Buffalo Technology is a company familiar to most. They are very popular in the SOHO NAS and networking market with popular products such as the Linkstation Pro. Buffalo may not be the first name that springs to mind when thinking about enthusiast DDR2 but with their new line of Firestix modules, that is sure to change. Today I’m going to be taking a look at the most affordable dual channel kit in the DDR2 FireStix lineup — the FSX800D2C-K2G.
The DIMMs came in a typical plastic package. There is no literature included except for the cardboard insert. You’ll find the following on the back of the package:
- Lifetime Warranty
- Aluminum heat spreader with thermally-conductive adhesive
- Extremely high over-clocking capabilities and extended RAM life
- Exceptional performance in latest gaming motherboards
- ISO 9001, ISO 14001 and RoHS Compliant
The FSX800D2C-K2G kit has the following technical specifications:
- Memory Type: DDR2 240pin, Unbuffered, Non-ECC
- Capacity: 2GB (2x1024MB)
- Rated Frequency: 800MHz PC2-6400
- Timings at Rated Frequency (CAS-TRCD-TRP-TRAS): 5-5-5-18
- Voltage: 2.1V
- Banks: 4 (2 per DIMM)
Although FireStix are targeted to the enthusiast, almost their entire lineup of DDR2 is rated for rather loose CAS 5 timings at 2.1V. As you’ll see later in the review, you should never judge modules by their default specifications. On a positive note, these rather loose SPD timings will undoubtedly improve ‘out of the box’ compatibility with a large number of mainboards. Many manufacturers are setting SPD values below advertised values to ensure board compatibility — it is then up to the enthusiast to set the rated timings and voltage manually.
The FireStix are good looking, simple modules. The heat spreaders may not be as elaborate as some on the market, but they are indeed functional. Buffalo claims to use ‘thermally conductive’ adhesive for the heat spreaders. I did not brave trying to remove them as they are secured very well to the DIMMs.
Test Setup
I’ll be putting the FireStix through their paces on an AM2 based system comprised of the following:
- AMD Athlon 64 X2 6000+
- DFI Infinity NF570-M2/G
- Thermalright SI-128 SE w/Panaflow ‘High’ fan
- 2x1GB Buffalo Firestix PC2-6400 (FSX800D2C-K2G kit)
- BFG 8800GTS 640MB OC (6.14.11.6218 drivers)
- PC Power and Cooling 510 Express
- Lian-Li PC-65B
- Windows XP Professional SP2 (32 bit with all hotfixes installed as of 9/14/2007)
Underneath that behemoth of a heatsink, there are a couple of red DDR2 DIMMs. My DFI Infinity NF570-M2/G had no issues booting up with the FireStix kit. Upon visiting the BIOS, vDIMM was defaulted to 2.0V. I manually increased it to 2.1V and proceeded to Windows. It should be noted that the heat spreaders got nice and toasty — although this may not sound like a positive thing, it is. This means heat is successfully being transferred from the ICs to the surface of the heat spreaders where it can be dissipated. The powerful 100+CFM Panaflow ‘high’ kept the DIMMs only warm to the touch when at full speed.
As can be seen above, there are three SPD configurations programmed. In case you are wondering why the DIMMs show as made by Melco — Buffalo Inc. is a wholly owned subsidiary of Melco Holdings Inc.
Overclocking
DISCLAIMER: As always, overclock at your own risk. Icrontic, its owners and authors do not take any responsibility for damage caused to your system while overclocking. Remember that overclocking and or overvolting hardware will almost certainly void your warranty. Check with the manufacturer if you are ever unsure. We do try to be as realistic as possible during our testing but we can make no guarantee that you will be able to achieve the same results. There are simply too many variables to consider while overclocking.
