Hi guys,
I have a Threadripper 3990x and under full load (rendering with V-Ray Next) it drops under the base clock (2.9 Ghz).
Im using a Gigabyte Aorus Master TRX40 Rev 1.0 with the latest bios update. 64 gigs of RAM.
Temps are good - in full load it rarely reaches 55' Celsius (cooled with liquid cooler).
It used to render above 3.0 Ghz (when full load 100%), I just recently noticed that it drops the clock. Happens the same with Corona Renderer as well.
I noticed that when i activate embree (which in v-ray is impletented by using SSE instructions) this happens. But if i disable embree the render slows down by 20% even if i get over 3.2 Ghz clock speeds all core.
There is no OC in BIOS only the memories are set to XMP profile.
Somehow it doesn't utilize 100% of the cores. It used to be 100% when rendering, not anymore.
I'm attaching a screenshot with HwInfo so you guys can have a look.
What should I do?
Thank you
Solved! Go to Solution.
Your processor is working perfectly normally with the stock settings. You're running a workload which draws more power, meaning the cores have to clock lower. If you want higher clock speeds, you're going to need to allow the processor to consume more power, which means either doing a fixed manual overclock or turning on PBO. The latter is a better solution, as it will continue to let the processor ramp higher than the all-core speed on fewer cores with lighter loads, while a manual overclock will not exceed the value you set.
PBO isn't actually overclocking, mind you. You are increasing the power limits only, and the CPU still decides how high it will clock. It will just be able to clock higher on all cores under heavy load. So long as you have sufficient cooling and are using a motherboard with sufficient power delivery (which should be true with all TRX40 boards), enabling PBO will net you more performance under all-core loads with no downside.
leoberti, please do not use HWinfo - it is not to be trusted. Please post a screenshot of Ryzen Master (RM) - simply drag-n-drop the image into your reply. Also please post more specifications - OS, power supply, video, etc. Thanks, John.
Thank you for the reply
My Pc Specs are:
MB: Gigabyte Aorus Master TRX40 Rev 1.0 with latest bios F5K from 2020/09/25
Graphics Card: Gigabyte RTX 2080 OC
HDD: 2 x 500 GB Samsung NVME 970 plus in RAID 0, 2x SHDD Seagate
RAM: 64 Gigs 4x16 GB G.Skill Ripjaws 3200Mhz CL14
Power Source: Seasonic Prime Gold 1300W
Cooler: Enermax Liqtech II TR4 (refurbished with coolant replaced).
OS: Windows 10 Enterprise 1909 with latest updates Build 18363.1256
PowerPlan: I'm using AMD Ryzen Balanced Power plan , but I tried with AMD Ryzen High Performance as well.
Sometimes it drops to 2.7Ghz with 93-94% all core utilization...
I'm attaching a Ryzen Master screenshot:
Thanks, leoberti. I did lots of things to view your image and cannot read it. Please post ONLY the RM image. You should update W10 to 20H2.919042.685 as I am running. Probably will not fix anything but I think it best to be up to date. I have the same cooler (280 mm) as you (3970X) and it has been RMAed twice. Hope it hangs in for you - and me too. The image looks like TDC, PPT and EDC are all red. This means the processor is almost throttling or actively throttling. The MB set limits can be changed but may be considered overclocking (warranty voiding). Please DL the latest QRG (Reference Guide) from the RM site - https://www.amd.com/en/technologies/ryzen-master
I also run the Ultimate power Plan.
I will give you a reference when I find it. Are you running your memory at 3200MHz? Use XMP in BIOS to do this. I run with NUMA mode enable. See the BIOS option in ACPI - worth a try. See https://forum.gigabyte.us/thread/9003/trx40-designare-numa-nodes When I see an RM image I can read I may have more suggestions. Here is mine:
Please run Cinebench R23 all cores. I get about 44,000 on all cores. Thanks, John.
Hey misterj,
I'm using the Liqtech TRII 360. And after the first RMA I have decided to fix it myself following these steps :
Works like a charm so far.
I can't update windows to 20H2 right now, the option is still not available for my PC yet. Only the advertise that it will come in the near future :).
I have also tried to use the Ultimate Power plan.. the results are the same.
It doesn't make any difference if I set the memory with D.C.P.O.P profile or manually set the timings to CL14. Anyway .. the memory is not overclocked (you know what I mean :) ).
And of course NUMA mode is enabled, otherwise it wont see all the cores.
I will have a look on the RM site. Thanks for that.
