Print this article

Buffalo FireStix 2x1GB PC2-9600


Date
: 07/25/07 – 03:27:53 AM

Author
:

Category
: Memory

Page 1 : Index

Manufacturer:
Buffalo Technology (USA), Inc.

Price:
$288.99USD (Buffalo On-Line Store)

As we wind down the life cycle of DDR2 memory, prices have hit the absolute lowest they are ever going to be and the frequencies being offered from some manufacturers are just incredible. Among the manufacturers that have decided to offer extremely high binned modules is Buffalo Technology. I was a bit surprised to hear that Buffalo was offering a 2x1GB DDR2 kit at a higher frequency than their PC2-8000 modules that I looked at a while ago but jumped at the chance to have a look at this new PC2-9600 2x1GB kit.

Buffalo has taken giant leaps and bounds over the last couple of years moving their memory products from the IT department into the cases of the forum crawling enthusiast. With 30 years experience of supplying businesses with high end products from every segment of the industry, Buffalo has developed a great understanding of how to set a goal and accomplish it. Since the introduction of the FireStix line of performance memory, the name Buffalo has become common-place in many discussions involving memory overclocking.

The Buffalo Technology FireStix memory line is targeted at the enthusiast and that means overclockers. The overclocking crowd is a hard one to impress. They want industry leading technology, unbelievable performance, rock solid reliability, and above all…value. The fact that the FireStix line has inserted itself into this crowd of consumers with high praise says something about this memory. Today I will look at the cream of the FireStix DDR2 crop, the FSX1200D2D-K2G which includes two 1GB modules of PC2-9600 memory rated for this incredible DDR1200 frequency at 5-6-6 timings with 2.3v.

Page 2 : Package & Contents

Here we go, let's get this review underway with a quick look at the modules and the package that they come in.

If you are familiar with Buffalo memory and their FireStix like I am, you will notice that there are no significant changes from this package to any in the past. The modules, like so many other manufacturers, are secured in a molded plastic blister pack with a cardboard insert.

There is one minor change to this blister pack when compared to the last FireStix kit that I received from Buffalo. This package comes with the entire outside edge of the blister pack heat-stamped and sealed tightly. This is in fact the first manufacturer that I have seen use this method to seal the blister pack with memory inside.

The only thing to identify these modules as the ultra high-speed PC2-9600 are the pair of stickers seen here. Each module itself has a sticker on the heatspreader outlining the specifications but the cardboard insert also has a UPC sticker on it with the part number disclosed and visible to the outside world.

The back-side of the cardboard cutout contains typical marketing copy outlining all the features and benefits of the FireStix line of memory. More importantly, the blister pack has a slight improvement over previous packages I have seen. There are two grooves molded into the plastic and these give the package a very firm feel. Overall, this is probably the most secure package I have seen memory come in despite looking very much like all the others.

Of course, there is one down-side to the heat stamped ring around the package. Getting in to remove the memory is a complete nightmare and all but ruins the package. The one benefit of traditional plastic blister packs for a user like myself is that they are re-usable and provide excellent storage for memory when not in a machine. This package looses that as you have to virtually destroy it to get the sticks out.

Page 3 : Specifications

The PC2-9600 2GB kit of memory we are looking at today is the highest frequency Buffalo Technology offers in their FireStix line, like the PC2-8000 kit was at the time that I reviewed it a few months ago. Also like the PC2-8000 kit of memory, I fully expect these PC2-9600 modules to utilize the highest binned Micron IC's. Before looking under the heatspreaders though, let's have a peek at the specification sheet that accompanies these modules.

The FireStix modules are pretty conventional compared to some other PC2-9600 offerings from the likes of OCZ with their Reaper HPC heatspreaders, Team Group with their Thermalright heatpipe heatsinks, and Corsair with their Dominator line and the elaborate heatsinks that come on those. Buffalo has by-passed the hype and released these DDR1200 modules with their standard fire red aluminum heat spreader. Come installation time, this will be nice because these will not interfere with any setup un-like the others I mentioned.

