Edit: I have done a bit of additional testing per whiskey-foxtrot's request. I have included results for the Gears of War 4 benchmark at 2133MHz (command rate 2T and 1T), as well as 3333 MHz (command rate 2T and 1T). I also included 3466 at a command rate of 2T. All other settings were identical, and the benchmark was repeated 3 times at each setting and the average used. All testing was done with 4x8GB of RAM installed.
1T command rate does confer an advantage over 2T but it is pretty minor. At 2133MHz the effect of command rate is more pronounced, gaining 3.5% in CPU Game fps and 2.5% in CPU render. At 3333MHz those gains drop to 0.2% and 1.3% respectively. Running 3466MHz at 2T does produce more fps than 3333MHz at 1T (3.2%, 2.1%).
Based off this, guess I'll stay at 3466 2T since it is stable for me.
Thank you ajlueke, I love the graphs - and this is also good information for folks with Corsair Dominator Platinum 3466MHz DDR4 32GB. I've seen a few people still struggle with this, and noticed that most of them are trying to push 1T.
Thanks whiskey-foxtrot. It is really interesting data actually. Benchmarks that I run that test the CPU in a vacuum, like the physics test in Firestrike, don't gain from increasing the RAM speed. However, the Gears 4 benchmark lists the CPU stats as the following:
Average CPU Framerate (game): The average framerate the game thread can achieve. The game thread encompasses all work needed to simulate one frame. The work is spread across multiple cores using a thread pooling system.
Average CPU Framerate (render): The average framerate the render thread can achieve. The render thread encompasses all work needed to translate the simulation data to visual data that the GPU will process. The work is spread across multiple cores using a thread pooling system.
The "Latency" then, is actually the average time the CPU needs to complete one "game" or "render" thread. As you can see, that goes down as the RAM speed increases.
I gain over 30% fps in both game and render threads just by increasing RAM speed. There are a couple of different reasons that could happen. The Ryzen processor may be empting the thread pool as fast as it can be filled from RAM, meaning increases RAM bandwidth will improve performance until the processor's ability to execute threads becomes the bottleneck, at which point fps gains will plateau. Or it could be related to the infinity fabric speeds which run at RAM speed as well.
For the people who are still struggling, I was able to boot into Windows with 3466 1T, but it wouldn't pass HCI memtest. I wasn't stable in memtest until I set the command rate to 2T and the ProcODT to 48 ohms. There is a solid video where AMD's own rhallock discusses these features. I found it helpful.
With one dimm I have a stable system running at 2933 (R7 1700, Gigabyte Aorus GA-X370-Gaming K7 latest F4 BIOS Agesa 10.0.0.6, 2xRX 580). It does not boot at 3200. But the problem is when I use both dimms on dimm slots 1-2 or 3-4 (dual channel recommendation in the manual) I have random BSOD at 2133 and more frequent at 2933. Using CPU-Z Aorus version triggers the BSOD 50% of the time at 2933. I want to be sure what HW is bad to do a RMA and buy a replacement for the weeks I will have to wait. Any other tip is welcome.
Just to make sure I'm understanding correctly. The X370 only has 2 memory "channels", A and B. When using two DIMMs those should be populated in A2/B2 first (slots 2 and 4), followed by 1 and 3 (A1/B1) for a 4 DIMM set up. If you are putting your DIMMs in 1,2 or 3,4 you are actually loading both DIMMs on the A channel or B channel and putting nothing on the other, which would certainly cause problems.
Now for memory testing, you can also use GSAT for windows 10 (Google stress app). It is the application Google uses to test memory stability in their servers. It can be installed via the following method.
kxuping this is probably not related, but have you seen this post from jaandoh: Ryzen Master Utility Shows wrong clock speed for memory? Check your memory with the free version of Thaiphoon Burner. Also, update your CPU-Z version (if it's below 1.79.1) as it has caused some issues with SPD data corruption.
Now if you've tested both sticks individually and they check out fine, it may not be the RAM. Have tried flashing back to an older BIOS where you did not have these BSODs? As for memory testing tools, I prefer memtest86 as it's a low level test outside of the OS. It's slow (much slower than the built-in Windows Memory Diagnostic tool), but very effective at pinpointing errors and free.
I have only tested one dimm individually for 4 days without BSOD (gaming, web,benchmark). I will test it for 7 days and then switch dimm.
Yes I read your post about problems with CPUZ and SPD minutes after you posted. I always download the latest CPU-Z/GPU-Z/MSI Afterburner/HWinfo64.
The last BSOD were so bad that I had to reinstall windows 4 days ago. I always had BSOD even with the first BIOS, but were random and a few days between them.
This is my only DDR4 dimm system, so I cannot swap HW to test what is wrong. I like Memtes86+ for the same reasons you describe too, but you cant use your PC while testing. On previous HW I saw dimm, dimm slots and CPU problems with ram.
Are you always testing the single DIMM in position 1? What happens if you test it position 2? It is possible that there is a defect with the second channel. If your DIMMs pass testing alone in both channels I'd be hard pressed to blame the DIMMs.