Overclocking the FSX800D2C-K2G was a very enjoyable experience. As I suspected, the rated specifications are very modest. It appears that the vast majority of FSX800D2C-K2G kits contain none other than the very popular and overclockable Micron D9GMH ICs. D9GMH is well known for its good response to vDIMM and high frequency potential. There have been some isolated reports of Elpida AG-6E-E in this kit, but rumour has it that was only a small batch. As always, your mileage may vary. Buffalo’s higher end FireStix kits are reported to contain D9GKX — which is higher binned D9 than D9GMH. Sticking with D9 for their performance line is definitely a wise move by Buffalo as this is highly sought after RAM.
To conduct the overclocking tests, I set the CPU multiplier to 10x to eliminate the processor as a potential limit. I have confirmed that the DFI Infinity NF570-M2/G is stable beyond 400MHz reference clocks, so the board will not be the limiting factor. CPU voltage was increased to about 1.425V for testing to help protect the on-die memory controller from high vDIMM (a large delta in vDIMM/vcore can cause damage to the memory controller). The Thermalright SI-128 SE as can be seen in the previous section actively cooled the DIMMs (quite a bit of overhang). Adequate airflow is important with DDR2 when pushing vDIMM beyond 2.1V. Command timing was also left at 2T for all tests. This is done because 1T is often very mainboard and CPU/memory controller dependent and is rarely a factor of the memory itself. Conducting testing at 2T (or CPC off) paints a more realistic and achievable picture for the majority of individuals. Secondary timings are all left to Auto during testing.
Without further ado, lets get to the results.
I started by pushing the modules at their default 5-5-5-18 timings. This is what I was able to achieve.
Freq. | CAS | TRCD | TRP | TRAS | vDIMM | Stability |
800 | 5 | 5 | 5 | 18 | 2.1 | PASS |
902 | 5 | 5 | 5 | 18 | 2.1 | PASS |
940 | 5 | 5 | 5 | 18 | 2.1 | PASS |
968 | 5 | 5 | 5 | 18 | 2.1 | PASS |
988 | 5 | 5 | 5 | 18 | 2.1 | PASS |
1008 | 5 | 5 | 5 | 18 | 2.1 | PASS |
1040 | 5 | 5 | 5 | 18 | 2.1 | PASS |
1068 | 5 | 5 | 5 | 18 | 2.1 | PASS |
1086 | 5 | 5 | 5 | 18 | 2.1 | PASS |
1100 | 5 | 5 | 5 | 18 | 2.1 | FAIL |
1100 | 5 | 5 | 5 | 18 | 2.2 | PASS |
1120 | 5 | 5 | 5 | 18 | 2.2 | FAIL |
1120 | 5 | 5 | 5 | 18 | 2.3 | PASS |
1140 | 5 | 5 | 5 | 18 | 2.3 | FAIL |
1140 | 5 | 5 | 5 | 18 | 2.4 | PASS |
1160 | 5 | 5 | 5 | 18 | 2.5 | FAIL |
1160 | 5 | 5 | 5 | 18 | 2.5 | FAIL |
1160 | 5 | 6 | 6 | 18 | 2.5 | FAIL |
The FireStix were able to reach an impressive 1086MHz at the default voltage of 2.1V. Interestingly, the DIMMs had difficulty with frequencies above 1100MHz — it was as though they hit a wall that no amount of extra voltage could overcome. At 1160MHz, there were literally thousands and thousands of Memtest errors, even at 2.5V. Only 20MHz less yielded perfect stability at a more reasonable 2.4V and 1100MHz with only 2.2V is very impressive.
