Crucial Ballistix 1GBx2 PC3-16000Jul 3rd, 2008 | By Overclockers Online
Crucial (Micron Technology Inc.)
What is in a number? At one time, 300 was crazy, then 400 became stock and 600 was the new goal. That didn't last very long as stock then turned into 800 and the goal was 1000. Are you confused? Unless you are an utter memory geek like myself, you should be. I am of course talking about memory frequency and how not 'so' long ago 300MHz at 2.5-3-3-5 was the ultimate memory overclock and only achieved by the best of the best TCCD/TCC5. Today we usher in a new era of memory frequency numbers with a look at one of the highest rated kits of DDR3 memory available from the heart of the DDR2 and DDR3 dragon, Crucial and their Ballistix line of memory.
Crucial, also known as Lexar Media Inc., is a division of Micron Technology Inc. and has been putting out the Ballistix line of memory for years now. With each new generation of hardware, Ballistix memory kits are always at the top of their game satisfying the needs of its customers without questions. Crucial memory is as well known in the computer enthusiast forums as it is the IT department of your office. The Ballistix line of memory is not the only type of DRAM that Crucial offers as they qualify hundreds of different modules for all major OEMs so whether you build your own system or are upgrading your Dell, Crucial has you covered.
The focus of today is going to be solely on the Crucial Ballistix line of memory, in particular, the new 2x1GB PC3-16000 Ballistix kit. This flagship kit of DDR3-2000 9-9-9 modules are at the absolute top of the food chain on planet DDR3. At this time, there are a few manufacturers that offer DDR3-2000 modules but nothing faster thus far. Don't be misled by what seems like very loose timings because at 1000MHz, 9-9-9 is lighting fast. There should be some interesting results with a kit this fast and we have the perfect platform to push it to the limits, so let's get started.
Page 2 : Package & Contents
In the past we have displayed great enthusiasm in the Crucial packaging and that is going to continue today. The brown box is the ultimate package in our opinion and companies like Crucial and Thermalright apparently agree.
It is just unfortunate that more companies can't recognize that a solid brown box is the way to go because most, if not all sales are going to be through on-line vendors. On the shelf fancy displays are a thing of the past and shipping security should be priority number one with retailers. Of course, with memory, the standard package is the plastic clamshell and yes it is a solid solution but there are times when the plastic clamshell falls short. Sometimes they can come loose and the modules can show up strew throughout a package with potential damage to capacitors or resistors, it has happened to us here.
The interior of Crucial's cardboard box doesn't allow for that sort of movement. The accordion inside stabilizes the memory during transport and the tight confines of the box prevent any dancing around by the modules. This has been and always will be my favorite package for memory. I thank Crucial for keeping the cardboard box and look forward to many more kits showing up in this fashion. Of course, the other item of Crucial's package that differs from others is the fact that the modules are individually sealed in anti-static bags. This not only helps protect the memory further, it also provides 100% security from receiving a used 'new' set of memory. This is another huge bonus over a re-sealable plastic clam shell because with those, you really never know if you are the first purchaser of the memory.
As mentioned and seen, anti-static bags individually seal each module off from the outside world and any harm from static electricity. Once outside the anti-static bags we can see that these modules are identical looking to the previous DDR3 Ballistix kit we looked at over eight months ago now. Wow, it has been eight months already, how time flies when you're having fun…I digress. The un-mistakable orange of the Ballistix heat spreaders still looks great and the clip less design Crucial moved to back then is still used today and we couldn't be happier to see it.
Without clips along the top, the issue of the heat spreaders lifting off the ICs at the bottom of the modules is completely removed and a good solid contact can be found across the entire surface of the IC. The modules are also no thicker or taller than a standard memory module so installation will be a breeze and no fitting room issues should pop up, un-like when you go shopping with your girlfriend and you get caught looking at another female who is also trying on clothes. Admit it, we have all been there and faced the wrath of an upset woman because of it.
There really isn't much else to discuss so here are a couple more photos of these wonderful looking modules. For those esthetically proficient in their builds, finding a motherboard to match the orange of the heat spreaders is going to be impossible but at least everyone will know what memory you have in your system with just a simple glance. The use of the dark Levin PCBs continues with Crucial and after our last experience torturing the PC3-12800 Ballistix with the same PCB, we can safely say that there should be no concerns with the quality of the PCB. They have stood up to months of abuse and these modules will likely see the same if not more.
