[SOLVED] Getting more out of the AMD 5950X above 5Ghz

[V8VENOM]

After reviewing several OC threads and YouTube videos and experimenting with various recommendations, I keep coming back to my simplest settings as provided the best stable results:

Reset HWINFO and run some benchmarks:

4.9-5.0Ghz on all cores, very stable (no Prime95 errors) and temps under control even at 1.494v.

So far the simplest OC approach seems to be the best. I've tried Ryzen Master and that produces lower performance results. I've tried manual tuning core/dram and so far I've not been able to beat the simple DOCP Standard and most everything else on Auto and RAM is being set to spec.

I'm curious to know how high on frequency can I go and at what point will I ever see temps closer to 90 C (which is about where I stop).

I should add I'm running my own custom Chiller setup (sub-ambient automatically control output according to environment dew point which in my part of the world at this time of year around 7 C.

Cheers, Rob .

 

[kerberos_20]

from your picture, youre hitting 140 ampers limits on EDC and TDC
until you rise that, it wont run any better, but enabling PBO voids warranty
but 1.5v rly? what kind of tasks are you running for all core boost?

 

[alceryes]

That's a great OC, but yeah, you're definitely givin' it a lot of juice! Can you post some 3DMark or Userbenchmark scores?

Note that temps should NEVER be the only factor in determining your 'high end' for overclocking. Just because it's stable and stays within acceptable temperatures at 5GHz and 1.5v doesn't mean you're NOT wearing down your CPU fast.
Even if you keep temps low I suspect that your CPU will show signs of degradation within 1 year at your current speed/voltage.

 

[V8VENOM]

Just to clarify, the values you see are simply a result of me setting up the "DOCP standard" for the RAM I have installed, nothing else (all Auto). I use Liquid metal on the CPU to coolant block for heat transfer and my vapor chiller unit is external to complete heat transfer. PBO is set to "Auto" so I'm assuming the logic (this is an Asus Crosshair VIII formula so VRM is chilled also) for Auto seems to be driven entirely on Temps ... does that sound like a valid assumption?

Most of my load testing tasks come from Prime95, 3DMark, PCMark 10, Cinebench R23, and CPU-Z Bench ... that's my "normal" suite or testing to determine thermals and/or performance improvements/degradation. For "normal" tasks the PC is built for Flight Simulation (MSFS, P3D, XP) which can use all available cores (primarily when loading scenery during flights) but there will be a higher load on the synchronizing main thread which can be designated to any one core.

3DMark for my vapor chilled AMD 5950X:

This is the chilled PC:

I'll continue with testing at PBO manual at higher values and see what I can do. FYI, I have no problem running 4.8 all cores at 1.28v (temps are lower).

FYI, I have a 2nd computer (my Dev PC) that is a dual radiator cooling setup (not chilled) and it performs lower (3DMark CPU score of 13276 on TimSpy) so it seems "temps" are extremely important for the 5950X "auto" settings.

As far as longevity, I've never had a CPU (AMD or Intel) fail from OC in my decades of doing this, but I do tend to swap out CPUs/Motherboard every 2 years.

Cheers, Rob.

 

[alceryes]

Your CPU score is fantastic but what was your GPU doing for the first ~40 seconds of 3D Mark?! Looks like it was sleeping.

 

[kerberos_20]

Your CPU score is fantastic but what was your GPU doing for the first ~40 seconds of 3D Mark?! Looks like it was sleeping.

thats more or less 25 seconds during demo loading

 

[alceryes]

thats more or less 25 seconds during demo loading

Ahhh, okay. I usually don't run the demo. (paid version)

 

[drea.drechsler]

...
I'm curious to know how high on frequency can I go and at what point will I ever see temps closer to 90 C (which is about where I stop).
....

It's not completely clear what you're doing. The screen shots of BIOS settings shows both CPU clocks and frequency settings are on AUTO, and the PBO section is collapsed so can't say if you've changed anything there nor elsewhere in AMD Overclocking/CBO sections of BIOS.

