Print this article

Buffalo FireStix PC2-8000 5-5-5


Date
: 02/7/07 – 01:54:34 AM

Author
:

Category
: Memory

Page 1 : Index

Manufacturer:
Buffalo Technology Inc.

Price:
$137 x 2 = $274USD(Newegg.com)

The Buffalo roams once again across the plains of Overclockers Online. Just over two months ago, a set of Buffalo FireStix made an appearance here with a 1GB memory kit. There was a lot of interest in the FireStix line then and even more so now. Buffalo Technology Inc. is a worldwide leader in network solutions, multimedia devices, storage hardware, and of course memory. Buffalo Technology is an international corporation and has offices in Taiwan, UK, Germany, Ireland, and Japan as well as Austin Texas in the USA.

Having supplied the IT industry for more than 30 years now, Buffalo Technology has interest in a lot of different markets and is primarily known in the business world for providing technology solutions for a broad range of needs. In their effort to increase brand awareness in the end-user market and primarily the enthusiast memory segment, Buffalo Technology released the much talked about FireStix line of memory.

Designed with extreme performance in mind, the FireStix line of memory is positioned to fight with the other big names in the performance memory market. Today I will be looking at a PC2-8000 part which means DDR1000 as the rated frequency at 5-5-5 timings with only 2.1v. I have the 2x1GB kit, part #FSX1000D2C-K2G, here for testing today and I am pretty excited about these modules. There should be plenty of overclocking headroom so if you are looking for your next kit of memory to push your system, have a look inside.

Page 2 : Package & Contents

Welcome to the graphical portion of the evening where I will be guiding you through a visual experience of the Buffalo FireStix and the package they come in.

We have all seen this before and will likely see it again, unless 80% of all memory manufacturers change their module package overnight…which won't happen. The plastic blister pack has become and continues to be the package of choice. I have no complaints with this as the modules are well protected and easy to transport.

The FireStix logo is proudly displayed along the top of the transparent package by way of a cardboard insert. There is no actual information on the package itself or the insert about the modules inside. These are generic pieces that are used with all variations of the FireStix module and help keep costs down. All indications of what modules are inside are provided by way of the stickers we see here on the modules heatspreaders.

There is a third identification sticker on the cardboard insert at the side seen here. Also notice the staple securing the two halves of the plastic package. Some manufacturers don't use a staple and that is fine but there is no such thing as too much protection. There are two staples, one on each side of the package that help keep it closed as some of these packages have the habit of popping open.

The backside of the cardboard insert is just typical advertising type stuff and most of it will be seen in just a bit in the specifications section. In the bottom left corner of the insert, Buffalo has their web site listed. I noticed that since the last FireStix review, Buffalo has done a major overhaul to their global and geographical sites. I highly recommend a quick visit as it is quite a nice upgrade from the old one.

I have pulled the modules and given them their first breathe of fresh air after being couped up in that plastic blister pack for the long journey here. These modules are certianly quite stunning to look at. The red heatspreader is bright and vibrant and the Firestix logo that graces them is very appealing.

The only identification of the modules specifications come by way of the stickers that are present on the right hand side of each module. All the important elements are there including part number, size, and basic frequency/timings of the modules.

I really don't need this many photos of the modules but for I know many of our readers here at Overclockers Online.com thoroughly enjoy a good photo session so I like to bring a lot of eye candy to my reviews for those people. Especially when the product in question is as good looking as these modules.

Having looked at the backside, front side, and everywhere but underneath, I can't think of much else to go over or explain about this memory. The markings on the PCB, 2DUE28F-AA, don't really lend much information about the PCB specifications except that they are produced by Dynamic I believe, more on that later.

The heatspreaders Buffalo uses are pretty much standard issue aluminum pieces slipped together at the top and held in place with a pair of clips. The clips really aren't necessary though because the adhesive thermal material between the heatspreaders and the ICs holds on pretty tight.

The adhesive material is basically a thermal tape that adheres to the heatspreader and then to the ICs. It appears these modules use the rubber like thick material that seems to have become standard in the industry. Some manufacturers are turning to a thinner material that is actually cured to the modules though so it will be interesting to see if companies like Buffalo follow suit.