Our last photo before moving on to the specifications section is a macro shot of the contact between the heat sinks and the ICs underneath. We can see Crucial has stuck with the grey fiber thermal pad and that these modules, like all Micron D9 based 1GB sticks, are single sided. We will now go over the outlined specifications of these modules, their EPP2.0 profiles, and of course we will also have a look under the heat spreader to see which Micron IC we will be working with today.
Page 3 : Specifications & Features
There really isn't a whole lot that goes into the specifications of a kit of memory that affects us and most of which we have already seen on the modules stickers. What isn't shown on the sticker, however, is a few minor details that may be of importance to some including those in the overclocking world. Let's first just recap the outlined specifications of the Ballistix BL2KIT12864BE2009:
The one specification we want to discuss aside from the blazing fast DDR3-2000 (1000MHz) frequency is the fact that this memory is SLI ready. In reality, this has nothing to do with SLI which is the acronym depicting NVIDIA's multi GPU technology. Instead, this terminology is used to describe this kit as having EPP2.0 profiles. When paired with a 7 series NVIDIA chipset, such as the 790i Ultra-SLI that we will be using in today's review for testing, these modules have the ability to adjust the BIOS settings accordingly in order to have the memory run at the above outlined specifications…all by simply turning on EPP. Let's take a look at the SPD table of this kit and then we will test to make sure the EPP profile is working as advertised.
At first, we can see the SPD Ext. field is filled in with EPP2.0 which means that the system is correctly identifying the modules as having EPP profiles. But once we move down the CPU-Z tab, we don't actually see the EPP profile in the timings table. This we found quite interesting and at this point, we weren't sure we were going to be able to enable EPP and have the system automatically overclock the memory to be running at DDR3-2000. We then went back into the BIOS and with the defaults loaded, simply enabled EPP and saved changes with our fingers crossed.
Upon booting, it is quite clear that the system has absolutely no problem picking up the EPP profile and within a couple seconds, we were up and running at the specified frequency and timings. In addition to setting the memory un-linked and at 1000MHz as well as the timings, the vDIMM was also automatically adjusted to 1.90v in the BIOS and the SPP (north bridge) voltage was raised to 1.40v in order for the system to be stable with the higher memory frequency.
Essentially, the EPP2.0 profile makes these modules 100% plug & play. The key to all of this is the fact that the 790i chipset allows the setting of a large amount of memory ratios and dividers and allow this memory to run at DDR3-2000 without the CPU being overclocked a single MHz. In the above screenshot, our FSB is still 1333 and the E8400 used for testing is still at the default 3GHz. Of course, we can also manually adjust the FSB and memory ratio to get the memory running at spec. We are still curious about the lack of the EPP profile being listed in the timings table like XMP profiles are but clearly, it doesn't matter because the modules definitely have the correct EPP2.0 profiles programmed on them. Perhaps that is the fault of CPU-Z and not the SPD table. Now, let's take a peak under the hood of these fantastic looking modules to see what we get to play with.
Buried under the Ballistix etching we can make out that these modules are Micron D9GTS. This is a bit of a surprise because D9GTS has been out of production for some time we thought and Micron has basically replaced it in the market place with D9JNL which are cheaper to produce and tend to run higher frequencies with less voltage required, albeit at slightly looser timings than D9GTS. Either way, there have been many kits out in the wild sporting D9JNL ICs on the modules so obviously both ICs are being used on this memory and after discussing it with Crucial, there is no way to know which your kit is going to come with. Both ICs have bins that meet the specifications of this memory so both get used. Here is a link to the full specifications of D9GTS ICs from the Micron Inc. web site:
- Density: 1Gb
- Part Status: Production
- RohS: Yes
- Depth: 128Mb
- Width: x8
- Voltage: 1.5V
- Package: FBGA
- Pin Count: 86-ball
- Clock Rate: 667 MHz
- Cycle Time: 1.5ns
- Op. Temp.: 0C to +85C
- CL: CL = 9
- Data Rate: DDR3-1333
Rated for DDR3-1333 CL9 at 1.5v, these ICs obviously scale quite well with voltage because the jump from DDR3-1333 to DDR3-2000 with only a 0.04v in vDIMM is quite a substantial gap to fill. Obviously the ICs can handle it otherwise they wouldn't be soldered to the PCB.
