The launch of Broadwell-E last summer confirmed that Intel's X99 LGA 2013-V3 is still the platform to be used for benchmarks like Vantage, Catzilla, 3DMark11 and them tough Firestrike runs; especially the Futuremark versions can benefit from a well-dialed in memory subsystem, reflecting in a better total score thanks to improved PhysX and combined scores, so time to explore what these B-Die memory kits can mean for you, besides burning another hole in your wallet :)

 

 

Everybody knows that ultra-high memory kits are nice but your system must be able to handle them. With each new platform we have seen an increase the memory speeds. Especially the processors Internal Memory Controller (IMC) has to cope with the insane bandwidth some of these memory module are able to deliver; while initially DDR4 timings looked utter crap, things have evolved nicely and the million dollar question is always which sticks are the best for my needs.

 

This article is for the die-hard tweakers and Overclockers that want to get the utmost performance out of their setup. A lot of questions have arisen at the HWBOT.org forum about using these B-Die sticks on Intel's current High-end platform X99. Some knowledge is required as one has to wade through the numerous BIOS pages to get everything dialed in accordingly.

For our testing purposes we tested three different candidates based on Hynix MFR, Samsung E-die and the latest and greatest Samsung B-Die. No quad channel AFR memory kit was available at the time of testing; maybe we can update the current article in the future. Also the popular D-Die (slightly tighter main timings than E-Die) was left out of the test as I only had one Dual Channel kit available.

Our test platform  comprised of the following hardware:

• Intel i7-5960X CPU cooled by Noctua NH-D15
• ASUS Rampage V Extreme motherboard (1902 bios) & MSI Godlike (V1.7 bios)
• Memory running at 3200MHz, timings as tight as possible
  
• ASUS GTX 570 DCII
• Seasonic 1200W Platinum PSU
• Samsung 250GB Evo SSD
• Windows 8.1 64Bit Professional Edition

 

The DDR4 memory kits we tested are the following:

• Quad channel 16GB G.Skill Ripjaws 4 3000C15  (MFR based)
• 2 x Dual Channel 8GB G.Skill RipjawsV 3200C16 (E-Die based)
• 2 x Dual Channel 16GB G.Skill TridentZ 3600C17 (B-Die based)

 

 

I'll include some BIOS screens with the G.skill B-Die memory ICs. Consider them screenies as a Bios starting profile; nany users on the forums have been asking questions to stabilize their memory on X99. thus hoping the below will serve you well.

I've tested both settings on the standard ASUS Ramapge 5 Edition and the MSI Godlike Gaming board. Although the latter is not the greatest in memory overclocking, 3200MHz memory speed is achievable, however it will require some hefty SA and IO voltages compared to the ASUS' offering.

 

 

On both motherboards we used the 100 strap and selected the 3200 Memory divider. Make sure you up the IO and SA voltages to needed level as these are needed to allow your processor IMC to cope with the extra bandwidth you are gonna throw at it.

 

 

It is very important to dial in properly the tCKE and tWCL setting as these can be the difference between a proper post or not; play with these till your board starts to post. Memory tweaking takes some time, learn how your own board, CPU and memory react to each new BIOS input.

The ASUS R5E board and the Intel Core i7-5960X combo needs around 1.2 SA (CPU System Agent voltage). On the MSI Godlike Gaming this needed to be raised above 1.365 to enhance stability.

 

With an added 133MHz things start to get more tricky for both the motherboard and the processor's Integrated Memory Controller. The ASUS Rampage V Extreme had no issues at 3333MHz with the B-dies at Quad Channel.

The MSI Godlike Gaming board started to loose randomly some channels if we pushed above 3200Mhz. The ASUS board behaved the same when we wanted to try any memory divider above 3333MHz, this no matter if we used the Haswell-E or Broadwell-E precessor. In all honesty this is not bad at all, however if you really want to explore new heights in memory clocking the ASUS flagship Rampage V 10th Edition is the only way to go. The latter has been proven to be way more versatile with high speed and large quantity memory than any other X99 board out there.

