I finally got my 3950X CPU, but I would like some advice

GWGeorge007
GWGeorge007
Joined: 8 Jan 18
Posts: 3043
Credit: 4948754354
RAC: 1211621
Topic 222310

I finally did it. I broke down and got a Ryzen 9 3950X CPU, and it took me a little while but I finally have my 2nd system up and running. Here's what I have for now with Hardware Monitor (v1.41.0):

I am presently running Windows 10 Pro w/latest updates, mostly because I am not yet comfortable with Linux (but I'm learning).

AMD 3950X CPU @ 3.7GHz @ 125W w/Noctua NH-U12A running @ 1900 RPM while Utilization is at ~85% (in BOINC I have it set to 80% CPU use, 25 threads).

G.Skill Trident NEO DDR4-3200 DRAM (CL-14) running @ 3200MHz (in BOINC I have set 50% memory usage when active).

2 - RTX 2070 Super GPUs w/driver version 432.00, both churning out work (in BOINC Manager I am presently running Einstein @ 0.9 CPUs + 1 Nvidia GPU on "1", and Milkyway @ 0.995 CPUs + 1 Nvidia GPU on "0").

Here is my start up log from BOINC Manager:

5/2/2020 7:11:31 PM |  | cc_config.xml not found - using defaults
5/2/2020 7:11:31 PM |  | Starting BOINC client version 7.16.5 for windows_x86_64
5/2/2020 7:11:31 PM |  | Libraries: libcurl/7.47.1 OpenSSL/1.0.2s zlib/1.2.8
5/2/2020 7:11:31 PM |  | Data directory: C:\ProgramData\BOINC
5/2/2020 7:11:31 PM |  | Running under account George
5/2/2020 7:11:32 PM |  | CUDA: NVIDIA GPU 0: GeForce RTX 2070 SUPER (driver version 432.00, CUDA version 10.1, compute capability 7.5, 4096MB, 3553MB available, 9062 GFLOPS peak)
5/2/2020 7:11:32 PM |  | CUDA: NVIDIA GPU 1: GeForce RTX 2070 SUPER (driver version 432.00, CUDA version 10.1, compute capability 7.5, 4096MB, 3553MB available, 9062 GFLOPS peak)
5/2/2020 7:11:32 PM |  | OpenCL: NVIDIA GPU 0: GeForce RTX 2070 SUPER (driver version 432.00, device version OpenCL 1.2 CUDA, 8192MB, 3553MB available, 9062 GFLOPS peak)
5/2/2020 7:11:32 PM |  | OpenCL: NVIDIA GPU 1: GeForce RTX 2070 SUPER (driver version 432.00, device version OpenCL 1.2 CUDA, 8192MB, 3553MB available, 9062 GFLOPS peak)
5/2/2020 7:11:32 PM |  | Windows processor group 0: 32 processors
5/2/2020 7:11:32 PM |  | Host name: GWG-PC
5/2/2020 7:11:32 PM |  | Processor: 32 AuthenticAMD AMD Ryzen 9 3950X 16-Core Processor [Family 23 Model 113 Stepping 0]
5/2/2020 7:11:32 PM |  | Processor features: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 htt pni ssse3 fma cx16 sse4_1 sse4_2 movebe popcnt aes f16c rdrandsyscall nx lm avx avx2 svm sse4a osvw ibs skinit wdt tce topx page1gb rdtscp fsgsbase bmi1 smep bmi2
5/2/2020 7:11:32 PM |  | OS: Microsoft Windows 10: Professional x64 Edition, (10.00.18363.00)
5/2/2020 7:11:32 PM |  | Memory: 31.91 GB physical, 36.91 GB virtual
5/2/2020 7:11:32 PM |  | Disk: 930.51 GB total, 860.13 GB free
5/2/2020 7:11:32 PM |  | Local time is UTC -5 hours
5/2/2020 7:11:32 PM |  | No WSL found.
5/2/2020 7:11:32 PM |  | VirtualBox version: 6.0.14
5/2/2020 7:11:37 PM | Universe@Home | General prefs: from Universe@Home (last modified 02-May-2020 16:09:39)
5/2/2020 7:11:37 PM | Universe@Home | Computer location: home
5/2/2020 7:11:37 PM | Universe@Home | General prefs: no separate prefs for home; using your defaults
5/2/2020 7:11:37 PM |  | Preferences:
5/2/2020 7:11:37 PM |  | max memory usage when active: 19608.07 MB
5/2/2020 7:11:37 PM |  | max memory usage when idle: 32680.11 MB
5/2/2020 7:11:37 PM |  | max disk usage: 93.05 GB
5/2/2020 7:11:37 PM |  | max CPUs used: 25
5/2/2020 7:11:37 PM |  | suspend work if non-BOINC CPU load exceeds 80%
5/2/2020 7:11:37 PM |  | (to change preferences, visit a project web site or select Preferences in the Manager)
5/2/2020 7:11:37 PM |  | Setting up project and slot directories
5/2/2020 7:11:37 PM |  | Checking active tasks
5/2/2020 7:11:37 PM | Einstein@Home | URL http://einstein.phys.uwm.edu/; Computer ID 12829820; resource share 2000
5/2/2020 7:11:37 PM | Milkyway@Home | URL http://milkyway.cs.rpi.edu/milkyway/; Computer ID 847486; resource share 2000
5/2/2020 7:11:37 PM | Universe@Home | URL https://universeathome.pl/universe/; Computer ID 559081; resource share 2000
5/2/2020 7:11:37 PM |  | Setting up GUI RPC socket

I would like to know what you think of my rig as it is now?

I would also like to know what to put in my cc_config.xml file since I haven't run an AMD processor before, nor have I run two GPUs before.

I would also like to know what else I should do for my running Windows 10 Pro to get the most out of my system when running BOINC.  I am running 3 projects as you can see, and I would like to continue running all three.  My other computer is still running fine (i7-990X w/Windows 10) with the same 3 projects.

Thanks in advance for any help, or even constructive criticism.

George

Proud member of the Old Farts Association

Keith Myers
Keith Myers
Joined: 11 Feb 11
Posts: 4959
Credit: 18640574537
RAC: 5365238

I would just use the Manager

I would just use the Manager to populate a stock cc_config.xml.  All you have to do is to change or add one Event Log logging option and a complete cc_config file will be created.  I think that adding the sched_ops_debug to the standard logging options is a must for all users.  That way, you will always know how much work you are requesting for both cpu and gpu for all projects.  Only adds a few lines of verbosity to the log so doesn't increase the busy-ness of the log.

Depending on the memory kit you have, you probably can coax more performance out of the memory because the Zen 2 cpus have a great IMC which controls the memory.  If you haven't discovered this resource, please download the DRAM Calculator and input your memory for some suggested timings for Safe and Fast timings. As you should know, Ryzen responds very well to increased memory clocks speeding up the the cpu.

DRAM Calculator forum and application links

Also you might try using a fixed cpu multiplier instead of relying on the Auto settings in the BIOS.  You can get a lot better overall all-core clock with much less cpu voltage if you move off from Auto.  By reducing the cpu voltage you can decrease the cpu temps and power consumption.  The Auto settings send way more voltage to the cpu for multi-cpu core computing which you are doing.  The Auto is best for single core gaming, not distributed computing.

Watch out how you control your memory usage and cpu usage when running cpu applications.  Some projects can consume a lot of memory for each cpu task.  Here in Einstein, I mostly see 2.1GB per cpu GW application task and 760MB per GR application task.  Max concurrent statements in an app_config.xml file can be helpful here in Einstein to avoid surprises. With as many threads as the 3950X has, it is a shame to not use as many as you can.  If you have enough memory.  That problem arose for me when I signed on to Rosetta and quickly exceeded the capacity of my initial 16GB which induced me to add another 16GB to my host.

Just for an example, these are my current system specs for my 3950X host.

CPU = 4100Mhz all cores at 1.22V Temps typically between 65-72° C. depending on ambient room temps.

Cooling is a 360mm radiator and custom loop.

Memory (32GB) = 3600Mhz at 1.42V and 14-15-14-14-28-42-288 Fast timings. G. Skill F4-3200C14D-16GTZ times two kits.

 

GWGeorge007
GWGeorge007
Joined: 8 Jan 18
Posts: 3043
Credit: 4948754354
RAC: 1211621

Keith, once again you come up

Keith, once again you come up with some very good advice and suggestions. I'm going to need to sleep on this one tonight and do some work tomorrow.

I am as always very greatful...

George

Proud member of the Old Farts Association

GWGeorge007
GWGeorge007
Joined: 8 Jan 18
Posts: 3043
Credit: 4948754354
RAC: 1211621

Keith, before I try out some

Keith, before I try out some of your tips on 3950X and memory overclocking I have a few questions.

The following info is from Hareware Monitor:

My G.Skill F4-3200C14-8GTZN runs in the mid-40s °C

My ASUS Prime X470-Pro shows CPU VCORE @ 1.199v, temps @ TMPIN0=68°C, TMPIN1=44°C, TMPIN2=52°C, while...

My AMD 3950X shows CPU VDD (Node 0) @ 1.081v, temp Package (Node 0) @ 69°C, and Powers Package @ ~126 W, Powers Cores (total?) @ ~79 W

My two RTX 2070 Super GPUs (Hareware Monitor does not identify which one) show one of them has GPU #0 Fan = 975 RPM with GPU #1 Fan = 0 RPM, and the other GPU #0 RPM = 1575 RPM with GPU#1 = 1575 RPM.  Both GPUs are using ~6% - ~12% memory.

My questions are:

Both GPUs have both fans running, can I assume the one with the higher fan speeds is the one closer to the CPU? (on top?), and the one with GPU #1 Fan = 0 RPM indicates that that one does not have the RPM sensor working correctly (or not at all?)

Is there a setting I can make that will allow my GPUs to use more than ~12% memory?

While each indivdual core has ~5.2W - 6.0W, just 5w x 16 cores = 80W (?), and the total(?) core wattage shows 79.02 W.  I'm so confused!!

With the max TDP of 105 W and maximum power dissipation is usually 20% - 30% more, or 30% = 136.5 W, how close to Max TDP should I go?

What is the difference between CPU VCORE and CPU VDD (Node 0)?  Should it matter?

What do the TMPIN0, TMPIN1, and TMPIN2 mean, or stand for, in the motherboard readings?  Does TMPIN0 = CPU temp on the motherboard, while TMPIN1 & TMPIN2 indicate different temps at different locations on the motherboard?

In CPU-Z under the Memory tag my DRAM frequency shows 1599.6 MHz (which I know is 1/2 the actual memory speed).  But my CAS# Latency (CL) shows "22.0 clocks", while my base memory has a CL14.  I'm curious as to why?  Also, is mid-40s °C an OK temp for memory?

Again, thanks in advance for any information you can provide me.

George

Proud member of the Old Farts Association

Keith Myers
Keith Myers
Joined: 11 Feb 11
Posts: 4959
Credit: 18640574537
RAC: 5365238

Well for starters you need to

Well for starters you need to use the correct screen for gpu utilization.  The one in Task Monitor does not show the correct screen and always throws confusion for most people when they ask how to increase utilization.

You need to change the screen view from graphics to CUDA with a right-click menu on the memory utilization screen.  You might have to do the same in your monitor program which I am not familiar with.

As far as temps are concerned I am more familiar with the Hwinfo monitor program which reports the official AMD temps at Tdie and Tctrl.  Tdie is the correct one showing how the cpu is going to respond to the thermal and power limits that are hard wired into the cpu.

I would say you are probably in good shape if the 68° C. is the same as Tdie.  Normally the gpu that is both feeding the screen and doing crunching is going to be hotter than a card that is just crunching.  Also the card at the top next to the cpu is usually more hampered in airflow than a lower card that is out in the open with full airflow to the card.  Remember in the typical vertically mounted mobo, that heat rises and the lower card dumps its heat output onto the cards above it.   Modern gpus don't even run their fans until the card hits some manufacturer specified temp limit in the attempt to produce a silent running card until it is stressed by several hours of intense gaming.

Did you set your memory to XMP timings in the BIOS?  It looks like you did not if the primary timings are showing 22 clocks. Your primary latency timing should be 14 clocks at XMP.  I think you are running your sticks at default JEDEC 2133 timings. I think this is your kit since you show 32GB.  You have dual rank sticks.

F4-3200C14D-32GTZN

Go into the BIOS on the Extreme Tweaker tab and select the AI Overclock Tuner function and change it from Default to D.O.C.P Standard setting and then save the settings and reboot.  You should see your RAM identified on the main page of the BIOS running at 3200 and 1.35V. If you select the Memory Timings selection you see the primary timings set just like it shows on the G.Skill product page I linked as 14-14-14-34.

These sticks are made with Samsung B-dies, the best overclocking memory so far produced for Ryzen.  Those sticks are easily made to run at 3600Mhz CL14 at FCLK=1800Mhz.  The RAM Calculator I linked in my previous post will show the memory timings you should plug into the memory timings page for 3600@FCLK=1800Mhz.

 

GWGeorge007
GWGeorge007
Joined: 8 Jan 18
Posts: 3043
Credit: 4948754354
RAC: 1211621

OK, thank you Keith.  I think

OK, thank you Keith.  I think I have a little more understanding of what I have been looking at.  Now I can begin my tinkering with overclocking and hopefully be a bit more successful than I am now.