Installation is going to be a formality due to the standard sized heat spreaders but we are still going to have a quick peek at how these modules install in a couple of motherboards.
Page 4 : Installation & Test Setup
Let's keep this brief because, well, because it need not be long and drawn out. We have mentioned a couple of times that the heat spreaders on the Ballistix line of memory add very little bulk to the modules and therefore installation will be quite simple and straight forward. Let's take a look at the modules in a P5E3-Premium as well as the test platform, the EVGA 790I Ultra-SLI.
First up is the ASUS P5E3-Premium and clearly there are no issues here. The modules fit in the orange slots just fine, can run side by side, and certainly don't come anywhere near the outstretched arms of the Thermalright Ultra-120. Even had the layout put the memory directly beside the CPU socket, this memory has no problems hiding under the shade of the Ultra-120s branches.
Our 790i install photos are more setup photos to be honest but we have already seen that the memory isn't going to play any roles in installation in any setups so these are more just for the sake of saying we took installation photos in the 790i. It is clear that the Ballistix heat sinks won't be playing havoc with anyone, especially when the hold down on my phase change evaporator would actually clear the modules as shown in the second photo above.
Here now is a complete list of the hardware that will be used for the testing and benchmarks:
- CPU: Intel C2D E8400
- CPU Cooling: Chilly1 Single Stage Phase Change
- MB: EVGA 790i Ultra-SLI (P04/P05 BIOS)
- PWM Cooling: 8 x Swiftech MC-14s
- NB Cooling: Swiftech MCW30/PA 120.2/DD5
- SB Cooling: ASUS P5B-Dlx Southbridge heat sink
- GPU: HIS HD3870X2 1GB (Catalyst 8.2)
- PSU: Silverstone Zeus ST56ZF
- HD: 2 x Seagate SATAII 80GB 8MB NCQ
- OS: Windows XP-Pro SP2 (with all updates/hot fixes)
Crucial Ballistix 2x1024MB PC3-16000 (BL2KIT12864BE2009)
The two hard drives are not in a RAID configuration. Everything is loaded on the first drive, the second is simply for certain benchmarks to help alleviate hard drive reads and writes as a bottleneck. It won't completely eliminate it, a RAM drive would be required for that, but it will help a little.
Before we get to the overclocking, we want to do a quick stability test of the memory at the default EPP profiles. Just to make sure they at least run stock before we start pushing the memory.
As expected and hoped, this kit has absolutely no issues running stable at the default EPP profiles. Aside from enabling EPP, there were no adjustments to the motherboard for the above screen shot. It is all EPP at this point and it works absolutely wonderfully.
Page 5 : Overclocking
We now turn our attentions to the best part of the afternoon and that is the overclocking section. We are going to be doing things a little bit different today because not only are we going to show you what this memory was able to do with full stability on the 790i motherboard, we will also be throwing in a couple of the more 'fun' screen shots we were able to manage at less stable frequencies and timings. We have managed to do things with this memory that no other memory has done or at least been reported, and then we pushed it some more. First though, let's take a look at just what it is we consider stable.
The following is what constitutes stable for our testing purposes. Everything listed below has been run on each overclock listed or shown in the stability section. It is a bit of a lengthy stability test but it really does make sure the whole system is stable and that the overclocks shown are not just benchmark stable like the last section will be.
