Corsair Twin2X 2048-8500C5Aug 17th, 2006 | By Archive
Corsair Twin2X 2048-8500C5
: 08/17/06 – 04:26:02 AM
Page 1 : Index
Round two begins and my AM2 single core test rig is looking to clock high with a pair of Corsair DDRII modules that promise to impress. The onslaught of DDRII memory will hopefully continue on for some time because I cannot stress enough how much fun I get out of pacing memory modules and getting every bit of performance out of them at the various timings. With that said, being rated to run at DDR2-1066 means there are plenty of timings to find in the Corsair Twin2X 2048-8500 C5s.
That's right folks, the crème de la crème, a pair of PC2-8500/DDR2-1066 monsters. The C5 in the name dictates a CAS timing of 5 which is the standard for the small handful of DDR2-1066 rated memory that are out there right now. Corsair was also kind enough to send a pair of 1GB modules as opposed to a pair of 512MB modules which makes sense as most users are going that route, especially with the introduction of Vista looming overhead. This memory is also available in 512MB modules for 1GB kits as well.
Corsair needs no introduction for a review so I will skip all of the boring details. Put it this way, they have been a pioneer in the memory game for over 10 years, back when a Mac was still just a Mac and happy to be so. Focusing particularily on mission-critical environments and high-speed applications which includes the highest-demanding customers of all… gamers. There isn't much more high-speed right now than PC2-8500 so Corsair is sticking to their company creto being one of the earliest to market in the DDR2-1066 category with the EPP-ready Corsair Twin2X 2048-8500 C5. As usual, there is plenty to cover so sit back, relax, grab a drink, and take a ride on the DDRII train to possibly DDR2-1200 and beyond…
Page 2 : Package & Contents
I don't know why we bother even showing the memory anymore. They all look the same at this point and without extravagent heatspreaders, they fit the same as any other DDRII module out there. Either way, here is the eye candy starting with the in-package photos as usual.
Now that is some purple action that can't be denied. The prototypical Corsair blister pack holds the modules securely in place for distribution. My last pair of XMS Pro from Corsair came with a pastel colored teal insert and these XMS2 modules now come with yet another pastel color, an almost mauve-like purple.
The insert is a generic insert used for all the Twin2X 2GB modules, I would imagine, as the only description the insert has aside from the Corsair branding is 2GByte and DDR2. The modules themselves provide the specifications by way of the sticker on the right hand side of the actual module.
Again, there is nothing new or inventive to see, this is just how it is done for memory and will be for some time, I would imagine. The sticker on the module provides operating frequency, timings, model, and part number as well as serial number.
THe XMS2 logo dominates the top of the package and there is no mistake that DDR2 memory is what is inside this package.
The rear of the package is still the same as it has been for a while with a few review quotes and your basic legal lines of text. I think it is time to pull the package apart and get some finger prints on these little sticks of joy.
Imagine that, the modules look the same outside of the package as they do inside the clear plastic protection. The colors on the insert and module labels are both still a pastel purple as well. It is creepy how light can sometimes do that with clear plastic.
The backside of the modules is not without a sticker as well. I often wonder how much the stickers on heatspreaders affect the transfer of heat from the moudles to the open air. Corsair has stickers along both sides of these modules so they don't think it hinders it at all, I guess. Also note, no clips at all are used on these modules, I knew they looked "leaner" than other Corsair XMS memory.
The PCB used is, of course, a multi-layered one and we can actually see the layers in this photo here. The lack of a visual of memory chips peeking out from under the heatspreaders indicate small surface area ICs underneath. These XMS2 modules come off as light and airy with little weight or girth to them. I think the lack of a clip on top is the main reason for the sense of a smaller module and that is enhanced by the lighter weight of the DDR2 modules. The heatspreaders are most of the weight on memory so they just might be a thinner heatspreader than I am used to from Corsair XMS memory.
The other thing I like about this heatspreader design is that it is open at the top for hot air to escape out of. With thin and light heatspreaders like the ones Corsair is using for these 8500 C5s, installation will not be an issue in any motherboard.
Page 3 : Specifications
In the past, memory hasn't had a lot going on in the way of specifications and features. You would have your frequency and your timings, perhaps a different specified operating voltage from JEDEC standards, which is more common these days. Other than that, there wasn't a whole lot to memory without getting into the ICs used. Now, however, there is a bit more to discuss… such as EPP and SLI memory. More acronyms to roast on the fire…
I will start with
which is also being touted as
and the like. EPP stands for Enhanced Performance Profile which means the module has additional SPD profiles programmed into the module that NVIDIA nForce 590 motherboards can enable to automatically run the memory at the advanced timings/frequencies/voltage/drive strength or any combination of the four. In the testing section I will enable EPP and see what we get different from the standard SPD profile as well. Corsair mentions this about EPP when describing the TWIN2X 8500 C5s:
Here is the rest of the outlined specifications provided by Corsair at their website in the form of a PDF file. I am still not sure I like their method of delivery for specifications and features but the Corsair web site is always fast and reliable so I can't complain if I have to open a PDF file in my browser to look at memory specifications.
