I watched a Buildzoid video about installing 8000 MT/s RAM and he had some interesting things to say about it:
- The safest way is to start with a clean BIOS, but he also installed the kit without it and memory training did its job. I was not so lucky.
- 8000 RAM is the best for gaming, due to the lower latency of UCLK=FCLK combined with the higher memory bandwidth. There will be a performance hit (which I saw) by setting UCLK=MCLK/2 (which is necessary with RAM that fast.). But that is less important for most games. In my case MCLK=4000, UCLK=2000, FCLK=2000, and that’s optimal.
My 6400 RAM used an XMP profile with tightened timings. There was no Expo profile available.
I ended up setting it to 6000 MT/s and adjusted secondary timings. That was my baseline and was very stable. With that kit it was MCLK=3000, UCLK=3000, FCLK=2133. Setting FCLK higher did give me a measurable improvement. I tried FCLK=2200 with that kit and the system wouldn’t boot.
This new RAM (v-color Manta) has both XMP and Expo profiles available.
For sure! Like I said in the earlier post, everything was stable using Expo with no other tweaks.
I started messing with timings and had another HAL moment ("I’m sorry Dave, I’m afraid I can’t do that.) Q-flash didn’t even work to recover it, and I ended up having to put the computer on the bench and clear the CMOS. Ugh… 
I’ve been messing around with BIOS settings since, and I think that with 8000 RAM things are on a bit of a razor’s edge. I could downclock it in order to play with tightening up timings, but then I lose the FCLK=UCLK synchronization, which Buildzoid said is highly recommended.
One change I made that seemed to really help. My motherboard has two options that are supposed to improve DDR performance:
- Low Latency
- High Bandwidth
I’m not sure what either one actually does. ‘Low Latency’ had no measurable benefit.
But ‘High Bandwidth’ really improved my OCCT and AIDA benchmarks.
I had one pretty funny thing happen: I was running benchmarks and all of a sudden they tanked. I mean really tanked. Turns out Prime95 was still running, and using all 48GB RAM, even though I’d exited the program before running the next benchmark. Oops!
I think you’ll be very interested to see the data I’ve collected. Synthetic benchmarks are down a bit for the 8000 (Expo) vs. 6400 (downclocked to 6000 with tightened timings.) I kind of expected that with the asynchronous UCLK=MCLK/2.
I did some flights with the new setup, mirroring the exact same flight I’d done with the old RAM.
I saw what I’d hoped for - and bought this faster RAM to get - namely (at first glance) around 20% better 1% Lows.
I did 10 flights, and the 1% Lows were consistent and much improved. I need to tread carefully with trying to tighten timings further. Things look pretty darn good right now.
To reiterate: I’m running the sim @ 1080p, trying to stay CPU-bottlenecked.
I did have a few blue screens when loading Windows, and backed off of a couple of BIOS settings for the moment. I had set SPD PowerDown to ‘disabled’ and applied a PBO -20 all core offset. I changed PowerDown to ‘Auto’ and set PBO to ‘Auto.’
I’ll mess around with PBO later. Right now I’m seeing both CPU and GPU around 60°C during flight. Maybe I don’t understand PBO well enough, but I’m not sure I see the benefit of undervolting when I’m nowhere near thermal limits.