Once again, as always, I am thankful for your generous time devoted to just me.  I can only hope that others will make use of this also.

George

Proud member of the Old Farts Association

Keith Myers
Keith Myers
Joined: 11 Feb 11
Posts: 4959
Credit: 18640574537
RAC: 5365238

I see I was wrong in my

I see I was wrong in my guesstimate of your RAM.  I didn't notice the explicit kit number for 8GB sticks in your post.  So you have 4 sticks of 8GB for a 32GB total.  That actually makes it simpler in some cases to set optimum timings with single rank memory.

I also skipped over your memory temps.  Stay in the 40's is the goal.  Once you move into the 50's, the overclocking gets harder and usually requires some active memory cooling with increased airflow to the sticks.  Either with a dedicated memory cooler fan or just an AIO cooler installed in the top which pulls air past the sticks to cool the radiator.

The new Trident Neo sticks are supposed to be pretty good overclockers.  I have the Trident Z predecessors without the RGB bling.  The RGB sticks run a bit hotter since they have all the active RGB LED's and driver components adding to the stick power.

 

GWGeorge007
GWGeorge007
Joined: 8 Jan 18
Posts: 3043
Credit: 4948754354
RAC: 1211621

Keith, I tried every which

Keith,

I tried every which way I could, but I evidently cannot get anymore out of my CPU and RAM, at least with the ASUS Prime X470-Pro motherboard.  One thing I did was update my UEFI BIOS from #5007 to #5406 (latest).

I finally got to 4.00 GHz on my CPU and DDR4-3200 MHz @ 1.350v on my memory with CL14.  My memory won't go any higher, and I think it is because it's "certified" at 3200 MHz and I don't have access to the XMP in my UEFI BIOS.  There is an XMP Rev. 2.0 for my RAM being used, but it is greyed out and I can't change anything with it.

My CPU on the other hand is running at 146W & 79°C @ 4.00 GHz on all cores with 1.850v.  I'm leary of running it higher.  Maybe if I get another motherboard.

The 2x RTX 2070 Super GPUs on the other hand need a little help.  They are both using 3.6% CPU and no more than 330MB of ram.  I know I'm still running Windows 10, Version 1909 (the latest).  What can I do to make them similar to yours?

George

Proud member of the Old Farts Association

Ian&Steve C.
Ian&Steve C.
Joined: 19 Jan 20
Posts: 3930
Credit: 46052852642
RAC: 64278524

1.8V on the CPU?? are you

1.8V on the CPU?? are you sure? that's like liquid nitrogen level voltages. if it's really running that, you need to change that like now. that's wayyyy too much.

_________________________________________________________________________

GWGeorge007
GWGeorge007
Joined: 8 Jan 18
Posts: 3043
Credit: 4948754354
RAC: 1211621

OK, I was wrong.  My CPU is

OK, I was wrong.  My CPU is running at 1.45v, not 1.80v, and I didn't change anything. I don't know what I was thinking. 

Thanks

George

Proud member of the Old Farts Association

Keith Myers
Keith Myers
Joined: 11 Feb 11
Posts: 4959
Credit: 18640574537
RAC: 5365238

Quote:My CPU on the other

Quote:
My CPU on the other hand is running at 146W & 79°C @ 4.00 GHz on all cores with 1.850v.

I'm positive you are not running 1.85V or the cpu would be dead.  I suspect that your HWmonitor program is misbehaving.

Good that you got your RAM at least running its rated XMP specification.  But that spec is just the minimum that the manufacturer tested for.  I am sure it will overclock higher being B-die.

But every BIOS is different for every manufacturer.  All the BIOS features are normally reserved for the higher price tier motherboards and the lower tier boards often have a very cut down BIOS with features completely missing or locked out.  Nothing you can do about that unless you are willing to use a modified BIOS that a overclocker has unlocked for your board.  But the X470 Prime Pro motherboard is NOT a lower tier board, more midlevel and quite capable of running your 3950X well if tuned correctly. You should ask for help in the forums for pointers on where to find things in the BIOS.

Have you joined the forum for that board yet?

My experience with the Asus PRIME X370-Pro 

ASUS PRIME X470-PRO (Issues and Successes)

You can start with the old Prime Pro X370 threads.  There is very little difference in the two generations.  More older information there. Then move on to the dedicated X470 Prime Pro thread

 

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.