- 2048 Megabytes of DDR2 memory
- SPD includes Enhanced Performance Profiles (EPP) which allow automatic overclocking to aggressive performance settings
- Implemented using 64M x 8 DDR2 SDRAMs
- 100% tested at 1066MHz in high performance DDR2 motherboards
- Legendary Corsair reliability and service
- Lifetime warranty
We get a better understanding of what the EPP SPD profiles will do for the memory and what is enhanced from this image taken from Corsiar literature. It appears that these modules are programed to boot up with loose relaxed timings of 5-5-5-18 on the 800MHz divider and then tightening up slightly to 5-5-5-15 but run at a higher divider I think. Testing will tell for sure. I will now have a look at what is under the heatspreaders on the Twin2X 8500 C5s.
Warranty? What warranty? Oh, you are not supposed to take those off; otherwise you void warranty? Thanks for the information now.
Joking aside, taking your heatspreaders off not only voids warranty but it also gives you an excellent opportunity to rip an IC off and guess what? That isn't covered under warranty either. Not to mention that these heatspreaders don't go back on. Don't bother trying to explain to the Corsair rep that they fell off when you are trying to RMA later on down the road. I am not sure if this was intentional by Corsair but these heatspreaders seem to be adhered to the ICs by a heat process with the thermal tape applied to the heatspreader first and possibly heated before being stuck to the module. Either way, once you "un-stick" them, there is no adhesion left and I am left with naked sticks.
And naked they will stay because they are just too sexy looking, all naked and stuff. Needless to say these are D9GMH ICs supplied by Micron. At this point in time Micron seems to be leading the charge and has for some time with every pair of high-end DDRII modules having some form of D9 ICs on-board. My understanding is that the D9GMH are rated for 333MHz at CAS 5 with a 3ns recycle time but are binned at 400MHz at C4 with a recycle time of 2.2ns. This means they may perform well at low frequencies but also run very well at the high end.
The PCB is Brain Power much like many top end kits these days. BP is the king of multi-layered PCB for memory and has held that belt long through the DDR days. We should be armed with enough information about these modules now to perform some testing so next up is installation.
Page 4 : Installation & Overclocking
Corsair is the one company that will occasionally stuff a little memory module inside of a huge LED-encrusted heatspreader that is 'slightly' taller than your standard module. The TwinX and Twin2X modules are not of this girth and come, as we saw, in small sleek black heatspreaders. My dad always said you can't go wrong with black and Corsair certainly has been carrying on the dark tradition for a long time. We begin with the playing field:
As I did with the OCZ DDR2 PC2-7200 Platinum XTC EPP, I will be utilizing the memory clocking prowess of the Asus M2N32-SLI which continues to impress me and runs DDRII modules to limits that bring Integrated Memory Controllers to their knees.
The M2N32-SLI requires modules to sit side-by-side which is my only real gripe but easily overlooked when the modules still have a bit of space for that precious airflow to creep in and evacuate the hot air when frequncies reach the 600MHz mark. Notice the modules magically have heatspreaders back on again. This is because I do all testing and overclocking for the review before removing the heatspreaders to ensure my results are not compromised by a different setup than you can buy in the store.
My setup isn't anything fancy, your typical bench setup in cramped quarters with other motherboards and their components. A stock AMD retention bracket and the water block mounting bolts make the perfect perch for a 120mm fan to blow directly down on the modules. I will be more than comfortable stressting these beasts with solid airflow like this. Here is the full list of hardware used for testing.
Asus M2N32-SLI Deluxe
HIS X800GTO IceQ II 570MHz Core / 600MHz Memory
OCZ GameXStream 700W
WD Raptor 10,000RPM SATA
Windows XP Pro + Latest Updates
Corsair TWIN2X 2048-8500C5
I like to find the limit to a kit of memory at various timings that are offered to compare performance numbers. I will find the maximum overclock of the Corsair Twin2X 8500 C5s at CAS latencies of 3, 4, and 5 at the rated voltage of 2.2V. I will then turn the dial up and see what it takes to stabalize the memory at DDR1200 5-5-5-x.