There are no surprises this time looking under the heatspreader and finding
Micron D9GKX
ICs. The PCB is clearly marked Dynamic but I couldn't find any further specifications on the PCB used. I did ask Buffalo and have confirmed that they are using a 6-layer design. Here are the actual specifications of Micron D9GKX as outlined by Micron:

If you search the internet for D9GMH or D9GKX, you will find numerous forum discussions claiming one is better than the other. Yes, D9GKX is binned for 400MHz operation at 1.8v and CL5 which is slightly higher than D9GMH which is binned for 333MHz operation at the same 1.8v and CL5. When we run these ICs at the voltage of 2.2v or 2.3v as is the case with these Buffalo PC2-9600, there really is no clear cut winner. Some D9GKX are better than some D9GMH and vice versa.

Page 4 : Installation

With a standard sized heatspreader, an installation section for memory isn't really necessary to be truthfully honest. With that said, I am still going to have one with a couple pictures of the modules work area on the motherboard.

With 4GB of memory being hyped as 'completely necessary' for Windows Vista, there are a lot of questions coming up in forums on whether users should go 2x2GB or with 4x1GB of memory. With recent developments in heatspreader design, some users are worrying that fitting high-end memory in a 4x1GB configuration is somewhat difficult. This obviously isn't the case with Buffalo FireStix as they use your standard aluminum heatspreader and the image above shows that there is clearly plenty of room for side-by-side configuration.

The next problem with the large heatsinks on some other PC2-9200 and PC2-9600 modules is the taller heatsinks they use. Again, the Buffalo FireStix don't have this issue so regardless of what CPU heatsink you are running or the proximity of the memory slots to the CPU heatsink, the FireStix will fit with ease. Because of this I won't even bother showing you the modules in a handful of other motherboards I have lying around because is simply isn't necessary. It is safe to say that these modules should fit any motherboard in any configuration that any standard memory without heatspreaders would.

After installing the memory and the rest of the components in my test setup, the last piece of the puzzle is active cooling. As you can see, a 120mm fan is situated over the modules the entire testing time. Any DDR2 memory that is running over 2.0v in my opinion should be actively cooled, regardless of heatspreader design. The fan is also used to cool the passive Zalman northbridge that I run so it sort of kills two birds with one stone.

Here is an overview of the whole bench setup. Yes my CPU heatsink fans are mounted with a little bit of jimmy-rigging going on but that is because I use bolts that come up through my motherboard table to ensure even pressure mounts. It is mainly for when I am running my phase change or dry ice benches but also works wonders for a good mount with an air cooler. The important thing to remember is that all of my benchmarks and testing of this kit will be performed with active cooling.

Page 5 : Overclocking

We have come to that point in the tour where the fun and excitement begins. Please refrain from feeding the animals along the way and keep your hands and feet inside the ride at all times. Ensure your seatbelts are securely fastened as the ride may make sharp turns or sudden stops. Most importantly, enjoy the ride and be sure to join us again. Here is a list of hardware that I will be using in order to squeeze the most that I can from the memory:

The following is a rundown of the programs used and their length of run-time for testing stability of the overclocks listed in the graphs below. I have found this suite of stability testing to be quite strong at identifying a stable memory overclock.

*
Dual 32M runs of Super Pi Mod 1.5 (ran at the same time)

*
2 hours of dual Prime95 using Orthos Beta on blend mode

*
1 hour of dual MemTest in Windows using 850MB/instance

*
Multiple loops of each 3DMark 01 / 03 / 05 / 06

I will first present a chart outlining the various overclocks achieved at voltage intervals that the P5B-Dlx offers. Then present a few screenshots of said overclocks. Again, all of the below overclocks have passed my stability testing as outlined above.