Here is a new screenshot with the RM only:
Thanks, leoberti. I do not have to run NUMA to see all my cores (32). What is D.C.P.O.P? Almost sounds like Asus's DOCP. Please do not run anything like that. Run only the XMP profile. I also recommend not running any MB applications. Your memory is running 3200 MHz as seen in RM (1600 MHz memory clock). Mine is also. Who will not allow you to update W10? I have been running the latest for some time. If you go here - https://www.microsoft.com/en-us/software-download/windows10 you can DL a MS updater (tool). Power plans make little difference. I still have not found the limit changes for you in RM for PPT, TDC and EDC. It seems to be under Precision Boost Overdrive, page 32. I will update you, if needed. It works weird - when PBO is selected in a Profile, all values (PPT, TDC and EDC go to the MB maximum limit). Please lower them to above where they are turning red but probably not near maximum. I used this once to really get a good Cinebench score. Please ask any questions. Thanks, John.
Watching the Enermax cooler video - very interesting. Thanks much, will not RMA again. They paid no shipping on the second RMA.
EDIT: Really cool cooler video.
AOC, Auto OverClocking also allows adjusting PPT, TDC and EDC.
XMP is Intel terminology. DOCP is the ASUS name for AMD's XMP equivalent. Enabling DOCP is exactly the same thing as enabling XMP, in concept.
Thanny, I totally disagree. XMP, defined and supported by Intel and implemented (burned into SPD) my the memory vendors is a standard. It usually works well. DOCP (Direct OC Profile) is defined and implemented by ASUS and is quite unreliable. I think it is better now than when ASUS first released it when it was a true disaster! I have advised users for some time not to use it. ASUS pushes it as an AMD version of XMP but it is not. I suspect AMD has nothing to do with DOCP. You might ask AMD. Enabling XMP and DOCP are not the same thing. John.
Thanks. I was referring to DOCP and not D.C.O.C.P or whatever i've written in the previous post. It doesn't matter if I use DOCP or just manually set my DRAM, the results are the same.
I feel a little bit uncomfortable to use PBO to solve this issue, when it should work with default settings.
I didn't try to search for 20H2 version of the windows, I just used the Windows Update feature inside my copy of windows. I believe the Enterprise version may not be ready for update just yet... I don't know. It doesn't matter, it worked very well couple of months ago. No issues at all. Then all the sudden I get these power restrictions... very frustrating.
Glad you liked the cooler video. Its easier than it looks to DIY. You just have to have a little patience, especially for the filling with coolant part :D.
Interesting.
Running CinebenchR23 keeps my CPU over 3.0 Ghz and everything seems to be 100%
Thanks, leoberti. Your processor is still throttling. Please post your R23 score. Thanks, John.
leoberti, I strongly recommend you contact AMD support - https://www.amd.com/en/support/contact-email-form
The reason messing with memory timing has no effect is because your problem is throttling due to PPT, TDC and EDC limits. I could be wrong, but I do not think you are going to get past this without increasing the limits. Please let us know what you learn. I would expect response to be slowed by the virus and holidays.
Please make sure you are running all fans and the pump at full speed. I run all on 12 Volts directly from the power supply.
Where did you get the DOCP? I thought it came with ASUS not Gigabyte (I have a Gigabyte TRX40 DESIGNARE).
Thanks, John.
Your processor is working perfectly normally with the stock settings. You're running a workload which draws more power, meaning the cores have to clock lower. If you want higher clock speeds, you're going to need to allow the processor to consume more power, which means either doing a fixed manual overclock or turning on PBO. The latter is a better solution, as it will continue to let the processor ramp higher than the all-core speed on fewer cores with lighter loads, while a manual overclock will not exceed the value you set.
PBO isn't actually overclocking, mind you. You are increasing the power limits only, and the CPU still decides how high it will clock. It will just be able to clock higher on all cores under heavy load. So long as you have sufficient cooling and are using a motherboard with sufficient power delivery (which should be true with all TRX40 boards), enabling PBO will net you more performance under all-core loads with no downside.
Thank you for the answers.
DOCP naming appears in my BIOS and its a Gigabyte motherboard. I'm pretty sure that the memory has nothing do to with my problem.
The only think that its related somehow is that increasing the memory frequency also increases the FClock which means more voltage to the infinity fabric => more heat /power into the CPU.
I have actually tested this in the past. Cinebench R15 and R20 score was higher with the memory set to default (2133 Mhz), because the infinity fabric voltage was automatically set to 0.85V. Now is at 1.05V.
Either way, in the past I've never seen clocks under the base clock no matter what I did to the CPU (Prime95, Rendering, Single core load, All core load, gaming...). This means something happened in the meantime: windows updates, rendering software update perhaps.
BTW, I'm not using any motherboard software. They are all crap. I've learned this the hard way long time ago.
FANs are adjusted in BIOS and the pump from the liquid cooler is directly connected to the 12V rail from the PSU.
When I'm rendering for long periods of time the FANS do ramp up and you can actually feel the heat exhausting from behind of the case.
I only did 2 runs of CB R23 and with all the monitoring software open ( HW and RM) it did something between 62-63k score. Don't know if its ok as I couldn't find any stock score for my CPU on the web.
