What's new
  • Please do not post any links until you have 3 posts as they will automatically be rejected to prevent SPAM. Many words are also blocked due to being used in SPAM Messages. Thanks!

OCZ Blade 3x2GB PC3-17000 CL8 Memory Review

Status
Not open for further replies.

3oh6

Well-known member
Joined
Mar 18, 2007
Messages
1,049
Location
Edmonton, AB
<center><img src="http://images.hardwarecanucks.com/image/3oh6/ocz/ocz_logo-1.png" alt="OCZ logo">


<b>OCZ Blade 3x2GB PC3-17000 CL8 Memory Review</b></center>



<b>Price:</b> $433 CND+<a href="http://www.sohodiffusion.com/produit.asp?num=12036">Sohodiffusion.com</a>
<b>Manufacturer Product Page:</b> <a href="http://www.ocztechnology.com/products/memory/ocz_ddr3_pc3_17000_blade_series_low_voltage_triple_channel">OCZ Technology Group Inc.</a>
<b>Manufacturer's Part Number:</b> OCZ3B2133LV6GK
<b>Warranty:</b> Limited Lifetime Warranty



<p style="text-align: justify;">For years memory manufacturers have gone to great lengths to increase the "number" used in marketing memory. Memory manufacturers introduced the DDR terminology that duplicates the actual operating frequency for a theoretical one that was derived simply to increase the "number" over SDRAM specifications. We then have the PC3-xxxxx terminology which again is nothing more than a theoretical "number" based on an equation of what a memory's potential bandwidth is, implemented for ease of marketing. Coincidentally enough, this "number" is even higher than the DDR rating of a particular memory module. It has always been about the "number" with memory, and advertising a higher "number" has been the goal. This isn't limited to the system memory segment of the market, does a 9800GT really need a full 1GB of memory? No, but it is a higher "number" that the marketing department can push.

No matter what "number" you use to describe the OCZ DDR3 PC3-17000 Blade Series Low Voltage Triple Channel memory, these modules are fast. Whether it be PC3-17000, DDR3-2133, or their actual operating frequency, 1067MHz; these modules are as fast as they get right now. Just over a month ago we presented the PC3-16000 OCZ Blade triple channel kit and shortly before publishing OCZ announced this PC3-17000 Blade triple channel memory kit. We thought our hands were full with the PC3-16000 kit, boy did we have a meal in front of us when we sat down at the table with these PC3-17000 modules. These modules are specified for the highest frequencies of any memory you can buy, but what good is that if the system can't run them at their rated operating frequency. We found with the PC3-16000 OCZ Blade memory - a full 67MHz slower operating frequency - that our i7 920 processor simply couldn't run them at DDR3-2000. This is obviously going to be a primary source of investigation for us with the yet faster PC3-17000 memory.

The next question is that of a performance nature. Is memory this fast even necessary for the average consumer? This is the question we have visited in past memory reviews and the consensus has always been a resounding no. Forget the fact that it will be tough running memory of this caliber at its rated frequencies, the benefits in a daily machine are somewhat limited if any at all. Where this kit shines is when it comes to benchmarking. Overclocking and benchmarking go hand in hand and with the advent of HWBot.org, international competitions, and even small forum ran overclocking competitions, benching specific hardware is quickly becoming a hot topic. This is more of a segment where this memory falls into place. It is this need where the "number" means everything. We plan to not only explore what we are getting for 24/7 use with the OCZ Blade PC3-17000, but also to explore their worth as a premier benchmarking kit of memory.</p><center><img src="http://images.hardwarecanucks.com/image/3oh6/ocz/ocz3b2133lv6gk/index-1.jpg" alt=""></center>
 

3oh6

Well-known member
Joined
Mar 18, 2007
Messages
1,049
Location
Edmonton, AB
Package & Memory Overview

Package & Memory Overview

<p style="text-align: justify;">With the formal introductions out of the way let's take a closer look at the modules and the package OCZ ships them out in.</p><center>
package-1.jpg
package-2.jpg
</center><p style="text-align: justify;">Like our last kit of OCZ Blade memory, the PC3-17000 Blades come in a package that consists of a thin cardboard outer shell encasing two molded plastic blister packs. The outer package doesn't have any identifying markers of the modules inside as it is a generic package used to keep costs down. Instead, the stickers on the modules are visible allowing us to identify the memory inside. The rear of the package has some basic marketing copy as well as ever so important customer support information.</p><center>
package-3.jpg
package-4.jpg
</center><p style="text-align: justify;">Two molded plastic packages secure the modules inside the outer shell. These are the same packages we have grown accustom to seeing memory presented in. When combined with the cardboard outer package, this method of shipping product should provide a very safe environment. The molded plastic shell holds the memory in place and the cardboard package holds the plastic shells together.</p><center>
modules-1.jpg
modules-2.jpg
modules-3.jpg
modules-4.jpg
</center><p style="text-align: justify;">OCZ includes a cardboard insert that outlines handling and installation procedures for memory. Most people know how to handle memory but still something that is nice to see in a package. The modules themselves are obviously no different to the PC3-16000 modules we looked at just over a month ago. The almost flat black exterior with it's granite like finish presents the OCZ Blade modules as a very rugged module. The front side as well as back of the modules have a number of groves and design elements that help increase surface area of the modules. The cooling "fins" long the top do not extend too far up and allow for installation in most setups with large CPU coolers. Another subtle item we noticed was the serial number has each module labeled -1, -2, and -3 in a matched kit. This actually can come in handy when keeping track of modules orientation in the motherboard DIMM slots. As all overclockers know, arrangement of the modules in the slots can lead to higher overclocks. And if you didn't know, now you know...swap the order of the modules when testing to find the optimal order.</p><center>
modules-5.jpg
modules-6.jpg
</center><p style="text-align: justify;">As mentioned, the OCZ Blade heat sinks have a number of design elements to increase surface area to aid in cooling the heat sink. These elements include a hollow upper portion of the module. There is one rather disappointing element we found with this kit of OCZ Blade modules though. If you look close in the first photo you can see the back heat sink slightly pulling away from the module.