These modules overclocked remarkably well at all three timing sets of 3-3-3, 4-4-4, and 5-5-5. Most impressive was the overclocking at 3-3-3-8. Lately, all the higher binned memory such as PC2-8500 and higher have been rather poor performers at timings other than 5-5-5. So to get a kit at this point in time that clocked well at 3-3-3-8 was a nice surprise. Keep in mind, however, that every kit of memory is going to overclock differently. These results are just from a single kit from hundreds upon hundreds that have been produced. Below are a few screenshots of the overclocks closest to the rated voltage of 2.30v.

DDR840 3-3-3-8 2T @ 2.32v

DDR1080 4-4-4-8 2T @ 2.32v

DDR1270 5-5-5-8 2T @ 2.32v

Page 6 : Benchmarks

As I mentioned in the overclocking section, this memory did extremely well overclocking at 3-3-3. The first thing you will probably notice here in the benchmarking section is that I ran the benchmarks at 3-4-3 instead. This is because the P5B-Dlx actually runs faster at 3-4-3 when at 400FSB or less and for all of the 3-4-3 benchmarks below, I booted at 400FSB in order to stay on the 1066 strap and enjoy that performance benefit. Here is a chart with all the details of the various overclocks used in the benchmarking section.

Booting up with the memory set to AUTO in the BIOS would run the memory at 5-5-5 so I had to manually adjust the settings to run at 5-6-6 for the first two lines on the graphs. Even with the 1:2 divider, I still had to overclock the FSB slightly up to 300 in order to achieve the stock DDR1200 frequency in the top row. I then maxed out the memory at the rated voltage of 2.30v for the second row.

The bottom 3 rows of the benchmarks are a direct comparison between the three main dividers at a set FSB and CPU frequency. You will however notice that the 4-4-4-8 benchmarks were ran on the 1333 strap because the 4:5 divider causes problems with the 965P chipset in general and I had to boot at 420 instead of booting at 400 and clocking up to 420 in Windows like I did for the last and third last rows. Booting at 400MHz FSB on the P5B-Dlx allows us to stay on the 1066 strap and as you will see, maintain optimal motherboard performance. Let's have a look at the results.

Lavalys Everest Ultimate 07 v4.00 – Memory Bandwidth

I first want to look at the bottom three bars of the graph in Everest. You will notice the grey bar doesn't scale as it should with the other two bars at the same CPU frequency. This is the difference between the 1066 strap (booting at 400FSB or less) and the 1333 strap (booting at 401FSB or higher). Memory performance drops off drastically as a result of the looser internal chipset latencies aside for Write performance which appears to be chipset limited anyway.

SiSoft Sandra Lite XI SP4 – Memory (Buffered)

SiSoft Sandra doesn't show the same drop off in performance for the grey bar as Everest did above. This is because Sandra memory bandwidth is a true synthetic benchmark in that it doesn't even measure bandwidth. Instead, it simply calculates it based on memory timings as well as CPU and memory frequency.

ScienceMark 2.0 – Memory Bandwidth

ScienceMark 2.0 appears to calculate the same results that SiSoft Sandra came up with producing a gradual increase in bandwidth from top to bottom. I would have expected ScienceMark to come up with similar results to Everest showing a distinct bandwidth deficiency in the grey bar results.

Memory Latency

The latency results in SiSoft Sandra and Everest are closer than the bandwidth results but there is still a large discrepancy between the middle results of 3-4-3-8 @ DDR840 and the other settings. Overall though, we can see that the latency of the memory at 5-6-6-18 isn't too bad because the higher frequency really makes up for it. Had these results been on an AMD setup, the numbers would be quite different. On Intel chipsets, memory latencies are not that big of an issue due to the design of the Memory Controller Hub. On an AMD CPU with the integrated memory controller, memory latencies play a bit larger role in overall performance. With Intel chipsets, memory frequency is generally king.

SuperPi Mod v/1.5 – 1M

SuperPi Mod v/1.5 – 8M

SuperPi Mod v/1.5 – 32M

SuperPi is a benchmark I will always include in memory reviews because SuperPi is very dependent on memory performance. The SuperPi 32M result is the most obvious of the three results as the gap increases between results. We can see that the grey bar, running on the 1333 strap, suffers and is actually slower than the 3-4-3 result at a much lower memory frequency. At the same time, memory frequency reigns supreme with the DDR1260 result at the bottom being much faster than any of the other three results that were run at 3780MHz despite the looser 5-5-5 timings.

