Threadripper BUILDERS thread

[ub4ty]

Yes, I think it was 1.0 and I went to 1.3. Is that the RGB ram ? Like I said, that could be the issue.

Yep, it is RGB ram. Going to do the 1.0->1.3 update and see if this fixes the issue and then I'll go further :

• Bios update 1.0 -> 1.3
• 4x8GB RBG (3600@CL16 rated) stick test @3600 if fail :
  
• 2x8GB RBG (3600@CL16 rated) stick test @3600 (see if fail and then test other 2 sticks)
• If issues, I'll slap in 2 sticks of (2x8GB 3200@CL14) non rgb ram and bump it to 3600@CL16-16-16-36 and see what happens when I mix and match

Seems like a good way to try to isolate this issue...
Going to try ajc's test combo iterations.
Trying to avoid a return and downtime.

 

[ub4ty]

Also, I found improved scores with 3333@CL14 than 3600@CL16. Something you might check for your system.

Very nice, Yeah this is something I check for since my Ryzen 7 memory training/performance days. After I find the range of stability, I'll then check which setting is best for my workflows. The only reason I bought a higher speed rated ram is so I can more comfortably probe the ranges clock/timings. I'm guessing this is why you did the same?

 

[Markfw]

Welp, there it is, Flat out 100% CPU utilization 16-16-16-36 1T plus you have a serious OC on.
Can you grab a quick capture of your Wattage utilization? Did you change any of the power settings on your CPU in BIOS? Further, how is your OC enabled? From bios or using Ryzen master?

Wattage. From the wall using kill-a-watt, its 750. BUT thats with 2 x 1080TI@2ghz@100%load ! I think its idle@135 and 305@100%load CPU ONLY but don't hold me the that, I did the test a while ago. As I said, the only changes from default bios, are XMP profile, 3600 then came up by default, and 1.2 vcore, and 3900 speed. I didn't even enable any of the vdroop or whatever.

 

[ajc9988]

Thanks to everyone providing timely help as we try to narrow configs and brings some of our setups to stability.

Just poked my head in my case and checked. It is flipped [on]. Checked the mobo manual and it says this is the default setting.

I am going to try the bios update to 1.3 (best thing to do to start) and see if this brings stability. Will post updates as I will be trying to bring stability throughout the day.

So, when off, this takes off a lot of limits. If you don't know what you are doing (general warning, not specific to you), do not change it from default. This takes off the limits so that LN2, DICE, or phase change can be used without the current limits interfering. By turning it off, if you set a voltage too high our current too high, poof, there goes the game. But if you go allow, watch the voltages and current, as well as heat, that should side step the issue of shared current limit. Make sure you keep the VRM cool!

 

[Markfw]

EK will have a monoblock for the Zenith, the MSI, and the Gigabyte boards. Not the Asrock boards, which I have the Taichi.Nice. See my post above on HPET on and off on scores. On the win 10 ent. Fall creators update, the real time clock bug is present, although I don't get that warning from Realbench on the spring creators update, so it may just be that version being a beta. Also, I found improved scores with 3333@CL14 than 3600@CL16. Something you might check for your system.

I have to save up $4300 (or more) for property taxes, due in about 6 weeks. Once I pay that, I am going to REAL custom water cooling, like at least on 480 rad, and a case that will handle 2 of those, later I will do the GPU's. THEN I OC to 4.1 or 4.2 and tweak. ddogg already beat me to that.

 

[ddogg]

In that screen copy, 63c, at that temp I have found no problems. I did a CB15 run@4.1, but I am sure it was throttling, but I still beat that 7900X@6 ghz on LN2, so I am happy. I think it was like 3408, but with proper cooling I think I can do 3500 and 4.2

I just tried 4.2 and it was a near instant crash at 1.38 when running cinebench. This likely means I would need 1.4V and my temps are probably going to be in the high 80s. I'll keep trying and post if I can get any sort of stability.
So far 4.1 is rock stable with 3443 CL16 ram speeds.

 

[ub4ty]

So, when off, this takes off a lot of limits. If you don't know what you are doing (general warning, not specific to you), do not change it from default. This takes off the limits so that LN2, DICE, or phase change can be used without the current limits interfering. By turning it off, if you set a voltage too high our current too high, poof, there goes the game. But if you go allow, watch the voltages and current, as well as heat, that should side step the issue of shared current limit. Make sure you keep the VRM cool!

