Does RAM Speed affect Plotting Speed on AMD?

high angle view of people on bicycle
Photo by Pixabay on Pexels.com

It’s only been about two months since MainNet launch on Chia and I feel like there is still much to be discovered with the plotting process. I recently decided to upgrade the RAM on my plotting machine from 32GB (4 DIMMS) to 64GB (2 DIMMS). I’ve learned in the past that when using Big DIMMS (32GB per stick in this case) that you need to pick RAM that is in the Qualified Vendor List (QVL) for the motherboard. In order to not get caught again with low RAM like before, I wanted to leave two slots available this time. This will leave an upgrade path to 128GB if that is needed in the future. This is why I choose two 32GB sticks. The kit that I went with was the Thermaltake TOUGHRAM RGB 64GB (2x32GB) DDR4 3600MHz C18 (amazon affiliate link). This was the only one I could find for my motherboard.

I got the kit and installed into the plotting machine. I did a single plot with the same settings I’ve done before (6 threads, 3416 RAM, Solo NVMe). To my surprise, my times improved as shown below:

Plot Times32GB
3200CL16
64GB
3600CL18
F1119s165s
Phase 14929s4680s
Phase 22627s2530s
Phase 35304s4995s
Phase 4387s372s
Total Time13248s12578s

There is roughly a 700s speed up. So what could be happening here? There are a couple things that can cause this:

  • AMD processors are known to perform better with faster RAM. Ryzen processors use Infinity Fabric which is tied to the RAM speed. Faster RAM results in faster Infinity Fabric speed.
  • This could also be an effect of going from 4 DIMMs to 2 DIMMs. It has been known that 2 DIMMs can overclock better than 4 DIMMs. Although I’m not over clocking here, just using XMP.

But there are more questions that needs more testing:

  • The RAM timings are worse with the 3600 Kit. Does this mean that RAM timings don’t matter that much? Is RAM MHz the king?
  • Will this improve parallel plotting times?

It takes a while to test these things so I will try and answer these questions. The reason for releasing this blog early was so that people know that if you are going to build an AMD plotting machine to get the fastest RAM that you can get. Right now, its really hard to hit 4000Mhz with RAM and have the Infinity Fabric match it at 2000Mhz, it seems like 3800 RAM is the sweet spot? AMD has said that it should be possible to do this, so they’re working on the AGESA fixes to get it there. You will have to do your own research on your specific motherboard and processor combo.

Another bit of information is that I’m currently testing a machine that has an AMD 5700G. I will release that information once I have it.

Keep on Farming on!

4 thoughts on “Does RAM Speed affect Plotting Speed on AMD?

  1. Hi Alex,

    Congrats on the upgrade.
    It will be interesting to watch your update on 2-thread parallel plot numbers.
    A quick question- are the two numbers for six thread from same chia version ?

    I had trouble getting my build to run the RAM (GSKIL RIPJAWS V 16 GB 3200 MHz which was on QVL) at the labeled speed. I too am not over clocking the CPU. However I ran into problems (computer was not getting past POST) when I tried DOCP (Direct Over Clocking Profile) setting, which I think is the AMD equivalent of Intel’s XMP. (I have an ASUS PRIME PRO X570). I checked with ASUS tech support and their faq* for trouble shooting and nothing solved my problem. I tried a different brand of RAM (Crucial Ballistix). Finally I just ran w/ bios optimized settings since I did not want to get behind. My RAM runs something like 2999 MHz even though it is labeled 3200 MHz. This is my first AMD build. I do not recall having this problem with my first intel build.(* https://www.asus.com/support/FAQ/1042256/).

    1. Hi Gigi, thanks! The versions were different, yes, v1.1.2 vs v1.1.5. I’ll have to check if any plotter improvements were made between those. My gut feeling says no, but I need to check. My X570 board says XMP in the BIOS, instead of DOCP. once you finish plotting, maybe tinkering with the ram settings is worth it.

  2. nice test.
    3200 CL16 and 3800CL18 have same nanoseconds latency times: 10ns.
    (see pic: https://www.profesionalreview.com/wp-content/uploads/2018/07/tabla-RAM-latencias-calculadas.png )
    as you said, 1800FCLK vs 1600FCLK is better for overall performance.
    In case of zen2 ryzen 3000 (i have 3900) better ram’s latency affects more the performance than your zen3 5900x.

    In zen3 case, More MHZ memory have more system performance. You can go High mhz, high latency in zen3, will perform better.

    Have a look https://www.youtube.com/watch?v=kEP8iVVQR-g
    In zen 3, performance at minute 15:10 (CL18 at different MHZ) Versus min 10:11(3600Mhz at different CL).
    For zen3 we should be go as high as Mhz posible, with 1:1 clk. Timings doesnt affect .

    PD : You still have “RAM: 2 kits of Crucial Ballistix 3200 2×8” in your /about url 🙂
    PD2: have a look your memory RANKS. can be 1 ranked or 2 ranked. Can use Thaiphoon to show all your ram specs.

    1. Hi lokus, thanks for the nice comment. The Crucial was single rank. This new kit is dual rank. Micron E vs Hynix A chips. Someone left an amazon comment saying that they got it to run 4133, so I will try to get it to 4000 to see if what it does to plot time.

Comments are closed.

%d bloggers like this: