List of Plotting Performance Tweaks

I thought all XMP did was run the ram at a compatible overclock? Basically with an OC the ram gets squirly. I should have noted this is only when plotting to a RAMDrive. Stable for cad, video editing and gaming. Just not sustained 90% ram use. As the the board it’s high end, the ram is 128 Gb Ripjaw and my drive is an PCIe enterprise SSD. I’m guessing the ram drive software is causing an issue with XMP and/or OC.

XMP is much more complicated that a simple overclocking.

That G.Skill RipJaws is a solid memory, so should not be causing problems (I actually also have it, and run it at 3600MHz - via XMP). As your mb is high quality as well, then you should not have any problems. I would run MemTest86 on that box, just to make sure that your RAM is not defective (takes long hours).

Also, what RAM drive are you using? Maybe that is the issue? I am plotting with ImDisk on that box, and have no issues. Although, I had to specify 112GB for it, as below that MadMax was also croaking on me for some reason.

1 Like

Thanks for the feedback. I’m using RAMDisk program for Windows10. I had it set for dynamic sizing. It weird in that I had a crash that was so bad it bricked the computer. Couldn’t even get fans to power on. So I stripped the motherboard, reset the BIOS, disabled XMP and set the RAMDisk allocation to “fixed” at 112 or something. Has been running stable 24/7 since. I’m blaming XMP and/or OC. Ut maybe it was the dynamic setting in RD? Thanks again for the feedback.

TLDR RAMDisk windows 10 disable XMP and set memory allocation to fixed for a stable ram plot.

Try that ImDisk. It is a very good program. Also, recommended by quite a few farmers.

Also, there is rather no point of setting RAM drive with dynamic allocation, as you know that it will need that 110GB with every plot. So, you just introduce an additional overhead, as the program needs to stall, every time that RAM drive is trying to increase the allocation.

It is also a bad sign, when a RAM drive tries to manage XMP, or actually any memory settings. The point of the XMP is that it should be transparent to the OS, so if a program is trying to fiddle with that, there are potentially plenty of ways to screw things up.

Maybe you can consider changing your TLDR to just “caution when using RAMDisk” :slight_smile:

1 Like

Good tips !! Im starting to think maybe the dynamic was the root cause. I do CAD, video editing and gaming on this PC hence the thought of dynamic. But its easy enough to disable. Happy plotting !

I am using Softperfect RAM Disk. It is only free for a trial period but works perfect for me since a month+ 24/7. I tried others as well but could no get them working stable.
I used fixed amount of mem no dyn. Obviously the machine then needs to swap a bit when using other application in parallel - but with fast SSDs not that big pain.

1 Like

Great recommendation ! So far RAMDisk is stable once I implemented fixed size. That dynamic is bad bad bad !

Are all of these points with respect to the built-in plotter? Because using something like MadMax would be far more effective

I have 128GB Ripjaw DDR4-3200 kit CL14-18-18-18-38

I OCed it to 3600 CL16-20-20-20-42

Plotting time:

before ~23 min
now ~22 min

~4.5% improvement

Since this kit is pretty much maxed out…any pimping of timing may render it unstable…corrupted plots for another 5% extra plot speed is undesirable for me.

Not to mention, finding limits of RAM takes days :wink: You have to think about profitability…if you spend 100 hours * 30 CHF (3000CHF) of your time for 5% increase in just 100 plots…well, it is better to focus on something else :wink:

2 Likes

Most important performance tweak is to use Linux, especially Clear Linux for max performance.

Nothing beats “from-box” performance than Clear Linux, fast NVMe (such as Corsair MP600, WD SN850…do not know if “modern” NVMe is faster) with XFS

1 Like

UPDATE few hours later

poll_queues in Clear Linux crashes MadMax plotter. No idea why. Did anyone managed to successfully use it?

Latest plotting record of ~19.7 min per plot

-u 512 (from default 256)
3600 CL16-20-20-20-40, FCLK forced to 1800 MHz
mitigations = off (little or no difference on latest Clear Linux)
/sys/block/nvm*/queue/scheduler = none (mq-deadline vs none almost no difference)

After two years of plotting, there is still no tourist SSD capable of sustainable write speed over 1500MBs. Samsung 970 Pro allegedly did 2GBs, but they are not sold anymore…980/990 Pro still 1500MBs.

RAID 0 has little or no improvements…never tried though, I have no second MP 600 2TB disk.

The only real improvements will be at least 330GB RAM with both temp, and final plot in RAM. Guys doing 12-15 mins. It would mean 3000-5000CHF investment in Threadripper…I need to plot just 140 TB right now…20 days with current setup, that is already paid off.

so MM, 2x2699v3, 352GB RAM total (8x32+6x16GB, I have to see if the server will accept, its what I have) can get me to a non-SSD solution? Can you elaborate on the 330GB setup and config?