After a year of wanting to learn how Ryzen Master and all the OC functions works, I tried right now because of your nice explained videos. Thank you! If I’m happy with the results, I’ll try to OC my RAM.
@@sup3rnov400 Hi, I am two minutes into your video and already have two questions. What is the curve optimizer supposed to achieve, and what is this "offset".? when I get a reply, I will watch the rest of the video. Thanks.
@@sup3rnov400 Thanks, but I still do not know what a "negative offset" actually is, and it is sadly missing in this video. I installed RyzenMaster and had a look around and found nothing that I was comfortable using, mostly because it assumed that I (being the Master) knew what I was doing, and the information given within the program (tooltips) was essentially useless.! . It is obviously not your fault, not your program, but what I would have loved to see was a basic overview of what the meanings were being used, such as: offset, mv, and so on. I am not suggesting that you re-do this video, or make a new one, but IMHO it would be super useful for noobs to get the absolute basics right at the start, I have no doubts that many people found this video and have no idea what OC stands for, what an overclock is, or why they are even watching the video, except that a friend suggested they use RyzenMaster, and they found themselves watching a video and was overloaded by acronyms and never went further because they don't watch foreign language videos. Just someone trying to help.
Thanks for actually explaining WTH curve optimizer actually is. Everyone tells you their settings but I wanted to know what they actually meant. Now I do! much appreciated.
hey, I know this video is a year old but like myself and many many other people that are looking for insight on the subject your video was very helpful. as in how you explained everything and the amount of time you took to explain everything. So thank you very much!
Well..I just found out about your video. It helped me a lot. My R9 5950x was so unstable @ -30 PBO on all cores, while streaming and gaming at the same time. You light up a highway for me. Thanks
This was hands down the best video on this topic. You explain this in a very simple and understandable way. hats off to you, my god man. You just earned a subscriber!
FYI - I first did a BIOS all core negative undervolt of 30. The system would crash when loading windows, I then changed all core to 25 and would get idle crashes. I then used OCCT to stress test the CPU and it told me that 2 of the 8 cores had errors. So back in the BIOS I increased to power to the problem cores to eventually get a stable undervolt with no errors in OCCT but when gaming for 5-10min I was getting pc restarts. So if anyone is experiencing the same issue, you need to use WHEA within Windows Event Viewer and look for WHEA error logs and that was telling me that a further 2 cores had power issues with OCCT didnt detect. I finally got a "perfect" undervolt for my system ranging for 20 down to 5 within the BIOS and a peak clock speed of 5.3GHz on my 5800x and average of 4.8GHz at low 70c.
How do i go about viewing this in windows 11? with the OCCT program theres an option in the settings that stops the test if there are any WHEA errors as well as the other ones. I have all 24 per cores set all at -30 offset and ran three tests for 10 minutes a piece. No errors. I have a 5900x cpu paired with a 5700xt -3200mhz ram everything on auto except for curve optmizers and AMD Smart Acess.
I’m running total stable on my R7 7800X3D with per core (-) core 1@ -25, 2@ -20, 3-8@ -30 (-28 average all core) Used OCCT CPU stability to dial everything in- worked flawlessly.
Man your video deserves to be on top list of Curve Optimizer. Very well explained tutorial. I went crazy for the sudden blackouts, restarts, crashes of my pc while not understanding what's going on, but I have found out that -30 on all cores on Ryzen 5600x is killing it. Thanks for your efforts!
Me too. I just followed the standard "logic". It simply does not work. PC might be stable under 100% load for days on end but will fall over even at the lightest load.
I've watched many Curve Optimiser videos online, however only your video made sense and answered the questions I had. Thank you! Looks like you've healed up well after the surgery too. :-) Keep up the great work. Subbed!
Thank you for watching! I am slowly recovering, still have some challenges (the anesthesia reaction really did a number on my muscles and endurance!). I really enjoy making content though and everyone's great comments are always a nice motivator! Thank you!
