Question Threadripper freezes when rendering using Adobe Media Encoder

[JediRyan]

So I built a 1950x TR with the intention of it being able to handle pretty much anything I throw at it editing wise. I was having the same freezing issues a while back and someone on another forum said that I was having these issues due to my cpu being undervolted. So I upped it from 1.350v to 1.375v, which made no difference, so I pushed it to 1.40v and everything seemed to work fine. I recently have been taking on heavier work loads doing batch renders now and the freezing issues have started happening again. I know the easiest solution would be to just undo my OC, but I feel like I would be losing out on quite a bit of performance if I drop my cpu from 4.0 ghz down to stock 3.4 ghz. I also like to game on my PC from time to time and there is a noticeable difference at stock speeds. I use an EK Performance 360 water cooling loop so I have adequate cooling, temps never go above 60c even under full load so I don't think heat is the issue. I am wondering if pushing the voltage to 1.425v is safe or should I be going at this a different way? I can also provide screenshots if needed, don't hesitate to ask! I would greatly appreciate any suggestions that I can test. Thank you for taking the time to read this!

 

[DrMrLordX]

The first two questions that come to mind are:

1). What stability testing did you do while dialing in your overclock?

2). Have you had any problems with Adobe Media Encoder at lower clockspeeds?

 

[JediRyan]

The first two questions that come to mind are:

1). What stability testing did you do while dialing in your overclock?

2). Have you had any problems with Adobe Media Encoder at lower clockspeeds?

1. I use CINEBENCH and OCCT for stability testing.

2. No problems what so ever at stock clock speeds

 

[DrMrLordX]

Try downclocking from 4 GHz to see if that improves stability, then. Maintain voltage, and lower clocks in 50 MHz increments. Use Adobe Media Encoder for your stability testing instead of Cinebench and OCCT. BTW, I do not think those are great stability tests myself. I would recommend a combination of Prime95 Blend, Prime95 Small FFTs, and y-cruncher (calculate pi to at least 2 billion places, or calculate to 500 million 5 times in a row).

 

[Markfw]

So I built a 1950x TR with the intention of it being able to handle pretty much anything I throw at it editing wise. I was having the same freezing issues a while back and someone on another forum said that I was having these issues due to my cpu being undervolted. So I upped it from 1.350v to 1.375v, which made no difference, so I pushed it to 1.40v and everything seemed to work fine. I recently have been taking on heavier work loads doing batch renders now and the freezing issues have started happening again. I know the easiest solution would be to just undo my OC, but I feel like I would be losing out on quite a bit of performance if I drop my cpu from 4.0 ghz down to stock 3.4 ghz. I also like to game on my PC from time to time and there is a noticeable difference at stock speeds. I use an EK Performance 360 water cooling loop so I have adequate cooling, temps never go above 60c even under full load so I don't think heat is the issue. I am wondering if pushing the voltage to 1.425v is safe or should I be going at this a different way? I can also provide screenshots if needed, don't hesitate to ask! I would greatly appreciate any suggestions that I can test. Thank you for taking the time to read this!

Put everything back to 100% stock and try it FIRST. If it works, then go up on the overclock until it fails.

 

[JediRyan]

Put everything back to 100% stock and try it FIRST. If it works, then go up on the overclock until it fails.

I have tried it at stock and it works without issue. For the amount of money I spent on this machine I should be able to reach 4.0, I'm not asking for 4.1 lol

 

[JediRyan]

Try downclocking from 4 GHz to see if that improves stability, then. Maintain voltage, and lower clocks in 50 MHz increments. Use Adobe Media Encoder for your stability testing instead of Cinebench and OCCT. BTW, I do not think those are great stability tests myself. I would recommend a combination of Prime95 Blend, Prime95 Small FFTs, and y-cruncher (calculate pi to at least 2 billion places, or calculate to 500 million 5 times in a row).