I then proceeded to set timings to 4-4-4-12. This timing combination can be found in many higher end PC2-6400 kits. This was the result:
Freq. | CAS | TRCD | TRP | TRAS | vDIMM | Stability |
800 | 4 | 4 | 4 | 12 | 2.1 | PASS |
902 | 4 | 4 | 4 | 12 | 2.1 | PASS |
940 | 4 | 4 | 4 | 12 | 2.1 | PASS |
968 | 4 | 4 | 4 | 12 | 2.1 | PASS |
988 | 4 | 4 | 4 | 12 | 2.1 | PASS |
1000 | 4 | 4 | 4 | 12 | 2.1 | FAIL |
1000 | 4 | 4 | 4 | 12 | 2.2 | FAIL |
1000 | 4 | 4 | 4 | 12 | 2.3 | PASS |
1020 | 4 | 4 | 4 | 12 | 2.3 | FAIL |
1020 | 4 | 4 | 4 | 12 | 2.4 | PASS |
1040 | 4 | 4 | 4 | 12 | 2.4 | PASS |
1060 | 4 | 4 | 4 | 12 | 2.4 | FAIL |
1060 | 4 | 4 | 4 | 12 | 2.5 | PASS |
1064 | 4 | 4 | 4 | 12 | 2.5 | PASS |
1080 | 4 | 4 | 4 | 12 | 2.5 | FAIL |
Things really start to get interesting at tighter timings. Setting timings to 4-4-4-12 allowed for almost as much headroom as 5-5-5-18, topping out at 1064MHz, which is just a sliver off of PC2-8500. They did require a rather high 2.5V to achieve this frequency but they were still able to clock up to PC2-8000 (1000MHz) at a more reasonable 2.3V. The DIMMs scaled a little better with voltage at the tighter timings. Very impressive!
As can be seen above, the DIMMs craved a bit more vcore to maintain stability at higher clocks and 4-4-4-12 timings.
Although 4-4-4 timings are considered ‘tight’ for DDR2 at any frequency above 800MHz, I wanted to see just how far these DIMMs could go with extremely tight 3-3-3-8 timings. Timings like this are in the DDR1 league and are rare to find default on anything other than expensive enthusiast grade 667MHz kits. Because of this, I started clocking them up at 667MHz. Amazingly, this is what I was able to achieve:
Freq. | CAS | TRCD | TRP | TRAS | vDIMM | Stability |
667 | 3 | 3 | 3 | 8 | 2.1 | PASS |
700 | 3 | 3 | 3 | 8 | 2.1 | PASS |
940 | 3 | 3 | 3 | 8 | 2.1 | PASS |
732 | 3 | 3 | 3 | 8 | 2.1 | PASS |
749 | 3 | 3 | 3 | 8 | 2.1 | FAIL |
749 | 3 | 3 | 3 | 8 | 2.2 | PASS |
772 | 3 | 3 | 3 | 8 | 2.2 | FAIL |
772 | 3 | 3 | 3 | 8 | 2.3 | PASS |
799 | 3 | 3 | 3 | 8 | 2.3 | FAIL |
799 | 3 | 3 | 3 | 8 | 2.4 | FAIL |
799 | 3 | 3 | 3 | 8 | 2.5 | PASS |
804 | 3 | 3 | 3 | 8 | 2.5 | PASS |
815 | 3 | 3 | 3 | 8 | 2.5 | PASS |
832 | 3 | 3 | 3 | 8 | 2.5 | FAIL |
832 | 3 | 4 | 3 | 8 | 2.5 | PASS |
849 | 3 | 4 | 3 | 8 | 2.5 | PASS |
I was amazed to see very tight 3-3-3-8 timings rock solid at 815MHz. It did take the maximum voltage my DFI Infinity NF570-M2/G could provide, but impressive none the less. I got a bit of extra headroom by increasing TRCD to 4. Not only is the FireStix PC2-6400 kit stable at tighter 4-4-4-12 timings at its default frequency of 800MHz, but also at 3-3-3-8.
As can be seen above, the DIMMs get very voltage hungry at 3-3-3 timings.
A 32M SuperPI test at 418MHz, 3-3-3-8 reaffirms just how stable this memory is at this configuration.
To complete my overclocking tests, I wanted to get a complete picture of the DIMMs stability at varying timings. Voltage was set to 2.5V since the FireStix always seem to benefit from this high level. As can be seen above, I was able to maintain a TRP value of 3 all the way to 1000MHz. The FireStix have no problems keeping the timings tight when fed ample voltage. Please note that TRAS did not play a large role in stability and was included in the graph only for completeness sake.