So if I understand settings correctly it's hitting almost 1.5V because the algorithm is doing it, with all settings on AUTO. If it's coherent with it's FIT parameters (one hopes it is) it should be safe at the low temps you're running. The sub-ambient cooling is holding temp down so it boosts higher and for longer and can use a higher voltage, right up to the 1.5V design limit. That's effectively overclocking while leaving the algorithm in full control so it really only boosts (overclocks) briefly and when it needs to while able to pull back on clocks and voltage when it doesn't.

Assuming you haven't I agree removing boost limiters (EDC and TDC in particular) with PBO is the only way to get more out of it. But if you do, also use the curve optimizer with some undervolting. I've seen marvelous results from that so could only surmise it would be better yet when holding temps in the 30's-40's under full boost in heavy workloads.

BTW, I would use Cinebench 23 for making your benchmark assessment of performance improvements. It may not show the higher clocks but being kept nicely chilled the algorithm will be able to hold cores at higher mid-range clocks throughout the BM run. That should result in very much improved scores.

If simply trying to hit highest possible clocks for gaming then look into disabling cores or use per-core and per-die overclocking tricks. There's no reason to get all 16 cores to the highest clocks since no game in existence is capable of using even half of them effectively. 1Usmus' overclocking utilitiy for Ryzen should be helpful for that.

Would be interested in your sub-ambient cooling arrangement. I don't see any insulation around the CPU block so how are you dealing with the bug-a-boo that seems to do them in? By that, I mean condensation of course.

 

[V8VENOM]

Thanks for the responses, my return to AMD has a learning curve. After some PBO curve tuning with PPT, TDC and EDC and I'm stable with Negative 30 on Scalar:

Under heavy loads temps are ok:

It seems AMD don't really allow anything beyond about 5.1Ghz or am I missing something? So how is der8auer getting up to 6Ghz (other than LN2 of course) ... is there some special Asus app and/or does it require the "Dark" series MB from Asus?

Would be interested in your sub-ambient cooling arrangement. I don't see any insulation around the CPU block so how are you dealing with the bug-a-boo that seems to do them in? By that, I mean condensation of course.

I've built my own external custom chiller and controlling circuit (around Arduino Uno board) to control condensation based on current humidity/temp values from internal and external sensors ... here are some videos from the unit I created using some off the shelf parts in combination with my own controller/design. My video (hopefully I don't get moderators deleting my account):

View: https://youtu.be/7TByB8CgS3A

View: https://youtu.be/7UmEU238d9I

View: https://youtu.be/_gYva2jom6A


Cheers, Rob.

 

[drea.drechsler]

Thanks for the responses, my return to AMD has a learning curve. After some PBO curve tuning with PPT, TDC and EDC and I'm stable with Negative 30 on Scalar:
....
It seems AMD don't really allow anything beyond about 5.1Ghz or am I missing something? So how is der8auer getting up to 6Ghz (other than LN2 of course) ... is there some special Asus app and/or does it require the "Dark" series MB from Asus?

....

I've built my own external custom chiller and controlling circuit (around Arduino Uno board) to control condensation based on current humidity/temp values from internal and external sensors ... here are some videos from the unit I created using some off the shelf parts in combination with my own controller/design. My video (hopefully I don't get moderators deleting my account):
....

I don't know that AMD expressly prevents it, it's probably just how the boost algorithm works in conjunction with the FIT operating parameters for assured life.

I'm pretty sure what der8auer does is with fixed frequency overclocks, dynamically applied after startup using Ryzenmaster. And of course, he starts it up with a heated pot and then cools it down to target temps with LN2 before laying on the insane voltage needed for the extreme clocks he uses.

I also remember Steve @ GamersNexus running experiments with the first Matisse chips (3900X/3950X probably). He found the boost algorithm works better with cold temps but even LN2 didn't get close to 5G at the time, for those CPU's of course. I have to imagine the same holds true with Vermeer today but using appropriate parameters.

That looks like a very clever cooling arrangement if it's able to keep the condensation from forming while offering subambient temperatures for cooling.