I am pretty sure that this is a memory issue because I only started having problems after upgrading from 32gbs to 64gbs of RAM. I bought the exact same kit and tested it with memtest86 to make sure there are no bad modules. I'm looking for some suggestions on things I can do to my memory because my CPU was completely stable at 4.0ghz with 32gbs of RAM. My logical way of thinking is that it has something to do with the memory settings, right?

 

[JediRyan]

Try downclocking from 4 GHz to see if that improves stability, then. Maintain voltage, and lower clocks in 50 MHz increments. Use Adobe Media Encoder for your stability testing instead of Cinebench and OCCT. BTW, I do not think those are great stability tests myself. I would recommend a combination of Prime95 Blend, Prime95 Small FFTs, and y-cruncher (calculate pi to at least 2 billion places, or calculate to 500 million 5 times in a row).

I used y-cruncher and this is what it showed, looks to me like a memory issue?

 

[Markfw]

I have tried it at stock and it works without issue. For the amount of money I spent on this machine I should be able to reach 4.0, I'm not asking for 4.1 lol

So because it won't overclock as much as you want it to, you are upset. I can't get ANY of mine 100% stable@4.0 ghz. 3.9 is the best for 24/7

 

[DrMrLordX]

I used y-cruncher and this is what it showed, looks to me like a memory issue?

Could be. y-cruncher does catch memory instability. Also make sure you run y-cruncher as administrator (though if you have problems there, it would simply fail to run at all).

If you suspect memory problems, you need to isolate CPU + cache stability from memory stability. Prime95 small FFTs should fit in CPU cache meaning it can test your CPU + cache without touching RAM. Prime95 Blend hits your RAM. So does y-cruncher. If you can pass Prime95 Small FFTs but not Blend or y-cruncher, then you are looking at RAM trouble.

You can run a dedicated RAM test as well (memtest86 is the classic, though you have to reboot to run that).

 

[JediRyan]

Could be. y-cruncher does catch memory instability. Also make sure you run y-cruncher as administrator (though if you have problems there, it would simply fail to run at all).

If you suspect memory problems, you need to isolate CPU + cache stability from memory stability. Prime95 small FFTs should fit in CPU cache meaning it can test your CPU + cache without touching RAM. Prime95 Blend hits your RAM. So does y-cruncher. If you can pass Prime95 Small FFTs but not Blend or y-cruncher, then you are looking at RAM trouble.

You can run a dedicated RAM test as well (memtest86 is the classic, though you have to reboot to run that).

My PC is stable through Prime95 tests, but still failing y-cruncher even when running as an admin. Pretty sure this is a memory issue. Wondering if I should be lowering the RAM speed or uping the voltage?

 

[JediRyan]

So because it won't overclock as much as you want it to, you are upset. I can't get ANY of mine 100% stable@4.0 ghz. 3.9 is the best for 24/7

Just tried dropping it to 3.9 and it made no difference.

 

[H T C]

Just tried dropping it to 3.9 and it made no difference.

Instead of dropping the CPU clocks or raising the CPU voltage, have you tried to slow the RAM speed instead?



This pic shows the non-OCed attainable RAM speeds for Zen 1 chips: it's outdated but it should get the point across: the more RAM you have, the more likely you'll have to run it @ lower speeds than the advertised on the kit.

 

[JediRyan]

I use Quad Channel RAM and I have dropped it from it's rated speeds of 3200mhz to 2133mhz and it didn't make a difference

 

[Markfw]

Just tried dropping it to 3.9 and it made no difference.

That was an example. Try the ram at 2933@1.35 volts, and the cpu@3.7 at whatever you think for vcore (go low to start) then, see what happens, and try raising the vcore until stable. Then try for 3750, 3800, 3850, etc.

