today is 64-bits. However, the nForce2 Platform Processors utilize a new and higher performance 128-bit (effective) memory bus, theoretically capable of delivering 100% more peak bandwidth.
Based upon the first-generation nForce Platform Processors, NVIDIA’s DualDDR design again employs two 64-bit memory controllers, providing an effective 128 bit memory interface, while utilizing new proprietary algorithms for pre-fetching and pre-processing memory requests.
More than just a single “128-bit” memory controller, DualDDR consists of two independent, complementary, and intelligent memory controllers. Both memory controllers operate concurrently to each other to “hide” latencies associated with typical “chipsets”. For example, controller “A” reads or writes to main memory while the controller “B” prepares for the next access, and vice versa. The complementary nature of the two memory controllers cuts the effective latency in half.
Equally important is the second-generation dynamic adaptive speculative pre-processor (DASP), which has been re-architect with a more aggressive algorithm. Write buffers and draining algorithms has also been re-architected to reduce latency.
To sum it up, the nForce2 utilize two 64-bit memory controllers, providing a 128 bit effective memory interface and more aggressive timings to reduce latency, boosting the performance by 100% over the typical 64-bit DIMMs. In addition, DualDDR also doubles the addressable memory space to 3GB.
With the introduction of the NForce2 chipset with DCDDR and the ability to raise your FSB while locking down the PCI/AGP bus for the first time with an AMD platform, came some extreme FSB overclocks. Asus was the first to release their NForce2 motherboard, and with that release came some memory issues that arose when using certain brands of memory. Corsair has attempted to relieve the issues with their release of their XMS series TwinX, available in PC3200 and PC2700 flavors. These are matched pairs of memory modules said to be tested at 2-2-2-6 timings with default memory voltages of 2.6V (PC3200@400Mhz and PC2700@333Mhz).
Fig. 1 Corsair TwinX 3200 LL DDR RAM modules.
Here's what Corsair has to say about the TwinX memory kits:
"The TWINXTM memory kits are composed of two low latency memory modules that have been tested as a pair in an Asus A7N8X dual channel motherboard. The matched pair of modules are then physically packaged together, guaranteeing that the customer receives a product that has been verified in the dual channel environment. These memory kits are available in sizes of both 512 MByte (two 256 MByte modules) and 1 GByte (two 512 MByte modules), at speeds of either 333 MHz (also known as PC2700) or 400 MHz (also known as PC3200)."
"TWINX memory kits take the guesswork out of dual channel motherboards," stated John Beekley, Vice President of Marketing at Corsair. "By using modules that have been 100% tested, as a pair, in this environment, the user can be confident in the performance and stability of his or her dual channel system."
Today we take a look at two modules of the 256Mb TwinX 3200 LL DDR RAM variety and see how they match up against two 256Mb modules of Mushkin PC3000.
Do the TwinX kits deliver as promised? Read on to find out.
Addendum to the review on the last page.
- Asus A7N8X Bios 1001e
- Unlocked Athlon XP 2400+ @ 10 X 200 (2.0Ghz)
- 2 X 256Mb Corsair TwinX PC3200 (2-2-2-6 email@example.comVDimm)
- 2 X 256Mb Mushkin PC3000 (2.5-3-3-5 firstname.lastname@example.orgVDimm)
- EVGA Ti4600
- Maxtor 40Gb 7200Rpm HDD
- WinXP SP-1
- NVidia Det 42.86
We compared the Corsair TwinX PC3200 modules up against the Mushkin PC3000 we had in our labs already. All benchmarks were run at the same FSB for both pairs of memory modules with their respective default timings. All video Benchmarks were run at 1024 X 768
Fig. 2 The 3DMark 2001 SE benchmark result.
The enthusiast's favorite benchmark to show off the performance of their system, 3DMark 2001SE. Here we see the Corsair taking a fairly healthy lead by 491 points. It appears that timings do give you a fairly significant edge when benching 3DMark.
Fig. 3 The Code Creatures benchmark result.
Code Creatures is another high end synthetic 3D benchmark that requires DX8 support and gives us a good idea just how games will fare on any given system. While the Corsair did edge out the Mushkin, it was hardly noticeable.
