Mushkin Ridgeback 998826 6GB 1600Mhz 6-8-6-24 Review

[Lsdmeasap]

Mushkin Ridgeback 998826 6GB 1600Mhz 6-8-6-24 Review





Special thanks to the great team at Mushkin Memory for providing this review sample.


Before we get started looking at the new Ridgebacks, let's learn a little bit about Mushkin.


If you have ever wondered how Mushkin should be pronounced, this is what Mushkin's Wiki Page has to say.

Although some people pronounce Mushkin /ˈmʌʃkɪn/ "mush-kin", the proper pronunciation is /ˈmʊʃkɪn/ "moosh-kin" (rhymes with "push").

Mushkin - Wikipedia, the free encyclopedia


Mushkin has come to be a widely known brand among enthusiasts, and has been around for a very long time, believe it or not. I can remember purchasing Mushkin RDRAM (Rambus DRAM) in 98-99, it is very likely that many of you have never seen or heard of this type of memory. Although Mushkin was around before then, that's a good example of how long Mushkin has been providing quality memory to the masses, and how long I have been using their products.

Mushkin was established in 1994 by Bill Mushkin, purchased in 2000 by Ramtron International, and then finally in 2005 George Stathakis bought the company, and became the new owner and president. Soon after purchasing the company, he sold the stock to employees, and Mushkin is now a fully employee-owned company.

Mushkin has been located in Denver, Colorado for the past 14 years, but has recently relocated its base to Inverness, the suburban tech center south of Denver in Englewood, Colorado.

They are well known for providing "Enhanced" memory, but they also manufacturer "Value" ram, memory specifically designed for Apple/MAC systems, various server and laptop memory modules, a small lineup of modular Power Supply Unit's (PSU's), and have even ventured into the graphics card market. Mushkin has also recently ventured into the SSD market, releasing an Indilinux-based lineup - the Io SSD drives, and the more recently released Sandforce-based Callisto Deluxe line of solid state disks.

Their memory products are available in several categories and levels of performance, from standard to extreme. These memory kits are labeled from ES/EM (Essential or Value Series) for standard usage, HP (High Performance) for mild to moderate overclocking, and XP (eXtreme Performance) designed for overclocking and manufactured with some of their best IC Chips for tight timings at high clock speeds. Finally, the Redline Series, which is the cream of the crop when it comes to Mushkin's memory lineup. The Redline series is designed with the very best IC's Mushkin has to offer, which means you get the very best timings, superb bandwidth, and low latency when using these modules. They are intended for the extreme overclocker in all of us.

Mushkin has made a few revisions to the heat spreader and naming scheme recently. The above mentioned heat spreaders and/or names are still available and used to an extent, but the naming scheme and some of the heat spreaders are being phased out. These changes also add the new Silverline series and heat spreader, taking the place of the ES/EM series. The Blackline name has replaced HP labeled kits with not only the new name, but also the blue heatspreader is being replaced the with black frostbyte type. The Redline lineup remains unchanged at this moment, although some Redline modules are also now available in Ridgeback or Ascent heat spreaders as well. An example of this is today's Ridgeback review sample, it is the same kit (inner) as the 998805 Redline Frostbyte.





Heat spreaders and heatsinks are all the fuss these days, from huge CPU heatsinks, massive dual or triple fanned GPU heatsinks, multiple motherboard component heatpipes and heatsinks, and finally the many various types of memory heat spreaders. Memory heat spreaders today come in a large variety of styles and forms, from the thin simple coverage type to the large heatpipe add-ons and finned styles. The Ridgebacks fall in between the various types combining great looks and superb cooling. They are based on a finned style spreader, with several differences that I am sure the Mushkin design team tested and retested until they found the ultimate medium for the best of both worlds - aesthetics and cooling capability.

