To accurately compare this platform with the previous LGA 2011 one, is almost impossible to achieve; no matter the settings it will still be like comparing apples with oranges. Though some adjustments can put the outputs on a more balanced scale.

 

We have to take into account the different processor architectures, the difference in stock and Turbo clocks, more on-board L3 cache, DDR3 versus DDR4, 2 extra cores,...

To make it a more even battleground we intervened in the following ways:

• To rule out the difference in MHz processor wise we overclocked both the Intel i7-4930K and i7-5960X, so all cores are running at +/-3900MHz.
• To keep data aligned two cores of the i7-5960X were disabled in the bios, so we are comparing 6/12HT
• The different quad channel ram speeds compared are:
  
  - 16GB DDR3 1600C9-9-9-27 2T (Samsung IC)
  - 16GB DDR3 2666C10-12-12-31 2T (Samsung IC) XMP profile loaded
  - 16GB DDR3 2800C11-14-14-31 2T (Hynix CFR IC) XMP profile loaded
  
  - 16GB DDR4 2133C15-15-15-35 2T (downclocked Corsair Vengeance LPX kit)
  - 16GB DDR4 2400C15-15-15-35 2T (downclocked Corsair Vengeance LPX kit)
  - 16GB DDR4 2666C15-15-15-35 2T (downclocked Corsair Vengeance LPX kit)
  - 16GB DDR4 2800C15-15-15-35 2T (tightened Corsair Vengeance LPX kit)
  - 16GB DDR4 3000C15-15-15-35 2T (overclocked Corsair Vengeance LPX kit)

 

 

Starting off with Superpi 32M, a single threaded application that can quickly reveal if higher specced memory is worthwhile. With The Ivy-Bridge E going from the standard rated DDR3 1600Mhz to 2666MHz yields a healthy 18secs gain. On the DDR4 powered LGA 2011-v3 platform the gains are less impressive at first glance. A 14 sec advantage for DDR4 3000MHz versus 2133MHz running memory speeds.

 

 

 

The memory bandwidth test in the AIDA64 software reveals the Achilles heel of the Haswell-E platform. While the Copy and Read significantly improve with added memory speed, the write performance stays almost put. On the DDR3 side of things the Read, Write and Copy performance receioves big boosts from any added memory speed. The extra generated bandwidth by the DDR4 modules is crippled by the standard stock Uncore speed of 3000MHz of the brand news Haswell-E processors. On the next page we will investigate a bit further how to boost performance by raising the uncore clocks.

 

 

 

Maxon's Cinebench Release 15 scales a little bit with higher clocked memory. However take note that the different DDR4 memory speeds were all running with the same primary timings of C15-15-15-35 2T Command Rate. The X264HD encoding test scales similar to the previous tests, however again the effect over 2800MHz seems to be less impressive.

 

 

2D scales mildly with increased memory clocks, how does the brand new Haswell-E platform and its DDR4 respond in a gaming environment? Looking at the Physx test which is extremely bandwidth dependent, we see again mild scaling with increased memory speed, however the total gain is significantly less than with IVY-E. We ran each test three times to rule out any abnormal scores, though time after time the ASUS Rampage V Extreme board produced lower GPU scores when upping the memory speed over 2800Mhz.

 

 

GPU-wise the same things happened with the 3DMark Firestrike benchmark, we have to verify if this behavior is board related ( less performance or buggy memory divider ) or if this is a platform issue by testing other high OCing motherboards. For benchmarking it seems that even 2666MHz with tight timings will be more than sufficient for 3D benching. Now how big is the performance impact on games?

 

 

 

In our three games tested, being Bioshock Infinite, Sleeping Dogs and Tomb Raider we see no significant reason to invest money in high rated memory. The standard Intel supported DDR4 2133MHz speed will do more than nicely. High speed memory required for gaming, no thank you...