I just did a CB R23 run with clean windows restart and just the CB open.
I have enabled PBO. The results are insane :)
I think i will limit the EDC , PPT and TDC in PBO, because these numbers are just....too high.
Found this website that did do a Cinebench R23 on your processor: https://www.cpu-monkey.com/en/cpu-amd_ryzen_threadripper_3990x-977
Yes I saw that but I'm pretty sure that the multicore score is with PBO enabled. It goes to 4.0Ghz all cores when you hit 76000 score.
I have imposed some limits in PBO: TDC = 250A (215A default), EDC=350A (300A default) and PPT=350W (280W default). In full load in 3ds max rendering it stays between 3.07-3.10 Ghz all core with 54'C temps. I also get a 67000 CB R23 score with these settings.
This pc should be reliable as it constantly does hours and hours of rendering jobs. So I think is very reasonable to leave it like this.
I will leave it like this for the moment and I will consider this problem solved.
Thanks everyone for all the help.
Couple of notes:
I have updated to the latest Ryzen Master version and it looks like I shouldn't have done that as it constantly crashes and freezes my pc. One time it even reset my BIOS to default settings.
In the overall scheme of things this issue that I had (and with your help I found a way to avoid it) shouldn't happen. A base clock is a base clock and a CPU should not go bellow it unless a thermal limit is reached. Somehow in my case power limits where reached before reaching the base clock. The AMD engineers should assure that within the default power limits + thermals, the CPU will be capable to reach its base clock.
I'm aware that there are all sorts of variables that must be taken into account. Such as BIOS auto voltage, which kinda sucks. Its set to AUTO in my case and it seems that All motherboards overvolt the CPU , SOC ... when set to AUTO. Capacitors, power delivery form motherboard VRM, clean electricity from PSU... these all might be a factor.
Also very poor written bios code. I've found a lot of very visible bugs in my BIOS. Gigabyte quality...
leoberti, you can decouple fabric/memory clocks in RM and test your theory. I strongly urge you to contact AMD support. We have seen many posts here about PBO being OCing or not. I take the position that PBO IS OCing. Only AMD can say for sure (officially). Limits set in PBO are set by the MB vendor. BTW, BIOS is released by the MB vendor not AMD - have always been suspicious of MB vendors. AMD does releases AGESA and other microcode. Here is my PBO limits (not applied):
Please post the setting you before you lowered them. My EDC limit is 960 Amperes. Thanks, John.
EDIT: Your Cinebench look good. I was thinking 80,000 or so to compare to my (3970X) 44,000.
I am running RM 2.6.0.1702 and Update says none. I am wondering if you need a later version of W10.
Thanny you are correct. Asus DOCP or Gigabyte EOCP or AMD AMP are all the same as XMP.
They all do the exact function of using the SPD found in all RAM to automatically Overclock the RAM.
I read several Threads and articles that all says that DOCP is the same as XMP. They both do the same function of automatically overclocking RAM according to the RAM's SPD. SPD which tells BIOS what speeds, including Overclocking, it can safely run automatically the RAM Stick in certain BIOS settings. It also tells BIOS other data about the RAM stick and its Characteristics and speeds it can safely run.
CPU-Z will list the SPD data and shows all Speeds that the RAM can run safely including Overclocking.
Here is mine CPU-Z showing what Speeds it can run safely including XMP. I have a Asus Motherboard by the way:
I have my RAM set for 3600 Mhz via BIOS Setting.
From Wikipedia concerning DOCP:
Extreme Memory Profile (XMP)
A similar, Intel-developed JEDEC SPD extension was developed for DDR3 SDRAM DIMMs, later used in DDR4 also. XMP uses bytes 176–255, which are unallocated by JEDEC, to encode higher-performance memory timings.[21]
Later, AMD developed AMP, an equivalent technology to XMP, for use in its "Radeon Memory" line of memory modules optimized for use in AMD platforms.[22][23] Furthermore, motherboard developers implemented their own technologies to allow their AMD-based motherboards to read XMP profiles: MSI offers A-XMP,[24] ASUS has DOCP (Dynamic Over Clock Profiles), and Gigabyte has EOCP (Extended Over Clock Profiles)
elstaci, you are wrong and so is Thanny. John.
Well since you ask everyone else to prove their opinions why don't you prove you opinion by posting links that show you are correct and we are wrong.
But since I already know you from past threads you can't prove anything validating what you say because you know you can't while I have done so many many times.
But its fine if you still want to believe 2+2=5 that is your right.
DOCP is the ASUS name for reading the XMP overclocking information stored in a DIMM. It's that simple. Supporting evidence has already been posted, so it's rather odd that you think you can simply say it's wrong. You are, in fact, in error.
Your screenshot clearly shows you're power constrained. The obvious answer, if you're comfortable with your cooling, would be to enable PBO so that it allows itself to draw more power.