In the second photo, it is quite obvious due to the change in viewing angle. Thanks to the light we can actually see that the bottom 1/3 of the IC isn't even in contact with the heat sink. At some point, it looks like someone tried prying up on this heat sink. Perhaps OCZ had to remove them at some point but either way, it should have been flattened before installation. The bend in the heat sink is obvious even when off the module. In the end, we simply removed the back heat sink and used a bit of force to flatten it out. While we had the heat sink off we took a quick couple photos.</p><center>
modules-7.jpg
modules-8.jpg
</center><p style="text-align: justify;">The thermal material OCZ uses is very sticky and should do a great job transferring heat. Even with the bent heat sink it was still a bit of a battle getting it off. Standard heat sink removal methods were invoked and it ended up coming off without any issue. "Standard heat sink removal methods" include liquid nitrogen and not something we recommend the end user doing. The second photo above shows the manufacturer markings of the PCB indicating an 8 layer PCB. We couldn't find an online source for this information, "but I know some people that know some people that robbed some people", and we were able to get the information from them.

During the course of the PC3-16000 OC Report, it was mentioned that some users were receiving OCZ Blade memory with black PCB's. Clearly ours here are green, as were our PC3-16000 OCZ Blade modules. We are not sure why some come with green and some with black PCB's, but we would definitely prefer to see black PCB's on all modules. Talk about a sexy memory module getting even sexier.</p>
 

3oh6

Well-known member
Joined
Mar 18, 2007
Messages
1,049
Location
Edmonton, AB
Specifications

Specifications

<img src="http://images.hardwarecanucks.com/image/3oh6/ocz/ocz3b2133lv6gk/specs-1.png" alt="" style="float: left; margin: 4px 7px 4px 0px"><p style="text-align: justify;">In previous memory reviews we haven't tried to disguise the fact that memory really doesn't have a lot of specifications to it, yet memory specification questions continually come up in forums. I think the fact that there is conflicting information about frequency/speed of memory is part of the problem. There are three ways to identify memory frequency/speed and getting them all straight is a bit confusing at first. Add in the timings and the confusion just multiplies.

With that said, seeing the specifications to the left of 2133MHz means this memory runs at DDR3-2133 (we won't go into a rant about speed naming scheme OCZ uses as we think our point was made in the Blade PC3-16000 review).<img src="http://images.hardwarecanucks.com/image/3oh6/ocz/ocz3b2133lv6gk/specs-2.jpg" alt="" style="float: right; margin: 4px 0 4px 7px"> At DDR3-2133 the memory is specified to run 8-9-8-24 timings at 1.65v, nothing confusing about that really. What DDR3-2133 means though is that this memory runs at astonishingly fast frequencies, so fast that many i7 processors will not even be able to run the memory at the specified clocks. We will discuss this in length in the Stability Testing section a little later on.

The image to the right is a photo of the now infamous IC's that run this particular kit of memory. Elpida J1108BASE-MNH-E, most recognizable as Elpida Hyper. The last couple weeks have been headline grabbing for Elpida Hyper due to the recent announcement by many memory manufacturers to cease production of modules with Elpida Hyper because of abnormally high failure rates in the field. OCZ hasn't seen the same issues as other manufacturers but has also decided to cease production of modules with Elpida Hyper for the time being. Here is the official statement from the OCZ support forums regarding this issue...

OCZ Support Forum - <i>"In regards to the potential issues relating to Elpida Hyper IC’s, OCZ has not yet seen extended failures in the field, and the Elpida Hyper IC’s we have used came from earlier batches so we do not yet know if this is the reason why we are not seeing field issues but as a proactive measure based on the latest information from media reviewers like AnandTech we have officially held the use of any Elpida Hyper IC’s at this time. Any customers that are seeing issues are encouraged to contact us and we will absolutely take care of all customers with a no questions asked refund or exchange for another OCZ product of equal value, whichever the customer prefers. We want to make sure that our customers have peace of mind and the confidence that we will take care of any and all issues related to modules that utilize Elpida Hyper chips."</i>

This obviously isn't a good situation for anyone involved but here at Hardware Canucks, we haven't had any issues with our kits of memory sporting Elpida Hyper IC's from either OCZ or Corsair. We have only three kits but they have seen extensive use and are still as good as the day they came out of the package. So there are two ways to look at it, either be scared of buying any remaining stock of memory with Elpida Hyper IC's because you are worried they are going to die. Or buy up all the kits you can find because they are going to be very rare and potentially a collector's item...not to mention the best benching memory out there. With that said, other manufacturers have already announced the re-introduction of Elpida Hyper based memory. So this "issue" looks to be corrected or at least on it's way to being corrected.

We will now move on to the Installation & Test Setup section but if you have any questions regarding "Hypergate", feel free to pop into our forums and let us know what is on your mind. We will do our best to provide accurate information and not speculation or sensationalized news.</p>
 

3oh6

Well-known member
Joined
Mar 18, 2007
Messages
1,049
Location
Edmonton, AB
Memory Installation & Test Setup