Yep, VRMs hit 47*C on a short run of Prime95 with AIO .. hit 44*C with Noctua.
One thing I noted is, at boot into windows, with no OC, the CPU clocks read 4.0ghz and 3.7 and float around then drop back to 2125Mhz. Within this time I see voltages on CPU swing from 0.85 to 1.44V. I've noted this and another user has as well as a youtube reviewer. I think this is the CPU doing a probing run to see what its ranges are?

Once this settles, the voltages stay in normal windows. That being said, when you push the CPU, I do so it swing up to 1.40V from time to time. I think this is also normal. I am watching all of my temps/settings like a crazed mother during this testing.. haha, so indeed I'm being quite watchful.

Updating to 1.3 Bios as we speak and then going to try out 3600CL16-16-16-36 RAM. Super jelly that mark got this working out the box . I was hoping for the same last night and then bootloops. So, hunkering down.

 

[Markfw]

I just tried 4.2 and it was a near instant crash at 1.38 when running cinebench. This likely means I would need 1.4V and my temps are probably going to be in the high 80s. I'll keep trying and post if I can get any sort of stability.
So far 4.1 is rock stable with 3443 CL16 ram speeds.

When you get a chance, tell us your CB15 score@4.1 and temps (max)

 

[ajc9988]

Wattage. From the wall using kill-a-watt, its 750. BUT thats with 2 x 1080TI@2ghz@100%load ! I think its idle@135 and 305@100%load CPU ONLY but don't hold me the that, I did the test a while ago. As I said, the only changes from default bios, are XMP profile, 3600 then came up by default, and 1.2 vcore, and 3900 speed. I didn't even enable any of the vdroop or whatever.

That is a Hell of a Vcore, as the LLC is set at level 5 by default (level 1 is where it matches what you set, level 5 has the droop). No I'm jealous, as my chip needed 1.225 to get that. And 3402@3.9GHz is a good score. You are doing it right. Give your thanks to chew at xtremesystems or overclock.net, he helped to hone this bios/uefi.

Very nice, Yeah this is something I check for since my Ryzen 7 memory training/performance days. After I find the range of stability, I'll then check which setting is best for my workflows. The only reason I bought a higher speed rated ram is so I can more comfortably probe the ranges clock/timings. I'm guessing this is why you did the same?

Yep. When 3600 costs about $200 for a kit and 4133 for $220, I'll pick the faster ones to give me more room to play. If you can't hit the rated speed, you can always hit the lower one at tighter timings...

 

[ajc9988]

Yep, VRMs hit 47*C on a short run of Prime95 with AIO .. hit 44*C with Noctua.
One thing I noted is, at boot into windows, with no OC, the CPU clocks read 4.0ghz and 3.7 and float around then drop back to 2125Mhz. Within this time I see voltages on CPU swing from 0.85 to 1.44V. I've noted this and another user has as well as a youtube reviewer. I think this is the CPU doing a probing run to see what its ranges are?

Once this settles, the voltages stay in normal windows. That being said, when you push the CPU, I do so it swing up to 1.40V from time to time. I think this is also normal. I am watching all of my temps/settings like a crazed mother during this testing.. haha, so indeed I'm being quite watchful.

Updating to 1.3 Bios as we speak and then going to try out 3600CL16-16-16-36 RAM. Super jelly that mark got this working out the box . I was hoping for the same last night and then bootloops. So, hunkering down.

Remember, at stock, this CPU does 1-2 cores up to 4.2, 4 cores at 4.0, and all 16 cores to 3.6 (although I see up to 3.7 all cores sometime, or it is switching then out fast enough to look like it). The lower range is checking lower p states and core parking.

As to the voltages, it will change by number of cores engaged and the p state, so it seems like cycling, as you mentioned, or just entering them at different boot points.