- Multiple loops of 3DMark 01 / 06 (30 minutes of looping the full tests each)
- Dual 32M runs of SuperPi Mod 1.5 (ran at the same time using HyperPi 0.99b)
- 2 hours of dual Prime95 using Prime95 v2.55 on blend mode
- 2 hours of dual MemTest Pro in Windows using 800MB/instance for the first instance and the second using all available memory
Like we said, a little more than a CPU-Z screenshot is required in order for a memory overclock to be considered stable around here. So without further adieu, I present to you the stability portion of our revamped overclocking section with the results at three timing sets of CL6, CL7, and CL8:
There really isn't a whole lot to say other than, WOW! The motherboard is really the limit in memory overclocking these days as it takes a good motherboard, or in my case, a good soldering iron and water cooling loop to get these types of clocks stable. With that said, this particular kit appears to be quite stellar itself and absolutely loves the black slots of the 790i reference board. CL7 clocking was a bit tough forcing us to loosen up to 7-7-7-21 for full stability for anything over 900MHz. At the top of the CL8 clocks there is plenty of headroom for less than Prime95 stability but the stress Prime95 Blend puts on the north bridge is massive at 1060MHz and anything higher is just impossible at this point.
The real star of the show though is the clocks at 6-6-5-18. You'll notice we went up to 2.1v with the CL6 clocks and that is because the memory just kept going and going with the more volts we gave it. The memory controller on the north bridge had an easier time with the clocks at this low a frequency so the memory was truly able to stretch its legs and it shows. DDR3-1680 at 6-6-5-18 1T fully stable is not something that just any kit of memory is going to do. Of course, we wouldn't recommend the memory be run at 2.1v for 24/7 operation without effective airflow and even then it is well above warranted voltage and potentially putting the memory in danger of damage. Below are the screenshots of the listed overclocks with details about secondary timings thanks to Memset in them for those that might be interested.
Now, on with the real show…as if that wasn't enough. The screen shots below are again, only stable in the benchmarks they have completed in the screen shots. The memory may or may not be more stable at the given frequencies. First up is a quick look at the highest frequency single 32M SPi we have so far ran with this kit at 8-7-6:
DDR3-2240 8-7-6-18 @ 2.15v 32M SPi
Going over 1100MHz with this kit on this setup is quite easy, getting 1120MHz stable for 32M didn't take too much work either. Really, these modules are really pushing the envelope as far as what a motherboard is capable of these days. It has been mentioned a couple times, the motherboard used in this review is modified and the modifications have helped get this memory to run faster. We are simply showing these for those into the benchmarking scene and have to warn that not everyone is going to be able to come even close to these results, there is a lot involved in making all of these last few screenshots possible.
DDR3-2016 7-6-5-18 @ 2.15v 32M SPi
We couldn't leave out some CL7 1000MHz action either and not to mention the second best 32M SPi for an E8400 in the world. The best time belonging to myself with another kit of memory which is only 2 seconds quicker. Needless to say, this sample of ours is up for pretty much anything we can throw at it.
DDR3-2260 8-7-6-18 @ 2.15v 1M SPi + 15K Read Bandwidth Everest
This is the sort of crowning achievement of this memory. Personally the 32M 1120MHz result is far more fantastic, but other folks have run that or gotten very close. 15K Read in Everest is not something that anyone else has done though so this result tends to get more attention. Everest bandwidth isn't really much of a benchmark but in the overclocking community, breaking this barrier has only happened once before…with this setup and another kit of memory, but the Ballistix are a very close second.
That is going to wrap up our overclocking section, hopefully you enjoyed reading through it as much as I have doing the benchmarks for it. Let's now get back to reality and see what some of these memory frequencies and timings do in benchmarks comparing the EPP2.0 settings to the stable overclocked settings.
Page 6 : Memory Benchmarks
We start off with our benchmarks with a look at a few bandwidth and memory latency benchmarks. For all of our benchmarks today, the versions are listed on each graph and the results are averaged to the same decimal point they are achieved in. So if the benchmark does not provide a score with a decimal place, the results will reflect this. All benchmarks are ran three times with UT3 being the exception as it is run ten times then averaged. Here is a chart of the various timing sets we used for the benchmarks and their settings.