I'd strongly suggest to explore using PBO with the curve optimizer. What it does is lower the Voltage/Frequency operating curve for the processor...so for any given voltage it hits a higher frequency, or for a given frequency it uses a lower voltage.

The algorithm is still using temperature to govern boosting, though, so as long as temp is in the box it will boost merrily even under heavy loads. Ideal core temps are low 70's. Or lower of course: if your cooling is able to maintain Tdie in the mid 50's even under heavy load I'd like to imagine it keeps boosting as well as it does for light processing loads. That would be great since you don't have to dial in any more VCore to get it.

I just got a 5800X ($299 at Microcenter!) and been checking it out. It's now boosting to 5050 on best cores, the rest 5000. I'm only using a 240mm AIO for cooling. It's doing this without pumping more volts to the cores, just letting the algorithm manage it. Of course, when I run a super heavy load like Folding@Home it drops cores to 4725 or so since Tdie is now running in the 78-80C range. That's still holding max rated boost clock in heavy realistic workload so I'm happy, and this for only a few minutes work to achieve it.

But one word of advice: don't focus too much on frequency when letting the boost algorithm handle processor boost functions. Reason is it's so dynamic in the way it dithers clocks to manage temp you don't really get a clear picture of what's going on. The only way is to run repeatable benchmarks: CineBench 20 or 23 are the best. Also be sure to run both single thread and multithread BM's as it's easy to 'break' performance for one while maximizing the other.

EDIT added:
I suggest you check www.overclock.net AMD/CPU forums. There's a very active thread on Vermeer dual-die CPU's (5900/5950) on overclocking. There are differing strategies for getting the most out of those CPU's due to the two dies and more 'best cores'.

 

[V8VENOM]

I've been using CineBench R23 for Single and Multi-core testing. Prime95 for stability testing. My last tests results above were with the PBO curve optimizer under heavy multi-core loads.

If I turn up EDC I increase temps but testing results are lower with lower CPU frequency on all cores ... I balanced (PPT, EDC, TDC) those out with lots of trial to get the best possible frequency, but never really got sustained above 5050 with a few logs at 5100Mhz even with 57C max thermals.

It seems as if there is some other frequency road block?? I just don't know where to look.

My Silent Chiller has been in operational regular usage and extreme usage testing for 4 weeks now no condensation (cold but not too cold). I spent a day tuning my code/firmware for my Silent Chiller for optimal chilling with my AMD 5950X. I give myself about a 2C buffer from actual dew point (tolerance for error in the temp/humidity sensors I use). In my environment at this time of year the dew point has ranged from 4.4 C to 11 C. I don't allow the Vaper chiller compressor to run at 100% duty all the time, too much wear and would lead to early failure ... I am at 50% duty for the chiller and allow up to about 90% for short cycles.

I have allowed for a 2nd Vapor chiller option to be added to my Silent Chiller and run either serial or parallel ... can be used to reduce duty cycle or add more cooling potential for additional PC components (GPUs, CPUs, RAM, VRM, etc.).

Cheers, Rob.

 

[USAFRet]

I have a question....

Disregarding temps, and disregarding artificial benchmark numbers...

What specific performance benefit have you seen, vs a stock component?
Numbers, please.

Something like:
"Rendering this particular video took 45 secs, now it takes 30 secs."

 

[RodroX]

I have a question....

Disregarding temps, and disregarding artificial benchmark numbers...

What specific performance benefit have you seen, vs a stock component?
Numbers, please.

Something like:
"Rendering this particular video took 45 secs, now it takes 30 secs."

I would love to know this too please! I been thinking about upgrading my cooling system for a while now (I actually already started getting a better case, much better in terms of cooling, compared to my old one).

