Dual channel memory

[Guest]

Interesting results when fiddling today. I built a system with a Ryzen 2700 on a gigabyte X470 Ultra today for a friend. I had some vengeance memory kicking around but could only find one stick of 16gb 3000hz memory so put it in and did some benchmarking with 3DMark and PCmark. Knowing I had another stuck at the office of identical memory I popped in to get it, installed and re-did the benchmarking. Repeatedly I was getting 5-7% lower benchmarks with correctly configured dual channel memory on these benchmark tools. Real world a render in Photoshop was about 2% faster with 32gb in a dual channel setup but was that just because there was more memory or because it was dual channel? I have never seen a slowdown before when testing high end servers, mainly on Xeon. This is the first time I have tested a desktop in this way. I did have a little confusing moment where the Dual bios kept switching to the other instance (at least that is what I assume was going on) and the memory profile was not active but other than that is there anything else I am missing or is this normal for this kind of setup

[Elisis]

Sticks not within the same kit (even those of the same model) aren't guaranteed to run dual-channel flawlessly. That is likely to be your problem. Also check whether your sticks are actually running at 3000 and see if the timings are in order. More RAM can cause faster rendering speeds in Photoshop, though. What are you using to compare the current setup's performance against?

[Guest]

The sticks are from the same kit but really that should never be a problem surely? In the server world it used to be the case where if a stick failed you would need to replace the whole set. Back just 10 years that could cost 50k on larger servers, even more on a mainframe. Now it is not the case, on all of our servers for the company I work for there are no issues running memory from multiple kits etc. With some servers having 48 sticks of ram that has to be the case. Sure, the memory costs more so is probably better matched but modern design methods negate the problems of old. I would hope the same applies to desktop memory now. If a stick dies the home user cannot be expected to replace all their memory and I am certain no vendor would want to be that wasteful. 
 

I have checked and all memory is running at 3000 once I got the stupid gigabyte dual bios sorted. Checked in various tools on the desktop. I have today done a couple of blender renders and the dual channel memory performs 2% faster than a single 32gb stick and 3% faster than just one of the 16gb pair. So I wonder if PCmark and 3DMark are giving lower scores for a different reason. I have yet to run those with just the 32gb stick I have.