I agree. I've read so much posts about CO on Reddit, and watched so many videos on YT, and your video so far made the most sense. I didn't even know I can use Ryzen Master to automatically optimize my curve. It gave me -15 on all cores, and a different one on each core (-29 Core 1, -13 Core 2, -30 the rest). The most stable that I got was -20 all cores, but that even gives me the occasional app crashes and restarts. Now I'm using -13 on Core 2, and -20 the rest. So far no issues, but we'll see. Thank you for this video!
THANK YOU, FINALLY someone mentions this issue. I have watched literally 100s of video on undervolting using the curve optimizer and NOT ONE PERSON mentions the stability issues at lower clocks and at idle. I've personally experienced this when undervolting my R9 5900x. My CPU ran great at load but once my system was idle I would get random reboots. I ended up giving up on undervolting because no matter the offset I always ended up with random reboots at idle. I just bought a 5950x yesterday and I'm hoping with the BIOS and AGESA updates since I last tried undervolting that maybe it has gotten a little better. I guess I'll see within the next few weeks. Right now I just need to find an offset that will work and not screw up. I currently have it set to a -12 offset which seems to be working (atleast for one day so far) but strangely my temps have increased a bit with this offset.
Really awesome video, thank you! A total aside... re 14:03 into vid, the "high tech" screenshot, I loved it... why? Not because of the informality, but because your vid is generous with informative content in short order for someone new to OC. I've seen top names in fields who use the proverbial "pencil and paper" because their talent and technique wins the day, not the shine/sheen of tooling. Whether you continue with vids (the audience hopes you do) or do something else, never underestimate your mind's value and content ... your informal screenshot surrounded by comprehensive meaning is right up there with big names... Thank you again!
Wow! Thank you so much for the comment! I'll be back at the videos soon! (In Argentina for a much needed vacation after a long tough year). Can't wait to get back to it! I love everyone's support and am happy to be contributing meaningful content! Thank you!!!
currently on the 6 heatpipe burst assasin 120 with a ryzen 7600 65w chip hitting 90w at 5.35ghz all core via boost and i never pass 87c (arctic mx6 thermal paste) ( i hit 5.5ghz stable, 5.6ghz couldnt stay cool, but i didnt like the power draw either over 100w) i cant remember the temps i was getting with the 5600x but i had a massive arctic freezer 50 direct touch heatpipe and it was always cool as far as i can tell, i was more hitting a cpu limit than a thermal limit i dont recall ever seeing more than 75c ever with that cooler and that was under cinebench r23 after 15+mins
Your video is gold. I agree with your analysis as I independently came to the same conclusion today after spending 21 days of tuning. Here are my observations and conclusion: I did exactly same visualization while tuning my curve. I bought a ryzen 5950x recently and took it as a challenge to stabilize the bottom part of the curve while maintaining the top. Finally came to a value of -12 on all cores except the worst 2 cores. Contrary to popular belief of maxing out on neg offset for worse cores (ridiculous values like -30 or so), these need to have more voltage to be stable. I found the worst core to be stable at -8 and 2nd worst to be stable at -10. The trick here is to play a minimax sort of game. You want your best cores to boost higher, but not so high that the worst core eat too much voltage but not perform. But you also want the curve to have enough voltage so the worst cores dont crash at idle. That means there will be a upper bound and a lower bound that will exist. And within these bounds, you can simply assign lowest neg offset (-12 in my case) to the best core and higest neg offset (-8 in my case) to the worst. I spent 21 days extensively testing a lot of strategy i found on internet and none use to be stable at idle. This is a classical case of general wisdom being wrong. Remember, there is only a single curve and all cores operate on this single curve based on their offset. I have tested for 7 days continuously with various stress test at all possible kind of scenarios to come to this conclusion. My cinebench R23 scores are ~29200 multi and ~1612 single core. Note that I believe that I have not won the sillicon lottery. I would say I have an average CPU here. My RAM is also 3200Mhz (mixed 2 Pairs and running on timings on which both are compatible, ie the timings of the worst pair) and I use NH-D15 for cooling (so no liquid cooling). My mobo is Asus Dark Hero. However, for me, I am not using dynamic OC as I am getting best of both worlds just by curve optimizer. The best tool to check per core stability is corecycler. All the best.