The top of the heatspreaders is pretty much open for the most part and will allow fresh air to enter if using active cooling or hot air to escape if not. That about finishes off our look at the modules and we will now take a look at the specifications page where I will have a look under the heatspreaders and discuss the ICs underneath.

Page 3 : Specifications

Since the FireStix line of memory from Buffalo Technology is designed for the enthusiast, there are a few features that should be standard including a lifetime warranty, aluminum heatspreaders, and overclocking headroom. Check, check, and check…here is a brief specifications sheet:

Here is what Buffalo has to say about their FireStix line of memory, this is what was printed on the back of the cardboard insert:

Now, because Buffalo does not make it apparent what ICs are used on these modules, I am going to take a peek under the heatspreaders. I am expecting to see the familiar face of Micron D9GMH but one can never be certain.

This is a pleasant surprise, Micron D9GKX are the ICs being used on these modules. I really wasn't expecting to see this because D9GKX are rather rare these days. Although, the date code on the ICs "6A" means they are rather old stock from Micron. Either way, D9GKX are Microns highest binned DDRII ICs right now and are ever so slightly ahead of D9GMH as far as stock ratings go. Here is a list of specifications that can be found at the Micron web site that can be found here:

As you can see, they are rated for operation at 400MHz CL5 with 1.8v. D9GMH are rated for 333MHz at CL5 with 1.8v. This doesn't necessarily mean they are better and in my experience with both D9GMH and D9GKX, I have found that GKX tend to overclock slightly better with less volts but don't seem to respond to higher voltages as well as GMH.

The PCB is marked with the part#2DUE28F-AA which appears to be made by Dynamic but I couldn't find any more information than that. I personally haven't had a lot of good experience with PCBs other than BrainPower with Micron D9 based modules but regardless, the overclocking results should definitely be interesting now that I know what I am working with.

Page 4 : Installation

I feel as if I am repeating myself for some reason when I say, the heatspreaders on these modules are no different than standard size heatspreaders and aside from a couple millimeters of added height and width, they don't increase the size of the modules so installation should be 100% straight forward.

As you can see, even with the poor mounting orientation that I am left with on the SLI-M2R/G and the Thermalright Ultra-120, I can still easily install and remove the modules in the yellow slots. The orange slots are also accessible but take a bit of angling to get the first module in.

I will of course be actively cooling the FireStix during the entire testing and overclocking process. If I have said it once, I have said it a million times…actively cooling DDRII modules is always a great idea. Even if you aren't pushing them, a little airflow goes a long way.

I decided to test fit the modules side-by-side in the P5B-Dlx just to show that they clearly fit without issue. I actually think that this motherboard states that dual-channel memory should be installed in this fashion but I have always used both yellow slots and always ran in dual-channel.

This will be the testing setup while running the memory on the Intel platform and again active cooling is also being used here. I will also be utilizing the efforts of a Hipro-Tech.com DDR Maximizer like I do in all reviews. That is the little silver box in the lower right hand corner. it simply allows better voltage regulation and control over motherboard supplied VDIMM.

Page 5 : AMD Overclocking

With the introductions and pleasantries out of the way, we can now get to the sections that most people came to see. I always show AM2 overclocking results first, and today will be no different. This will actually be the last of the AM2 testing that I will be doing as I will be replacing this bench setup with another Intel system based on the NVIDIA 680i chipset. Sorry AM2 users, there just isn't enough interest to continue with memory testing on the AM2 platform. With that said, let's take a look at the complete list of hardware that I will be testing with.

All the overclocks on this page are considered extremely stable due to the stability testing they have all passed. There is no exact set of stability tests that can be run to guarantee 100% 24/7 stability from memory, but the list of programs I use tends to be a great judge of stability. Here is what every overclock listed on this page goes through:

• 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 750MB/instance
• Multiple loops of each 3DMark 01 / 03 / 05
  

  I have broken the memory overclocking results down into three popular sets of timings. There are many more combinations of timings that I could have tested the overclocking at but that could take a couple months so I stick with the standard timing sets of 3-3-3-8, 4-4-4-8, and 5-5-5-8. Let's take a look at the tightest of those timing sets first.

