Buffalo FireStix 2x512MB PC6400 5-5-5

Dec 1st, 2006 | By

Print this article


Buffalo FireStix 2x512MB PC6400 5-5-5


Date
: 12/1/06 – 02:46:20 AM

Author
:

Category
: Memory


Page 1 : Index

Manufacturer:
Buffalo Technology

Price:
$169.99 USD (MicroCenter.com)

Today in my lab, also known as my computer hardware littered workspace, I will be looking at some memory from Buffalo Technology. Despite being known for memory, Buffalo offers so much more than just that. Buffalo Technology USA is based in Austin Texas and has an extensive portfolio of products including storage solutions, multimedia hardware, networking and wireless networking technology, and external antennas amongst others. Buffalo Technology has been around longer than my fast talking mouth has been on this planet and is so entrenched with small businesses, system builders, and integrators; they will likely still be innovating long after I am gone.

OC Online Image

Buffalo however is a name not normally associated with the big boys of the enthusiast and memory overclocking market such as Corsair, OCZ, GSkill, Mushkin, and others. Buffalo has concentrated on providing solid mainstream memory products under their Certified and Select lines of memory. These lines of memory are for specific compatibility for hassle free use by end-users and builders alike…but what about us enthusiasts and overclockers? What does Buffalo have to offer us?

OC Online Image

One word, FireStix. The FireStix line of memory is Buffalo's endeavor into the enthusiast world of overclocking which consists of memory designed for running above JDEC standards. Offering everything from PC3200 DDR modules up to PC9200 DDRII memory kits, the FireStix line of memory has the starting line-up to challenge for a spot in every enthusiasts system. Today we will be looking at a pair of their PC6400 CL5 sticks. The 1GB dual-channel kit consists of a pair of 512MB modules and is rated for operation at DDR800 5-5-5 with only 2.1v, let's get started.


Page 2 : Package & Contents

This is a memory review so there is sure to be a clear blister pack at some point, and that point is now.

OC Online Image

Like pretty much every memory manufacturer at this point in time, the Buffalo FireStix come in a clear blister pack displaying the modules front and center with a cardboard insert doing the advertising. Immediately the red aluminum heatspreaders jump off of the white background.

OC Online Image

The rear of the package has a bit of information about the FireStix line but doesn't go into specifics about the modules in the package. I will provide all of the information printed here in the Specifications section on the next page.

OC Online Image

The FireStix logo is the prominent markings here and I have to admit, the logo is quite well done. If Buffalo is targeting the enthusiast and overclocking with this line of memory, the logo and heatsink color is sure to help their efforts.

OC Online Image

The only details about the modules in the package are found on the front side of the package in the bottom left hand corner where the UPC code is. The modules themselves also have the standard sticker with specifications listed on them.

OC Online Image

No one can complain about these modules not looking good because they really are sharp. The red is obviously Buffalo red and that FireStix logo with the Buffalo logo makes these sticks unmistakable. Now for some more eye candy.

OC Online Image

The backside of the modules is the same as the front, sans the specification sticker. If only Buffalo could pull off a dark PCB, the red heatspreaders would be that much cooler looking.

OC Online Image

You can sort of see here that the top of the heatspreaders are not closed off completely, there is some gaps for hot air to escape.

OC Online Image

I just couldn't stop taking photos of these sticks, the camera just loves them.

OC Online Image

As I mentioned, the top of these heatspreaders are pretty much open. Personally I think this is a good thing because some modules have heatspreaders that are almost closed right off and this I feel will trap some amount of heat right near the ICs. With the open top, that heat can get out. Plus, fresh air from a fan above can wash over the PCB a little bit.

OC Online Image

Speaking of heatspreaders, I had to take them off at some point. You can see the black foam on one side and the grey thermal tape on the other heatspreader. The black foam goes on the backside of the module as there are no ICs back there and the foam fills the gap and keeps the heatspreader off the PCB. We will take a closer look at the ICs in the Specifications section which is on the next page.


Page 3 : Specifications

Specifications wise, these FireStix don't really have much to talk about. They don't appear to be EPP compatible and Buffalo doesn't really go into great depth about what makes these FireStix special. Buffalo does have this to say about their FireStix line of memory:

FireStix's impressive features enhance PC performance while maximizing stability. They allow over-clocking the memory bus at the same speed and bandwidth as the Front Side Bus (FSB), eliminating bottleneck issues and allowing maximum data transfer efficiency. In addition, the heat-spreading, aluminum-cooling skin extends RAM life while facilitating faster memory bus speeds for superior performance in gaming and over-clocking environments.
This is the same quote on the back of the package we saw earlier. Here are the listed specifications for this kit from the Buffalo web site. I have highlighted the kit we are looking at today.