But back to the starting profile for 3333MHz action:

 

 

Basically there's not that much difference between the 3200 and 3333MHz strap. Just note that we are using the 125 strap now, of course the 3333MHz memory divider and had to increase the System Agent Voltage to 1.275 on the ASUS Rampage V Extreme motherboard.

 

 

Several pre-tests were conducted to verify if we had properly dialed in the right primary and secondary timings. Secondly these also needed to be stable to run the bench suite multiple times without crashing.

 

First up is the what we Overclockers consider as the mainstream memory kit, Hynix MFR for Intel's high end X99 platform. We took one of the popular kits of the lab, the G.Skill Ripjaws 4 3000C15Q-16GRR, stock rated at 15-15-15-35 2T 1.35VDimm.

For the G.Skill Ripjaws 4 3000C15 MFR Quad Channel we reached at 1.75Vdimm 3000/3200C12-13-15-18 1T Command Rate. Running 3333MHz was a no go anyway on our MFR kit, no matter the timings nor how high we set the Vdimm it was never stable enough to conduct testing.

 

The G.Skill Ripjaws V 3200C16 E-Die Samsung based kit runs stock at 3200C16-18-18-38 2T Command rate. It was pre-tested at 3200MHz CAS 11. Take note these kits can easily reach way higher frequencies on other platforms alike one Kaby Lake (3866+speed).

 

 

For 3200C11-16-16-26 1T Command Rate we also needed 1.75 Vdimm; tightening the primaries, especially the tRCD and tRP to 15 was a no go below 1.85Vdimm on our sticks, so we opted to remain in our voltage comfort zone and run at the below settings.

 

Finally the G.Skill TridentZ 3600C17 B-Die based kit ran happily at 1.725Vdimm at 3200 C11-11-11-24 1T Command Rate. Same as for the E-DIE sticks there's way more potential left, though 3333MHz would drop from time to time a channel, ending up in Triple channel and even in worst case scenario Dual Channel mode. totally crippling performance. With the Broadwell-E processor this phenomena was not observed.  Concluding our processor's IMC is mainly to blame as we have seen screenshots at 3400+ speeds, but more on those in the conclusion.

 

 

 

When running 3200C12-12-12-24, we only needed 1.52Vdimm on the Haswell-E platform. The MSI Broadwell-E setup ran 3333C12 at 1.59VDimm.

 

Time to start off with the benchmarks as this what we are all keen to see if the B-DIEs will rip the other two candidates apart. All tests were executed 3 times to avoid any bugged read-outs, executed each time after a fresh reboot. Since we could not run the MFR memory at 3333MHz speeds, all test were conducted at 3200MHz, including a few runs at 3000MHz for the Hynix MFR kit.

First up is SuperPi 32M, a 2D favorite for most. Okay admitting this is not really a benchmark that is being benched heavily on the X99 platform; the CPU frequency is one of the major limiting factors here versus the more mainstream platforms that clock easily over 6Ghz on Liquid Nitrogen. Nevertheless it will provide a good indication what is going on behind the scenes with the three different memory kits.

Superpi 32M:

 

 

The 3200C11 B-DIE is the fastest kit for this benchmark, though we are talking only about a few seconds, however these could mean the difference of course between ranking in the top 3 or just hovering somewhere in the top 20. The 3200C11 E-DIE is the slowest in this test, the high tRCD and tRP value are probably the main culprit for its inferior performance, Even versus the good old Hynix MFR kit the E-Die memory stood no chance.

 

Maxon Cinebench R15:

 

Maxon's Cinebench R15 also gets a minor boost with added memory bandwidth as one can see below.

 

 

MFR again giving B-DIE a run for its money, though the new kid on the block still pumps out the best scores, not mentioning anymore the again lack luster performance of the E-DIE kit... Last but not least to finish of the 2D tests we are gonna throw in Intel's Xtreme Tuning utility, especially this benchmark has seen major improvements with the newer and faster memory on the Sky/Kaby Lake platform.

 

Intel Extreme Tuning utility (XTU):

 

 

Due to the above, I re-ran this benchmark more than the other tests, however the best result were always topping at 2146. Only the E-DIE based set was always scoring less than the MFR and B-DIE kits. Seems XTU is not properly using the extra available bandwidth of the installed G.Skill B-DIE. Interestingly enough this was also noted by the numerous OCers that are benching the i3-6320. The high scores were usually a combination of a proper dialed in OS, preferably 32 bits and just a good run. No matter if they ran 3733C12 or 4000C12, the score didn't budge at all... So we are observing something similar on the X99 platform with XTU.