With the NVIDIA 790i chipset allowing us the ability to adjust memory frequency independent of FSB, we are able to do a direct comparison between the stable overclocked settings. Of course we are limited to certain ratios still but we were able to get close to all of the timing sets maximum stable settings except for the 6-6-5 timing set which only allowed us to run at DDR3-1600 instead of closer to the stable DDR3-1680. Let's look at the results…
Lavalys Everest Ultimate 07 v4.20 – Memory Bandwidth
ScienceMark v2.00 – Memory Bandwidth
Our bandwidth results are basically exactly where we figured they would be. With Lavalys Everest, Write speed and thus Copy is primarily influenced by FSB. This is why our three results in the middle which are all at identical FSB settings result in almost identical Write speeds. The read results are what scales with memory frequency as it has in past DDR3 reviews. The EPP Copy and Write suffer because of this heavily meanwhile the Last result benefits tremendously from the higher FSB used. ScienceMark 2.0 confirms the Everest findings in pattern but produces slightly different numbers. This is simply due to the different way they calculate bandwidth.
Lavalys Everest Ultimate 07 v4.20 – Memory Latency
The latency results are almost identical to the bandwidth numbers as far a pattern goes. The middle three results are all tightly clustered together due to the same FSB frequency being used while the EPP and highly overclocked results at the top and bottom skew in opposite directions for obvious reasons.
Futuremark PCMark 05 – Memory Bench
With one exception, the PCMark 05 Memory Suite further replicates the previous programs. The one exception is the 6-6-5 results which takes a very substantial jump out in front of the other two timing sets at the same FSB. This is a bit of a curious result and something we will keep our eyes on going forward through the rest of the results.
Page 7 : System Benchmarks
The system benchmarks will rely on more the entire system and not just the memory sub-system. The memory will still have a substantial impact on results in some benchmarks but other factors are also going to play a role in how things shake out.
WinRAR 500MB Benchmark
We start off with the WinRAR benchmark up to 500MBs and this is where things start to get interesting. Remember how PCMark 05 picked up a bit of an advantage to the 6-6-5 timing set? Well look what WinRAR found as well in the difference between the three middle results. Now WinRAR relies heavily on the memory sub-system for performance and higher bandwidth has always resulted in better times on Intel chipsets. Apparently the 790i is showing a bit of an advantage to running at 1:2 versus some of the other ratios the navy and light grey results are running on. A 20 second win by the 6-6-5 results is extremely large and very indicative of an issue with the other two settings. Come to think of it, the Everest Read bandwidth results did seem kind of low. Perhaps the fact that the system is only at 400FSB is bottlenecking the memory from showing any further gains from higher memory frequency? As mentioned, it might just be the memory ratios playing games on us as well with different chipset latencies changing that we don't have control over.
Lame Front End – WAV => MP3 Encoding
The madness ensues with the Lame Front End audio encoding task. Not only do the middle results jump all over the place, but the last dark grey result is only a couple seconds faster than the 6-6-5 timing set which again, spanks the other two results at the same 3.6GHz CPU frequency. This tight battle between the light blue and dark grey might indicate the hard drives becoming the bottleneck because the CPU frequency of the dark grey result is through the roof compared to the rest, not to mention the FSB, and yet it barely eeks out a victory here. Lame encoding should scale with CPU frequency, which clearly isn't the case here.
SuperPi Mod v/1.5 – 1M
SuperPi Mod v/1.5 – 32M
All theories really get pressed hard with the SPI 32M results we are now looking at above. SPi 32M is one benchmark I know more about than any other and the influence memory should have on the results is pretty much exactly what we see above. Perhaps the gaps between the middle three timing sets should be a slight bit more but not a whole lot. So SPi 32M doesn't seem to have as big and advantage with the 6-6-5 timing set and the higher frequency of the middle results wins like it should. But as mentioned, the margin of victory should really be a few seconds more based on years of SPi benching experience.
x264 Benchmark HD
The x264 Benchmark is a new benchmark created by an enthusiast to measure a systems ability to encode HD content into x264. You can find the benchmark at Tech ARP. We were interested in seeing if memory frequency plays much if any role in the performance so we decided to throw it in this review for comparison. Obviously the results tend to lean toward the fact that memory doesn't seem to play much of a role in this particular benchmark. All three middle results are all but identical which can all but conclude that the x264 benchmark is isolated to the CPU, or at least not influenced by the memory sub-system.
Page 8 : 3D & Gaming Benchmarks
At this point, we simply don't know what to expect from the 3D and gaming results. The lack of performance from the 7-7-7 and 8-7-6 timing sets when compared to 6-6-5 at the same frequency and FSB is a bit perplexing as we have already discussed, so let's just look at the numbers and see if anything jumps out.