Gaming Benchmarks – Half-Life 2 / UT 2004

The gaming benchmarks start off by showing preference to CPU frequency with all 3 of the last 3 bars giving much higher FPS than the first two, this I fully expect every time I run the custom time demos. What I am a bit confused with is the HL2 results showing a clear advantage to the DDR1050 4-4-4-8 result despite it being on the 1333 strap which is suppose to hamper system performance. The only reason I can think this is possible is that the 4:5 divider that it is running on to achieve DDR1050 at 420MHz FSB is making up for the slower strap. Or HL2 just prefers the 4:5 memory divider. Either way, this advantage is not there in Unreal Tournament 04 where memory frequency gives a slight advantage but nothing drastic.

The last little bit of information I will mention from testing of this memory is that from BIOS defaults, my kit had absolutely no problem booting on the P5B-Dlx. Many P5B-Dlx owners, as well as other recent motherboard users, know what I am talking about when discussing booting at BIOS defaults with performance memory. A lot of the time the default 1.8v provided for vDIMM after a BIOS reset won't be enough to get performance memory to POST. This forces you to boot with just 1 stick in any of the 4 slots to get into the BIOS, raise the vDIMM to whatever is required, then shut down and put the second stick in. With these PC2-9600 FireStix modules from Buffalo, booting with both modules in any orientation from a BIOS reset was not an issue. The motherboard booted up with the memory at DDR800 with 5-5-5-18 timings every time without fail.

Page 7 : Conclusion

The presence of DDR3 memory in the market has a lot of people wondering if DDR2 is already on its way out. The answer to that is no, not even close. DDR2 memory has really stretched the limits of what SDRAM can do over Jedec standards. With many modules running at well over the specified 1.8v, such as these Buffalo PC2-9600 modules at 2.3v, DDR2 has surpassed what many of us would have thought the performance to ever be with DDR2. What this has done is put a lot of pressure on DDR3 because it is going to be quite some time before it outperforms DDR2. Plus, DDR2 right now is as cheap as it is ever going to be and DDR3 has a long ways to go before it can even compete in a cost/performance ratio against DDR2 modules.

This is especially true with modules that provide the versatility of stunning 5-5-5 clocking and very impressive 3-3-3 clocking. This sample I received of the FireStix PC2-9600 truly is a jack of all trades being able to clock very well at all timing sets. Can you recall 2x1GB kits of DDR memory ever clocking above 300MHz at 3-3-3? No…why? It is because there were none that did it with any kind of regularity. Now with DDR2 clocking well over 400MHz at 3-3-3, like we saw this kit do, it is safe to say DDR2 is far and above what DDR ever was and looks to hold that crown for a good while longer; despite what strides DDR3 has made in recent weeks.

As for this kit in particular, I am very impressed if you haven't been able to tell. I have had my hands on a ridiculous amount of DDR2 memory ranging from PC2-6400 CL3 modules to PC2-9600 CL5 modules and many in-between. I have pushed every kit to its absolute maximum at various voltage ranges and so far, this FireStix PC2-9600 kit has been the best clocking kit overall. The only drawback is the $289USD price tag. Now it is not as expensive as many modules out there in its class, but there are a few other PC2-9600 kits can be had for less. Despite this only flaw, I still have to give the Buffalo FireStix PC2-9600 2x1GB kit of memory the O2 Editor's Choice as it is the best all around DDR2 kit I have had the pleasure of working with.

Advantages

• Exceptional versatility overclocking CL3, CL4, and CL5

• Voltage requirements were very low for high clocks
• No fancy heatspreaders that will cause fitment issues
• Buying direct from manufacturer is always a bonus in my books

Disadvantages

Overclockers Online would like to thank Buffalo Technology for the review opportunity.