I rarely post but there are some misconceptions in this video (not blaming Tech, you're doing what you can and the content is relevant and entertaining!). I have binned 5+ 7800X3D and while it seems initially that all of them can do -20 all core CO, they rarely pass real stress tests. Only 1 out of 7 can do -30 throughout the cores, and in some instances, some individual cores barely can do -5. PLEASE TEST YOUR CPUS PROPERLY. Having said that, I dont think CO is too relevant for 7000s, especially how YES also noted, you dont necessarily save on temp, it just boosts higher/longer. Therefore I suggest getting a very good cooler coupled with CO and dont worry if your CPU doesnt CO too much. ANOTHER thing Ive noticed is that some CPUs are just binned higher (better VID), where they run at 4890MHz out of the box (CB23) while others barely break 4850MHz with -20 CO. Finally, there is no correllation between CO and Max FCLK or MAX Bootable RAM Frequency. I much rather have a CPU that can do 6600MHz 1:1 at 2200MHz than one that can do -40 CO All Core. My best 7800X3D can do avg -25 CO (some cores -35, others -10), does 6400MHz 1:1 C28, FCLK 2133MHz with 100MHz or -5 CO All Core @ 105MHz (eCLK). EDIT: Also read about CLOCK STRETCHING.
yeah, curve optimizer stability is much harder than intel offset or lite load undervolt to test in my experience. I also cant seem to get Core Cycler to ever error on me with zen5, might be missing something. It did work for zen4 and actual curves were in the range of 5 to 30 on most CPUs... pretty wild actually and that -5 core would cause maybe 1 crash every other month.
I totally agree, thats what we call Sillicon lottery. For me personally every core has different values that i found out by running Prime95 with every new offset for atleast 2hrs, once a core crashed i just set it to the next stage. Startingpoint was -40 ended somewhere between -21 to -29 for different cores. Did the same for my friend he had complete different results. One core could manage to run -38 while core 2 always crashed over -10. Ofc after running prime95 successful you should also try other things to somehow stress the cpu.
yeah I pointed this out in his 5700X3D and in this video. you rarely stress test CO results. as most of the cores primarirly crash at idle states and he also ignored the most glaring issue that in lot of cases can net you less performance and thats clock stretching. so while maybe -30 looks stable and doesnt crash, it doesnt also mean its the best result as you could be clock stretching to a point your losing performance. overall its a bad video imo, 2 crucial pieces of information were just left out, which is using Corecycler or ycruncher to test cores individually over several days of continious testing and clock stretching NEEDS to be observed when setting CO offsets
@@123TheCloop there is nothing wrong with the video. It shows the test improvements and how to test. With any instability you dial back. There is no way you will face clock stretching if you do it as the video. He also mentioned that is best to start with -10 probably a setting that most cpus will be fine. You set and test. He never said set it to -30 and forget about it, test it.
Long time overclocker with experience. Cinebench stable with these chips using curve optimiser is easy. However, they can seem stable but will crash infrequently if there is no load at all or dropping from high to low load. You can test with CoreCycler, but it's much more time consuming than with Zen 3 non X3D. TL;DR plenty of these are cinebench stable at -40 but will crash under no load until you dial them back to around -30.
I was so happy when i used -30 offset on my ryzen 5900x , only to have it error it immediately when testing with singlecore and bluescreened me. In the end after a lot of testing , not 1-2h on all core ,which its pointless, 2+3 h on each core, i could say that -10 offset was stable on all cores, regardless of load.
thank you, CO results NEED to be tested in idle conditions because ryzen cores down volt which is why they can crash when you offset (drop the voltage curve) so in idle situations the core literally isnt getting enough power (think it like a GPU crashing when down volting) its a little complex than that but im annoyed he thinks just -30 and cinebench are "how you test curve optimizer"
I disabled the 2 weakest cores to make it a 7600X3D and set the power limit at 35 watts with a -30 CO. It's an efficiency beast and stays so cool and quiet I love it (I just use it for games where 6 cores is plenty and they boost higher)
I think its very important to always double check if the undervolting worked by using cinebench. Because undervolting slightly too much wont make you immediately blue screen, but you get so called Clock Stretching, meaning it looks/"acts" like you get the desired clock speeds, but in an absolute test (like cinebench) you get a lower score.
I agree. I always use hw info and look at the effective clock while doing an all core workload. Whichever cores are not reaching the boost clock is not good on the negative offset. I have -30 on all but 2 cores at -25 due to this effective clock seen on hw info.
Very Good video ! I have an ASUS ROG STRIX B650E-F Gaming Wifi motherboard with Ryzen 7900x. I will try to undervolt it using this method. The cooling is done by a Be quiet! DARK ROCK PRO 4. I love Undervolt videos :) I hope to be successful. Thank you !
