Slower Firestrike Bench results with Dual Channel vs Single stick?

[flaystus]

Just finished a Brand new Skylake build.
Asus Z170-AR motherboard. Matched Dual Channel Kit.
Crucial Ballistix Sport 8GB DDR4 4gbx2
GTX 970

The problem:
RAM is properly installed and the timings appear to be correct They are also in the A2 B2 sockets (I think, which everone is correct is what they are in). CPUZ confirms its in dual channel mode. But if I have EITHER single chip in and run a Firestrike I get over 9000. If I put both sticks in it drops to over 600 points.

Has anyone seen this before?

here is a screenshot of the different results:
http://i.imgur.com/hl47vhV.jpg

 

[KingFatty]

So it's faster in graphics, but slower in physics.
That's really weird.

Why would a GTX 970 slow down with less system ram?

 

[flaystus]

You're reading it backwards I think.
Two sticks in dual channel makes the Physix go up and a 8 to 10% loss everywhere else.

 

[AlienTech]

Maybe dual channel is adding latency. But the fetch size would be double than that of single channel. So the code also has to match. If data is fetched in a dword size that would not use the entire 128bits.. It would just discard the extra data and not use it but instructions that contain data to use for execution might use the entire 128 bits.. Depending on the code, the speeds would be much faster or slightly slower in this case.. I did notice that when I went form single channel to dual, it added an extra wait state for some reason. Memory fetch showed it was twice as fast though as it was fetching twice as much data on each clock cycle even if it was running 5% slower. This is why the x43 or x64 operating system must have equality in the memory size as well. other wise a x64 fetch would require double the amount of time to do 2 fetches of 64+64 bit data instead of a single fetch of 128bit data.

 

[flaystus]

Issue resolved.

I feel silly.

Totally reset my BIOS then changed nothing except turn on the XMP profile.
Now its benching perfectly.