Corsair Vengeance LPX 2666 CL 15, Surprisingly Fun Memory

Today we will take a look at some DDR4 from Corsair, the Vengeance LPX 16GB (4x4GB) CL15 memory kit.  DDR4 is brand new and is used motherboards featuring Intel’s new X99 chipset using s2011-3 processors. DDR4 uses 1.2v standard as opposed to the 1.5-1.65 that is now standard in DDR3.  It can operate at higher frequencies with much less voltage needed, being much more efficient.  Running in quad channel the memory will provide much more bandwidth than that of dual channel systems by a hefty margin.  This kit is some of the entry level DDR4 from Corsair, but as you will see they have much more to offer if you do some tweaking.

Specifications:

• CMK16GX4M4A2666C15R
• 16GB (4 x 4GB)
• DDR4 2666 (PC4-21300)
• Cas Latency – 15
• Timings – 15-17-17-35 2T
• Voltage – 1.2V
• Quad Channel Kit
• Anodized Aluminum Heat Spreader
• Low-profile heatspreader design
• Intel XMP 2.0

Design and Build Quality

The build quality of these RAM modules is top notch as expected with Corsair products that are currently available.  The heatsinks are of high quality and look great in the motherboard with no defects visible.  The PCB from visual inspection looks very clean and made of high quality materials.  The kits have single sided Hynix MFR for the ICs and are now the type that is most coveted by overclockers.

Performance

This kit was really impressive as the performance that was achieved with the XMP profiles and with overclocking was much better than was expected.  In these tests you will notice that they are done with varying uncore frequencies.  This was done by adjusting the frequency in the bios to achieve the overclock.  This is very important as it has a huge effect on the performance of your memory. Even at the same speed and timings there will be a huge performance gain from the simple act of overclocking the cache/uncore.  It should be noted that these tests were run on the ASUS Rampage V Extreme, as it is the best board on the market currently for achieving the highest RAM and cache/uncore clocks.  This is attributed by the inclusion of its OC Socket that is a non reference s2011-3 CPU socket.  The results with the 4000MHz and 4500MHz cache/uncore speeds may possibly be harder or impossible to replicate on other boards. be addressed is cache/uncore speed .  The first XMP profile is 15-17-17-35 2T at 2666MHz with 1.2v, the second XMP profile is 15-17-17-35 2T at 2800MHz with 1.35v.

 

When checking to see what these modules were set to auto and then run while increasing the speed until it could no longer be stable and boot into windows.  The max that was achieved was 3070MHz with the RAM set to 1.5v.  Running the memory at these speeds would have required setting the memory to 2T and having loose timings, which would not equate to the best performance possible from the memory.  To find the best possible performance that would be possible a memory profile from the Rampage V Extreme for single sided Hynix was needed to be loaded.  Selected was a profile for 3000MHz 1.65v, though it could not load it at 3000MHz.  This was expected since this isn’t exactly a highly binned set of Hynix MFR, so 2750 was tried.  This was a success at 12-14-16-15 1T with the 1.65v, and proved to be better performing than the other combinations that were tried.  Please bear in mind that if running DDR4 at over 1.4v additional cooling is a must or damage to the modules may occur.

Test setup

• Asus Rampage V Extreme
• Corsair Vengeance LPX 2666 C15 #CMK16GX4M4A2666C15R
• Intel i7 5960X @ 3.5GHz
• Thermaltake Water 3.0 Extreme
• Cooler Master V1200 PSU
• MSI R9 290X Lightning

AIDA64

 

Here it is seen that the write speed is impacted more than anything else with the increase in the cache/uncore speed.  The read performance also makes a big improvement, though not as drastically as the write did.  At 3000MHz  the read was 58532MB/s, the write was 46945MB/s, and copy was 60979MB/s with a latency of 71ns.  At 3500MHz it was 62801MB/s – 5480MB/s – 62948MB/s with a latency of 68.7ns.  At 4000MHz it was  65685MB/s – 62046MB/s – 62853MB/s with a latency of 67.1ns.  At 4500Mhz the results were 67654MB/s – 69344MB/s – 63552MB/s and the latency was the lowest at 66.2ns.  These results are good and fall in line with what you would expect from running at these speeds

The write and read speeds increase at the same rate as the other XMP profile, but here the speed is a decent amount faster due to the speed increase of the RAM. The results at 3000MHz cache/uncore were 61124MB/s – 47798MB/s – 64022MB/s and a latency of 68ns.  At 3500MHz the  results increased to 66168MB/s – 55563MB/s – 66290MB/s with 66.1ns latency.  The 4000MHz results came in at 69620MB/s – 62917MB/s – 67165MB/s and 64.6ns latency.  4500MHz returned the best results of 71367MB/s – 68829MB/s – 67348 MB/s and 63.9ns.  Being able to obtain a read speed of over 70000MB/s on the XMP profile was very good and was better than expected.