The new Ridgeback heat spreader is designed in thick black porous metal, which we'll see better in an upcoming image, providing more surface area to help aid in removing heat from the memory itself. They have a 10.5mm raised fin in the middle of each heat spreader, positioned at the top of the module, between each side of the heat spreader. The fins themselves are also porous, and are formed into Mushkin's logo < which runs along the top portion of the memory modules. Recessed into the middle of each set of fins is again the < Mushkin logo, this time in colored print. This creates a very cool look, while at the same time allowing everyone to know you are rocking the world's best memory - Mushkin.


This is a 3x2GB kit rated @ 1600Mhz (PC3-12800) 1.65V using 6-8-6-24 timings, further info can be found at Mushkin's product page:
Mushkin Ridgeback 998826 Product Page



The modules and packaging:




Mushkin packages this kit in a plastic clamshell, creatively designed in green and black with the Mushkin Logo symbol coming up from the bottom in a "Ridgeback-like" fashion. The package is very sturdy and the modules are securely held in place inside by snapping in from each side. On the reverse, quietly designed in green and white, there is a bit of information about Mushkin at the top, and 5-step install directions for the memory itself in the middle.



Up close and personal with the set you can see how the Ridgeback's design is a little bit taller than your average module. The "Ridge" part of the heat spreader measures in at 10.5mm, resting atop the 30mm main part of the heat spreader. In comparison, a Mushkin Frostbyte heat spreader measures in at only 33MM total, with 8mm of that being above the upper PCB Edge. The total size of the module from the gold lead insert edge to the top of the Ridge measures 46mm or 1-13/16", compared to the 38mm total size of a Frosbyte module. This size is considerably larger than most modules, yet not as obtrusive as some, so be sure to do your homework and measure your clearances before making a purchase.



Here is a closer look at the single module from both sides. I used different lighting as you can see, in an attempt to better show the Ridgeback's heat spreader design surface. Note the edge of the surface is not smooth, but rather rough or porous. This design allows for better and quicker transfer of heat to the surrounding air, which keeps the modules running cool at all times.




Ahh the Ridgeback, this heat spreader creates a very interesting look, and it is a nice change over other rather plainly finned heat spreaders. It's cool that Mushkin decided to forge the fins into their logo symbol, instead of using a plain fin design. They did a great job, and the final product looks very sharp and cools well at the same time.




These are a few up close images of the PCB itself, for anyone who may be interested. The PCB is marked with K0-8177 markings, and is an 8-layer PCB that delivers a stronger voltage to the IC's and other components of the memory, cuts down the risk of resistance changes and/or interference meaning the signal integrity is kept clean and electrical noise is minimized. At the same time, this also helps keep the module from overheating. The IC's have been purposefully obfuscated to keep in line with Mushkin's IC policy. They do not discuss which IC's are used in their modules for multiple reasons. If you are not sure which IC's are used, and would like to know, feel free to contact me.

[Lsdmeasap]

And finally, here we have several images of how the modules look when installed on a motherboard. The Mushkin Ridgeback modules installed on the GA-X58A-UD7 test board look very sleek, from any angle you can see the Mushkin logo and the awesome looks of the Ridgeback design. These would look fantastic on any motherboard!




Testing

Hardware used in tests:
GIGABYTE GA-X58A-UD7 1366 Motherboard
Intel i7-965 EX ES Quad-Core CPU
Intel i7-980X EX ES Hexa-Core CPU
Mushkin Ridgeback 998826 6GB 1600Mhz (PC3-12800) 6-8-6-24 Triple Channel DDR3 Memory Kit
Western Digital 1TB Caviar Black WD1001FALS
GIGABYTE 5870 SOC GV-R587SO-1GD 1GB Graphics Card
Ultra X3 ULT40312 850W Power Supply


The Mushkin Ridgebacks come pre-programmed with XMP settings for use in non-overclocking situations, here are the SPD/XMP programmed settings.