3D time... This is where this platform is far more interesting for benching, especially the 10 core Broadwell-E is the CPU of choice to work with, thanks to the added CPU power.

 

Futuremark 3DMark Vantage:

3D Mark Vantage is up first, just to see if we can see anything moving with the CPU score.

 

 

Surprise, surprise, each time the 3200C12 MFR setup came out as winner here, grabbing the highest CPU score and total score. Not a total surprise as the two other 3D tests have a dedicated PhysX test which is far more memory bandwidth dependent.

 

Futuremark 3DMark 11:

 

 

The 3DMark 11/Firestrike results are more in line with some of the 2D tests. B-Die on top, followed closely by the 3200C12 MFR. Honestly I expected a bit more difference in output for the PhysX tests between the mighty ultra tight Samsung B-DIE and the older Hynix MFR ICs... Seems it is not gonna happen on this platform.

 

Conclusion:

Looking at the few tests we conducted it is clear that B-DIE based memory also has a performance advantage on the X99 platform. Yet it is far from mind blowing as it is on the Intel Z170/270 Sky or Kaby lake platform, showing it's real potential is not possible mainly due to architectural limitations, especially pointing fingers at the processor's memory controller which is sometimes really hit or miss on this platform. The used motherboard and its BIOS will also play an important role whether you can tweak the memory or not.

Ow what about them 3600+ screenshots then? Well a pro tip from an old dog in the scene, don't always believe what you see. Some stuff is "ahum" forced to look good. Remember them CAS 10 runs when X99 was released? In my book not really happening on air sticks... Most of them screenies were tricked via software OC tools. Today some of these high clocked B-DIE runs are just executed on Dual or Triple Channel, even though CPU-Z will still report Quad Channel and 32GB installed. Judging by the performance of some screenshots it is mostly crystal clear that it looks awesome, but performance is inferior to a true Quad channel setup. Always remain skeptical of what you see, especially in this niche hobby :)

 

 

That little rant aside, so is B-DIE the way to go for X99? Well yes and no; if you are chasing World Records it might be your only option to nail them needed extra points. Nevertheless don't expect miracles for X99 and Haswell-E. Thanks to the launch of Broadwell-E, the playground changed a tiny bit, yet again it will all depend if the processor IMC is able to handle it or not. Sadly we can't vouch that each Broadwell-E CPU's IMC is superior to previous generations processors.

Now the million dollar question what to buy...

A properly dialed in MFR kit is still more than plenty for most OCers who are benching on this platform. Far cheaper and less stressful for your IMC than these expensive Samsung based B-DIE kits. Also most renown MFR kits can manage the tested timings with ease. B-DIE like mentioned is either apples or oranges: either your kit can handle the voltages, speeds and timings or you just bought yourself a cat in a bag. Hopefully the production process continues to improve and increases the overall quality and especially the reliability of these Samsung B-DIE ICs. Fingers crossed for that, as it is utter disappointment if your brand new kit can't even properly manage +3866C12 rock stable (not on X99 of course).

I still have some mixed feelings of the results. Maybe a retest with 4500/4500 clocks is possible in the near future? The higher uncore speed might yield a bigger performance difference between the different kits. My other sentiment is that the platform is pretty maxed out, even when using the MFR memory at appropriate speeds and timings.

If ever you want to buy new memory, let it be B-DIE based. When tweaked it will be the most powerful memory IC to date. Secondly it can be used on other platforms too like AMD's latest Ryzen generation. Hence why the below X99 memory for Tweakers buying list would be made of:

 

1. B Die based memory

2. Hynix MFR based memory

3. Don't even think about E or D memory :p

 

 

 

Thanks for reading and hopefully see you soon at some obscure OC event for some OC trash talk!

Leeghoofd logging off...

 

Thanks to ASUS and MSI for the provision of motherboards and processors. And the crew from the TONES.BE shop, without this support it would not have been possible.