Memory Installation & Test Setup

<p style="text-align: justify;">Since we have already tested the Blade heat sinks for fitting in an EVGA X58 3X SLI motherboard, we figured we would use the EVGA X58 Classified for testing this kit with a Thermalright Ultra-120 eXtreme (TRUE).</p><center>
install-1.jpg
install-2.jpg
</center><p style="text-align: justify;">With the heat sink reaching out over the memory slots we can see it would be a tight fit for the first slot. Using the primary red slots, however, is just fine. There is plenty of room regardless of where we have fans. In the second photo we are showing that a memory module will fit in the first slot with the heat sink in this position, but look closely at the down turned fin of the TRUE. This fin barely touches the OCZ Blade heat sink and in order to install the module with the heat sink already in place, we have to pry this fin up just a little. It is possible to install the module with the TRUE mounted but it is very tight and the back of the Blade module did get scratched up a little.</p><center>
install-3.jpg
install-4.jpg
</center><p style="text-align: justify;">Turning the heat sink around to an East/West orientation leaves plenty of room for the memory, even with a 32mm fan installed. With no modules in the first slot we have lower the 120mm x 38mm fan down right to the heat sink screws so we get more air flow over the PWM heat sink. Obviously the fan would have to sit higher with a module in the first slot, or we could simply use a 120mm x 20mm fan.</p><center><table><tr><td>
memory-1.png
</td><td>
memory-2.png
</td><td>
memory-3.png
</td></tr></table></center><p style="text-align: justify;">The first part of our stability testing is going to discuss the fact that this memory doesn't exactly run at stock...ever. You see, there is no memory ratio that will allow the memory to run at the specified DDR3-2133. There are also no XMP profiles with these modules, like the PC3-16000 OCZ Blade modules. We mentioned with that kit that the variables involved to get DDR3-2000 to run stable were too many to trust static settings to be able to run the kit on various setups, that goes double for DDR3-2133. Needless to say, the memory boots up fine at BIOS defaults and the screen shots above show what the memory boots at with the BIOS cleared on the Classified. Let's now have a look at the complete setup we will be using for the testing, and overclocking of this memory.</p>


Test Setup

<center><table cellpadding="0" cellspacing="0" width="735px"><tr><td align="left" colspan="2">
setup-1.jpg
setup-2.jpg
</td><td align="right" colspan="2">
setup-3.jpg
setup-4.jpg
</td></tr></table><br /><table border="0" bgcolor="#666666" cellpadding="5" cellspacing="1" width="735px"><tr><td colspan="4"><b><font color="#ffffff">Test Platform:</font></b></td></tr><tr><td align="center" bgcolor="#cc9999" width="25%"><b>Memory:</b></td><td align="left" bgcolor="#ececec" width="75%"><b>OCZ Blade 3x2GB PC3-17000 8-9-8-24 (OCZ3B2133LV6GK)</b></td></tr><tr><td align="center" bgcolor="#cc9999" width="25%"><b>Motherboard:</b></td><td align="left" bgcolor="#ececec" width="75%">EVGA X58 Classified</td></tr><tr><td align="center" bgcolor="#cc9999" width="25%"><b>Processor:</b></td><td align="left" bgcolor="#ececec" width="75%">Intel Xeon W3540</td></tr><tr><td align="center" bgcolor="#cc9999" width="25%"><b>Processor Cooling:</b></td><td align="left" bgcolor="#ececec" width="75%">Thermalright Ultra-120 eXtreme-1366 CU<br>Scythe Ultra Kaze 120mm x 38mm 2000RPM 87.6CFM (DFS123812L-2000)<br>Adda 120 mm x 20mm 2050RPM 80.5CFM (AD1212MS-A73GL)</td></tr><tr><td align="center" bgcolor="#cc9999" width="25%"><b>Thermal Paste:</b></td><td align="left" bgcolor="#ececec" width="75%">Arctic Cooling MX-2</td></tr><tr><td align="center" bgcolor="#cc9999" width="25%"><b>North Bridge Cooling:</b></td><td align="left" bgcolor="#ececec" width="75%">Stock</td></tr><tr><td align="center" bgcolor="#cc9999" width="25%"><b>South Bridge Cooling:</b></td><td align="left" bgcolor="#ececec" width="75%">Stock</td></tr><tr><td align="center" bgcolor="#cc9999" width="25%"><b>PWM Cooling:</b></td><td align="left" bgcolor="#ececec" width="75%">Stock</td></tr><tr><td align="center" bgcolor="#cc9999" width="25%"><b>Power Supply:</b></td><td align="left" bgcolor="#ececec" width="75%">Corsair HX1000W</td></tr><tr><td align="center" bgcolor="#cc9999" width="25%"><b>Video Card:</b></td><td align="left" bgcolor="#ececec" width="75%">EVGA GTX295 (NVIDIA GeForce 186.18 WHQL)</td></tr><tr><td align="center" bgcolor="#cc9999" width="25%"><b>Additional Fans:</b></td><td align="left" bgcolor="#ececec" width="75%">Scythe Ultra Kaze 120MM 2000RPM 87.6CFM (DFS123812L-2000)</td></tr><tr><td align="center" bgcolor="#cc9999" width="25%"><b>Hard Drives:</b></td><td align="left" bgcolor="#ececec" width="75%">Seagate 7200.9 80GB SATAII 8MB cache</td></tr><tr><td align="center" bgcolor="#cc9999" width="25%"><b>OS:</b></td><td align="left" bgcolor="#ececec" width="75%">Windows Vista x64 SP1 (with all updates)</td></tr><tr><td align="center" bgcolor="#cc9999" width="25%"><b>Ambient Temperature:</b></td><td align="left" bgcolor="#ececec" width="75%">23C ~ 25C</td></tr></table></center><p style="text-align: justify;">Our test setup has changed a little bit, but it wasn't our choice. We have mentioned a couple times that not all processors are going to be able to handle running memory at DDR3-2133. This is because of the on-die memory controller of the i7 processor. We definitely ran into this issue during testing, but not only that, we had a motherboard tragically pass during this process as well. We will go over the details next in the Stability Testing section.</p>
 

3oh6

Well-known member
Joined
Mar 18, 2007
Messages
1,049
Location
Edmonton, AB
Stability Testing & Overclocking