It is very difficult to keep the CPU speed at the same range for all tests, but I did try to keep things within the closest range possible 3.95-4.24Ghz. Uncore, as you will read more about below, is also hard to keep the same frequency at times, I did so when possible. I kept the memory voltage within 1.6-1.74 actual, depending on what timings and speeds were used. Something to take note of with these modules is that you should not always assume more voltage is needing when pushing them and running into errors. There can be times that a higher voltage will hurt your overclocks, and in actuality a much lower voltage than you would expect turns out to be what you really needed instead.

The following are the CPU and memory speeds for each test, and my testing result images. Images are presented in the same order as the text list from L to R, top to bottom


965 EX:






3.99Ghz - 1480Mhz - 5-7-5-18
4.19Ghz - 1600Mhz - 6-8-6-24 @ Spec
4.14Ghz - 1713Mhz - 6-8-6-24 @ Spec
4.07Ghz - 1812Mhz - 6-9-6-24
4.03Ghz - 1920Mhz - 7-9-7-24
4.17Ghz - 2000Mhz - 7-10-7-24



980X EX:






4.24Ghz - 1573Mhz - 5-8-5-18
4.20Ghz - 1600Mhz - 6-8-6-20
4.20Ghz - 2000Mhz - 7-10-7-24
3.95Ghz - 2049Mhz - 7-10-7-24
4.00Ghz - 2240Mhz - 8-11-8-26
4.10Ghz - 2300Mhz - 9-12-9-29


Programs used for testing:

Each set of speeds and timings in the graphed results below were tested with the following programs/methods and considered to be stable.

Memtest86+ 4.15b2



This is a common DOS based memory test, and should be a part of every overclocker's arsenal. Tests using this program were ran from 50 passes per set, or up to 100 passes, depending on the time I had each day for testing. An hour or so using test #5 is generally considered stable in my book, and is about 50 passes when using test #5 with 6GB, which is what I use for memory testing in reviews. Tests #5 and #7 are the more complex tests, and thus create the most heat, making it easier to find if there is instability.

Memtest86+ - Advanced Memory Diagnostic Tool


Memtest HCI Design 3.6 (Used older version due to the smaller window size)



This is a great little program, unlike Memtest86+, this program can be run within windows and in multiple instances. My testing methods with this program vary, depending on time allotted, memory amounts used in each instance, and how many instances are used. Normally, I use 3-6 instances with 600MB or more to 1024MB, depending on memory free at the time, and again the time I have to let the tests run. Passing 200-300% with zero errors, using 3-4GB testing total, usually takes about an hour. If more time is available, I sometimes test until 500-750% or above. If you are stable past 200% using 3-4GB, you are normally good to go.

MemTest: RAM reliability tester for Windows

Everest Ultimate V. 5.50.2100



This also is a great program, everyone should have a copy. It is not so much a testing tool for our purposes, but it provides multiple stability tests, various bandwidth test, and in-depth system hardware information. For this review it will only be used to provide simple bandwidth and latency test results at each frequency.

Computer Diagnostics & Network Audits Software | Lavalys.com

RightMark Memory Analyzer V. 3.80



Same applies here as above, a great tool for quick bandwidth and latency tests, but also offers other information and tests.

RightMark Memory Analyzer. Products. CPU Rightmark


MaxxMEM² Preview V. 1.82



Similar to the above two programs, this tool provides memory bandwidth and latency tests.
MaxxPI² - Download (MaxxMEM² - PreView)


Winrar Benchmark tool (V. 3.80 used)



This test compresses data, but is affected by other hardware, and will be used as a general comparison in the results below. At the following link you can find more information about this tool, one way to use it, what hardware it tests and what affects it.