Performance
Measuring performance of memory is always a controversial subject. Graphs and charts are always nice to look at, but there are some important variables that need to be kept in check to ensure an apples-to-apples comparison. The biggest challenge with the AM2 platform lies in one of its greatest benefits — the integrated memory controller. As the reference clock is increased to overclock the memory, so is the CPU clock. Memory performance increases proportionally with the CPU clock and this can throw results off quite a bit. If you were to compare two Sisoft Sandra XII bandwidth benchmarks with the exact same timings and memory frequency, but different CPU clock speeds, you would have something like this:
Freq. | CPU Freq. | CAS | TRCD | TRP | TRAS | vDIMM | Memory Bandwidth (Sisoft Int Buff’d iSSE2) |
800MHz | 2400MHz | 4 | 4 | 4 | 12 | 2.1v | 8313MB/s |
800MHz | 2800MHz | 4 | 4 | 4 | 12 | 2.1v | 8544MB/s |
Although this may not look like a significant difference, a 200MB/s increase in bandwidth is almost as significant as decreasing memory timings from 5-5-5-18 to 4-4-4-12 and thus, cannot be ignored. I will only be comparing various timing/frequency combinations with the CPU clockspeed kept constant.
As you can see, the greatest impact to bandwidth came from increasing the frequency. There was still a significant 400MB/s increase in bandwidth from the timing changes alone, but an incredible 2.2GB/s increase in bandwidth is realized by moving from 400/800MHz to 560/1120MHz.
We see a similar picture with the CPU at 3GHz. As mentioned earlier, bandwidth clearly increases as the CPU clock speed increases. About 320MB/s is realized by tightening up timings from 5-5-5-18 to 4-4-4-12 at a PC2-8000 frequency.
The greatest impact to random access latency came from the increased frequency. A 6ns decrease is realized by decreasing timings to 3-3-3-8 and a 13ns improvement by increasing the frequency to 1120MHz while leaving timings at their default SPD values.
At PC2-8000 and a 3GHz CPU frequency, a 4ns delta is realized between 5-5-5 and 4-4-4 timings.
SuperPI relies on pure number crunching ability and benefits by almost a full second at 1120MHz. This is significant considering the only change was to the memory frequency. A smaller benefit is realized by tightening timings both at 800MHz and 1000MHz. High bandwidth and low latency results in better SuperPI performance.
When it comes to gaming, memory performance does indeed make a difference as you can see in the HL2 Lost Coast benchmarks. With my 8800GTS, HL2 is not GPU limited and benefits from a pretty large 17 FPS overall improvement at 1120MHz. Higher bandwidth and lower latency memory will not improve frame rates much in a heavily GPU bottlenecked system but improvements can be seen when CPU/Memory limited.
Final Thoughts
The Buffalo FireStix FSX800D2C-K2G kit is an excellent product at a great price. They were purchased from NCIX.com for a very reasonable $114.99CDN — you will have a difficult time finding more bang for your buck at that price point. The FireStix had no issues surpassing PC2-8000 frequencies at timings much tighter than the SPD defaults. At their default timings, they could be overclocked by a very impressive 42.5 percent or 340MHz. There are not many PC2-6400 5-5-5 kits able to run at super tight 3-3-3 timings, either. The heat spreaders are not flashy but are definitely functional and aesthetically pleasing. Pair up all of these benefits with a lifetime warranty and Buffalo Technology has a clear winner on their hands. I would highly recommend this product.
Pros:
- Lots of overclocking headroom (tight timings and high frequency potential)
- Good vDIMM scaling
- Good price
- Functional heat spreaders
- Lifetime Warranty
Cons:
Unimpressive SPD default timings. Not a good choice for someone who does not (or can not) overclock.