Stop testing all cores workloads, it s pointless, you can test on a processor -30 offset and will not get an error for days, the moment you test each core individually it will error in seconds/minutes. Use tests like OCCT single core test or Prime95 with core cycling.
Great Video, thanks. I tried this with an 7800X3d on an Asus X670-E-F and Gskill 6400 EXPO with a 4090. In a quick CPU-Z Benchmark I do get more performance with the values ranging from 20 to 30. While in the CS2 Benchmark I got around the same fps (730fps +-2fps) but the P1 value is dropping from 300fps to around 277fps. (I tested with 20, 25 and 30) So in my case I stay away from it, it only helped me in CPU-Z Benchmark but not in my main game I play CS2.
@@hansmaker1236 How long has it been running like that? Doing that is "usually" okay, but you're still going to run into crashes. Even if it's just once a month. You'll also get all sorts of other random errors from anything else running on the bus.
Nice video. Thank you. Its crazy mobos and amd have defaulted so high. Just -30 and you get smoother and consistently higher fps with lower temps. Thats an insane gain so easily done.
I just built my 7800x3d setup and running -30 all cores stable. I'm using a B650E Taichi. Thank you ohhh i gota try windows 11 pro link as well tomorrow.
@De2t3ny 0 crashes great temperatures . Why do you say that? Ran every benchmark and always passes. My ram is at 28 37 37 30 is that unstable as well lol. I'll give you my secondary timings as well if you want. Hynix M die 2 x 24.
I did -20 CO on mine and it runs way better than optimized defaults. All cores can sustain 4.8ghz and boost to 5.25 ghz. I was blown away by the increases
@@arc00ta If you want to lower the X3d's temperatures you have to lower the soc voltage to 1.2 and limit the voltage spike. Mine with factory configuration reached 80plusºC and with some adjustments it doesn't go above 72ºC whatever i do.
@@starlightHT Why do you need to lower it? 72C and 85C are both fine, and at 1.2 SOC you're not going to be bumping up the fabric and RAM speeds. Since we're trying to squeeze the last 2-3% out of this thing might as well go all the way.
I'd like to wish for an updated version, which includes the Curve Shaper, which was initially introduced with the Ryzen 9000 Series. Nontheless, withy my newly purchased Asus ROG Strix B650-A Gaming Motherboard and Ryzen 7 7800X3D I also have the opportunity to use the curve shaper. I've played a little bti around and combining both the curve shaper and the curve optimizer and I can rock -40 Curve Optimizer. I've also set some curve shaper values, but since I ain't got what the feature works really I test them out currently. So far I have to say my 7800X3D is running with -40 Curve Optimizer undervoltage and clocks ingame like cyberpunk majority of the time with 5050 Mhz, and pulls around 60-65 watts. Therefore I can runs my AiO cooler and all fans in my system slower then with my older 5800X non 3D CPU. But it gets a lil warmer around 60-64° Celsius
@pwn3426 We'll have to wait and see, as zen 5 now has a new "curve shaper" function that allows much finer control of voltages according to cpu (or individual core) load/temperature and "recent history", ie is the core speed/load/temperature ramping up or down compared from a few (fractions of) seconds ago. This could be a game changer as crashes often occur when voltages over/under-shoot with big changes in load.
In Ryzen Master it will show your best and second best core with golden star and grey star. Best way is to give them lower offset then all other cores and u will actually improve performance. For me my 7600x is currently stable at best core -15 second best core -20 rest of cores -25.
I wonder why you channel isnt 5x bigger. Your channel is solid AF. Your so slept on its crazy. Keep it up man, ive been watching quite regularly since i found your windows 10 optimization video, which was gold. But i admit ibdid find an insane optimization block of info that blew it out of the water. LoL still, im a big fan, and i hope you blow up in popularity atleast quite a bit.
So I was stable at -30 but would going to -20 give me a bit more performance? I’m new to this, cinebench score went down from -20 to -30 but gaming seems unaffected as I tested them.
Higher SoC voltage is needed for higher RAM frequency, while lower SoC voltage is better for higher FCLK speeds like 2200MHz. I have set SoC at 1.26v and run stable 6400cl30-38-30-44 with all other timings manually tuned. It's in 1:1 ratio with FCLK at 2133MHz.
7900x3d with a Peerless Assassin 2 running with a -20 x3d ccd/-28 nonx3d ccd offset on a 100W limit. Scores the same as stock in multithread but showed about 8% gains in single thread performance. At the stock power level (120W) I got an additional 4.5% multithread performance boost, but I didn't feel it was worth the power and heat. I can now run looped benchmarks indefinitely at 71 degrees 60% fan speed vs 78 degrees 100% fan speed at 120W(stock). There is absolutely no performance reason to buy a watercooler for these cpu's when a decent cheap air cooler works just fine.