WinRAR Compression Test
WinRAR download and support. WinRAR is a powerful Windows tool to compress and decompress zip, rar and many other formats: Download now

[Lsdmeasap]

And finally

LinX 0.6.4



This is THE hot one, this tests and measures (In GFlops) your CPU and memory floating point computing power using Intel's LinPack Binaries with linear algebra. Many consider this program to be a standard of stability, however, often you may see my advice against the use of it in long or large period runs. This program is very hard on systems, and may overheat the CPU, or push it into extremes it would not otherwise see, and can cause unintended permanent damage or degradation. My advice on this program is to use it sparingly, as you will see I only use 10x passes for this review. From time to time I will run 20 passes, but that would be few and far between as there are other less harmful ways to test for stability.

LinX - A simple Linpack interface - XtremeSystems Forums
Intel® Math Kernel Library



Other thoughts before getting to the main results:

I decided to use 2 different processors for this review because of the bandwidth differences, overall memory speeds, and various performances allotted between the two. This did increase the time it took to complete this review, due to the multiple ranges of tests I ran. The use of two processors was necessary due to the vast differences in the way each CPU affects memory as a whole.

While the 980X is a great CPU for many reasons, it would only be fair to the memory itself, and to you my readers, if I also performed testing using a 965 CPU. The 980X does have some huge advantages over the previous generation i7's (920-975), but this comes at a cost, which is part of the reason I choose to test using both types. The quad core i7's have a jump on the 980X in memory bandwidth performance, depending on settings used this can range from 4% to 15% or more, but that is where it all ends for these CPU's. Putting memory bandwidth aside, the 980X can make massive leaps in raw Mhz over the quads, due to the uncore allotments.

The gigantic frequencies the 980X can achieve are staggering in comparison. I was able to reach an outstanding 2634Mhz @ 9-12-11-30 using these Ridgebacks and a 980X CPU, this was on a water cooled setup (CPU + NB/SB only) with no additional cooling on the memory. Voltages used to reach these speeds were not excessive, 1.43 QPI/Vtt and 1.72-1.74 memory voltage.

Highest validation obtained was at 2634Mhz triple channel using 9-12-11-30 timings.



CPU-Z Validator 3.1


Let's talk about Uncore -

What is uncore and why does it matter?

Uncore is a term used by Intel to describe the part of a microprocessor that is not the core. The core contains the components of the processor involved in executing instructions, including the arithmetic logic unit, floating point unit, L1 and L2 cache. Uncore functions might include L3 cache, the on-die memory controller, and other bus controllers such as PCI Express.

Uncore - Wikipedia, the free encyclopedia


I have thought long and hard about this section, and decided it must be a part of this review. My recent experience with 980X has shed a new light on memory overclocking and how uncore is a huge part of how far memory overclocking can be taken, and how performance and efficiency are greatly affected by varying uncore speeds.

980X's uncore design allowances blow the usual X58 memory overclocking experience out of the water, allowing the ability to use 1.5-x times memory multiplier, in comparison to the previous generation of i7's having to be at least 2 times the memory multiplier. Specifications of a memory kit are no longer required When using a 980X CPU, the specifications that a manufacturer labels per kit no longer apply. The new uncore changes in the 980X allow for a vast amount of possibilities with any kit of memory from 1333Mhz to 2000Mhz+ - specifications can now be disregarded.

Now, the ability to use less uncore on an X58 setup, compared to previous generation CPU's allows for a drastic increase in raw frequency possibilities. Previously, if you owned a decent CPU, your uncore (under Non-Extreme conditions) would be limited to somewhere in the 4Ghz to 4.4Ghz range, depending on the processor itself, the motherboard used, and of course the skills you possess. This is where memory overclocking previously halted, when you ran into your uncore's upper limit, that meant you could no longer push the memory speed any higher without causing your system to crash due to the uncore failing. Now, with i7-980X the general range of uncore's speeds remains the same (again depending), but now being able to take advantage of a lower uncore speed at any given time is where things change dramatically. Lower uncore speeds mean you can now push the memory speed much higher, while still keeping the uncore under the given limit for your CPU. This opens a whole new level of memory overclocking possibilities, you can now push much farther - passing 2000Mhz with ease. You are no longer hitting the uncore wall in that 4-4.4Ghz range, because you can now use a lower uncore multiplier, which means you will end up with an uncore value less than your CPU's limit.