To add: I took overclocking ryzen 5950x with curve optimizer as a challenge. Otherwise I would not spent so many days to max out my chip. Note thaf the difference between all core -8 and my optimized curve is 2-3% in benchmarks. It will not translate to visible benefits in real life scenarios for most parts. So unless you take this activity as a challenge, I would not recommend you to spend sleepless nights tuning this. Btw, on my MOBO, with asus's boosting ON and PBO enabled (not the one under AMD) and set to auto, I get the same results as manually changing the PBO under AMD to MOBO limits and all core neg 5 and 10x scalar. So even a default PBO on will give great perf for most. And dont care about high voltages and stuff...These CPUs are designed to survive these voltages with realistic workloads.
Thanks for the comment! You are bang on that this is really the kind of tuning only an enthusiast would undertake it just takes so much time, patience, and real-world workloads to truly prove it all out. That being said, it's a fun learning experience and the little extra performance boost is no extra cost so why not right?
Exactly..Btw, this analysis of yours should be shared as much as possible so that community can upgrade their strategy to tune ryzen processors, especially the ones with 16 cores (32 logical cpus)
Thx so much! Just upgraded to the 5700x from my 2700x and working with PBO2 and Curve Optimizer via Ryzen Master exactly as you show. I was finding it a bit confusing exactly what the problem was but suspected it was a case where the auto tune tool was not very accurate as is typically the case with an “easy button” solution. RM was setting me to -30 on 6 cores and -28 on two others but I realized immediately I was failing any and all stress tests. Yours was the first video I found that explained this in such a clear and concise way. I went to a -10 on each cores and if this proves stable I’ll probably called a day. By the way, the 5700x is proving to basically be a 5800x once you get PBO2 dialed in. Steady at 4.65ghz single and all core. The general “snappiness” all around vs the 2700x is very noticeable which I honestly did not expect. You have a new subscriber sir !
Well just for those don't know there's actually a tool called CoreCycler, it stress test each core using a test mode you chose from a list. And just recently I found some redditor posted additional code he created that automatically increase CO curve value by 1 once a core fails the test, that means the user doesn't have to go into BIOS to change this setting every time it needs to be adjusted. The 5950x I own has huge difference in its cores: most cores on its CCD2 can go -30 without any problem, but most cores on CCD1 can go as low as -18 or -19. One has to be adjusted to somewhere +9 and another one stays at 0 as previously tested (talking about losing silicon lottery...). I'm now re-running the whole curve using this new automatic feature to fine tune it. This example means using one universal number that fits all cores is impossible. The system can be unstable because the all-core CO value may be very far from optimized for some of the cores.
People will get different results with CoreCycler. I've had it work okay on some chips and just cause constant hard-locks on others. Unfortunately there is no perfect "one size fits all" solution that guarantees stability for all use cases on all chips. That being said core optimization will likely go away in future generations as AMDs built-in features and out-of-box performance get better generation after generation. Side note, almost all high performance builds I have done in the last 6 months people want 13900K over 7950X/X3D. Seems like the big spenders are getting AMD shy again.
Well it sounds like there are some more tweaks to be done in the CoreCycler config cos if you don't stress each core properly with all 3 modes (sse,avx,fma3) you may not get a true CO value. But since it's really time consuming and Zen3 is already a past so... I do agree that Zen4 generation didn't perform as good as many had expected and Intel actually had been doing better than before, but hey that's what exactly we want: competition. It's always a good thing that Intel brings us cheaper AMD and AMD brings us cheaper Intel.
Power plan balanced, grab 5% to 0% minimum usage. In ryzen master settings: Stress test: Time 600 seconds Type of load:All. Back to Curve Optimizer: Per Core. Slap the start optimizing button.With these settings it will test all the v/f in both single and multi core. On 5950x is 1 day 21h 30 min testing, for others less. Best tunes comes with default settings and scalar off. With the override one cores get a lot less curve shift and you get clock stretching on most cores. Focus on effective clocks, multiplier are not real sustained clocks. Ryzen 9 works best with no override
Thank you for the explanation. RM curve optimizer set an offset of 30 to all cores after the 1h40m test. And my system ran fine for 2 days, until the screen went black after closing a game. Now I know a little more what RM is doing (what my unknowledgeable self is making RM do, no excuses) and how to proceed experimenting.
