I'm new to Intel support forum, since I didn't have any issues with Intel boards until this one (or which I couldn't sove by my own knowledge).
I have DX79SR board, 3930K, 16GB kingston 2133 RAM (4x4) and GTX570 (and A bunch of SSD's - no HDD). I use PC for video editing and vas majority of time I render video using NEAT video (very CPU intensive filter for noise reduction which can use CUDA processing too). I upgraded my sistem from Z68/2600K and LOST PERFORMANCE! I'm soo sad.
My issue is slow RAM performace, or Quad vs Dual = the same. It is no difference if put 2 ram inside or 3, or 4, regarding throughput. I do measure performance on 3 different ways: SiSandra2012, AIDA64 and real life rendering speed in one of my application. Please read my background, to understand why I think something is wrong.
NEAT video is very CPU intensive so I upgraded from Q9550 to 2600K. I put it on Z68 (Asus V-PRO) and 1600 Kingston HyperX ram. I overclocked my (24/7 OC was 44) system. Since NEAT video can take advantage of CUDA processing I bought GTX460. It helped just little. few % more using CPU+GPU. I bought 550ti, it was better, but not much. I bought GTX570, it was even less improvement. Funny is that CPU rendered at 9 fps, GPU at 8 fps. But together it was 11 fps maybe. Than I overclocked GPU, and little change was seen (3-5% at 30% GPU overclock). I blamed software.
Than I overclocked RAM to 1866. Boy that was improvement in rendering speed! So I bought 2133 HyperX I have now. Again 15% improvement on rendering speed. Not only CPU and GPU alone performed better, but aggregate performance was much better too. Filter now started to use HyperThreading "virtual cores" too, not only real ones. I saw that GTX570 was not reached its limits yet. So SB-E with 4 channel RAM would rock (I thought).
But when I transplanted RAM and GPU to new PC, rendering speed was much slower than on Z68! CPU did few % better than 2600K (overclocked the same to 44) but that very same GPU performed 30% less than on Z68. Even aggregate rendering speed was now slow as with Z68 with 1600RAM. I tried to everclock RAM and CPU to the limits, but all I can achieve is 90% of performance what my previous Z68 achieved.
Than used AIDA64 and saw the problem. RAM "reading speed" test showed 22000MB/s on X79, but same stick of ram put in Z68 scored at 28.000MB/s. "Writing speeds" are even more different between plaforms. Latency Z68=40ns, Z79=45 (at best). If I calculate this values they reflect the same %-tage in real rendering speed differencies. The speed scores in AIDA are the same if I use two sticks of ram, or 4. But it is halwed if I put inside only one RAM. So what I get on X79 is effectively dual channel, regardless CPU-Z sais Quad. I have RAM in blue slots, and they are QuadKIT.
Than I downloaded SISoftwareSandra2012. Now this software showed 26GB/s on Z68, but 50GB/s on X79. With two sticks half of that. Ok, that proves I have quad channel engaged and it works (in theory). But in real life is slower than Z68's dual channel.
Because I needed to pay for new CPU and motherboard, I sold Z68+CPU+cooler. SO I can+t test any moore. But as I have well documented all my previous sistems, in real life all my work is now slower 25% than on Z68 (Because SB-E does not want to be stable at same speeds than 2600K + it does overheat sooner). Before all 4 cores was mostly at 100% when rendering. Now 6 cores are anywhere from 30-70%, and only Prime95 make them 100%.
So clearly RAM is NOT capable of deliver enough data to both GPU and 6 cores of CPU, not mentioning virtual cores, to be 100% occupied. Or bottleneck is maybe elsewhere??
If anybody knows what solution might be, I woul give him a hug:) Intel claim great memory performance boost with SB-E, but all I can see is lees than dual channel with exact same RAM (and XMP settings + tightened timings). I hoped for at least 25% improvement if not 50%, over 2600K on Z68. But -10% is a big shock.
Sorry for long post, please take my opology for my english too (if needed).