These new uncore changes make room for huge leaps in memory overclocking, and the possibilities are seemingly endless. I have been testing this kit for a little over a month, and while I was working to get this review published, I found myself each day thinking, "What about testing this way, or with these timings or speeds". Needless to say the possibilities with this kit and a 980X CPU could keep you occupied for months trying to find the sweet spot of a low overclock, or pushing the envelope and tweaking a super high overclock into a most rewarding state of stability.

Due to how uncore affects bandwidth this is also something you need to keep an eye on when comparing your benchmark results with others, or with your own results. I ran a few quick tests to compare this and show you how memory bandwidth and related scores in benchmarks can vary depending on the uncore used.

In all of these tests the Uncore was the only setting changed. The following is the memory speeds and timings in these tests, for further info please see the result images provided.

Uncore Comparison:


1100Mhz 5-5-5-15 (If only my BIOS allowed CAS 4, I think these could do 4-5-4 or 4-6-4 around 1200-1300Mhz, that would be nice to see on DDR3)





1312Mhz 5-6-5-15





1573Mhz 5-8-5-20





1600Mhz 6-8-6-24





2049Mhz 7-10-7-24





2350Mhz 9-11-9-29




MaxxMEM² Memory Score




Super PI 1.5 XS




Everest Latency




Everest Ultimate Bandwidth

1100-1300Mhz




1573-1600Mhz




2049-2350Mhz

[Lsdmeasap]

Benchmark and Testing Results:


965 EX Results:


Everest Ultimate V. 5.50.2100 Bandwidth




MaxxMEM², Everest Ultimate, & RightMark Latency




MaxxMEM² Preview V. 1.82 Memory Score




RightMark Memory Analyzer V. 3.80

Real Bandwidth Read-Write (Synthetic Test)




Real Bandwidth Maximal (Performance Test)




Winrar Benchmark Tool V. 3.80





980X EX Results:


Everest Ultimate V. 5.50.2100 Bandwidth




MaxxMEM², Everest Ultimate, & RightMark Latency




MaxxMEM² Preview V. 1.82 Memory Score




RightMark Memory Analyzer V. 3.80

Real Bandwidth Read-Write (Synthetic Test)




Real Bandwidth Maximal (Performance Test)




Winrar Benchmark Tool V. 3.80





tRCD - The heart of these modules



tRCD is where everything happens with these modules. CAS timing isn't really the setting that needs changed when you run into a wall overclocking, it might help you to an extent, or in certain instances, the main thing to tackle with these modules is the tRCD Value. The default value of the stock specifications is CAS +2 = tRCD, and this is generally what will be required. However, you can use more such as CAS +3 or +4, and this is where things begin to open up. If you raise the tRCD to +3 or +4, you can generally leave the CAS value at a much lower than expected setting, allowing much higher overclocks while still having a low CAS Value. Using less tRCD than default gets a little bit tricky, but it is possible once you obtain a good grasp of how the modules work. It is also possible to use a tRCD value equal to CAS, but this is only easy to do at a higher CAS value such as around your given tRCD limit for any range of speeds. For example, if you find you can use tRCD 9 at 1900Mhz you can probably also use CAS 9 there as well, resulting in 9-9-x timings. This is degradative to performance compared to using 6-9-x or 7-9-x, so there is no real need to do so.

Knowing how tRCD affects everything with the Ridgebacks, and building a clear understanding of how things work together can take some time to learn. I've learned plenty while testing for this review, yet I also know there is still more knowledge to be gained. The Ridgebacks are quite different compared to much of the memory you may have used in the past, so it might take you a while to learn how they work, but once you do - the rewards will be worth the time.