Thanks for this explanation. I think it helped me to understand what Project Hydra is trying to achieve. It's trying to adjust the voltage on the lower end of the curve by bracketing the curve, monitoring the load on the cores dynamically moving the curve to guarantee that under low load there is still enough voltage.
Great content! The delivery, explanation, and video quality are great. Question, how long did it take for you to initially “optimize” when first setting up Curve Optimizer? My ETA was 12 hours for a 5900X (per core) to start and about 5 hours in, it says I have another 5 hours to-go.
It took around 6-8 hours overall. It isn't a fast process by any means. I usually set it and walk away for several hours. It's especially a pain when you are testing using several CPUs and different BIOS settings before optimization. Patience is a virtue! Thank you for the question!
An important companion video to create is one about clock-stretching: what it is, what its impacts are, and how to test for and prevent it. Without that piece, this video (while extremely useful) is missing something very important. Cheers!
Great idea! I'll do a separate video on that as well. Honestly if I did a fully comprehensive video on overclocking it would be 2-3 hours long and no one would watch it. Most boring full-length movie ever!
@@AquaStevae I will be soon! Had a nasty bout with Covid, lungs still working through it! New videos soon! (I'm working on a themed mega-build right now that's eating all my spare time, but it's going to be awesome!)
Thank you! It's funny I have been told that many many times, even at Star Trek conventions. Oddly enough I was always jealous of Wesley Crusher being on the Enterprise when I was a kid haha! Thanks for the comment!
Ryzen 5 5500, negative 28 at all cores (Ryzenmaster said I could do -30 on some cores) and Boost positive 200Mhz, the CPU runs above 5600X acording to CPUID bench, PassMark Bench, UserBenchmak. In boost seats stable at 4.4Ghz
excellent video, very well explained. do you have to assign your prefered core for each apps ? we can see cinebench dont use your prefered core by default so im wondering, but also for gaming for example it will more multicore than single core so not sure if its really usefull to assign prefered core for games. i know there is a CPPC prefered core option in bios which should allow windows to use your prefered cores by default, not sure if it work well tho
My pleasure! Thank you for the thanks! I really hope to do some more videos soon. The last year has taken a toll, so hard to manage work, life, hobbies, relationships, and still find time to make content!
Awesome, newcomer to your channel and you really helped me understand what curve optimiser is, I’ve just got a 5900X and I I have the Dark Hero motherboard. So I’m going to be trying this offset combined with increasing pbo max settings. Great video, well explained, you got a new sub 👍
Thank you for explaining this clearly with plenty of detail and examples and without talking down to folks like me new to OC! I finally understand thanks to you. ASUS and AMD documentation basically non-existent
Thanks for this detailed video. in 2 weeks ill have 1 month vacation and time to try it to the bones. i got an tuf x570 and an r5 5600x, i think i´ll start low with an offset of 5 and check the results. does ryzen master recognize the favourite core or do i have to find it for myself? sorry for digging up this old video ^^'
I found a good way to test your curve optimizer under-volt is to repeatedly start-stop a full CPU render in Cinebench R23. Start a render and after each thread completes a single block hit cancel. Repeat. Keep doing that about 20 times or more. This fluctuates the cores through most of the voltage curve. An unstable CPU will eventually black screen on you.
OCCT has a stability testing mode that does this automatically and works really well. CPU stability test, set it for a small data set, normal mode, variable load type, and core cycling thread settings. It basically cycles through each core, causes it to ramp up to full, and then puts it back to idle for several seconds. It does a pretty amazing job at finding unstable idle undervolts. Only pay attention to the first error though, once it finds one it seems to spew them out almost at random.
Nice tutorial. My 7060x is stable at -15 all cores so far (just doing regular stuff and gaming now to see how it gets along). However, I do want to do CO for each core one day...just don't want to spend the time now. However, I would like to find out my preferred cores. How did you determine your preferred cores? Did the AMD software pick them?