Fig. 4 The Quake3D benchmark result.
Quake 3D Arena is yet another gamers benchmark and the Corsair beat the Mushkin by a margin of 7.6Fps. Not much, but Corsair still have an edge.
Fig. 5 The Unreal Tournament 2003 Retail benchmark result.
What can we say about one of the most popular FPS games today. This game is fast moving and taxes the CPU, memory as well as the video card. While the difference was less noticeable, the Corsair did best the Mushkin.
Fig. 6 The SiSoft Sandra Memory result.
In the Sandra memory benchmark there is a clear winner and it's the Corsair TwinX modules. The Mushkin just can't compete with the faster timings.
Fig. 7 The PCMark 2002 benchmark result.
Here, as in the Sandra Memory bench, the difference was clear. Yet again, the Mushkin just can't compete with the aggressive timings of the Corsair modules.
Fig. 8 The MemTak ver 0.93 benchmark result.
MemTach is another good synthetic memory benchmark to test a systems memory structure.
Fill xxx measures the memory fill rates for double/int/short/char data types using optimized C code.
Sum xxx measures the rate at which optimized C code can sum up a very large vector; note that the double result also involves a floating point operation for every double accessed.
PSum xxx Same C code as Sum, adds prefetch instructions.
ASum xxx Same as Sum except innermost loop recoded in assembler.
APSum xxx Same as ASum, adds prefetch instructions.
Random int measures random memory access performance.
Analyze 1..14 measure the time taken to access bytes of memory under circumstances designed to torture test the memory subsystem.
It's no surprise that we see Corsair coming out on top here as well.
Keep in mind yet again that these benches were run at the same FSB for both pairs of memory modules. While I wasn't expecting to see a large difference between the two, it's abundantly clear that running faster timings do give you the clear edge in performance.
Let me start my conclusion by stating that while the Corsair TwinX did do 200Fsb @ 2-2-2-6 timings, it didn't do it at the default memory voltage of 2.6V. It took every bit of 2.8V to do it. I asked our contact at Corsair about this and his reply was this:
"Any modules that ship from Corsair were tested at default voltages (2.6 for most NVIDIA boards). It's possible there's something about your configuration, a weak power supply or problem board, that could cause the need to elevate voltage in order to run the modules."
I responded by stating that I had tried both an Enermax 550W and an Antec 400W PSU and both showed normal 3.3V, 5V and 12V rails.
We may very well have received a couple modules that just weren't up to the task. I just can't say, as these were the only modules that I was given to test. Does it bother me personally to run 2.8V through my memory? Not a chance. Might it bother some people? I'm guessing so.
On a related note, Corsair will send us a new set of TwinX modules, and we'll be updating the result as soon as we have tested the new modules. About the voltage issue, Corsair further said - "The bit about the voltage issue isn't what we wanted to happen, and we're disappointed that our module performed this way. (I should probably mention that we only void the warranty if users go above 2.8V, but that's not the point is it). We will do what we would do for any user and replace the parts quickly."
That being said, I wasn't disappointed in the least with the performance of the modules. In fact, after I finished the benches I decided to crank up the FSB to 211, my video card to 325/745 and see what it would do in 3DMark. I was pleasantly surprised to finally break the 15K Mark with my Ti4600
Fig. 9 Finally 15245 points in 3DMark 2001 SE
Copyright © 1999 -
2019, DWPG.Com. All rights reserved.
Does running faster timings make that much of a difference in your system's performance? Not really. Will it give you the edge over running memory that won't run at these speeds with aggressive timings? A resounding YES! I'll have to say that I have yet to run any ram that will run at these speeds with these timings. I was very pleased with the results as you will be should you decide to run a pair of these modules in your DCDDR system. A big thumbs up to Corsair for the release of their TwinX modules.
We received two more ram modules and are pleased to announce that they did in fact do the rated speed at the rated voltages. The benchmarks all came at approximately the same numbers (give or take a few points), with the Corsair still on top. I'd like to thank Paul Watkins at Corsair for all his help.