Stability Testing & Overclocking

<p style="text-align: justify;">For new readers to our memory reviews here at Hardware Canucks, this is our little spiel on memory stability. There are many different views on testing stability and we don't claim ours to be the best, we are simply very open and forward with what we think constitutes stable:</p><b>Stability Testing Methodology</b><p style="text-align: justify;"><i>Memory stability, what constitutes stable? What is not considered stable? These questions get hotly debated in enthusiast forums all over the internet like little brush fires on the fringe of an inferno that play havoc with forest fire crews. Everyone has their own opinion about stability, especially when it comes to memory stability. For some, stable means they can do whatever it is on their computer without it crashing, blue-screening, or restarting; whether that means gaming or just surfing the internet. To this user, stable means simply using the computer as they normally would.</p><center><img src="http://images.hardwarecanucks.com/image/3oh6/ocz/ocz3b2133lv6gk/setup-5.jpg" alt="Corsair Dominator 3x2GB PC3-12800 C8"></center><p style="text-align: justify;">Of course, that is not what we would call stable and we do a lot more thorough testing before labeling a memory frequency/timing stable. One of the toughest types of programs on a system has been found to be distributed computing projects such as Rosetta@Home, Folding@Home, World Community Grid, and more. Running 24/7 crunching for one of these great causes is a sure way to find holes in a system if there is truly some instability, unfortunately it takes a considerable amount of time to use them for stability testing so we use the list of programs below to all but guarantee the system to be 24/7 distributed computing stable</i>:</p><p style="text-align: justify;">As mentioned, there are many schools of thought on stability testing, and that is ours. We certainly won't be presenting CPU-Z screen shots and saying the memory passed our benchmarks so we call it stable. Each overclock we are about to look at has passed a rigorous stability testing procedure. Just keep in mind these results are from a single sample of memory. Like some many things in life, when it comes to memory, there are no guarantees.</p>

<b>Specification Stability Testing</b>
<p style="text-align: justify;">Let's be clear right from the start, there is nothing stock about running memory at DDR3-2133. Intel doesn't specify their processors to be able to run DDR3-2133, motherboards aren't specified to run DDR3-2133, the only component specified to run DDR3-2133 is the OCZ Blade PC3-17000 memory. Unfortunately, the memory is just 1/3 of the hardware needed to run memory. The motherboard plays a role and the most important component in this equation is the CPU. Yes, we were able to run the memory at the specified frequency and timings, with 1.65v...but only one of the three processors we tried could, and in the process a motherboard died trying to make another processor.</p>Click for full size screenshot...
<center>
stable_spec-1.png
</center><p style="text-align: justify;">The screen shot above is our successful processor mounted in the EVGA X58 Classified. The Intel Xeon W3540 was the only processor capable of accomplishing the task of running our stability testing at the specified DDR3-2133 with 8-9-8-24 timings. Our incredible clocking i7 920 can not even run DDR3-2000 so it didn't stand a chance with these sticks, and when trying our i7 975 processor in the EVGA X58 3X SLI, we had to turn VTT up so high that the stock cooling couldn't handle it and we blew the VTT PWM on that board. Needless to say, DDR3-2133 isn't going to be 24/7 stable for a lot of users, we would go as far to say many users.

The stress on the memory controller, which is located on the CPU, is so great at DDR3-2133 that many CPU's will need either too high a VTT or simply won't be able to run 24/7. The reason for this insane amount of stress isn't just the memory frequency, but mostly due to the uncore frequency that needs to run to reach DDR3-2133. The uncore runs at a minimum 2x the memory frequency. At DDR3-2133, that means an extremely high uncore clock of 4266MHz is required. Getting a CPU that runs 4266MHz at ambient temperatures for 24/7 stability is going to likely require at least 1.50v VTT if not more, even if it is an exceptional CPU.

This is why you will see substantially higher memory overclocks with this memory in our Extreme Overclocking section. When running a single threaded benchmark or one that isn't near as demanding on the memory or the memory controller, these modules will run as high as you want them to pretty much. The below overclocks are going to be our 24/7 overclock results that pass our full stability testing methodology, and are 100% limited by the CPU.</p>


<b>Stability Overclocking:</b><p style="text-align: justify;">If you have been reading the review up to this point, you will already understand that these overclocks are a product of our best CPU's memory controller, not so much the memory. Without a great CPU with a great memory controller on it, we wouldn't be able to run specified settings, let alone overclocked ones. Because of the limitations in memory clocking at this level by the CPU, we have to ask that you take the results with a grain of salt. No amount of volts or tweaking will accomplish these results with many i7 processors.</p><center>
oc-1.png
</center><p style="text-align: justify;">The results above are really quite an impressive set of stable results. Keep in mind, these aren't simple benchmark stable, these are full 24/7 stable overclocks. The 6-7-6 results are about the only ones that aren't being limited by the CPU and would scale with further vDIMM.

The 7-7-6 and 7-8-7 results are so close because the IMC finds both timing sets to be very similar both being CL7. Opening up CAS Latency to CL8 allowed us to clock the PC3-17000 OCZ Blade's up to their specified settings, but we simply couldn't get a single base clock higher stable. At 1070MHz, regardless of voltage Prime Blend would just gas out after a few minutes. So even though we were able to run the specified clocks and timings of this memory, we couldn't get a sliver more. Again indicating that we were very fortunate to get to the specified operating clocks of DDR3-2133. Below are the screen shots from the other three timing sets.</p><center><table align="center" bgcolor="#666666" cellpadding="10" cellspacing="1" width="90%"><tr><td align="center" valign="top" bgcolor="#ececec" width="33%">6-7-6 Stable Overclock
click for full size...
oc-2.png
<br>OCZ Blade @ 927 6-7-6-18 1T<br>@ 1.754v under load</td><td align="center" valign="top" bgcolor="#ececec" width="33%">7-7-6 Stable Overclock
click for full size...
oc-3.png
<br>OCZ Blade @ 1035 7-7-6-20 1T<br>@ 1.754v under load</td><td align="center" valign="top" bgcolor="#ececec" width="33%">7-8-7 Stable Overclock
click for full size...
oc-4.png
<br>OCZ Blade @ 1047 7-8-7-20 1T<br>@ 1.754v vDIMM under load</td></tr></table></center>
 

3oh6

Well-known member
Joined
Mar 18, 2007
Messages
1,049
Location
Edmonton, AB
Benchmark Methodology

Benchmark Methodology

<p style="text-align: justify;">As has been mentioned a couple times, there isn't really anything remotely related to stock settings with these modules.<img src="http://images.hardwarecanucks.com/image/3oh6/ocz/ocz3b2133lv6gk/method-1.png" alt="Just a sample graph outlining where the results will be coming from in the up-coming benchmarks" style="float: right; margin: 4px 0px 4px 7px"> We have also all but ran out of ideas for comparisons between various clocks resulting in question answering results. These two factors have resulted in us deciding to simply bench the exact setups that we found to be 24/7 stable in the previous Stability Testing & Overclocking section.

