Answering What DDR4 Clock Speed is Best

Shortly ahead of the Intel Z170 platform launch we heard that the memory controller Intel developed for Skylake was amazing and that the new Intel Core i7-6700K and Core i5-6600K could come close running DDR4 4000 MHz memory kits in dual-channel. Just days before Intel launched their two new K-Skus that are aimed at enthusiasts and gamers we noticed that G.Skill released a dozen new dual-channel DDR4 memory kits with speeds ranging from 2133 MHz to 4000 MHz. G.Skill is the first company to release retail memory kits at 4000MHz and it took the release of the latest 6th generation Intel Core processor and Z170 motherboards to make that happen! G.Skill DDR4 Memory Kits For Intel Z170 The new G.SKill Ripjaws V 8GB (2x4GB) memory kits are already all available on Newegg and range in price between $62.99 and $399.99. The lowest priced kit is actually the DDR4-2400 kit at $62.99 and you can move up to a DDR4-2666 kit for only two dollars more. The price differences between kits is actually reasonable as you can go from a DDR4-2400 and DDR4-3200 kit for only $17 extra. If you wanted to go with a DDR4-3466 or DDR4-3600 kit you better open your wallet though as the price jumps up to $230 and $400, respectively. The point of this article today is to look at all the above clock speeds and then overclock our kit even faster to see what the sweet spot is for DDR4 memory on the Intel Z170 platform. There will be many thousands of people building a new system and looking possibly their first DDR4 memory kit ever and we wanted to help them find the best DDR4 memory kit! To answer this question, we needed to look at DDR4 frequency on the LGA1151 platform to see how memory acts with Intel's latest Skylake processor microarchitecture. G.Skill Ripjaws V Obviously, the memory controller used on Skylake is rumored to be good, so we asked G.Skill if we could get one of their news kits of memory in to investigate on our own.  G.Skill graciously sent over a RipJaws V 8GB (4GB x 2) 3600MHz DDR4 memory kit over for us to play around with that is sold under part number F3600C17D-8GVK.  G.Skill Ripjaws V DDR4 memory kits are available in Blazing Red, Steel Blue, Radiant Silver, Gunmetal Gray, and Classic Black color schemes. G.Skill Ripjaws V DDR4 3600MHz 8GB Kit This memory kit operates at DDR4-3600 with CL17 timings of 17-18-18-38 at 1.35V and is plug-and-play ready thanks to have an Intel XMP 2.0 profile. G.Skill Ripjaws V and ASUS Z170 Deluxe Most all Intel Z170 boards and 6th Gen Core 'Skylake' processors should be able to run DDR4 memory kits ranging in speed from 2133 MHz to 3200 MHz without issue, but getting beyond 3200 MHz might involve some tweaking. We picked the ASUS Z170 Deluxe motherboard for testing as it is one of the best boards for memory overclocking according to the memory manufactures. We were able to run the G.Skill F3600C17D-8GVK memory kit on this board at clock speeds of 2133 MHz and 3866 MHz with the memory running at its stock voltage of 1.35V. The fastest memory kit available in the Ripjaws V line is clocked at 3700MHz, so we were able to overclock this memory kit beyond any Ripjaws V kit that you can purchase today. [gallery ids="170365,170364,170363,170362,170361"] To get such high clock speeds we did have to go into the UEFI and tweak some things manually to get things stable as pushed the clock speeds up. We learned that the VCCIO/System Agent voltage are both important for memory clocking on the Z170 platform. We found that when running this particular dual-channel kit of memory at DDR4-3400 or higher that we needed to bump up the CPU VCCIO Voltage to 1.250V from 0.968V and the CPU System Agent Voltage to 1.300V from 1.056V.  Increasing the voltages higher than this doesn't appear to help memory overclocking, so we don't suggest trying higher than that. You can see the ASUS Z170 Deluxe UEFI 0603 menus above with the settings we had to change to overclock both the processor and DDR4 memory.  

DDR4 Memory Frequency Impact on Intel Z170 and Skylake

Now we’ve reached the meat and potatoes of this memory scaling which we'll be showing you our test results from the benchmarks we selected to test with. The benchmarks that made the cut were AIDA64, 3DMark 11 Sky Diver Physics Test, Handbrake, Metro Last Light and Grand Theft Auto V. We feel these benchmarks should give a fairly good look at how the memory subsystem on Skylake's new Z170 platform performs with some game titles as well as video encoding and of course memory bandwidth tests. There are hundreds of DDR4 dual-channel memory kits on the market and many don't know what kit to buy, so we tested as many DDR4 clock rates that we could and ended up benchmarking our Skylake-based platform with memory clock speeds from DDR4-2133 all the way up to DDR4-3866 using the same exact kit of memory set to 1.35V!

ddr4-3866

We tried our best to use the most popular clock speeds and timings for each clock rate. Here’s the full list of our memory configuration we used for benchmarking:

We overclocked our Intel Core i7-6700K Skylake quad-core processor to 4.6GHz (46 x 100MHz) on our test bed and only adjusted the memory divider and never touched the baseclock speed on our processor. This will help ensure that any performance differences we see are due to the memory subsystem and not the overall clock frequency of our processor.