@@Methodical2 Did you happen to figure it out? All my cores run at -30 because its an X3D chip I guess but Ryzen Master has decided core 2 is preferred despite all being the exact same frequency.
Load line calibration didn't help my stability when playing with the curve optimizer settings unfortunately. That's not to say it wouldn't help in other CPU/motherboard combinations. I'll add LLC to my video list! (I'm on vacation for a few weeks, then I'll be back to it again!)
Great explanation of the problem with the optimizer. I ran it on all cores and just spent several hours trying to figure out the crashes just trying to watch other RU-vid videos after banging out a perfectly stable game session last night. It was bad enough instability that I had Windows file corruption that required starting over with default settings. Hopefully a thing of the past . .
Great explanation, thank you. If I run core optimizer in RM to determine the offset and then set that offset in BIOS, will RM recognize that when Windows launches? In other words, what happens if curves are set in BIOS and then RM profile is loaded in Windows?
Thank you for this video/tutorial/explainer. Much appreciate! One tip - instead of doing manual per core testing, run Ryzen Master Per Core test which will determine the offsets, then you can manually add them via UEFI/BIOS. Also it tells you about your best and second best cores, on which you should have lesser offsets and they will tend to reach higher/sustained Ghz. Am I right in my assumption? Question 2: With CO set, how about setting PBO Manual/Motherboard with Positive 200Mhz to get that extra?
This was very informative, useful and entertaining. I cut a lot of time off of this, by finding my baseline allowing RM to do an "AUTO" curve optimizer, which did get me to an all core negative 30, and I haven't had a single black screen, flicker or crash. I've run several benchmarks, and still no instability at -30. It took the auto curve optimizer about 45 minutes to complete, and my system runs an all core oc at 4.65ghz while benching, using my 5900x. I realize I can probably push this processor a bit faster than that, but my temps are idling in the 40's, and benching in the high 70's and low 80's. It appears to be sustaining its performance, and running very stable. I've scored in the mid to high 20k in CBR23, which is a strong showing. What I don't know after running the auto optimizer in Ryzen Master, is how to tweak it without destroying the current setup? Not a RM guru as you can see. But learning as I go. Thanks for another great vid.
Thanks for the awesome comment! I appreciate it! You can use one of the profiles to set/play with your own settings, but if you are sustaining -30 that's pretty awesome!
@@UnhingedSystems I had no idea about the auto optimizer in RM, and just saw it in a vid. So I decided to try it, just to see where it got me. 45 minutes later, I'm well ahead of the game, and can do smaller incremental tweaks to improve moving forward. But I will run with it as is for a few days, just to see if the speed and stability continue. I also downloaded the corecycler script, and will start it tonight and let it run while I sleep. It will be very interesting to see what the results are. Cheers!
If you just change half the cores and test for a period then repeat with the other half you could get through it much faster than incrementing just one core at a time. When you get an instability issue then divide the suspect cores in half and test. This will save a lot of steps by repeating this process to find which core is causing issues. My 5950X is unstable at 0, setting all cores I have had it on +8 to be stable. Curve optimize had most at -27 one @ - 13, another @ -11 and -7, then the computer would not even boot.
Thank you for this wonderingly informative tutorial. I ran the curve optimizer after watching your video and applied the recommended core settings on my R7 5800X which ended up being C1 -19 C2 -27 C3 -27 C4 -22 C5 -27 C6 -27 C7 -27 C8 -8 I don't know what happened to the last core, it might be a bit dodgy perhaps. System has been stable for 24 hours that that includes lots of load time playing games, cinebench and running Folding@home, as well as idle time thrown in as well. CPU seems stable and I'm completely happy for my boost clocks to now be 4962mhz across the cores. C1 & 4 even got up to 5007mhz a couple of times.
I set all cores to -30, no crash, but single core not stable and so many errors. my highest core0 clocks 4.557 MHz only . please share with me your numbers so I can better tune my 5800x. thx.