The results identified with the <b>blue</b> bars in the chart will be that of our 6-7-6-18 overclock of DDR3-1854 with the CPU at 3556MHz. The two <b>grey</b> bars will represent our CL7 overclocks of 7-7-6-20 at DDR3-2070 with the CPU at 3972MHz and 7-8-7-20 at DDR3-2094 with the CPU at 4016MHz. The last result in the charts will be the <b>red</b> bar that represents our closest match to the specified operating frequency of DDR3-2130 at timings of 8-9-8-24 alongside a CPU clock of 4085MHz.

Despite not really thinking the results would be interesting, our top three overclocks all have similar clocks on both the CPU and memory, that it might result in interesting results due to the different timing sets. The last of our methodology consists of how we setup the OS for benchmarking and the run the individual’s benches:
  1. Windows Vista x64 w/SP1 is installed using a full format
  2. Intel Chipset drivers and accessory hardware drivers (audio, network, GPU) are installed followed by a defragment and a reboot
  3. At time of benchmarks the latest drivers were downloaded from their official web sites as the latest drivers, most notable, NVIDIA GeForce Release 186.18 WHQL
  4. Programs and games are then installed followed by another defragment
  5. Windows updates are then completed installing all available updates followed by a defragment
  6. Benchmarks are each ran three times after a clean reboot for every iteration of the benchmark unless otherwise stated, the results are then averaged
</p>
 

3oh6

Well-known member
Joined
Mar 18, 2007
Messages
1,049
Location
Edmonton, AB
Memory Benchmarks

Memory Benchmarks



Everest Ultimate v5.02<p style="text-align: justify;"><i>Everest Ultimate is the most useful tool for any and all benchmarkers or overclockers. With the ability to read most voltage, temperature, and fan sensors on almost every motherboard available, Everest provides the ability to customize the outputs in a number of forms for display on your desktop. In addition to this, the memory benchmarking provides a useful tool of measuring the changes to your memory sub-system when tweaking to measure the differences. Unfortunately with the i7 processors, the results aren't always consistent and we can receive variations as much as 1000MB/s at any given time. Because of this we use multiple runs and drop any of the "high" scores from the averages.</i></p><center><img src="http://images.hardwarecanucks.com/image/3oh6/ocz/ocz3b2133lv6gk/mem_bench-1.png" alt=""></center><p style="text-align: justify;">Well that about sums up the difference in bandwidth between these setups, or should we say, lack of difference. With the exception of the 6-7-6 timing set, all of the setups are darn near equal. Variations in runs crossed over each other and these three setups are very similar. The biggest reason the 6-7-6 timing set has fallen behind is simply the CPU clocks are much lower than the other three.</p><center><img src="http://images.hardwarecanucks.com/image/3oh6/ocz/ocz3b2133lv6gk/mem_bench-2.png" alt=""></center><p style="text-align: justify;">Latency is a slightly different story but again, all results are within the same ballpark. The 8-9-8 timing set is actually a little bit slower as far as latency goes compared to the two CL7 setups. Looks like the memory bandwidth just was high enough to make up for the looser timings.</p>

SiSoft Sandra 2009.SP2<p style="text-align: justify;"><i>SiSoft Sandra is a popular and well used benchmark in the industry but not really a friend of serious benchmarkers. The results SiSoft Sandra produces have been suspect at times basing the numbers it comes up with on system specs and not actual testing. The latest version of Sandra seems to be one of the few programs that appear to calculate memory bandwidth consistently so we decided to include it in today’s benchmarks. Like we have always said with SiSoft Sandra though, take these results for what they are and nothing more.</i></p><center><img src="http://images.hardwarecanucks.com/image/3oh6/ocz/ocz3b2133lv6gk/mem_bench-3.png" alt=""></center><p style="text-align: justify;">Strangely enough, Sandra agrees with Everest. My word, what is the world coming to? Well, in all fairness to Everest, Sandra results look more "scripted" than Everest so I think Everest doesn't have to worry about being grouped in with Sandra just yet.</p><center><img src="http://images.hardwarecanucks.com/image/3oh6/ocz/ocz3b2133lv6gk/mem_bench-4.png" alt=""></center><p style="text-align: justify;">Like the Everest latency results, the Sandra latency picks up a slight hiccup in the 8-9-8 timing set indicating a lack of frequency difference between it and the CL7 setups to make up for the looser timings.</p>

ScienceMark v2<p style="text-align: justify;"><i>ScienceMark is an almost ancient benchmark utility at this point in time and hasn't seen an update in a long time. It is, however, still a favorite for accurately calculating bandwidth on even the newest chipsets.</i></p><center><img src="http://images.hardwarecanucks.com/image/3oh6/ocz/ocz3b2133lv6gk/mem_bench-5.png" alt=""></center><p style="text-align: justify;">And finally we can close the case on this one as Sciencemark backs up both the Everest and Sandra results with a virtual tie between the bottom three result sets. A mere five base clock separates the 7-7-6, 7-8-7, and 8-9-8 results and the bandwidth numbers show that with near identical numbers.</p>
 

3oh6

Well-known member
Joined
Mar 18, 2007
Messages
1,049
Location
Edmonton, AB
System Benchmarks

System Benchmarks



SuperPi Mod v1.5<p style="text-align: justify;"><i>When running the 32M benchmark of SPi, we are calculating Pi to 32 million digits and timing the process. Obviously more CPU power helps in this intense calculation, but the memory sub-system also plays an important role, as does the operating system. SPi 32M has been a favorite amongst benchmarks for these very reasons and is admittedly the favorite benchmark of this reviewer.</i></p><center><img src="http://images.hardwarecanucks.com/image/3oh6/ocz/ocz3b2133lv6gk/sys_bench-1.png" alt=""></center><p style="text-align: justify;">What there lacked in a gap in memory bandwidth doesn't lack in SuperPi 32M. The 8-9-8 timing set pulls out a nice advantage by a solid 9 seconds over the two CL7 results. At these CPU clocks, that is rather substantial, further confirming that memory frequency owns 32M over timings...unless of course you can run the same frequency with tighter timings. Another factor is the higher uncore and CPU so the difference between the three timing sets isn't only from memory.</p>