  
  3-3-3-8 2T
  
  

  

  This memory is off to a blazing start at 3-3-3-8. The overclocker standard for Micron D9 based memory is to see how much voltage is required to run DDR800 3-3-3. In this case, it was only 2.29v and that got us to DDR810. That is a very acceptable number and is actually quite impressive. PC2-6400 (DDR800) kits of memory that are rated for 3-3-3 timings are few and far in-between and have a much higher price tag than that of these FireStix. Being able to run this frequency at these timings really makes this memory versatile. Let's see if DDR1000 at 4-4-4 was as easy for these Buffalo modules.

  DDR820 3-3-3-8 2T @ 2.29v

  

  
  4-4-4-8 2T
  
  

  

  Having run DDR800 at 3-3-3 with relative ease, I fully expected DDR1000 at 4-4-4 to be obtainable and that is certainly evident. In fact, I got awfully close to DDR1100 at 4-4-4-8 with this memory and that matches some of the higher priced kits that I have tested in overclocking results. Generally speaking, 4-4-4 seems to perform the best overall on the AM2 platform and being able to run up to DDR1070 with low volts means this kit is a great option for AM2 users that have a highly overclocked AM2 setup.

  DDR1000 4-4-4-8 2T @ 2.10v

  

  
  5-5-5-8 2T
  
  

  

  The last of the memory timings that I tested for overclockability is 5-5-5-8 which also happens to be very close to the rated timings with this memory. Based on the earlier results on this page, I was hoping for DDR1200 and despite not reaching that high of a frequency, DDR1160 at 2.29v is still very nice. I mentioned earlier that 4-4-4 tends to be the performance sweet-spot on AM2 so it will be interesting to see how the 5-5-5 clocks match up to 4-4-4 in the testing section.

  DDR1160 5-5-5-8 @ 2.29v

  

  
  Page 6 : Intel Overclocking
  

  To accompany my AM2 overclocking and benchmarks, I will also run this memory on my Intel platform which currently consists of a P5B-Dlx and an evil little E6300 that has no problem dishing it out at 4GHz. I did say that I was going to be replacing the AM2 platform with a 680i one after this review so that means all Intel based reviews from here on out…I never thought the day when I would say that. Anyway, here is a look at the list of hardware on my P5B-Dlx test bench:

  CPU: Intel C2D E6300 (L630A864)
  CPU Cooling: Chilly1 Single Stage Phase Change
  MB: Asus P5B-Dlx Wifi-AP Edition
  GPU: Biostar 7600GS 256MB DDRIII
  PSU: Silverstone Zeus 560W
  HD: Seagate SATAII 80GB 8MB NCQ
  OS: Windows XP SP2 (with all updates)
  Hipro-Tech.com DDR Maximizer v1.2

  Memory:

  Buffalo FireStix 2x1GB DDR1000 5-5-5 @ 2.1v (FSX1000D2C-K2G)

  Nothing will change with the jump to the Intel testing platform. Since both AM2 and C2D platforms are using dual core processors, stability testing for the two remains the same. The same criteria as before is required for the overclocks to be considered stable and listed on this page in the charts below. Here again is that list:

• Dual 32M runs of Super Pi Mod 1.5 (ran at the same time)
• 2 hours of Prime 95 using Orthos Beta on blend mode
• 1 hour of dual MemTest in Windows using 750MB/instance
• Multiple loops of each 3DMark 01 / 03 / 05
  

  I will again be testing the overclocking at the three standard timing sets, like I did on the AM2 platform. There is just one slight adjustment though. The P5B-Dlx doesn't like 3-3-3 as far as performance goes. The system runs slower at 3-3-3 compared to 3-4-3 so I do the overclocking and benchmarking at 3-4-3 instead. For those that are interested, 3-3-3 clocks memory about 5-10MHz slower than 3-4-3 in all the testing that I have done on the P5B-Dlx.

  
  3-4-3-8
  
  

  

  The 3-4-3 clocking here on my C2D setup appears to clock almost the same as 3-3-3 on AM2. This is nothing new as the last few kits of Micron D9 based memory have done the same. Again, this type of overclocking is very impressive for the voltage used and this is no less impressive than the AM2 overclocking at 3-3-3.