I have run the optimizer three times now. It always takes my voltage offset to -30. I have a 5950X and this is so unstable, I have to reset cmos to get the pc to run. I finally ran Ryzen Master in simple mode and set the voltage at 1.39375 and the clock at 4.675. This is very stable and I ran an all core cinebench R23 test and scored 30714. WOW!
Really good video explaning curve optimzer, I enjoyed it!! I have a 5600x set to -30 all core and is stable. Best video Ive seen on this topic. Thanks.
Thank you very much for the video, very well explained everything but I have a very important question. How can I put the curve optimizer back on Ryzen Master, with the Start Optimizer option instead of Validate Offset, so that it can test me and get values by selecting Per Core? Thank you very much and greetings.
I'm sharing your video in a bug report to AMD about this same thing. I'm running a 5800x and just tried the same thing. I got all cores recommended to -30 offset. It crashed while idling overnight. But I'm perfectly stable long term at -27 curve for all cores idle or full load (full load for me is running Folding@Home or Cinnebench)
What I don't understand is why the curve offset has to apply globally to the entire curve? Why can't you leave the defaults say under 50% load and only modify the upper end that you're actually trying to optimize? And then blend it over some transition range?That would completely eliminate the issues of instability at lower clock speeds where you're not trying to push the frequencies up for a given voltage. Is the BIOS and optimization really that simple? Just seems odd to leave performance on the table just to support stability on a default curve at lower frequencies.
I'm certain it has to do with the actual architecture of the design of the circuitry of the chip itself. The chip needs parts of itself to be "tuned" in a specific way like voltages, voltage sensing, load and load sensing, etc. The challenge here is that changing how the curve works is probably only possible at levels integrated into the design of the chip so it isn't as easy as writing software to adjust the curve at different load points. Technically you could write software that runs outside of the chips own firmware and attempts to do that, but it needs to respond fast enough to load and voltage changes to not cause instability. They could definitely design chips to enable more granular support of voltage curves, but the stock algorithms are already quite good. I believe we are only a few years away from there being no real point in tweaking CPU setting anymore. The threshold for user tweaks to get performance benefits has been getting smaller and smaller over the years. As an example, GPU overclocking beyond stock overclocks is practically pointless now with barely even single digit % available before instability. In the early 2000s beefing up the cooling and overclocking your GPU could get you 15-20% gains in the right scenarios. Even the 13900K is very practical by just releasing power limits and providing superior cooling. Individual tweaks or overclocks are becoming obsolete. It's fun to play with though! Let's hope they always leave us a little bit of headroom to enjoy the hobby!
Thanks for the thorough reply. So just setting a minimum voltage would not resolve the lower voltage instabilities. I definitely agree, it is fun to play with so I'll be a bit sad if they optimize them too well, haha. @@UnhingedSystems
Just applied a conservative -10 all core curve optimize to my 5800x3d, and I'm seeing a pretty substantial drop in temps and a slight performance gain.
I’m pretty amazed how well Curve Optimiser worked.. here my results on Cinebench 23 Stock: score 15025, 82,4 celsius, pulling 94,3W. With CO - per core: Score 15848, 67,2 celsius, pulling now 72,2W. More then 20% of power reduction and 15 degrees cooler!
I currently run my 5700x (b550 tomahawk) in eco mode, do I need to return to stock or is there any suggestion points before running RM for the first time?
Old video now and only briefly skimmed through the comments, so maybe you've already found this out but - I'd like to add that you can see which core is causing the crashing , in windows event viewer the crashes will come up as a WHEA event id 18 - I have setup a custom view that shows me WHEA errors (whea 18 generally shows up after a crash when the curve opt. has been set too steep and the core doesn't get enough voltage, I see whea 19 alot when overclocking ram and it suggests an unstable infinity fabric) so if you set an all core -30, and the system crashes - you can check the WHEA 18 and it will speciffy what core is the issue - APIC ID: x - so APIC ID: - 0-1 will be core 1 in the bios, 2-3 will be core 2 etc, really helps when trying to dial in problematic cores. I will mention that sometimes* the pc will just black screen/crash and windows wont show the core that caused it.