 

[V8VENOM]

It'll depend on your use case and environment (high humidity locations vs. low humidity locations) and if you want to use one or two vapor chillers. The primary objectives of my Silent Chiller are:

1. Sub-ambient cooling down to environment dew point but NOT below (avoid condensation issues)
2. Lower noise levels, Vaper Chillers can get very noise 64 dB open vs. 42 dB in my remote sound dampening enclosure (per my videos)
3. Plug and play
4. Simplified install were the case doesn't require space for radiators and pumps

As far as performance numbers, I do have two computers based on the AMD 5950X (both 64GB 3600 CL14 RAM) . One 5950X on non-chilled custom water loop vs. my 5950X using my Silent Chiller rendering the same video project with Adobe Premiere (set to Software not CUDA) to 4k UHD 60 , and the very same sample segment from the video I rendered above, took 23.28 minutes on my standard non-chilled 5950X and 18.19 minutes on my 5950X using my Silent Chiller ... about a 21% gain.

The application I currently have my Silent Chiller operating on is CPU, RAM, VRM and I have not yet explore benefits on the GPU side ... which if used set to GPU acceleration (CUDA) under Adobe Premiere I'd expect to see similar gain with having my Silent Chiller operating on the GPU.

Another application I plan to gain metrics from is 3DSMax 2019/2020/2021 animations to see the benefits from my Silent Chiller, but those 3D tools are also both CPU and GPU centric these days so it's hard to do a direct compare.

An additional application where "time is money" would be crypto-miners. If a crypto-miner could reliably run their GPUs at 20% or higher performance that would translate directly into profits provided they have their energy infrastructure correctly setup.

Since my Silent Chiller is component independent it's advantages will vary, for example the Intel 12900K has good overclocking room but fairly significant current (AMP) requirements which again is where my Silent Chiller would benefit.

Cheers, Rob.

 

[V8VENOM]

Here is one "use" case ... my flight simulator setup with a "station" operator PC combining two computers into a single case ... you can see the coolant lines running from my Silent Chiller to my computer(s) on the floor left side. The temperature/humidity sensor is located in this room and governs chiller output to prevent condensation in "my environment" which changes day to day.

The record lowest DEW point I've had in my location is 24F (-4.4 C) ... below freezing for water but not for the coolant I use which can remain fluid as low as 5F (-15 C). In my control unit I created, I also compensate my dual pump output (serial) based on water temperatures as the colder coolant gets the less viscous and requires more pump energy to maintain steady flow rate.

There are obvious an infinite number of applications for my Silent Chiller.

Cheers, Rob.

 

[drea.drechsler]

....
It seems as if there is some other frequency road block?? I just don't know where to look.
....

I have read that the highest freq. is 5050-5100. Not sure it's a cap as much as the way the algorithm works. So not being a cap it's not something that can be lifted as you'd probably have to alter the algorithm itself. Itself not easy to do but if it entails incorporating in the microcode I'm not sure it can be done post production. At the best it would have to incorporate it into an SMU and create a custom/hacked BIOS.

If anyone was doing this sort of thing it would the likes of der8auer, theStilt, Elmor or 1Usmus, overclockers who IMO have done the most to expand our knowledge of the inner works of Ryzen. Some of them have signed NDA's with AMD to get the scoop, so they're not going to do or say anything that will cross the line and get themselves in trouble.

 

[RodroX]

Here is one "use" case ... my flight simulator setup with a "station" operator PC combining two computers into a single case ... you can see the coolant lines running from my Silent Chiller to my computer(s) on the floor left side. The temperature/humidity sensor is located in this room and governs chiller output to prevent condensation in "my environment" which changes day to day.

The record lowest DEW point I've had in my location is 24F (-4.4 C) ... below freezing for water but not for the coolant I use which can remain fluid as low as 5F (-15 C). In my control unit I created, I also compensate my dual pump output (serial) based on water temperatures as the colder coolant gets the less viscous and requires more pump energy to maintain steady flow rate.

There are obvious an infinite number of applications for my Silent Chiller.

Cheers, Rob.

Thats an insane steup, congratulations. Im a truck simulator fan, but I don't have the money or the access to "pro" level wheels, pedals, etc (not available in my country). In any case I enjoy a lot using a normal Logitech G29.

I can only imagine what it feels to have all the controls, dials, sticks, screens and what not to control the aircraft.