Here the RAM was set up by using the settings that was able to get the most bandwidth from, very tight timings and sub-timings were used.  It is visible to see just how much proper tuning of the timings can affect the performance when comparing to the XMP2 results. It can be seen that with some work these can perform on par or better than kits that are much more money, though a better binned set would likely require lower voltage at similar speeds and timings.  The results at 3000MHz were 62777MB/s – 47441MB/s – 68236MB/s and a latency of 63.4ns.  At 3500MHz they were 69794MB/s – 55169MB/s – 70848MB/s and 61.4ns latency.  The 4000MHz results were 74104MB/s – 62848MB/s – 72438MB/s and 60ns latency. 4500Mhz results were 76194MB/s – 70482MB/s – 71528MB/s and the lowest latency of all at 58.9ns.

 

Super Pi mod 1.5  32m

It is visible in this benchmark that is heavily influenced by RAM just how big of an effect that boosting the cache/uncore speed will have on the performance of a system.  The results for the super pi benchmark was 8:58.620 at 3000MHz, 8:54.301 at 3500MHz, 8:51.562 at 4000MHz, and 8:48.809 at 4500MHz. these results fall in line with the increased performance of the memory with the cache/uncore speed increase.

It is visible that XMP profile 2 at 3000MHz uncore/cache frequency is about as effective as the first one at 4500MHz.  The reduced time of about 10 seconds is substantial in this benchmark.  The results were 8:48.747 at 3000MHz, 8:44.192 at 3500MHz, 8:39.953 at 4000MHz, and 8:38.684 at 4500MHz.  Again, this is in line with what is to be expected from the memory performance increases like the previous test.

This result shows just how helpful to the benchmark memory speeds are to Super Pi, though in this one it looks like there was a bigger gain by increasing the cache/uncore than the other two. The results are 8:52.772 at 3000MHz, 8:47.063 at 3500MHz, 8:45.019 at 4000MHz, and 8:42.833 at 4500MHz.  It is interesting to see that the jump from 3000MHz to 3500MHz was far greater than the others.

 

SiSoft Sandra

Here it is clearly seen that increasing the cache/uncore frequency that it is possible to increase the bandwidth greatly, and the results are on par with what was expected.  The results are 52.68GB/s Aggregate, 52.36GB/s Integer, and 53GB/s Float at 3000MHz.  At 3500MHz the results were 54.65GB/s Aggregate, 54.1GB/s Integer, and 55.21GB/s Float.  At 4000MHz the results were 55.78GB/s Aggregate, 55.27GB/s Integer, and 56.3GB/s Float.  The results are in line with what was to be expected, though the jump from 3000MHz to 3500MHz showed the biggest improvement.

Looking at the results here it is possible to see that the increase gained from increasing the speed from 2666 to 2800 resulted in an increase of about 3GB/s across the board. The results at 3000MHz are 56.3GB/s Aggregate, 56GB/s Integer, and 56.64GB/s Float.  At 3500MHz the results are 58GB/s Aggregate, 57.52GB/s Integer, and 58.61 Float.  At 4000MHz the results were 59.2GB/s Aggregate, 58.76GB/s Integer, and 59.65GB/s Floar.  At 4500MHz the results were 59.71GB/s Aggregate, 59.15GB/s Integer, and 60.28GB/s Float.  Again, it is seen that the jump from 3000MHz to 3500MHz is by far the greatest leap of gain made.

Now with taking a look at these results it is clearly visible that properly configured timings at a lower speed can greatly outdo itself running at higher speeds with looser timings. The results are 60.39GB/s Aggregate, 60.49GB/s Integer, and 60.29GB/s Float at 3000MHz.  At 3500MHz the results were  63GB/s Aggregate, 62.7GB/s Integer, and 63.2GB/s Float.  At 4000MHz the results were 64.33GB/s Aggregate, 64.11GB/s Integer, and 64.55GB/s Float.  At 4500MHz the results were 65.14GB/s Aggregate, 64.91GB/s Integer, and 65.37GB/s Float.  What is able to be seen here compared to the XMP profile 2 test is that the tightened timings and sub-timings plays a great deal in helping the overall bandwidth.

 

User Experience

The RAM exceeded expectations in both how it performed with the XMP Profiles, but also with overclocking.  This set of RAM is fun to tweak and will reward the user with great results from the time they spend tweaking things to perfection.  This RAM performs well in 1T and will go quite far before it needs to switch to 2T to become more stable.  It was clear to see why overclockers enjoy working with the Hynix MFR DDR4, it performs great and is so tweakable with more voltage applied.  When testing to see what these could do, I tried to find the settings for the best bandwidth that I could get.

Value

Being essentially an entry level kit and being able to overclock so far and tweak it to perform so much better makes this a killer deal.  It would definitely be something to consider when it comes time to build a new DDR4 system.  Seeing what fellow overclockers were able to achieve with this kit has been astonishing since it perform better than other kits that cost hundreds more.

Should I buy this?

If you are looking to get something that you want to tweak a bit or be able to get more performance out of then this is it.  For $316.37  it is a great buy indeed, and this kit has received the Value Award from BSN*.  It has been a lot of fun to test this kit as I was surprised about how much performance can be squeezed out of these, and with more time with them I know that they can do even better.  I like to push my equipment to the max and see what it will do and these definitely proved themselves during testing.