  DDR800 3-4-3-8 @ 2.20v

  

  
  4-4-4-8
  
  

  

  DDR1000 was not an issue with it being surpassed somewhere in-between 2.10v and 2.20v. Being able to pull this off on the P5B-Dlx keeps this PC2-8000 kit of Buffalo FireStix up near the best clocking DDR2 memory I have had the pleasure of working with. It isn't the best overclocking at this voltage that I have seen, but these modules definitely use some quality components in getting this high.

  DDR1060 4-4-4-8 @ 2.30v

  

  
  5-5-5-8
  
  

  

  The last of the overclocking leaves us with 5-5-5-8 timings again, and one just like the AM2 overclocking section, DDR1160 is where this memory seems to stop. Despite the exact same maximum, the P5B-Dlx allowed for higher clocks at voltages below 2.30v. In fact, the lowest voltage I tested of 2.00v allowed for a very nice DDR1080 which is much higher than the rated speed of DDR1000 and with fewer volts even.

  DDR1140 5-5-5-8 @ 2.20v

  

  
  Page 7 : AMD Benchmarks
  

  With overclocking out of the way, we can now focus on some performance numbers from the Buffalo FireStix. Here is the color coded key with the settings for the results.

  

  From the five setups I decided on, we should have a nice little comparison between the blue and black result being both at the same CPU and memory frequency and the grey and yellow results are close too. Normally I have the last three results at the same CPU frequency but the SLI-M2R/G didn't want to run the dividers properly again. That was the last straw and one of the major reasons I am getting rid of it.

  SiSoft Sandra 07 SP1 – Memory Bandwidth

  

  Lavalys Everest Ultimate 06 – Memory Bandwidth

  

  The majority of my comments will be directed toward the blue and black results because it isn't often I get a direct comparison between 5-5-5 and 4-4-4 with the exact same CPU and memory speed. Both Sandra and Everest show a clear advantage to the 4-4-4 timings when compared to 5-5-5. I wasn't sure if bandwidth was going to show much of a gap but it clearly has.

  Lavalys Everest Ultimate 06 – Memory Latency

  

  I was anticipating a healthy margin between the black and blue results with the latency numbers and that is what we naturally see here. We can see with the white result just how much CPU speed and memory frequency are needed to overcome the tighter timings of 4-4-4 and nothing can touch 3-3-3 timings, even at the lowest CPU and memory frequency.

  FutureMark PCMark 05 & 3DMark 01 SE

  
  

  With PCMark 05, in the past I haven't experienced much of a difference with memory timings on the results and with the direct comparison between 4-4-4 and 5-5-5, there is only 35 points separating them. This technically is a difference but certainly not a significant one. CPU and memory frequency clearly play a larger role in PCMark 05 results than memory timings do. 3Dmark 01 is very much the same producing a slightly higher score for 4-4-4 but again, the margin isn't great and CPU speed combined with memory frequency increases provided greater score increases.

  SuperPi Mod v/1.5 – 1M

  

  SuperPi Mod v/1.5 – 8M

  

  SuperPi Mod v/1.5 – 32M

  

  If there is such a thing as a professional SuperPi benchmarker, I am it. I won't even admit to how much time of my week goes into tweaking for SuperPi. With that said, I was obviously eagerly awaiting to see the blue and black results from the SuperPi runs. As always, CPU frequency is king in SuperPi with memory frequency coming second. With the same CPU and memory speeds though, timings do actually make a reasonably noticeable difference. I would say the timings have about the same influence on times as the 3DMark 01 scores.

  Gaming Benchmarks – Far Cry / Half-Life 2 / UT 2004

  

  The gaming results are about what I expected with higher CPU frequency showing much better FPS. In comparing the blue to the black results, a solid 2~3 FPS is what we see with the tighter timings. This again equates to the same advantage we saw in SuperPi and the FutureMark programs. Overall there wasn't any surprises in the benchmarks on the AM2 setup, let's see what the Intel platform has to show us.

  
  Page 8 : Intel Benchmarks
  

  The Intel side of the board tends to benchmark a little bit differently. Sometimes results aren't always as easy to predict because of the hidden latencies in the north bridge which houses the memory controller. As always, I start with a quick look of the timing sets and their details that I will be running for the benchmarks.