PCMark Vantage<p style="text-align: justify;"><i>The latest iteration of the popular system benchmark is PCMark Vantage from the Futuremark crew. The PCMark series has always been a great way to either test specific areas of a system or to get a general over view of how your system is performing. For our results, we simply run the memory benchmark suite which involves a wide range of tests on primarily associated with media management and testing.</i></p><center><img src="http://images.hardwarecanucks.com/image/3oh6/ocz/ocz3b2133lv6gk/sys_bench-2.png" alt=""></center><p style="text-align: justify;">PCMark showed a little more of a gap than we expected giving the 8-9-8 timing set a solid 4~5% higher score than the 7-8-7 timing set. What we also see is that the 7-7-6 timing set upset the 7-8-7 timing set with the 6-7-6 timing set extremely close behind.</p>

DivX Converter v7.1<p style="text-align: justify;"><i>Next up is a real life benchmark where we simply time a common task done on the computer. Encoding DVDs for viewing on the computer or other devices is an increasingly important task that the personal computer has taken on. We will take a VOB rip of the movie Office Space, and convert it into DivX using the default 720P setting of the new DivX converter v7.1.</i></p><center><img src="http://images.hardwarecanucks.com/image/3oh6/ocz/ocz3b2133lv6gk/sys_bench-3.png" alt=""></center><p style="text-align: justify;">Moving away from the synthetics we can see that the real world results are pretty equally matched to the synthetic results we just looked at. DivX shows a 26 second gain going from the fastest CL7 result to the 8-9-8 timing set and this is primarily going to be influenced by the CPU clocks. On a 22 minute calculation, 26 seconds is a mere 2% gain in performance.</p>

Lame Front End<p style="text-align: justify;"><i>Un-like the DivX conversion we just looked at, Lame Front End is not multi-threaded and only utilizes a single core of a processor. This will obviously limit performance but we should still recognize significant time savings going from the stock settings to the overclocked results. We will be encoding a WAV rip of the Blackalicious album, Blazing Arrow and converting it to MP3 using the VBR 0 quality preset.</i></p><center><img src="http://images.hardwarecanucks.com/image/3oh6/ocz/ocz3b2133lv6gk/sys_bench-4.png" alt=""></center><p style="text-align: justify;">Lame Front End shows identical gaps between the results percentage wise from top to bottom. These benchmarks are really showing that bandwidth results can be a good indicator of performance in the real world. These are about the same patterns in results we saw in the first benchmarking section right through this entire system suite of benchmarks.</p>

Photoshop CS4<p style="text-align: justify;"><i>Adobe Photoshop CS4 is fully x64 compliant and ready and able to use every single CPU cycle our processor has available including the implementation of GPU support utilizing the GTX 280 in our test system. It is just a shame it can't fully utilize all 8 threads of the i7 processor yet. We have changed our Photoshop benchmark to more of a standardized test configured by DriverHeaven.net. Their Photoshop benchmark utilizes 15 filters and effects on an uncompressed 109MB .JPG image that will test not only the CPU but also the memory subsystem of our test bench. Each portion of the benchmark is timed and added together for a final time that is compared below.</i></p><center><img src="http://images.hardwarecanucks.com/image/3oh6/ocz/ocz3b2133lv6gk/sys_bench-5.png" alt=""></center><p style="text-align: justify;">The DriversHeaven Photoshop benchmark puts out somewhat erratic results at times but on the whole, the numbers fell in line for what we have seen so far. The slightly higher CPU clock and memory clock, regardless of memory timings, out performs the other setups.</p>

WinRAR 3.90 Beta 4<p style="text-align: justify;"><i>We all know what WinRAR is and does. It is a compression and decompression tool that has a built in benchmark, a way to tell just how fast a system can do this programs given task. We simply run the benchmark up to 500MB processed and time how long it takes.</i></p><center><img src="http://images.hardwarecanucks.com/image/3oh6/ocz/ocz3b2133lv6gk/sys_bench-6.png" alt=""></center><p style="text-align: justify;">The last of our real world tests is WinRAR and the newest beta doesn't seem to much care about the rest of the benchmarks as it puts the bottom three results in a virtual tie. There is no way to mistake these results, they all perform the same.

We will now have a quick look at the gaming benchmarks before moving on to the real exciting benchmarks.</p>
 

3oh6

Well-known member
Joined
Mar 18, 2007
Messages
1,049
Location
Edmonton, AB
3D/Gaming Benchmarks

3D/Gaming Benchmarks



Futuremark 3DMark Vantage / 06<p style="text-align: justify;"><i>We have forced ourselves to step up to 3DMark Vantage results for all reviews because the public demands it. 3DMark Vantage is the newest in a long line of 3D benchmarking software from Futuremark and is the most elaborate to date. Featuring multiple presets for various system configurations, Vantage is the culmination of all 3DMarks past relying on system and GPU power for its results. We will stick to the Performance preset as it seems to be the most popular at this point in time. 3DMark 06 is the previous iteration of this successful 3D benchmark suite.</i></p><center><img src="http://images.hardwarecanucks.com/image/3oh6/ocz/ocz3b2133lv6gk/3d_bench-1.png" alt=""></center><p style="text-align: justify;">The numbers here are pretty similar to what we have seen up to this point. Memory timings typically don't have a huge effect on 3DMark and we feel the slight CPU increase is the primary factor in the score differences but the memory and uncore play a role in this as well. The results in both Vantage and 06 are pretty much exactly what we were expecting to see at these clocks.</p>