My definition of memory stability is being able to run any benchmark repeatedly without reboots, lock-ups, or other issues. Being able to repeat results consistantly is also important. These overclocks may or may not be good enough for 24/7 use but that would require days of further Stress Prime and plenty of Rosetta testing but these clocks should definitely be game stable for extended periods…very extended periods. The screen shots will show the various programs used for testing which include:
- 32M digit calculation of Super Pi Mod v/1.4
- Future Mark 3DMark 01se
- StressPrime 04 for a good couple of hours
Again, this is by no means adequate testing for 24/7 use but time constraints limit my ability for that. That said, with tweaking of secondary timings, I could make any of the following overclocks 24/7 Rosetta, Prime, or 3D stable on the hardware I am using now.
DDR758 3-3-3-5 @ 2.2V 2T
The 3-3-3-5 performance wasn't impressive to say the least although DDR800 was almost achieved. Lowering Trcd to 4 may have provided that ability but the performance loss isn't worth it in my opinion for the small amount I could gain. CAS 3 isn't what these sticks were built to run anyway, and definitely not DDR7XX.
DDR1020 4-4-4-8 @ 2.2V 2T
At 4-4-4-8 the Corsair 8500 C5s start to warm up and stretch their legs. Remember folks, this is at the rated voltage by Corsair, 2.2V. DDR1000 is a nice achievement for Cas-Trcd / 4-4 and with enough voltage I am sure I can stabilize DDR1100 at these timings. All that is left is to shoot past the stock rated timings and frequencies without crossing stock voltage.
DDR1160 5-5-5-12 @ 2.2V 2T
Done, and done. Stock frequency surpassed handily and the timings are a hefty pinch more taut than the stock timings. Being so close to DDR1200 I was expecting a slight bump in voltage required to max out my CPU at 3GHz on the 800 divider but that wasn't the case. It took a full .2V increase to the memory to stabalize DDR1200.
DDR1200 5-5-5-8 @ 2.4V 2T
I wouldn't recommend daily use or constant usage of memory over what is recomended by the manufacturer and many will actually void warranty if they are overvolted so proceed at your own risk when crossing that line, knowing that damage and death can easily occur. The above screenshot also happens to contain my personal best for a 32M Super Pi run at 3000MHz or less. I wasn't even trying and just noticed while writing this up. That means there is definitely small differences there that Super Pi can take advantage of with the memmory at a whopping DDR1200 over crazy tight latencies at DDR600 on socket 939, which my previous personal best was done with.
Page 5 : Performance & Testing
As it turns out, I have a 3GHz single core AM2 CPU now so I might as well take advantage of it and do a set of benchmarks at the DDR1200 settings we saw on the previous page. I will also run
, and the stock AMD speed of 2GHz with the timings as if I had dropped them in without a trip to the BIOS alongside the EPP cranked up frequencies. Here is the chart of the timings and settings used for the benchmarks:
All benchmarks are run 3 times and the results are averaged for the charts below. I like to start with bandwidth numbers so let's see what SiSoft Sandra 07 and Everest Ultimate 06 have to say about these modules.
SiSoft Sandra 07 – Memory Bandwidth
Everest Ultimate 06 – Memory Bandwidth
The bandwidth numbers here tell the story of "higher is better". The last three lines lay it out very easily with the processor staying at the same speed and the memory timings and frequency the only thing changing. Obviously with higher frequency the bandwidth is going to go up. It has scaled appropriately in both Sandra and Everest, but will the latency also scale higher with frequency?
Everest Ultimate 06 – Memory Latency
What is interesting here is that the latency also scaled equally with higher frequency. This means DDR1200 @ 5-5-5-12 not only pushes more bandwidth but also has lower latency than DDR1000 @ 4-4-4-8. Or at least it does in this particular setup. Again, this leads us to the conclusion that memory performance increases with memory frequency despite the looser timings on AM2… so far it does anyway.
Futuremark PCMark 04/05 & 3DMark 01 SE
I like to run these three programs when testing memory because bandwidth and latency actually seem to affect results as they are system orientated benchmarks. For the PCMark benchmarks I only run the memory suite.
The results are very similar to SiSoft Sandra scaling with frequency but you will notice no hiccup at the Orange and Black bars like we do in Everest. This makes me question either Sandra and PCMark or Everest and something tells me Everest is the only one that actually tests the memory.
Clearly the system performance is king in 3DMark 01 SE, that is why I use it for motherboard and memory reviews in place of the newer versions. We can see that the processor speed has more impact on the results than anything but the memory increases show a steady climb in 3DMark scores as well.