 

[ddogg]

When you get a chance, tell us your CB15 score@4.1 and temps (max)

This is the highest I'm getting right now at 3433 RAM. I was actually able to boot once at 3600 and got close to 3500 but considering that memory speed is not really stable for me this is where I'm at right now. Temps peak around 78-80C after about 30 mins or so with Prime. Pretty acceptable temps to be honest. I'm trying 4.2 at the moment just to see if I can get a cinebench run atleast

 

[ub4ty]

That is a Hell of a Vcore, as the LLC is set at level 5 by default (level 1 is where it matches what you set, level 5 has the droop). No I'm jealous, as my chip needed 1.225 to get that. And 3402@3.9GHz is a good score. You are doing it right. Give your thanks to chew at xtremesystems or overclock.net, he helped to hone this bios/uefi.Yep. When 3600 costs about $200 for a kit and 4133 for $220, I'll pick the faster ones to give me more room to play. If you can't hit the rated speed, you can always hit the lower one at tighter timings...

Quick update.
I updated the mobo bios to 1.3 (1.0 Stock shipped version to 1.3)
Stock CPU settings, I set the memory to 3600 XMP profile which is 16-16-16-36 timings.
I got boot loops and it wouldn't take.
I then went in and set the CPU to 3700Mhz w/ a fixed voltage of 1.21V
It took the setting w/o a single boot loop and booted to windows. I ran Cinebench, score was 3200 first run 2192 second run. CPU popped up to 55C from 31C idle. Room temp is 78.8F/26F. Power usage idle is 47W .. Jumps to 218W upon running cinebench

Someone correct me if I'm wrong but it seems, from what I saw the CPU doing when I booted into windows at stock that the CPU probes speed/voltage ranges. I'm assuming, if you have the CPU at stock settings and you OC your ram that heavily up to 3600Mhz then you likely hit instability when the CPU tries to balance out the extra wattage from the increased infinity fabric and the memory controllers when it tries to probe the speed range but with an 180W lock limit on the power usage. Once you OC the CPU and set the voltage to 1.2V , it disables the wattage lock and allows for a higher floor voltage. I guess this is also what the Level settings are for in the Taichi bios, to ensure the voltage doesn't droop at any point causing instability especially when you're causing far more draw by having a high memory setting.

Running Passmark burnintest now

 

[ddogg]

Quick update.
I updated the mobo bios to 1.3 (1.0 Stock shipped version to 1.3)
Stock CPU settings, I set the memory to 3600 XMP profile which is 16-16-16-36 timings.
I got boot loops and it wouldn't take.
I then went in and set the CPU to 3700Mhz w/ a fixed voltage of 1.21V
It took the setting w/o a single boot loop and booted to windows. I ran Cinebench, score was 3200 first run 2192 second run. CPU popped up to 55C from 31C idle. Room temp is 78.8F/26F. Power usage idle is 47W .. Jumps to 218W upon running cinebench

Someone correct me if I'm wrong but it seems, from what I saw the CPU doing when I booted into windows at stock that the CPU probes speed/voltage ranges. I'm assuming, if you have the CPU at stock settings and you OC your ram that heavily up to 3600Mhz then you likely hit instability when the CPU tries to balance out the extra wattage from the increased infinity fabric and the memory controllers when it tries to probe the speed range but with an 180W lock limit on the power usage. Once you OC the CPU and set the voltage to 1.2V , it disables the wattage lock and allows for a higher floor voltage. I guess this is also what the Level settings are for in the Taichi bios, to ensure the voltage doesn't droop at any point causing instability especially when you're causing far more draw by having a high memory setting.

Running Passmark burnintest now

I'll give this a shot. I have the 1.3 BIOS but haven't set a fixed voltage for the CPU.

 

[ub4ty]

Remember, at stock, this CPU does 1-2 cores up to 4.2, 4 cores at 4.0, and all 16 cores to 3.6 (although I see up to 3.7 all cores sometime, or it is switching then out fast enough to look like it). The lower range is checking lower p states and core parking.

As to the voltages, it will change by number of cores engaged and the p state, so it seems like cycling, as you mentioned, or just entering them at different boot points.

K, you have a solid understanding of this. Question : Every reviewer (when the CPU was at stock settings) noted poor performance in certain heightened RAM clocks when they started ratcheting up the RAM clock timings above 2133. This is likely due to the 180W ceiling the CPU imposes whereby at higher ram clocks, it uses more power and thus must cut power to cores and not clock them as high to stay within the stock 180W ceiling.

I think anandtech defined the phenomenon as operating exactly like this. Higher RAM clocks when the CPU is stock starves the cores for power. Thus, to keep things balanced, you essentially have to OC in order to get things back in balance. I stated prior w.r.t to boot loops and instability that my bios OC appeared not to be taking. I am not sure what allowed for it to take but it took after I did the 1.0 -> 1.3 update. When I didn't OC the CPU I got boot loops and 3600@CL16 XMP setting wouldn't take. After I applied a mild OC to the CPU 3700Mhz @1.21V the memory settings took, i booted to windows, ran cinebench, am now running burnintest and will switch to more extreme testing.

So, is the instability due to the CPU doing probing of the voltage range/clocks on Post and in windows with a 180W limit but being unable to supply sufficient settings to run the RAM at 3600Mhz while giving adequate power to the cores simultaneously? Trying to root cause why stock CPU + 3600RAM @CL16 is unstable.

 

[ajc9988]

This is the highest I'm getting right now at 3433 RAM. I was actually able to boot once at 3600 and got close to 3500 but considering that memory speed is not really stable for me this is where I'm at right now. Temps peak around 78-80C after about 30 mins or so with Prime. Pretty acceptable temps to be honest. I'm trying 4.2 at the moment just to see if I can get a cinebench run atleast

Granted, I do a separate, bare OS for OC and tweaking timings, separate from the main one. But I wonder, what do you get at 3.9GHz in CB? Try that with lower voltage (like 1.2-1.3 volts) if you could. Also, I'll admit for 3382 score @3.9, I set priority to high (one under real time, which will shake your system if not fully stable). But you should be at least around 3340s without the priority setting if you don't have anything really doing background tasks without affinity.

I also do all OC in the UEFI, not the software.

 

[ddogg]

So, it took the 3600 the first time after setting a fixed voltage. It's back to boot loops again. Sigh. Going back a notch until it's verified 3600 is stable for multiple people.

 

[ajc9988]

So, it took the 3600 the first time after setting a fixed voltage. It's back to boot loops again. Sigh. Going back a notch until it's verified 3600 is stable for multiple people.

Don't give up yet. You have to find the timing that is giving the problem. What code does the board sick on before rebooting?

Quick update.
I updated the mobo bios to 1.3 (1.0 Stock shipped version to 1.3)
Stock CPU settings, I set the memory to 3600 XMP profile which is 16-16-16-36 timings.
I got boot loops and it wouldn't take.
I then went in and set the CPU to 3700Mhz w/ a fixed voltage of 1.21V
It took the setting w/o a single boot loop and booted to windows. I ran Cinebench, score was 3200 first run 2192 second run. CPU popped up to 55C from 31C idle. Room temp is 78.8F/26F. Power usage idle is 47W .. Jumps to 218W upon running cinebench

Someone correct me if I'm wrong but it seems, from what I saw the CPU doing when I booted into windows at stock that the CPU probes speed/voltage ranges. I'm assuming, if you have the CPU at stock settings and you OC your ram that heavily up to 3600Mhz then you likely hit instability when the CPU tries to balance out the extra wattage from the increased infinity fabric and the memory controllers when it tries to probe the speed range but with an 180W lock limit on the power usage. Once you OC the CPU and set the voltage to 1.2V , it disables the wattage lock and allows for a higher floor voltage. I guess this is also what the Level settings are for in the Taichi bios, to ensure the voltage doesn't droop at any point causing instability especially when you're causing far more draw by having a high memory setting.

Running Passmark burnintest now

Something isn't quite stable yet, as there should be a 5-10 point change from first to second run, with 20 points at the outside, unless you have programs that kicked on in the background.

I suggest start with a known stable 3200 ram oc or use 2133 stock on ram and find your CPU full stable OC first. Then move on to playing with the ram with the cpu oc in place, that way you find stable for your daily OC and rule out CPU instability from the ram.

K, you have a solid understanding of this. Question : Every reviewer (when the CPU was at stock settings) noted poor performance in certain heightened RAM clocks when they started ratcheting up the RAM clock timings above 2133. This is likely due to the 180W ceiling the CPU imposes whereby at higher ram clocks, it uses more power and thus must cut power to cores and not clock them as high to stay within the stock 180W ceiling.