While a higher tRCD value does affect overall performance, using a higher value does not hinder or hurt performance too much, and it's not as bad as having to increase to a higher CAS Value. Some users feel a higher tRCD kills performance, and in some instances this might, but with these modules the CAS Value is already low so an increase in tRCD isn't as bad as some may believe. Yes, tRCD equal to CAS, or one above may provide better results, but at timings such as 7-8-7-20 Vs. 6-8-6-20 things tend to be fairly even across the board due to the lower CAS 6 in this example.

Once you learn how to use the Ridgebacks, what they can/can't do, you will find this memory just as easy to use as any other.


Cooling - How does the Ridgeback heat spreader stand against the competition?

Well, I did mean Mushkin's own competition, the Frostbyte heat spreader





This is a general test, nothing scientific or anything like that. I planned to purchase a IR Thermometer to use in conjunction with my thermal probes, but never did get around to picking one up. Due to this these results were taken with my digital thermometer that I use for extreme benching.

Ridgeback Results:

Idle



Load




Frostbyte Results:

Idle



Load



I performed these tests using the middle module in a triple channel setup, in an open air case, providing the most heat for this test. Using thermal paste I affixed the probe onto the side of one module in the middle, then applied 3 layers of electrical tape over that, and placed a small clamp directly holding the probe in place tightly against the heat spreader. This is a good indicator of how well the Ridgeback heat spreader deals with heat. While these tests are not overly scientific, they provide ample data for valid results.

The Ridgeback heat spreader keeps the memory running very cool as you can see above. I noted the results with the temperatures not changing from 1.712 to 1.76 at load with either heat spreader, and I did re-test ending with the same results. I'm unsure if this means that this is the upper end of the heat range with either type of heat spreader, or if when more voltage is applied temps would again differ, but I ended my results there and did not use higher voltages to avoid causing damage to the modules.



Final Thoughts



The Ridgeback memory kit is simply amazing, each day tweaking and testing proved to be more astonishing than the last. While these modules do take some getting used to, once you get the hang of what can and cannot be done - the possibilities are mind blowing! Coming in at an impressively cool price of just over $200, these kits are the same as Redline #998805 and are very much worth the reasonable price. Spending the extra few dollars to enhance your system with a set of Ridgebacks will be well worth it, and you will not be disappointed!

With great looks, the new Ridgeback design, and the price Mushkin has a winner here, I would be happy to suggest these to any user!

Pros:
1. Superior Looks
2. Great Overclockability
3. Very Flexible Timings-wise
4. Ridgeback's Great Cooling Properties

Cons:
I have none, except to say that a custom colored black PCB would be great


Price of this kit as of 8/5/2010 is $212 directly from Mushkin
998826 - 6GB (3x2GB) DDR3 PC3-12800 6-8-6-24 Ridgeback (Triple Kit)



Special thanks to everyone at Mushkin Memory and Gigabyte for making this review possible!

[pr@$r1g]

gr8 review of 2300mhz of 1600

[mak1skav]

Very nice review with so many tests and informations, you put so much effort in the review. Thanks for your time.

[Lsdmeasap]

gr8 review of 2300mhz of 1600

Thanks pr@$r1g!

Very nice review with so many tests and informations, you put so much effort in the review. Thanks for your time.

Thanks, I appreciate your feedback! I know, so much info, I actually left out some of my testing results too so it wouldn't be too much.

I'm glad you enjoyed it, and thanks again!

[Calathea]

I have just briefly read this review. I must say, very nice work Lsdmeasap! For sure a lot of interesting data, Mhz vs timings

[Lsdmeasap]

Thanks for your comments Calathea, it's always great to hear what others think.

I hope you find the time to read it all and enjoy once finished!

[0xdeadbeef]

Great review! Thanks for all info.

[Lsdmeasap]

Great review! Thanks for all info.

Thanks, I'm glad you like it!

Took me a long time to get it all together!