Corsair, G.Skill, Mushkin, and Patriot X79 Quad-Channel Memory Reviewccokeman -
» Discuss this article (15)
To find out just what kind of performance these kits of memory will deliver, I will be running them through a series of benchmarks to see how they compare. 16 GB and 32 GB kits that range in speed from 1600 MHz to 2400 MHz will be tested at their native speeds as well as overclocked as far as the test platform and CPU will allow. Overclocking on the Intel's X79 series platform is much like that on P67 and Z68, with strict multiplier and bclock limits. However, the option to use the gear stepping of the Socket 2011 platform allows for some interesting combinations. The test setup used for this evaluation is listed below. Turbo Boost has been disabled to eliminate an uncontrolled clock speed increase that may skew the results. The CPU will be run at the default clock speed for the baseline testing and then bumped as close to 4.5 GHz as possible for overclocked testing. The operating system is Windows 7 Pro 64-bit with all current patches installed at the time of testing. The graphics driver is AMD Catalyst 11.12.
- Processors: Core i7 3960X @ 3.4 GHz and 4.5 GHz
- Motherboard: ASUS Rampage IV Extreme
- CPU Cooling: Corsair Hydro Series H100
- Memory: Mushkin, G.Skill, Corsair, Patriot
- Video Card: XFX Radeon HD6970
- Power Supply: Corsair AX1200
- Hard Drive: 1 x Seagate 1 TB SATA
- Optical Drive: Lite-On Blu-Ray
- Case: Corsair Graphite Series 600T
- OS: Windows 7 Professional 64-bit
- Comparison Module #1: Mushkin Redline (4 x 4 GB) 993997 9-11-10-28 2133 MHz 1.65 V
- Comparison Module #2: Corsair Vengeance LP (4 x 4 GB) CML16GX3M4X1600C8 8-8-8-24 1600 MHz 1.5 V
- Comparison Module #3: G.Skill RipjawsZ (4 x 4 GB) F3-17000CL9D-8GBXLQ 9-11-10-28 2133 GHz 1.65 V
- Comparison Module #4: Corsair (8 x 2 GB) CMGTX8 PC3-19200 10-12-10-27 2400 GHz 1.5 V
- Comparison Module #5: Patriot Division 4 (4 x 4 GB) PXQ316G1600LLQK 8-9-8-24 1600 MHz 1.65 V
- Comparison Module #6: Corsair Vengeance (4 x 4 GB) CMZ16GX3M4X1866C9R 9-10-9-27 1866 MHz 1.5 V
CPU-Z: This application shows us the settings that we have chosen in the BIOS. Items shown in this application include CPU speed and bus settings, motherboard manufacturer, BIOS revisions, memory timings, and SPD chip information.
Task Manager: We use this utility to show physical memory, kernel memory, page file, and processor usage.
Corsair Overclocked settings:
- Processor: Intel Core i7 3960X
- Memory: Corsair Vengeance LP 32 GB 9-10-9-27 @ 1898 MHz
- Memory: Corsair Vengeance 32 GB 9-11-10-28 @ 2016 MHz
- Memory: Corsair Dominator GT 16 GB 10-12-10-27 @ 2400 MHz
Overclocking the Corsair modules resulted in a totally different experience for each, with requiring varying approaches to getting the end result. Of course, the strength of the memory controller is also going to determine the ultimate speed potential on this platform. With the 32 GB set of Vengeance LP (Low Profile) modules, I had to work with a base clock speed of 1600 MHz using 1.5 V and the stock 8-8-8-24 latencies. The highest clock speeds I could reach without totally blowing the sub-timings out of the water was a prime95 stable 1898 MHz by increasing the voltage to 1.6 V, bumping the memory multiplier to the 1866 MHz mark, increasing the CPU bclock, and loosening the primary timings to 9-11-10-28. This will still yield satisfactory performance for everything short of hardcore benchmarking. In any case, it is an increase of almost 300 MHz over the baseline 1600 MHz rated speed, which offers increased bandwidth for higher performance. When the voltages were pushed, the modules did heat up due to how closely they were packed when in 8 DIMM configuration. Airflow helps, but the low profile heat sink is undoubtedly not as efficient in this configuration as the full size Vengeance design.
With the full size Vengeance modules, I expected a very similar overclocking experience, but it took a different route to get to my result. Starting out with bumping the memory multiplier divider to 2133 MHz resulted in a failed boot attempt and a BIOS recovery. Bumping the voltage up or loosening the sub-timings did not result in a successful boot at 2133 MHz either. On original Sandy Bridge systems, the only solution left would be increasing the bclock frequency. With their limited bclock tweaking ability though, only a small bump would be possible due to the maximum bclock potential of about 106 MHz on air with most CPUs. However on Sandy Bridge Extreme, Intel pulled some tricks out of their sleeve by allowing the use of Gear Ratios for the CPU bclock. By choosing the correct memory multiplier and 125 MHz bclock strap, I was able to tweak the timings (TRCD & TRP) and voltages (1.6 V) to reach just over 2000 MHz while running at 9-11-10-28 1T. This effort was worth only 150 MHz over the stock 1866 MHz clock speed, but nonetheless, offers a slight bump in performance for the time and effort. Compared to the low profile modules, the cooling performance was like night and day. The standard taller fin design is easily more efficient than the low profile design in all cases.