I think anandtech defined the phenomenon as operating exactly like this. Higher RAM clocks when the CPU is stock starves the cores for power. Thus, to keep things balanced, you essentially have to OC in order to get things back in balance. I stated prior w.r.t to boot loops and instability that my bios OC appeared not to be taking. I am not sure what allowed for it to take but it took after I did the 1.0 -> 1.3 update. When I didn't OC the CPU I got boot loops and 3600@CL16 XMP setting wouldn't take. After I applied a mild OC to the CPU 3700Mhz @1.21V the memory settings took, i booted to windows, ran cinebench, am now running burnintest and will switch to more extreme testing.

So, is the instability due to the CPU doing probing of the voltage range/clocks on Post and in windows with a 180W limit but being unable to supply sufficient settings to run the RAM at 3600Mhz while giving adequate power to the cores simultaneously? Trying to root cause why stock CPU + 3600RAM @CL16 is unstable.

That sounds like a firm hypothesis that needs further testing for confirmation!

 

[ddogg]

Granted, I do a separate, bare OS for OC and tweaking timings, separate from the main one. But I wonder, what do you get at 3.9GHz in CB? Try that with lower voltage (like 1.2-1.3 volts) if you could. Also, I'll admit for 3382 score @3.9, I set priority to high (one under real time, which will shake your system if not fully stable). But you should be at least around 3340s without the priority setting if you don't have anything really doing background tasks without affinity.

I also do all OC in the UEFI, not the software.

3324 @3.9. Didn't set priority to high. Scores fluctuate a little from test to test. I also do all my OCing in the BIOS.

 

[ajc9988]

3324 @3.9. Didn't set priority to high. Scores fluctuate a little from test to test. I also do all my OCing in the BIOS.

Fluctuation is fine. If you drop more than 100 points at the same multiplier, then it is time to see what happened. Event viewer becomes your friend. If nothing processes in the background, then it's looking for the cause. I take it this is an OS with your normal programs installed, not a clean install, you're down. Not a bad score.

 

[Markfw]

3324 @3.9. Didn't set priority to high. Scores fluctuate a little from test to test. I also do all my OCing in the BIOS.

hmmmm faster ram, same speed (3.9) and I get 3294

Edit: just ran it again, 3300

 

[ajc9988]

3324 @3.9. Didn't set priority to high. Scores fluctuate a little from test to test. I also do all my OCing in the BIOS.

To give you an idea, I'm at 3.85@1.2V, 3333@CL14 16 16 34 1T, have an excel spreadsheet open, 3 word documents, chrome with multiple tabs, Firefox with multiple tabs, a pdf reader open with a textbook loaded, steam and one drive in the tray, no priority set, and got 3211 for my CB15 score just now.

 

[ddogg]

hmmmm faster ram, same speed (3.9) and I get 3294

Edit: just ran it again, 3300

Yeah, I think our scores just vary by 25-50 points or so. I ran it maybe 4 times and got 3 scores above 3300 and one in the high 3200s.

 

[ajc9988]

hmmmm faster ram, same speed (3.9) and I get 3294

Edit: just ran it again, 3300

There is variance between OS build and installed programs. This can make a huge difference, especially if internet apps are open. But the 94 to 3300 is the type of variance that is statistically insignificant between runs.

 

[Markfw]

There is variance between OS build and installed programs. This can make a huge difference, especially if internet apps are open. But the 94 to 3300 is the type of variance that is statistically insignificant between runs.

Well, that makes me feel better. Also, the memory (me at 3600) vs 3466 does not seem to change the score. I just went with 3600 since thats what the motherboard was rated for.

 

[ub4ty]

This is the highest I'm getting right now at 3433 RAM. I was actually able to boot once at 3600 and got close to 3500 but considering that memory speed is not really stable for me this is where I'm at right now. Temps peak around 78-80C after about 30 mins or so with Prime. Pretty acceptable temps to be honest. I'm trying 4.2 at the moment just to see if I can get a cinebench run atleast

Since me and you share a similar issue, can you provide a link to newegg for the exact ram package you have?