Funny how i watched this video and i exactly bought ASRock motherboard, so it was easier to navigate with your video. I choose ASRock over MSI because i heard they are more stable on AM5 platform, but god ASRock UEFI is so confusing sometimes. It looks kind of like old bios with better graphics and more pleasing to eyes.
It's the ram that's been difficult to tune for me, but great video! I have -24 on mine, I could go lower but I slightly lowered the vsoc just to be sure.
nice. at first I was like "whats the point" on one of the best CPU's on the market right now right. but why not it's leaving performance on the table. get that thing tuned in and rolling. so it appears i've hit the silicon lottery -40 doesn't crash for me.
I think something more reviewers need to mention is their idle temps for the chips. I thought for the longest time that I was having some issue with my chip with it idling 50-60C but apparently, this isn’t abnormal. If you were coming from a mid range Intel chip, you’d think something is wrong. Lol
You rock man! you know how to undervolt and boost amd cpu… with little steps you can fix an Intel Cpu and stop crying like few weeks ago in some “copy/paste” video like many techtubers about Intel fails🎉
Same deal for the 5800X3D, this is absolutely essential if you own an X3D CPU. Mine is probably a golden sample. I have a -112mvtt undervolt, plus a -30 PBO offset (since you can only go to 30 with 5000 series, so im essentially cheating and making it -45/50 myself). It's only using 81W all-core. It's pretty toasty here now (28-30 ambient) and I was seeing 90C peaks. Now its more down to 70ish.
@@De2t3ny its not overheating, 90C is TjMax. Have consulted forums and everyone is seeing these numbers. Why do you care anyway, go be some adverserial cretin someplace else.
I set all cores to -30 with an PBO on 80 Celsius limit, and my temperature was the same in Battlefield 2042; I just don't understand the point of undevolt for the 7000 series! I have a 7800X3D, and I don't notice any changes in games; the CPU is running great, and the only heavy CPU game I observe is 70 degrees Celsius, the rest are 59-65 degrees Celsius!
@@techyescity -36 is just about OK for a CB23 run for a number, best I've managed is 19.2K. Really pushes all 8 cores to 5050MHz if you get the undervolt right down.
no he really doesnt, if you value a proper stable CO offset, go in intervals of -5 all core and keep testing using corecycler and watch for core crashing (and observe for clock stretching)
True and i swear i watched like 6-7 on this, every one of them barely did any stress testing and called it stable at -30 offset. My ryzen 5900x which some said undervolts really well, if you stress test with cinebwch or other all core workloads it works at -30, but its stable only with -10 offset.And i properly stress tested each core individually for about 2-3h each.
Pro tip, don t OC the 7800X3D, especially PBO, it causes instant stuttering, for example in CS2 it s terrible. It s a very good chip at stock settings, less headaches!
Ryzen Master is not good to test the stability. My 7800X3D is validated up to -40 by Ryzen Master but it's not stable even at -30, I have to keep it at -20 for daily use. OCCT and Warzone crash after a few minutes at -30. Garbage utility.
My 7800x3d can do -30 no problem but i ran OCCT ram and cpu extreme test and got WHEA error in 10 seconds, so i just run it at -20 passed 1 hour OCCT test no problem.
WHEA error means a core or multiple crashed........ this is why you test using corecycler and do it for extended periods. cinebench r23 is dog for stability testing CO values
@@ConnorH2111 doesnt matter what your using, -30 offset is not stable if your getting WHEA errors, (these are core crashing errors) should be Event ID 19 if its a core crash. His video is to be frank just poorly informed, any offset value YOU NEED to test under idle conditions (this is usually when core crashing occurs) which he just misses completely (either intentional or ignorance) If your getting WHEA errors and its not an Event ID 19 error (something else) then its more than likely not the CPU but a bios issue, it happened a few times on Ryzen 5000 with some bios revisions causing WHEA errors. But please, dont use OCCT/cinebench or pimre95 to "test offset values" download corecycler (which does use prime95 ironically) but its on a per core basis and does random work loads at varying usage levels per core. you need to run this for a good 3+ days when testing offset values and see if any cores are flagged as "crashing" or failing in the program. Then you need to use something like cinebench to watch for what is known as "clock stretching" which is basically the difference between your effective clocks and "core clocks" value in HWInfo, any difference greater than 15mhz means your CPU is reporting an incorrect clock speed mometarirly due to not enough voltage to the core/thread so as a result misreports the clock by throttling down to match the given voltage supplied. (again something he grossly doesnt mention) So while that -30 could be stable even after corecycler testing, it might actually be yielding worse performance than lets say a -25 offset due to clock stretching
I would adding corecycler and linpack extreme to makle sure the cpu is stable. I can get cinebench to run at -30 all core but during linpack stress test, i did get random resets until i got it down to -20 and -15 on my worst core.