Intel Core i7-6700K Overclock

Rather than just running each benchmark or application once, we took the average from running each test scenario five times. This means that we ran 25 benchmarks for each memory clock speed we tested at and there were nine clock speeds tested. All told we ran 225 benchmarks over a three day span this week to get the numbers that went into these charts.

Before we look at the numbers, let's take a really quick look at the test system. All testing was done on a fresh install of Windows 8.1 Pro 64-bit and benchmarks were completed on the desktop with no other software programs running. We will use an NVIDIA GeForce GTX 980 Ti reference card for discrete gaming performance tests. test-system The Intel Z170 platform that we used to test the the G.Skill Ripjaws V memory kit was running the ASUS Z170 Deluxe motherboard with BIOS 0603 that came out on 07/30/2015. The Corsair Neutron GTX 240GB SSD uses 19nm NAND and was using M311 firmware.

AIDA64 Extreme Edition 5.30 Beta:

AIDA64 Extreme Edition 5.30 software (websiteis an excellent benchmarking tool for looking at memory performance on any desktop PC. This version of AIDA64 implements full support for the upcoming Intel Skylake processors, adds advanced support for Microsoft Windows 10 and Windows Server 2016 operating systems, and supports the latest graphics and GPGPU computing technologies by both AMD and nVIDIA. AIDA64 Extreme 5.30 software is free to try with limited features or $39.95 to buy with 1-year of free updates. 

DDR4 Memory Scaling AIDA64

Benchmark Results: These charts are the largest that we've ever used for a memory article as we've never tested 10 different clock frequencies in one chart before, so be sure to click on them to open them full size for a better look. On the ASUS Z170-Deluxe motherboard with the Intel Core i7-6700K processor overclocked to 4.6GHz we saw the memory bandwidth increase rather dramatically as we increased the DDR4 frequency. With the slowest DDR4-2133 dual channel memory kit on the market we were just shy of 32,000 MB/s of read/write memory bandwidth and by the time we got up to DDR4-3866 we were running over 50,000 MB/s! The write memory bandwidth went from 31,923 MB/s to 56,467 MB/s, which is a massive 24,544 MB/s or 76.9% increase in the effective write bandwidth. The performance increase was found to be fairly consistent until the very end when we had to begin increasing the memory timings to achieve the clock speeds we were trying to get.

aida64-copy

Benchmark Results: Similar performance gains were seen on the AIDA64 v5.30 copy test. We started out at 28,700 MB/s with a DDR4 clock frequency of 2133 MHz and ended up at 48,400 MB/s when we topped out at 3866 MHz.

aida64 DDR4 Latency

Benchmark Results: The latency of the memory decreased as the DDR4 DRAM clock speed increased despite having to loosen the timings in the UEFI along the way. At DDR4-2133 we averaged 54.3ns then reached 43.4ns at DDR4-3000 and then finally 40.9nm at DDR4-3866.

3DMark Sky Diver and Handbrake

Futuremark 3DMark Sky Diver

3dmark-sky-diver-hero-wide
3DMark Sky Driver is the latest test in the world’s most popular benchmark for measuring the 3D graphics performance of gaming PCs. We'll be using the Physics test to look at memory performance. The Physics test introduces a new approach that extends the performance range for which the test is relevant. The test runs through four levels of work starting with the lightest and continuing to the heaviest unless the frame rate drops below a minimum threshold. The GPU load is kept as low as possible to ensure that only the CPU is stressed. The test uses the Bullet Open Source Physics Library.
skydiver
Results: The Physics score went up 5.7% by going from the DDR4 2133MHz kit with CL14 timings to the DDR4 3733 MHz kits with the CL17 timings, so there isn't a big performance difference to be had despite the 1600 MHz or 75% increase in memory clock on the Intel Z170 platform for this particular test. The 5.7% increase in the score is a significant improvement though. Not that the 3000MHz kit scored 47.31 FPS and the 3733MHz kit scored 47.65, so there is a very small difference in performance once you get to 3000MHz with somewhat aggressive 15-15-15-36 1T timings. 

Handbrake

handbrake-gui HandBrake is an open-source, GPL-licensed, multiplatform, multithreaded video transcoder, available for MacOS X, Linux and Windows. It is popular today as it allows you to transcode multiple input video formats to h.264 output format and is highly multithreaded.
Big_buck_bunny
We used Big Buck Bunny as our input file, which has become one of the world standards for video benchmarks. The 1080P clip was used in the MP4 format and the workload is encoded into h.264 output format using the preset - high profile. This benchmark test was setup to give you an idea of how these processors can take a 1080p BD rip and turn it into a 1080p H.264. HandBrake version 0.9.9 was used for benchmarking and we highly encourage you to download this MP4 clip and compare your system to ours with Handbrake!

handbrake

Benchmark Results: HandBrake version 0.10.1 showed nice performance increase from DDR4-2133 to DDR4-2400 and then we see performance pretty much plateaus.  