Interesting proposition. That might come with its own issues such as unnecessary heat and power draw, or potentially shortening the lifespans of the CPU. I know I used to do that as a regular practice for Intel all-core overclocks 10 years ago, but I'm not sure how well AMD will handle it. More testing to do.
Thanks for this video. I just turned off the curve optimization. :) The Stock setup is fine for me at the moment. The stability is the most important for me, it doesen't worth to win a few degrees, if the stability is gone. :) Maybe later, if I will have time, I try to optimize core by core. Thanks for the video!
My 5800x used to boost to 5050 mhz under maximum load with a peak cpu temperature of 76 degrees (420mm radiator AIO with Noctua 3x 140mm 3000RPM static pressure industrial fans) After following this guide it now boosts to 4850 mhz with a peak cpu temperature of 76 degrees. Thanks.
Max boost isn't the full story, it's more about sustained load and average sustained max clocks over time. Most of the time when you see a measurement of 5050 MHz that was actually measured after the load drops off between executions rather than during consistent load. Single core boost is where curve optimizer can increase the maximum sustained average clock speed. I've also noticed that with newer BIOS updates the boost behavior has changed. Older BIOS from 2021 boosted slightly differently. If you reset your BIOS your CPU should just resume its previous behavior.
I can access the curbe optimizer settings and change to negative but the cueve optimizer magnitude is locked. I enablex advanced PBO and disabled power limits. What else should I do ?
Someone needs to come up with a repeatable and consistent way to simulate the "idle crash" conditions from doing PBO negative offsets. I really can't be arsed to test each one of the cores on my 7950X without a consistent stability test to do anything other than an all core offset.
Indeed. AMD really out to do something that works since it's their tech. I've tried every stability test out there and I still run into stability issues outside of testing software. I don't fully understand why, but Stellaris (game) will crash within an hour or two while everything else appears stable. I'm running stable right now, but every time something weird happens I get rid of all the curve optimizer settings just to be sure it isn't the culprit.
N00b question: Do I have to keep Ryzen Master open and running to use the CO profile? If I close RM, will the curves be unoptimized? Or should I now just plug the values into the BIOS?
Ryzen Master will apply the OC when it loads, it will need to load each time. Once you are happy with the settings applying them in the BIOS will eliminate the need to use Ryzen Master 💪
@@UnhingedSystems Also, I didn't know Ryzen Master changed the BIOS values itself! Just been experiencing game crashes and BSODs (thought it was stable but no) and when I entered BIOS I found the auto optimized values there. (-25 all cores). I'm reverting back for now until I have time/patience to experiment more. Right now I want stability.
Hey, I recently got into PC gaming and do have some knowledge. With that being said, what would you recommend on a 5600x paired with a 3060ti and B550? Cooling is a CoolerMaster 92mm.
This is why I like the OC Scanner in ASUs GPUTweak for example, since it will test every single data point in the voltage frequency curve. My 5800x3D on the other hand is not stable at very low loads but of course whizzes through any maximum load benchmark you care to name effortlessly at minus 30. I am starting to think it is just too much trouble, especially since no application I will ever run actually uses all cores at once for more than a few seconds every 30 seconds at most.
Been looking around on reddit and youtube about undervolting Curve Optimizer settings and I usually see people set their "Per Core Negative" to something between 5-20. I've set most of my cores on my 5800X to 30 besides my 3rd and 8th core which read unstable when I set one of them to 15 and 25(thanks to OCCT), but is there a reason why most people set theirs between 5-20 and not most to 30? Am I fine setting most of my cores to 30? Am I "lucky" if I'm able to set most of them to 30?
Good video, the only explanation I'm missing is what is the difference between Scalar 1x and Scalar 10x? What does this setting do and is it safe? Thank you
Good video man, just bought a 5600 upgrading from a 2600 and I’m now intrigued to test it’s capabilities. My game focuses on single threaded workloads, so do you think PBO or Curve Optimiser would be better or should I use both?