Super Pi is very memory dependant for performance, that alone makes it a good tool to use as a stability gauge in combination with various other programs. It also provides a great playground for testing memory and that is why it is one of my favorite benchmarking programs for forum competitions.
Super Pi Mod v/1.4 – 1M
Super Pi Mod v/1.4 – 8M
Super Pi Mod v/1.4 – 32M
Needless to say there isn't much explaining to do, the charts speak for themselves. In Super Pi, memory latency plays a huge role in performance. Often times, tighter timings and less frequency will give better results like in the case of BH5 @ 2-2-2-X and TCCD @ 2.5-3-3-X. The BH5 will almost always beat the TCCD in Super Pi on one memory divider lower. With AM2 and DDRII we see that has changed… unless we get memory dividers closer together. A 5.5 memory divider would let me compare DDR1200 @ 5-5-5-X to DDR1100 @ 4-4-3-X… that would be an interesting battle.
Gaming Benchmarks – Far Cry / Half-Life 2 / UT 2004
I again base my choice for games to benchmark on the idea that I am looking for system gains to improve performance so I turn to games that tend to rely more on the system than the video card. In games where the GPU is the main factor in performance, memory won't play a big role. With multi-threaded games and dual core processors the norm now, system performance will start to play a larger role in performance, this includes memory. For now, size is most important so make sure to grab 2GBs for your next kit.
I really like AM2 for benchmarking because we have the ability to simply change memory frequency and timings without adjusting the CPU or any other clocks on the system. This allows us direct memory comparisons and the games even show noticable improvement with the higher memory clocks.
As we saw in the benchmarks, the EPP overclock max takes the CPU to a mind boggling 2680MHz to get the memory running at the rated DDR1066. Here are the screenshots of CPU-Z so you can see the timings with EPP enabled. Remember, this is all automatic adjustments, just a matter of turning it on in the BIOS and setting to max OC. Notice the raise in voltage on the CPU as well when it is set to Max OC, again, all automatic.
With EPP turned off, the system picks the modules up at 5-5-5-18-23 at the maximum AM2 divider of DDR800. You would have to manually set them up in a motherboard that didn't support EPP to lower the timings down when running at this frequency if you were not overclocking your system bus (HTT).
EPP Enabled – Performance
With Performance choosen in the BIOS we now get the timings shown in the first row of the timings table in CPU-Z. These are a fair bit tighter than the 5-5-5-x we saw earlier and we now get 4-4-4-12-22. The last of the profiles, Frequency, I fully expect to be the second column of the timings table.
EPP Enabled – Frequency
The frequency doesn't change but the timings do loosen back up again. These timings provide the best possible platform for high clocks so the only thing left to do is turn up the EPP overclock and see where it takes us.
EPP Enabled – Max OC
I can only assume that many people will be sadly reminded not all CPUs are created equal when shooting for this setting of Max OC. If by chance you had a 2.6GHz rated CPU then you are set as the overclock is only a mere 80MHz over the whopping 680MHz jump my 3200+ is having to make to run these. Of course my chip does this frequency without worries but be warned, your 3000+ or 3200+ may not, not to mention many other AMD CPUs aside from the FX line.
Page 6 : Conclusion
Everything about the Corsair Twin2X 2048-8500 C5 screams power and speed. Right from the on-set with sleek black heatspreaders that seem smaller and thinner than most, these modules are truly evil and dark, producing some staggering frequencies. I couldn't top the modules out with a 3GHz processor. Of course, that needed voltage that would void the warranty, but even at the specified operating volts, the 8500 C5s really hold their own.
The enhanced EPP profiles do make it very easy for the beginner to overclock their system with a compatible motherboard, to run the memory at its rated frequencies. Keep in mind an FX-62 or equivelent processor is required to keep up with the memory, making the Corsair Twin2X 8500 C5s the ultimate AM2 FX companion. Again, if you are willing to test fate like me, these modules can clock to the moon with a little more leg work.
I would like to have seen slightly better tight timing performance to provide that lower clock flexibility making one kit perfect for any setup or even platform, AMD or Intel. To be fair though, these sticks are meant for high frequencies and should be near the top of the list for any serious overclocker or benchmarker. Of course with being near the top of that list comes a hefty price tag. So ladies, take this bit of advice from me: the way to an overclocking man's heart is through Corsair Twin2X 2048-8500 C5s.
Small, for nice fit in tight slots
Great high frequency performance
EPP will help many with nForce 590 motherboards
$$$ – high performance still means high cost
I was left wanting more at 4-4-4-X
I would like to thank Corsair for providing the Twin2X 2048-8500 C5s for review.