OC Online Image

The only thing of interest is the 5-5-5 timings at DDR800 with 2.1v. Being only 512MB sticks, they are single rank which means only one side of the module has IC's on it. Speaking of IC's, it is time to have a closer look at the IC's on these modules.

In the Package & Contents section we saw that the modules only had ICs on one side of the PCB and the Heatspreaders did come off fairly easily but I would still advise against it as most manufacturers won't warranty your modules if you rip an IC off while pulling the heatspreaders.

OC Online Image

We find Micron supplied D9GMH ICs on these FireStix and I am happy to see them. I have had quite a bit of experience with D9GMH based modules and have always been impressed. Here is a quick list of specifications on D9GMH from the Micron web site:

Density: 512Mb
RoHS: Yes
Depth: 64Mb
Width: x8
Voltage: 1.8V
Package: FBGA
Pin Count: 60-ball
Clock Rate: 333 MHz
Cycle Time: 3ns
Op. Temp.: 0C to +85C
CL: CL = 5
Data Rate: DDR2-667
As we can see this particular IC is rated for DDR-667, CL 5 @ 1.8v. Many memory manufacturers use these D9GMH and as I said, I have been impressed with every kit with D9GMH. They run great timings at rated voltages and scale nicely when the voltage is increased. Right after installation we will see how this kit scales with D9GMH in the overclocking sections.


Page 4 : Installation

We are obviously dealing with memory modules today so the installation section isn't exactly exhaustive. I do however have a couple photos from both the AMD setup and Intel setup.

OC Online Image

My AM2 setup is pretty straight forward so the modules fit without problems having just standard heatspreaders. The SLI-M2R/G uses split slots for dual-channel but even if the modules were side by side, they wouldn't even be touching.

OC Online Image

Here is the system just after power up and you can just barely see the FireStix tucked in there behind the video card. I always mount a 120mm fan above memory modules, just like on the near side above the PWM area of the motherboard.

OC Online Image

The Asus P5B-Dlx lists memory slots side by side as being dual-channel but I have found that memory runs in dual-channel mode with any combination of slots so I choose to use split slots and with the Hipro-Tech DDR Maximizer, it is beneficial to have the VDIMM PCB in-between the modules so this setup works best.

OC Online Image

Again, like my AM2 setup above, I like to always keep a 120mm fan above memory modules. With the taller PCB and connection on the DDR Maximizer, the fan doesn't sit exactly how I would like it to but the fan serves the same purpose of pushing the hot air away from the modules in use. We are now going to have a look at overclocking and see just what this memory is capable of, starting with the AMD system.


Page 5 : AMD Overclocking

Overclocking memory is my favorite hobby. I admit it, I am obsessed with every aspect of memory overclocking and tweaking. The only way for me to get my fix is to constantly be working with a pair of modules. This bodes well for memory reviews as you will likely see more information in this section than most review sites. Keep in mind however, that I am working with a single kit of memory from thousands that Buffalo has produced. Overclocking results are not going to be the same with every set, especially when you take into account the other hardware used such as motherboard, power supply, and CPU. Here is a complete list of the hardware used in my setup.

CPU: AMD64 X2 4000+
CPU Cooling: Asetek WaterChill
MB: DFI LanParty NF590 SLI-M2R/G
GPU: Power Color X1650PRO 256MB
PSU: OCZ GameXStream 700W
HD: Seagate SATAII 80GB 8MB NCQ
OS: Windows XP SP2 (with all updates)

Memory:

Buffalo FireStix 2x512MB DDR800 5-5-5 @ 2.1v (FSX800D2C-K1G)

Overclocking memory is very time consuming, like any other type of overclocking, because stability testing is the most important aspect of the process. To test for stability I like to use a combination of programs and each program adds another level of stability. Anything I have listed as a stable overclock in this review will have gone through the following:

  • 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 410MB/instance
  • Multiple loops of each 3DMark 01 / 03 / 05

    OC Online Image

    I will organize the results by timings and list the screenshots and overclocks at various voltages. I decided to test from 2.00v up to 2.33v although I am not sure what Buffalo will warranty up to. These Micron D9GMH should be more than fine running 24/7 with only 2.33v but anytime you are over-volting memory, I would recommend some sort of active cooling. Let's see what the FireStix were able to achieve, I will start with the timings I got when booting on all auto settings.