Metro Last Light and GTA V

MetroLL-SS

Metro: Last Light is a first-person shooter video game developed by Ukrainian studio 4A Games and published by Deep Silver. The game is set in a post-apocalyptic world and features action-oriented gameplay with a combination of survival horror elements. It uses the 4A Game engine and was released in May 2013. metro-settings Metro: Last Light was benchmarked with very high image quality settings with the SSAA set to off, Tesselation on normal and 4x AF. We used the game titles built-in benchmark (seen above) and ran it 3 times at a 1080P screen resolution to ensure accurate results. metro-ll Benchmark Results: In Metro: Last Light with the NVIDIA GeForce GTX 980 Ti discrete desktop graphics card installed in the system we saw a jump in performance between DDR4-2133 and DDR4-2400 and again between DDR4-2666 and DDR4-2800. The performance between DDR4-2800 and DDR4-3466 was flat and then we saw 0.5 FPS increased with the move to DDR4-3600 and DDR4-3733 clock speeds. We were shocked to see performance gains at the end after no gains for several clock speeds, but we'll take it and it was repeatable.  GTAV Grand Theft Auto V, currently one of the hottest PC games, was finally released for the PC on April 14, 2015.  Developed by Rockstar, it is set in 2013 and the city of Los Santos.  It utilizes the Rockstar Advanced Game Engine (RAGE) which Rockstar has been using since 2006, with multiple updates for technology improvements. gta5 gta5-settings In Grand Theft Auto V we set the game to run at 8x MSAA with 16x AF and normal image quality settings as we didn't want the GPU to bottleneck the system too bad, but wanted a setup that your average gamer would actually play on. GTA V estimated that we'd use 2GB of our 6GB frame buffer with these settings, which isn't too taxing and should give us a good frame rate. V-sync was of course disabled. gtav Benchmark Results: In GTA V we found hardly any difference between running DDR4-2133 and DDR4-3733 as there was less than a single frame per second difference found after running the games built-in benchmark three times and averaging the numbers. The DDR4-3000 result of 106.0 FPS on average was about as good as you'll get as we only got 0.1 FPS higher by going up another 733MHz. 

Conclusion - Finding The DDR4 Sweet Spot For Skylake

G.Skill Ripjaws V DDR4 3600MHz 8GB Kit After spending dozens of hours benchmarking ten different memory speeds on the Intel Z170 + Skylake platform we must admit that we are too shocked by the findings. Our benchmarks show that the memory bandwidth increased, but there wasn't a tangible improvement in system performance with real applications. We ran other applications and game titles when we tested this memory kit and you mostly ended up with flat performance charts like you saw in Handbrake or any of the game titles that we tested today. We saw some nice performance gains from DDR4-2133 to DDR4-2666, but strangely the pricing for the G.Skill Ripjaws V 8GB 2133 and 2666 dual channel kits are identical. That is where you'll see the biggest performance gain in the real world benchmarks and you can get that benefit for free right now. There is only a $15 difference between 2133MHz and 3200MHz memory, so skip a meal out and opt for higher clocked kit if you are on a tight budget! The DDR4-3466 and DDR4-3600 kits are nice for overclockers trying to get high memory bandwidth scores, but there it's hard to justify spending $230 or $400 on a kit of memory that does little on real world applications. The one thing we didn't have time to test or maybe we just lacked the desire after spending three days benchmarking this is to test all these memory clock speeds over again on the integrated graphics (Intel HD Graphics 530) to see what that does for performance. We know from past experience that integrated graphics benefits greatly from higher clocked speeds, but at the same time not too many 'gamers' are going to spend $350 on a Core i7-6700K, $400 on the G.Skill RipJaws V 8GB 3600MHz DDR4 memory kit and $319.99 on the ASUS Z170-Deluxe to only run integrated graphics. If you're going to drop $1,070 on your CPU/Memory/Motherboard the chances are you'll have a discrete graphics card in your system. We've never doing a price per MB/s of memory bandwidth chart before, so let's average the read/write scores we got in AIDA64 and calculate that against the price above. band-vs-dollar This chart shows that you are getting 606MB/s of memory bandwidth per dollar spent on the 2666MHz kit, so from a price to performance aspect the best memory clock speed in the G.Skill Ripjaws V series would be that one! The 3600MHz hit actually gives you the lowest price vs. performance score of 128 MB/s per dollar. The 3000 MHz kit isn't too far from the top as it came in at 602 MB/s! G.Skill Ripjaws V and ASUS Z170 Deluxe We hope that you have a better understanding about picking the right DDR4 memory kit for the new 6th generation Intel Core processors and Z170 motherboards. Our advice for gamers and enthusiasts would be to get at least a DDR4-2666 kit as that is the sweet spot this very second when it comes to price versus performance! This new platform scales exceptionally well though, so if you like memory overclocking we have a feeling a number of you will be moving over to a Z170 platform soon!