Crysis - Sphere benchmark<p style="text-align: justify;"><i>We all know what Crysis is and how much it beats up systems but we wanted to add it to the gaming benchmarks to see how system changes can improve performance on a mid-level system. Detail levels are all set to Very High with the resolution at 1680x1050 with no AA. We ran the benchmarks with a demo of the Sphere level in DX9 and 64-bit. The game looks great with this setup and plays just well enough to keep us happy.</i></p><center><img src="http://images.hardwarecanucks.com/image/3oh6/ocz/ocz3b2133lv6gk/3d_bench-2.png" alt=""></center>

FarCry 2<p style="text-align: justify;"><i>Another fall release of this past silly season Far Cry 2 has some beautiful scenery but does lack that buttery smooth game play in places. A lot of moaning and groaning has occurred with FarCry 2 but acceptable frame rates are much easier to achieve than Crysis and the game play is plenty smooth enough to enjoy. We were really able to crank up the settings with this benchmark on this setup.</i></p><center><img src="http://images.hardwarecanucks.com/image/3oh6/ocz/ocz3b2133lv6gk/3d_bench-3.png" alt=""></center>

Left 4 Dead<p style="text-align: justify;"><i>The newest game in our testing sweet, Left 4 Dead was just added after we were asked to include a Source powered game in our memory benchmarks. Being based on the Source engine, there is definitely a chance that system performance will heavily influence the results. We used FRAPs to measure frame per second on a custom time demo of the rooftop level.</i></p><center><img src="http://images.hardwarecanucks.com/image/3oh6/ocz/ocz3b2133lv6gk/3d_bench-4.png" alt=""></center><p style="text-align: justify;">Well, let's keep this short and sweet...the smallest influence memory is going to have in your day to day life is going to be gaming. With the i7 platform, the system is just so strong and memory bandwidth is already saturating the CPU with data that increasing your memory frequency or lowering your memory timings are going to have very little effect on FPS. The fact of the matter is that GPU power is what determines gaming frames per second, not system power...again, once you are up to the level of system that an i7 rig is.</p>
 

3oh6

Well-known member
Joined
Mar 18, 2007
Messages
1,049
Location
Edmonton, AB
Extreme Overclocking & Benching

Extreme Overclocking & Benching


<center><table cellpadding="0" cellspacing="0" width="735px"><tr><td align="left" colspan="2">
xsetup-1.jpg
xsetup-2.jpg
</td><td align="right" colspan="2">
xsetup-3.jpg
xsetup-4.jpg
</td></tr></table><br /><table border="0" bgcolor="#666666" cellpadding="5" cellspacing="1" width="735px"><tr><td colspan="4"><b><font color="#ffffff">Test Platform:</font></b></td></tr><tr><td align="center" bgcolor="#cc9999" width="25%"><b>Memory:</b></td><td align="left" bgcolor="#ececec" width="75%"><b>OCZ Blade 3x2GB PC3-17000 8-9-8-24 (OCZ3B2133LV6GK)</b></td></tr><tr><td align="center" bgcolor="#cc9999" width="25%"><b>Motherboard:</b></td><td align="left" bgcolor="#ececec" width="75%">EVGA X58 3X SLI Classified</td></tr><tr><td align="center" bgcolor="#cc9999" width="25%"><b>Processor:</b></td><td align="left" bgcolor="#ececec" width="75%">Intel Xeon W3540 (3845B010)</td></tr><tr><td align="center" bgcolor="#cc9999" width="25%"><b>Processor Cooling:</b></td><td align="left" bgcolor="#ececec" width="75%">Chilly1 single stage phase change<br>MMouse Rev 3 CU Pot<br>w/Liquid Nitrogen</td></tr><tr><td align="center" bgcolor="#cc9999" width="25%"><b>Thermal Paste:</b></td><td align="left" bgcolor="#ececec" width="75%">Arctic Silver Ceramique</td></tr><tr><td align="center" bgcolor="#cc9999" width="25%"><b>North Bridge Cooling:</b></td><td align="left" bgcolor="#ececec" width="75%">Stock</td></tr><tr><td align="center" bgcolor="#cc9999" width="25%"><b>South Bridge Cooling:</b></td><td align="left" bgcolor="#ececec" width="75%">Stock</td></tr><tr><td align="center" bgcolor="#cc9999" width="25%"><b>PWM Cooling:</b></td><td align="left" bgcolor="#ececec" width="75%">Stock</td></tr><tr><td align="center" bgcolor="#cc9999" width="25%"><b>Power Supply:</b></td><td align="left" bgcolor="#ececec" width="75%">Corsair HX1000W</td></tr><tr><td align="center" bgcolor="#cc9999" width="25%"><b>Video Card:</b></td><td align="left" bgcolor="#ececec" width="75%">ATI PCI Mach64<br>Gigabyte GTX 260 (NVIDIA GeForce 180.48 WHQL)</td></tr><tr><td align="center" bgcolor="#cc9999" width="25%"><b>Video Card Cooling:</b></td><td align="left" bgcolor="#ececec" width="75%">Stock<br>Kingpincooling.com Tek9 4.0 Slim<br>w/Liquid Nitrogen</td></tr><tr><td align="center" bgcolor="#cc9999" width="25%"><b>Additional Fans:</b></td><td align="left" bgcolor="#ececec" width="75%">Scythe Ultra Kaze 120MM 3000RPM 133.6CFM (DFS123812H-3000)</td></tr><tr><td align="center" bgcolor="#cc9999" width="25%"><b>Hard Drives:</b></td><td align="left" bgcolor="#ececec" width="75%">Seagate 7200.9 80GB SATAII 8MB cache</td></tr><tr><td align="center" bgcolor="#cc9999" width="25%"><b>OS:</b></td><td align="left" bgcolor="#ececec" width="75%">Windows Vista SP1 / Windows XP Pro SP2</td></tr><tr><td align="center" bgcolor="#cc9999" width="25%"><b>Ambient Temperature:</b></td><td align="left" bgcolor="#ececec" width="75%">23C ~ 25C</td></tr></table></center><p style="text-align: justify;">Up to this point we have only been concerned with what this memory will do for the average consumer. Full stability testing, a lot of talk about being able to run this memory at spec, and so on and so forth. Well that all stops right here, this is where we talk about what this memory can really do. Show what this memory was meant to do, and trust me when I say, it has nothing to do with stability or the average consumer. I mentioned in the PC3-16000 review that those OCZ Blade sticks were becoming my favorite modules for benching, that needs to be adjusted because these PC3-17000 modules are just a hair better. We start this section off with a brief look at the OC Report found in the HardwareCanucks.com Forums posted a short time ago. Here is the chart of results from that single 32M SPi OC Report:

<center><img src="http://images.hardwarecanucks.com/image/3oh6/ocz/ocz3b2133lv6gk/xocss-4.png" alt=""></center><p style="text-align: justify;">From 6-7-6 up to 7-8-7, the results are remarkably similar to the PC3-16000 OCZ Blade OC Report at every voltage step of the way. What we added to this OC Report was the 8-9-8 timings and that is where things just get out of hand. With the new Intel Xeon W3540 processor and its much stronger memory controller, we were really able to push this memory in single 32M stability falling just a hair short of 1200MHz...that is DDR3-2400. And to accomplish this at only 1.75v vDIMM means this memory still has more in it once we can find an even better CPU with a stronger memory controller.</p><center><table align="center" bgcolor="#666666" cellpadding="10" cellspacing="1" width="90%"><tr><td align="center" valign="top" bgcolor="#ececec" width="50%">Single 32M @ 6-7-6
click for full size...
xocss-8.png
</td><td align="center" valign="top" bgcolor="#ececec" width="50%">Single 32M @ 7-7-6
click for full size...
xocss-9.png
</td></tr><tr><td align="center" valign="top" bgcolor="#ececec" width="50%">Single 32M @ 7-8-7
click for full size...
xocss-10.png
</td><td align="center" valign="top" bgcolor="#ececec" width="50%">Single 32M @ 8-9-8
click for full size...
xocss-11.png
</td></tr></table></center><p style="text-align: justify;">The screen shots just above are from the OC Report with the best result from the four timing sets tested with the exception of the 6-7-6 screen shot. Since I was so close to 1000MHz 6-7-6, I bumped voltage up another notch and went for it, that didn't even make the OC Report. The rest of the screen shots can be seen in the OC Report itself.

The last little piece of the OC Report puzzle was the 1200MHz validation. Not many have achieved it so it was something I just had to grab while I had the clocks that high, here it is...DDR3-2440 validated:</p><center> </center><p style="text-align: justify;">Needless to say, these modules are ready for the big time, so let's have some fun with them. Up next, a little 2D action including SPi 1M and 32M with the Xeon W3540 under LN2 cooling.</p><center><table align="center" bgcolor="#666666" cellpadding="10" cellspacing="1" width="90%"><tr><td align="center" valign="top" bgcolor="#ececec" width="50%">SuperPi 1M
click for full size...
xocss-1.png
<br>OCZ Blade @ 1182 8-9-8-24 1T</td><td align="center" valign="top" bgcolor="#ececec" width="50%">SuperPi 32M
click for full size...
xocss-2.png
<br>OCZ Blade @ 1172 8-9-8-24 1T<br>Top 15 Globally @ HWBot.org</td></tr></table></center><p style="text-align: justify;">That 32M time up there is a new personal best for me and by a healthy margin. That cleared my previous PB by 5 seconds. Running memory at 1172MHz definitely helps achieve crazy runs like that. I believe it is the fastest 32M SPi at below 5.4GHz in the world, or at least that I have seen. This is some serious memory and an equally serious CPU, and I am just getting started with the pair.</p><center><img src="http://images.hardwarecanucks.com/image/3oh6/ocz/ocz3b2133lv6gk/xsetup-9.jpg" alt=""></center><p style="text-align: justify;">The last of our benchmarks today are going to involve the help of a liquid nitrogen cooled Gigabyte GTX 260 that will run anywhere from 900MHz to 1080MHz on the core, depending on the benchmark. We have a couple results from 3DMark 01 as well as a very impressive result in Aquamark 3 and 3DMark 05.</p><center><table align="center" bgcolor="#666666" cellpadding="10" cellspacing="1" width="90%"><tr><td align="center" valign="top" bgcolor="#ececec" width="50%">3DMark 01 - Single GTX 260
click for full size...
xocss-5.png
<br>OCZ Blade @ 1095 7-9-7-20 1T</td><td align="center" valign="top" bgcolor="#ececec" width="50%">3DMark 01 - Single GTX 260
click for full size...
xocss-6.png
<br>OCZ Blade @ 1162 8-9-8-24 1T</td></tr><tr><td align="center" valign="top" bgcolor="#ececec" width="50%">Aquamark 3 - Single GTX 260
click for full size...
xocss-7.png
<br>OCZ Blade @ 1172 8-9-8-24 1T</td><td align="center" valign="top" bgcolor="#ececec" width="50%">3DMark 05 - Single GTX 260
click for full size...
xocss-12.png
<br>OCZ Blade @ 1132 8-9-8-24 1T</td></tr></table></center><p style="text-align: justify;">The first 01 result above was actually a Low Clock Challenge over at Overclockaholics.com that I entered and thanks to the PC3-17000 OCZ Blades, I won the single card part of the competition. This score was heavily influenced by the ability to run the high memory clocks that I did. The second 01 score is simply the same setup with the CPU maxed out. The AM3 and 05 results are amongst the top two in the single GTX 260 class over at HWBot.org and the AM3 score in particular is definitely aided by the crazy 8-9-8 clocks this memory can achieve.

In total we have had 5 or 6 really good sessions with these sticks thus far and they have been everything we had hoped they would be. The versatility to run super high clocks or uber tight timings allows us to tailor our memory for what each benchmark likes best...all without having to have multiple kits of memory on hand. There may be a problem with some Elpida Hyper, but for benchers, it is worth the time trying to find a kit that doesn't have issues like these ones. We have beaten on them from day one and they have done nothing but ask for more.</p>
 
Last edited by a moderator:
Status
Not open for further replies.

Latest posts

Twitter

Top