    Auto 4-5-5-18

    OC Online Image

    This is what the modules booted at with the memory timings set to "auto". I know the memory is advertised at 5-5-5-x but in two separate AM2 motherboards, the memory booted at 4-5-5-18 without any timing adjustments on my part. As you can see, there is quite a bit of headroom with these modules. At only 2.00v, which is less than the rated voltage of 2.10v, the modules climbed to an impressive DDR900 fully stable. With voltage increased incrementally, the modules scaled nicely all the way up to DDR1070 with the maximum voltage I decided on of 2.33v. Below is a screenshot of some of the stability testing at DDR1070:

    DDR1070 4-5-5-18 @ 2.33v

    OC Online Image

    3-3-3-8

    OC Online Image

    Just because memory is rated for CL 5 doesn't mean it won't run tighter timings. I decided to see what the FireStix would do at 3-3-3-8 and was quite impressed. At the rated voltage of 2.1v, the FireStix ran a very tidy DDR750 and made it to DDR800 at 2.33v. Keep in mind, that is keeping the secondary timings very tight as seen in the screenshot below:

    DDR800 3-3-3-8 @ 2.33v

    OC Online Image

    4-4-3-5

    OC Online Image

    Reaching DDR800 at a very acceptable 2.33v with 3-3-3-8, I was fully expecting big things from 4-4-3-5 and was not disappointed. The FireStix held stability of DDR800 at a low 2.0v. This I found very exciting because that is below rated voltage. Here is the screenshot of DDR800 at 4-4-3-5 with the ultra low 2.0v VDIMM.

    DDR800 4-4-3-5 @ 2.00v

    OC Online Image

    5-5-5-8

    OC Online Image

    The last of the timings I did stability testing at was the standard 5-5-5-8. This could be considered testing at stock timings because these modules technically are rated at CL 5. I am still not sure why the modules have an SPD table at 4-5-5-18 but either way, at CL 5, the kit went much further than the auto timings above. I was really hoping for DDR1200 but fell just slightly short. Since these modules scale well with voltage, they should only take a small bump to be stable at DDR1200. Again, a screenshot of the DDR1160 stability testing can be found below.

    DDR1160 5-5-5-8 @ 2.33v

    OC Online Image


    Page 6 : Intel Overclocking

    CPU: Intel C2D E6300 (L626A453)
    CPU Cooling: Chilly1 Single Stage Phase Change
    MB: Asus P5B-Dlx Wifi-AP Edition
    GPU: Biostar 7300GT 256MB DDRIII
    PSU: Silverstone Zeus 560W
    HD: Seagate SATAII 80GB 8MB NCQ
    OS: Windows XP SP2 (with all updates)
    Hipro-Tech DDR Maximizer v1.2

    Memory:

    Buffalo FireStix 2x512MB DDR800 5-5-5 @ 2.1v (FSX800D2C-K1G)

    The criteria for stability doesn't change going from AMD to Intel when it comes to memory so the same routine will be used on this Asus P5B-Dlx motherboard as it was on the AM2 setup. Here again is that list of stability tests run on each overclock:

  • 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 410MB/instance
  • Multiple loops of each 3DMark 01 / 03 / 05

    Intel Overclocking:

    OC Online Image

    The only thing I want to mention about this Intel setup is that I will be relying on the use of a wonderful little device called the DDR Maximizer to provide the VDIMM. The Maximizer is a piece of hardware that regulates and provides voltage for memory modules for use on pretty much any motherboard. You can find more details on the DDR Maximizer at Hipro-Tech.com. The layout of results is the same as the AMD Overclocking section but the results will be slightly different, let's have a look.

    Auto 4-5-5-18

    OC Online Image

    Much to my surprise, the overclockability of these modules on the Intel platform appears to be very similar. At the stock voltage of 2.1v and the auto timings of 4-5-5-18, I reached an impressive DDR980 that climbed all the way to DDR1060 with a bit more voltage. The Intel section is off to a great start so far in my opinion.