 

[JediRyan]

I put the CPU at 3.9ghz and the dropped the RAM to 1866mhz and it finally passed a 10 minute test in AIDA64 and OCCT. It also completed y-cruncher at 2,500,000,000 but it failed at 5,000,000,000. Is that normal? I think I wasn't lowering the RAM enough as it says in this chart from overclock.net https://www.overclock.net/forum/attachment.php?attachmentid=85289&d=1518612831

I am going to bump the CPU up to 4.0ghz and see if it is still stable. Running the RAM at 1866mhz when it's rated for and worked at 3200mhz in 32gb makes me quite upset. I have the GSkill Trident Z RGB. This chart is for the GSkill Flare X, but all the timings and voltages are the same (I know it doesn't show a 64gb configuration) https://www.overclock.net/forum/attachment.php?attachmentid=85297&d=1518612831

 

[thigobr]

If you're still seeing issues at 1866MHz could be that the new kit has a defect. Can you test the new kit in isolation?

Other things to test in any Ryzen system are increasing SOC voltage, enable 2T command rate and GearDown mode

 

[DrMrLordX]

I am thinking the new kit might be defective.

@JediRyan try setting the CPU to 3.4 GHz and RAM to 3200 MHz and then run y-cruncher 5G and see if it passes. Up CPU clock until it fails (if it passes at all in the first place). I want to see how all this pans out.

 

[Markfw]

Wait, you do know that speed will show 1600 when its set to 3200, due to ddr ? So 1866 would be 3733 ?

I only say this as 2133 is the slowest you can set memory to in the bios.

 

[JediRyan]

Wait, you do know that speed will show 1600 when its set to 3200, due to ddr ? So 1866 would be 3733 ?

I only say this as 2133 is the slowest you can set memory to in the bios.

I know how it is displayed, however my BIOS allows me to set the memory speed as low as 1066 mhz.

 

[JediRyan]

I am thinking the new kit might be defective.

@JediRyan try setting the CPU to 3.4 GHz and RAM to 3200 MHz and then run y-cruncher 5G and see if it passes. Up CPU clock until it fails (if it passes at all in the first place). I want to see how all this pans out.

I really don't want to run the CPU at stock speeds. I tried bumping it to 4.0 GHz last night and it froze during AIDA64 Extreme stress testing. At this point I think I am done messing around for a while as long as it is stable.

 

[JediRyan]

I ended up dropping the CPU to 3.9ghz and the RAM to 1866mhz and now I'm stable in all stress tests. I was also able to successfully render 9 videos today without issue, so I guess now I just have to decide if the extra RAM is worth the sacrifice in memory speed. I may end up taking 32gbs out, putting the CPU back to 4.0ghz and the RAM back to 3200mhz, since it was stable and compare render times. I also want to compare the gaming performance of the two configurations and see what the results are.

I attribute my success thanks to the discovery of this chart: https://www.overclock.net/forum/attachment.php?attachmentid=85289&d=1518612831

 

[H T C]

I attribute my success thanks to the discovery of this chart: https://www.overclock.net/forum/attachment.php?attachmentid=85289&d=1518612831

That's a "more up-to-date" version on the pic i posted, since it does actually refer to ThreadRipper. Here is the pic again:



If the program(s) you use require lot's of RAM to be faster, odds are the 64 GB will be better, even @ slower speeds.

 

[Markfw]

I know how it is displayed, however my BIOS allows me to set the memory speed as low as 1066 mhz.

I know how it is displayed, however my BIOS allows me to set the memory speed as low as 1066 mhz.

1066 is 2133 real speed, the lowest. I think you are overclocking the ram./

 

[JediRyan]

[QUOTE="JediRyan, post: 39768729, member: 440134"I attribute my success thanks to the discovery of this chart: https://www.overclock.net/forum/attachment.php?attachmentid=85289&d=1518612831

That's a "more up-to-date" version on the pic i posted, since it does actually refer to ThreadRipper. Here is the pic again:

View attachment 4299

If the program(s) you use require lot's of RAM to be faster, odds are the 64 GB will be better, even @ slower speeds.[/QUOTE]

I plan to do the tests I mentioned next weekend, as tomorrow is St. Patrick's and I'm Irish