  

  The stock speed isn't going to exactly be stock speed because there isn't a divider that lets me run the memory at the rated DDR1000 at 266FSB so the lowest CPU frequency I will be running is 2338MHz on the 2:3 divider which provides DDR1000. I will then be switching to the 3:5 divider and running the benchmarks at the same CPU frequency for the next set of results. The last three sets are going to be a mixed bag again because simply adjusting the divider was going to result in memory speeds I have run before and didn't feel like looking at basically the same type of results. Either way, there are a couple of nice comparisons at the same CPU speed tucked in there.

  SiSoft Sandra 07 SP1 – Memory Bandwidth

  

  Lavalys Everest Ultimate 06 – Memory Bandwidth

  

  With the bandwidth results and these sets of timings and frequencies, CPU speed is the champion as always and memory frequency comes in a close second place with Sandra results. With Sandra, Bandwidth isn't so much as measured as it is calculated and clearly the algorithm for the calculation favors those factors. Everest on the other hand actually measures the bandwidth and comes up with completely different results with the north bridge latencies playing an important role. Yellow and blue are below 400FSB. Grey, black, and white are above 400FSB and thus run at much looser NB latencies resulting in poor memory performance in comparison.

  Lavalys Everest Ultimate 06 – Memory Latency

  

  The NB latency differences really show up in the latency numbers and unlike the AM2 results, memory frequency is the champion in lower latency results. Because of the memory controller being off the CPU die, the tighter timings don't have as much effect on memory latency and the pure speed is what dictates the latency numbers.

  FutureMark PCMark 05 & 3DMark 01 SE

  
  

  The north bridge latencies don't affect performance in programs as much as they do in memory bandwidth and latency benchmarks, at least the effects are not as prevalent. In both PCMark 05 and 3DMark 01, the CPU frequency plays the biggest role in performance followed in the distance by memory frequency.

  Super Pi Mod v/1.5 – 1M

  

  Super Pi Mod v/1.5 – 8M

  

  Super Pi Mod v/1.5 – 32M

  

  In SuperPi, we see the gaps between the various CPU frequencies to be quite big but when you compare the results at the same CPU frequency, there is very little difference. This indicates that the memory frequency doesn't play a huge role and the timings do help a bit here. 14 seconds seems to be about the difference between the yellow and blue as well as the grey and black in 32M calculations. This is about the same type or performance gap we saw with the AM2 testing between 4-4-4 and 5-5-5 at the same CPU and memory frequency.

  Gaming Benchmarks – Far Cry / Half-Life 2 / UT 2004

  

  In the two games other than FarCry, it is pretty evident that we see a gradual increase in FPS as the CPU MHz increase and then as the memory frequency increases. FarCry seems to hit a wall at 130FPS at 2905MHz on the CPU. The other two games show us that the north bridge latency change at 400FSB doesn't seem to affect gaming all that much because the gap between the yellow/blue results and the grey/black numbers is just slightly less than that of the gap from grey/black to white where there is no chipset latency change.

  
  Page 9 : Conclusion
  

  The time has come to draw to a close another memory evaluation. There are so many 2GB DDRII memory kits available right now and the prices range from decent to over the top and then some. This Buffalo FireStix offering is definitely at the "decent" end of the spectrum coming in at under $300USD on NewEgg.com for a pair. They were in fact, the cheapest 2GB kit of the PC2-8000 lot with most of the 5-5-5 kits crossing over the $350USD mark.

  

  The performance and the ICs on the pair of modules that I received certainly don't portray the budget end of the enthusiast segment and the results back that up. Both the AM2 and Intel platform clocked really well with these modules and neither gave me a headache to achieve those clocks.

  I am quickly becoming quite fond of the FireStix line of memory but I wouldn't mind seeing a BrainPower PCB under the heatspreaders and of course, availability in Canada. I again ran out of luck looking for a set of these sticks north of the US border and I just know I am going to be getting some e-mails from Canadian readers asking for directions on how to obtain this excellent memory.

  Advantages

  Solid looking modules
  Excellent ICs under the hood of the kit reviewed
  Overclocking did not disappoint on AM2 or Intel
  Pretty decent bang for the buck in the overclocking segment

  Disadvantages

  Dynamic PCBs instead of BrainPower or an 8-layer design
  Again, no Canadian availability

  Overclockers Online would like to thank Buffalo Technology Inc. for making this review possible.