    DDR1070 4-5-5-18 @ 2.33v

    OC Online Image

    3-3-3-8

    OC Online Image

    At 3-3-3-8 I managed to stabilize DDR800 keeping at the self imposed 2.33v limit for this review. The Hipro-Tech.com DDR Maximizer lets me adjust memory to the hundredth of a volt but because the AM2 motherboard used for overclocking jumped from 2.00v to 2.33v, I decided to stick with the 2.33v limit on the Intel setup instead of settling at 2.30v.

    DDR800 3-3-3-8 @ 2.33v

    OC Online Image

    4-4-3-5

    OC Online Image

    Again, 4-4-3-5 proved to be very similar in overclocking on the Intel platform as it was on the AM2 setup reaching DDR940 at 2.33v. More impressive is the ability to run DDR820 at only 2.0v, my experience with D9GMH has 4-4-3-5 being the sweet spot and these sticks are no different as they responded quite well at these timings.

    DDR940 4-4-3-5 @ 2.33v

    OC Online Image

    5-5-5-8

    You may notice this section is empty and that is not my mistake. I have found some very in-consistent results when running CL5 on the P5B-Dlx over DDR1000. In fact, I have found that CL4 is more stable in some tests than CL5 at DDR1000 and up. This kit was another one of those cases where DDR1000 4-5-5-8 was MemTest stable in Windows while DDR1000 5-5-5-8 at the same voltage was not. Therefore all testing for this kit at 5-5-5-8 will not be published as it is clearly a motherboard/chipset issue.


    Page 7 : AMD Benchmarks

    With a lengthy look at overclocking complete, it is now time to focus on how that overclocking translates into performance. With the benchmarks I use for memory reviews, we should get a good overall feel for how the memory performs at the various clocks and timings that I have found stable. All benchmarks are run three times and averaged out for the graphs. We start with a look at the frequency and timings that will be used for testing:

    OC Online Image

    The last three sets of numbers were at the same clock speed of 2820MHz. I simply adjusted the divider on the motherboard to achieve the different memory speeds and choose the tightest timings offered at each frequency that we found in the overclocking section. We start off with a look at some bandwidth numbers from SiSoft Sandra and Lavalys' Everest.

    SiSoft Sandra 07 SP1 – Memory Bandwidth

    OC Online Image

    Lavalys Everest Ultimate 06 – Memory Bandwidth

    OC Online Image

    The bandwidth numbers in both Sandra and Everest seem to scale with frequency as it is expected to. Tighter timings and lower frequencies very seldom lead to better bandwidth numbers so these results are no surprise.

    Lavalys Everest Ultimate 06 – Memory Latency

    OC Online Image

    The Everest latency results are always fun because it is interesting to see if tighter timings at a given frequency will give better latency numbers than looser timings at a higher frequency. We can see the latency results scale with frequency but the gap between DDR806 and DDR960 is almost non-existent.

    FutureMark PCMark 05 & 3DMark 01 SE

    OC Online Image

    OC Online Image

    Having moved past the bandwidth and latency results we now get our first taste of some application performance. The first FutureMark program I ran was PCMark 05 and as expected, every result scales with CPU frequency and then memory frequency with the CPU at the same speed.

    Super Pi Mod v/1.5 – 1M

    OC Online Image

    Super Pi Mod v/1.5 – 8M

    OC Online Image

    Super Pi Mod v/1.5 – 32M

    OC Online Image

    Again, there are no surprises with the results coming in as they should but the difference between DDR940 at the tighter 4-4-3-5 timings kept closer to the DDR1126 CL 5 times than I was expecting. I spend a lot of time running SuperPi in the benchmarking world and seeing tighter timings perform almost as good as higher frequencies is common but the 32M times were really quite close.

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

    OC Online Image

    When it comes to the gaming results, nothing stands out except the FarCry results. Based on the average FPS of all the results, I think it is safe to say the results are slightly inconclusive but appear to favor memory frequency over everything else. Now let's see how the Intel system faired in the benchmarks.


    Page 8 : Intel Benchmarks

    Much like the AMD testing section, I will be running the Intel setup through the same benchmarks at a handful of different CPU and Memory frequencies and timings. Again, all benchmarks get ran three times with reboots in-between and averaged out. If there are any abnormal results they get tossed. Here is a chart of the overclocks used in this section.

    OC Online Image

    You will notice that I do have CL5 listed above and that is because CL5 seems to be stable for benchmarking, gaming, and programs, it is just not stable in MemTest or sometimes Prime95. Either way, let's have a look at the results starting of course with a peak and the synthetic bandwidth numbers.