Thank you! If you use PBO, tweaking curve optimizer will add some additional performance! Definitely recommend it if you have patience and don't mind a little trial and error!
@@UnhingedSystems I was thinking what if I Curve Optimise all the cores to the limit manually then go onto per core and boost the better cores. You think that would work?
We need an updated video on this in regard to the new Ryzen 7000 series. It seems Ryzen Master doesn't select optimal cores. It shows them all but nothing, highlighting what is the optimal cores.
@@UnhingedSystems Sweet! I recently upgraded from a 5600X to a whole new 7700X setup. I used Ryzen Master to check an all core curve, and It also said -30. I ran it for the most part, stable for a few weeks until I found and instability. I am now using -28, and it seems to be quite stable now. Granted, the 7700x only has one CCD, but that might be why it is doing well at -28.
I'm trying to figure out whether using the curve optimizer per core will lead to different VID values (voltages to core) when all the cores running at the same maximum frequency?
When I run the per-core curve optimizer on Ryzen Master, after a few seconds I get a system crash and it does not respond. The temperature rises to 89ºC. Some help?
Try resetting the bios to default settings and run it without making any changes in the BIOS. If it still locks up then you may have cooling issues or lower delivery inconsistencies. Let me know!
@@UnhingedSystems Good. I loaded the factory settings of my bios. When I turn on my computer and open the ryzen master program, the PC directly freezes. I pass tests of occt, cb R23 etc. No problem. But with ryzen master I don't understand what happens. I don't think it is due to a poor cooling problem since my temperatures do not exceed 75°c under stress test. Likewise, tomorrow I will change my mx-4 thermal paste for kryonaut. I will inform you. Thank you for your time.
Just what I was looking for. And good catch on turning off defrag, I recently crashed with being unable to load the OS at all, must had been been related to this.
That's the catch with the offset undervolting, compared to the classic cpu core voltage tweaking: the negative offset undervolts also the idle voltage of the cpu. Hence the instability when the cpu drops to low frequency.
Things were so much easier in the old days when CPUs only had one speed to worry about - flat out! Of course that was inefficient, wasted power and produced un-necessary heat. But I like the current ASUS OC Scanner for their GPUs, for example, since it will automatically create a new frequency for every single voltage along the curve - and you end up with something 100% stable no matter what the load is because every frequency and voltage combination possible has been tested. When I ran the scanner, the voltage frequency curve was identical to stock - up to 2,000. Beyond that the curve managed higher frequencies than stock for a given voltage, effectively undervolting at the top end or hitting higher frequencies depending on how it is utilised by an application.
Thanks for explaining this in actual detail with some test results. I was worried I was only going to find some LTT video's with soyface thumbnails. I was running a basic negative offset through voltage control, but that actually causes the clockspeed to continuously drop because the PBO curve stays the same. I will try this now, and see how it goes. That said, the 5800X3D is a bit of a fickle CPU and you can't really run wild with it due to thermal constraints. Not so much that it produces a lot of heat, but the 3D vcache causes an insulating layer. Too bad you can't edit the actual curve like you can with Nvidia's GPU's if you hit CTRL-F on MSI Afterburner.
Regarding 9:16 after a reboot or crash if you go into event viewer it will usually tell you which core failed in the error report. When my system was crashing with my 5900x it turned out to be 3 cores that kept causing issues. At the time I didn't really know enough about doing a per core offset so i just changed the offset for every core.
Finally, someone who did a decent rundown of how this quirky, annoyingly mysterious and vague program works. There's no way to update how the curve functions on the low end is there?
Any good way to cause the instability as fast as possible? For me OCCT helped at one point, but I'm wondering if there's a good way to cause instability and test for future crashes?
Prime95 and cinabench. As soon as a "worker" fails you know something is bad. And when you pass that for over a few minutes either continue undervolting or test it for complete Stability
Haha! Great question! The good old scientific method. Research, testing, measuring, gathering data and forming my own conclusions. I love tech, so this kind of thing becomes a bit of an obsession for me. Unfortunately I'm a working stiff like the rest of us and so I don't have nearly the time to make content that I would like to have. Soon...soon...