The Dominator-GT modules offered the largest overclocking challenge in the entire comparison set of modules. In its defense, just getting to 2400 MHz with 8 slots populated is a feat in itself – populating 4 slots makes it a little easier on the memory controller. What I found, after beating on these modules for ages, was that my memory controller was just not up to the task of getting them over the rated 2400 MHz by loosening primary or sub-timings, or increasing voltages on the memory controller, CPU, and modules themselves. Dropping down to four DIMMs had the same result of being incapable of bumping over 2400 MHz. With that done, I tried to tighten the timings a little to see if this would result in a higher level of performance. Again, I was up against a brick wall with these modules. Seeing as how the memory controller was good for 2400 MHz, I was expecting a little more, but it seems the high hopes I had were more limited by the memory controller than the modules themselves. Even without overclocking, the modules still deliver excellent performance when combined with a stock or overclocked CPU. Even when packed in an 8 DIMM configuration, these modules never warmed up whilst cooled by a single fan. The DHX+ cooling solution just works incredibly well, even with up to 1.7 V pushing through the modules.
G.Skill Overclocked settings:
- Processor: Intel Core i7 3960X
- Memory: G.Skill RipjawsZ 16 GB 10-12-11-28 @ 2414 MHz
The last G.Skill memory I looked at did quite well in the overclocking department and thankfully, this kit does not deviate from that path – the base speed of 2133 MHz was just the starting point for the kit. Making the jump to 2400 MHz, though, required some tweaking of the primary latencies and voltages. CAS latency was bumped to 10 with the tRCD bumped to 12 and the voltage to 1.67 V. The memory controller voltage was fine at 1.05 V with this configuration as seen by the long term (well, 7 hours at least) stability testing of the overclock. The higher speed, coupled with a decent CPU overclock, showed measurable performance gains in testing. The overclocking margin or headroom came in at 13+% or 281 MHz for the time spent tweaking the modules for maximum clocks without killing every day performance. This kit from G.Skill reached the highest overclocked speed in comparison to the other modules in this testing session.
Mushkin Overclocked settings:
- Processor: Intel Core i7 3960X
- Memory: Mushkin Redline Ridgeback 16 GB 10-12-11-28 @ 2364 MHz
I have never been disappointed with the memory kits that Mushkin brings to the table, so now is not the time to start a slide into the rat hole. Mushkin’s Redline kits are their high end modules designed for overclocking and in testing, these modules do not fail in that regard – I was able to reach a final mark of 2364 MHz or 231 MHz above the already high stock 2133 MHz speed. This is a bump in clock speed of almost 11%. Since I could not reach 2400MHz, I resorted to the same solution I used on some of the other kits and bumped the CPU gear ratio to the 125 MHz strap, while reducing the CPU clock multiplier to compensate for the increase in CPU clock speed. With that said, I was still getting in the ballpark of 4.5 GHz on the CPU. Add some bclock tuning and I was able to stretch up to 2364 MHz with a bump in voltage and loosening of primary latencies to 10-12-11-28 using 1.66 V. One thing to note; the Ridgeback-designed heat spreaders are far more efficient than the Frostbyte design that has served them for so long, keeping the modules cool with just a small bit of airflow over them.
Patriot Overclocked settings:
- Processor: Intel Core i7 3960X
- Memory: Patriot Viper Extreme Division 4 16 GB 10-12-11-28 @ 2202 MHz
Patriot’s Division 4 modules really surprised me with their overhead through simply loosening the primary latencies to a similar level as the Mushkin, G.Skill, and Corsair Vengeance modules. Even with the stock 1600 MHz frequency, this low latency kit stepped up with a 600+ MHz bump in clock speed. These modules were so much more flexible than the Division 2 kit I looked at after the Intel Socket 1155 Sandy Bridge launch – simply an amazing increase over baseline and on the order of a 75% bump in speed. This does not translate into a 75% increase in performance, but it definitely helps the cause if you spend the time to get there. Voltage on the memory was tweaked a bit to 1.67 V and the memory controller kept to the same 1.05 V as in the rest of testing. The latencies were moved to a much less palatable CAS 10 from the XMP profile CAS 8. When trying for CAS 9, I was only able to reach 1984 MHz. The 3960X seems to be a little latency tolerant for everyday use, since there is a nice size boost in overall bandwidth. The Viper Extreme heat shields also do their job as intended and keep the modules cool when pushed.
The maximum memory speed for each set of overclocked modules is indicative of how well the modules ran on this test system. As such, your results may differ in either a positive or negative way based on the capabilities of your hardware. That said, your mileage may vary!
The benchmarks used in this review include the following:
- CPU-Z Version 1.58
- Windows Task Manager
- PCMark 7
- PCMark Vantage
- Geekbench 2.1
- Super Pi 1.5
- SiSoft Sandra 2012
- Aida 64
- Battlefield Bad Company 2