    SiSoft Sandra 07 SP1 – Memory Bandwidth

    OC Online Image

    Lavalys Everest Ultimate 06 – Memory Bandwidth

    OC Online Image

    The bandwidth numbers are a bit interesting, especially the Sandra results. According to SiSoft Sandra, the sweet spot for memory on this P5B-Dlx is around DDR900 at 4-4-3-5 with very little gains being had by upping to DDR1000+ at 5-5-5-8. The reason the Sandra result for the 3430MHz result is because Sandra just calculates the bandwidth based on the specs and doesn't actually measure the bandwidth. Clearly a huge advantage is given to CPU frequency when it comes to bandwidth.

    Lavalys Everest Ultimate 06 – Memory Latency

    OC Online Image

    Again, the DDR875 4-4-3-5 numbers come out the winner here showing the lowest latency number in Everest. I am a little surprised thus far with the DDR700 3-3-3-8 results as I thought they would be more equal with the 4-4-3-5 figures but apparently DDR700 just isn't enough "umph" for the E6300.

    FutureMark PCMark 05 & 3DMark 01 SE

    OC Online Image

    OC Online Image

    With the higher CPU speed of 3430MHz, the results are a no brainer for programs like PCMark 05 and 3DMark 01. These two programs tend to be a good reflection of overall system ability so obviously a 1000MHz lead in CPU frequency will result in much better performance. Again though, the DDR700 3-3-3-8 scores appear noticeably lower than the other two setups at the 2450MHz mark.

    Super Pi Mod v/1.5 – 1M

    OC Online Image

    Super Pi Mod v/1.5 – 8M

    OC Online Image

    Super Pi Mod v/1.5 – 32M

    OC Online Image

    The SuperPi results are pretty predictable if you spend any time running SPi. CPU frequency is king followed by memory speed. The close neck and neck battles at 1M, 8M, and 32M between DDR875 4-4-3-5 and DDR1050 5-5-5-8 just goes to show how close those two sets of timings/frequency are in performance. On the Intel platform it really doesn't appear to be beneficial for higher CL5 frequency when compared to tighter timings around the DDR900 mark. Let's see if that holds true for gaming results.

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

    OC Online Image

    First off, toss out the Far Cry results. Clearly the dual-core factor is messing with those numbers and I didn't have the right patches/updates done to fix it. Time to get another game involved in benchmarking to replace Far Cry I guess. As for the other results, obviously memory isn't going to have a huge effect on gaming but there is some change amongst the three values at the same frequency that are worth noting.

    This wraps up my extensive overclocking and testing of these FireStix and I will sum up my overall opinion on the last page of this review in the Conclusion section.


    Page 9 : Conclusion

    I really wish I could have found a retailer that sold the 512MB Buffalo FireStix because then I could gush about how a low cost module can be so good. Based on the price of Buffalo's 2GB FireStix kits with the same specifications, I can only assume that the 1GB kit is priced to impress as well. I can still gush about the performance and more importantly, the overclocking performance of these FireStix.

    OC Online Image

    The rated operating frequency of DDR800 at 5-5-5-18 is very conservative and these modules, sporting Micron D9GMH ICs, could easily be rated for DDR800 4-5-5-15 operation at 2.1v or even 4-4-4-15 at 2.2v. 4-5-5-18 is actually the timings that they booted at in every motherboard I tried them in, both Intel and AMD, so you can almost call them a CL 4 rated piece. There was no compatibility issues with these modules and even with BIOS defaults loaded on my P5B-Dlx (1.8v VDIMM), the modules booted with ease unlike many others before them.

    Throughout the extensive overclocking and testing of these modules, I could not come up with anything to complain about. They look great, they perform as they should, and they have plenty of overclocking headroom for the enthusiasts out there. Again, without a retail price anywhere it is hard to give a very concrete judgment but something tells me that these modules should be a part of anyones buying decision if youre in the market for a 1GB memory kit. I can hardly wait to get my hands on a 2GB kit of some hot FireStix.

    Advantages

    Beautiful red hot heatspreaders
    Compatible in picky P965 chipset motherboards
    Plenty of overclocking headroom for the enthusiasts in the crowd
    Really good performance at low voltages with tighter timings

    Disadvantages

    I could only find one retailer on-line in North America

    Overclockers Online would like to thank Buffalo for making this review happen.

Leave Comment

You must be logged in to post a comment.