Having looked at a few dual channel memory kits, I'm a firm believer that if you're looking for the best performance, and stability in your dual channel motherboard, one of these matched pairs is a must. Up until TwinMOS contacted us, the dual channel kits we've reviewed were from manufacturers who target enthusiasts. Granted, the nForce 2, and Canterwood are enthusiast boards, but I was surprised "OEM" memory manufacturers didn't jump on this bandwagon, especially since the Granite Bay, which also supports dual channel, was a workstation board.

is very well known in Asia, and have a reputation of making quality, and well-priced ram. They've been manufacturing PC3200 modules for quite some time, which have been getting good reviews online, but today we'll be looking at their Dual Channel PC3200 kit.

There isn't any information on their site about it, but I was told by our TwinMOS contact that the specifications are identical to their .

Specifications

Double Data Rate architecture

MRS cycle with address key programs
- CAS latency: CL2, 2.5
- Burst length: 2, 4, 8
- Burst type: Sequential & Interleave

2 variations of refresh
- Auto refresh
- Self refresh

Serial Presence Detect support
2 Banks to be operated simultaneously or independently
Package: TSOP/CSP
184 edge connector pads
Clock frequency: 133/166/200MHz
SSTL-2 interface: 2.6 Voltage +/- 0.2V

Pricing looks to be about 50$ per 256MB module. Per module you say? Yes, that is exactly what I said. From what I understand, this kit is not a matched pair. I explain this much more clearly in our past Dual Channel reviews, but in a nutshell, other kits we've looked at took same sized memory modules and tested them as a pair. When two DIMMs are "certified" as being a matched pair, they are then sold as a kit. Here's what TwinMOS had to say when I asked them about the certification process…

"TwinMOS had a very good connection with all motherboard manufacturers, usually we join the early design of motherboard and get the early motherboard sample with Dual Channel function to test the performance for Dual channel. You can see the real performance under SiSandra test software. It is almost double performance comparing with single channel."

Now, not that I want to nitpick, but basically they are saying they tested and verified that Dual Channel is faster than Single Channel. From experience, yes, this is the case, but the fact is this does not really test for true compatibility between two DIMMs. In anycase, we'll explore if this is a problem when we get into testing.

Update: I just got word that their kit will be sold in pairs (which was something I wasn't sure about).

You should be able to find the kits as pictured above.

TwinMOS Dual Channel PC3200

The ram ships in a hard plastic shell, which does a good enough job of protecting the ram during shipping. There are no instructions, but installing ram isn't all that complicated. Just make sure you plug them into the proper DIMM slots as per Dual Channel specifications.

Unlike some other manufacturers, TwinMOS has elected to not go with heatspreaders on their products. Although this takes away from the "gee whiz" factor, it should save a couple bucks on production, thus saving the end-user some money. Heatspreaders never really did much for cooling in out testing anyhow, so I wouldn't sweat over this too much.

The modules themselves are fairly standard. There's a nice reflective sticker with the company logo, as well as another sticker for indicating the ram type and speed, which is rated at CAS 2.5.

The ram chips themselves are TwinMOS branded, rated at 5ns. Previous modules were rated at 6ns, so it looks like TwinMOS is refining their manufacturing process. In order to maintain a clean signal, the memory is made with a 6 layer, rather than 4 layers. Little things are what matters, and TwinMOS did their homework here.

Their 256MB memory modules are single-sided, as opposed to being double-sided. What this should mean is less noise, and overclock a little better. From past experience, when I've worked with 256MB sticks (double-sided), it's always overclocked a little better than two double-sided 256MB sticks. Whether or not two single-sided sticks will OC better is something we'll try to look into when we get into the overclocking.

Overclocking

Before we get into overclocking, the first thing I wanted to determine is what are the default timings of the ram.

The timings are nothing to go nuts over, as the settings are not all that aggressive. TwinMOS does emphasize stability though, so that's probably the reason for the rather conservative timings. At stock speeds of PC3200, I started to play around with the various timings. Right off the bat, 2-2-2-5 resulted in Windows not booting, as was the case with 2-2-2-6. At 2-2-2-7, the system was a little unstable, but by boosting the memory voltage to 2.7v, everything was fine. After 20 minutes or so, the system started crashing constantly, so I had to up the voltage to 2.8v. Memtest86 passed the test, and we began working on the clock speed.

At 2-2-2-7, the maximum stable overclock was 214MHz (DDR428). At 2.5-2-2-7, we managed a 220MHz (DDR440). Using the stock timings, we improved our overclocking to 225MHz (DDR450).

2-2-2-6 did work at DDR333 (which is below spec for this kit), but 2-2-2-5 did not.

Since these are not matched pairs, I was interested in seeing how they would do in overclocking as individual sticks. In terms of timings, both sticks had the same limitations at stock speeds. 2-2-2-7 was still the best we were able to manage, though I only had to boost the voltage to 2.7v. At those timings, the best OC we managed were the following…

Stick 1: 215MHz
Stick 2: 217MHz

Keep in mind that both sticks as a pair, we only managed 214MHz. What does this demonstrate? Well, if they were matched pairs, I would expect equivalent OC ability out of each, as was the case with the other Dual Channel kits we've reviewed. I did test if one stick of single-sided TwinMOS would out-OC a double-sided stick of Corsair TWINX (I don't have double-sided TwinMOS here unfortunately), but it did not. So much for that experiment.

Therefore, in summary, here's our results of our best, and most stable, overclocks:

	2-2-2-7	2.5-2-2-7	2.5-3-3-8
TwinMOS PC3200	214MHz (428DDR)	220MHz (440DDR)	225MHz (450DDR)

Test Setup

Abit NF7-M nForce2: Athlon XP 2500, 2 x 256MB TwinMOS PC3200 DC Ram, MSI FX5200-VTD, 120GB Western Digital SE 8MB Cache, Windows XP SP1, nForce 2 Unified Driver Package 2.0, Detonator 43.45.

Test software will be:

SiSoft Sandra 2003
PC Mark 2002 Memory
PiFast
3D Mark 2003
UT2003

We'll be presenting benchmarks at both the ram's stock speed (400MHz), as well as at the overclocked speeds, with the timings as shown earlier this page. Tests were done with the FSB and memory speed syncronous @ 166 (333DDR, 2-2-2-6), and asyncronous for 200MHz (400DDR) and up. Corsair TWINX PC3200 will be the comparison ram for these tests.

iSoftware Sandra 2003

Although a synthetic benchmark, it's a popular one, freely available if you wish to make comparison benchmarks. We will be testing the memory speeds.

Stock Speeds (DDR400)

At stock speeds, and default SPD timings, we can see that the TWINX holds an easy lead on the TwinMOS module. I do feel that TwinMOS clocks their ram rather conservatively, so for the remainder of the benchmarks, we'll be running the ram tweaked (as well as overclocked), though I will be "slowing" down the TWINX (except for the OC tests) to level the playing field. What we're trying to accomplish here is to see how the modules stack up against each other when all things are equal.

333DDR, 2-2-2-6

400DDR, 2-2-2-7

Max OC

NEXT