Futuremark 3DMark 01/03/05/06
There really isn't much to see here as far as comparisons are concerned which is kind of expected as 3DMark isn't going to show huge differences with memory adjustments. On a side note, I almost broke my personal best 3Dmark 01 score with the HD3870X2 and there were not even any tweaks involved. Not that it has anything to do with the review but I thought I would share. Let's move on to games where resolutions and detail levels were chosen based on what runs smoothly for gaming action. Focusing on the settings we actually play the game at makes a whole lot more sense to us than testing at super low resolutions that no one plays at anyway.
Gaming Benchmarks – Crysis
Crysis is clearly showing an advantage to the highest clocks and a dis-advantage to the lowest clocks but the difference in the middle three is somewhat erratic but at the same almost identical. The average frames per second for all three timing sets are within a single frame and the minimum frames per second are within just over a single frame so there is no single timing set that jumps out at us.
Gaming Benchmarks – UT3
UT3 on the other hand actually shows a bit of a pattern that we saw with the bandwidth results, except the 6-6-5 timing set almost look to be the overall winner here. Again, despite the larger gap in average frames per second, the minimum which is the most important of all, really shows no difference amongst the three timings sets.
Gaming Benchmarks – COD4
Last and actually least, is COD4, my favorite game. Talk about a dead heat race, and by dead heat we mean a complete neck and neck battle throughout the range at all three Min/Max/Avg. Clearly we are looking at a GPU bottleneck situation here so these results are all but useless in a memory review. COD4 will obviously get dropped from testing in memory reviews as memory, or system performance in general, plays little to no role at 1680×1050. Perhaps if an LCD upgrade happens and higher resolutions are ventured into we might give COD4 another try.
Page 9 : Conclusion
Well there you have it folks, Crucial once again knocks socks off and does it in style. The Ballistix line of memory has always been known to impress and rightfully they should. But what this kit of memory has done on the 790i test motherboard is just above and beyond what our expectations were…and our expectations were certainly a little bit lofty coming in. Needless to say, we weren't expecting to be able to benchmark at 1100MHz+ and certainly not at the little amount of voltage that was required to do so. Simply put, this memory out ran the motherboard and it wasn't even a close race.
This is the problem we will see going forward, motherboards with memory controllers onboard are going to be limiting memory clocks without question. The newest P45 chipset from Intel doesn't appear to alleviate this either and it is clear the 790i chipset is the best for clocking memory at this point, and it still isn't good enough. Of course there are some issues that come up with the 790i chipset like we saw in the benchmarks. Either the ratios aren't equally performing or low FSB speeds equated to low memory performance. In addition to the performance issues we found in certain situations, overclocking is held back by the motherboard as well at the high end of the frequency range.
Of course, these issues are all part of the motherboard and have nothing to do with the memory except that its full potential might not be filled. This is the sort of good news bad news situation we find ourselves in. The Crucial Ballistix memory we looked at today clearly is a beast. Our sample might have been at the higher end of the spectrum when it comes to overclocking but we would fully expect all samples of this memory to have some headroom above the already elevated stock settings of 1000MHz or at the very least, be able to run that frequency at tighter timings. In a nutshell, if you're looking for the ultimate in flexibility as far as memory frequency and timings go, the Ballistic PC3-16000 kit should offer that. From extremely high stock frequency of DDR3-2000 with loose timings or lower clocks with tighter timings, this kit has you covered. Just make sure you have a race track fast enough to handle this Ferrari of memory because it needs every inch of tarmac it can find.
- Incredible performance, even at stock settings
- Getting up and running at stock couldn't be easier on 790i motherboards thanks to excellent EPP profiles
- The Ballistix heat spreaders are great to look at, and even better to install in a system with a huge heat sink
- Very versatile across the timing range being able to handle anything from 6-6-5 to 8-7-6 and a nice range at each
- You are going to have to play for top binned memory
- Not guaranteed to be the coveted D9GTS by overclockers and benchmarkers
Overclockers Online would like to thank all of the Crucial/Lexar staff for helping make this review possible.