Validate Offset is not enough to find stability. I had to come in here and chime up. Been using offset for a while since back when it was capped at 30. I know my chip can only do around 20 all cores, but ryzen master validated at -40. This boots but is an almost instant crash in Cinebench R23.
AMD said it would be available ONLY for 9000 series cpus. Maybe consumer backlash will see them enabling it for 7000 series in the future kind of like intel with the apo locked on 14th gen at first till the backlash hit forcing them to expand to 12 and 13th gen.
I updated my Asus Crosshair Hero X670E BIOS to the latest one and I can see it in the PBO menu. I have Ryzen 7950x. I played a bit with the new feature, but either I don't know how to use it, or it doesn't work with Ryzen 7000 CPUs, as the commenter above said. I hope AMD enable it for older AM5 CPUs. Looks like artificial limitation to push people to buy the new CPUs.
mine hits 87c in a R23 multicore run when my room is pretty hot with a 240 AIO from arctic. When I play super cpu heavy games like warzone i max out at 74c so 86 is def. too hot
It's not true, we can do Oc at 7800x3d even without access to the multiplier. The old way is to gain 150 to 200mhz easily, move the memory back, FCLK and increase the bus speed... Technically it's not worth the time.
We need a better and faster way to quickly tell us the highest curve optimizer setting per core. It takes a fair amount of time, at least a few hours, to do it all manually to maximize per core settings.
I just did this is bios with my asus x670 motherboard in the precision boost overdrive tab. When make any changes on ryzen master my pc won’t boot after the restart and my d-ram light comes on. Maybe because I’m on windows 11 , either way I get cooler temps with just the motherboard settings
got my 7600 down to -50/ +200, runs 5.35 PBO All core all day at 75c max w/ a phantom spirit. It will push 5.5gz on a straight OC, but i hate loosing the efficiencies. I know its not 3d, just needed to share w/ some like-minded folk.
Did this with my old 7800X3D and either the cpu or motherboard fried, micro center got me with a new bundle for free and with this one I just left it stock not doing that again 😂
Curve optimizer is really frustrating to work with because when I run every benchmark for stability, things work out great, but after two or three weeks, my PC crashes just while idling.
Please help, I am using 7800X3D and 7900XTX Sapphire nitro+ but in CS2 my GPU usage isn't that good like the footage in this video. I enable PBO like this video, XMP, 4G Decoding, OC GPU, common window optimized. It turns out my FPS average is a lot lower. Like the Machine gun footage that your 7900xtx used 80%ish load but mine is 50-60% result in 100-200 FPS lower.
The "validate" thing doesnt work properly and -30 works but no in all workloads. If you dont have the time testing the system properly use hydra. You will get results which are more reliant overall.
The curve optimizer's variations are very difficult to test in stability. Stress test have absolutely nothing to do with proper testing, in this case. The real difference is in sudden frequency spikes in idle. Trust me, except rare cases, most chips can't handle -30 or even -20. Mine was stable at only -15 but the performance boost was negligible and it even increased temperatures. So it's a big no for me. Buy a better cooler instead. EDIT: typos
i got minus -30 seems stable but game or pc restart sometimes when only playing battlefield 4 random dont know if it my settings or the game but im running old bios 1809 rog strix gaming b50e-f gaming wifi. should i update bios? or need newer chipset driver.
I just bought a new mother board and a 7800x3d because my intel 13700k is basically dead. It stiaght hard crashes as soon as i try to load any game or CPU load. I currently have it down clocked to 4.2 on the pcore and 3.4 on the ecores and still cant do anything. My new parts get here in 4 days
I had 17100 score with default setting and 17850 with -30. Is it okay? Idk if i did something wrong, but my CPU going to like 83-84 degree on the full stress test over few minutes. In games it's like 75 degree. Is it alright? ASRock B650E PG Riptide, 6000mhz EXPO ram, Arctic Liquid Freezer III 360.
i have built 3 pc''s with 7800x3d excluding my personal pc . I really into overclocking undervolting I'm into oc since core 2 dou fsb shenanigan's. i have found that the 7800x3d can achieve full performance under 75w to change this use ppt to a max of 70 to 75 nothing higher. its recommended to force the pbo scaler to x1 anything else is considering overclock and lead to cpu degradation, the curve optimizer around -20 to -30 is doable for most chips. Also set the load line calibration to the lowest or second lowest! (if its stable you can try add some Mhz by using pbo - max cpu boost override positive values )
Maxing out the LLC will make the CPU run hotter. You will get a lower Curve Optimizer number, but your CPU will run hotter. And that's because High LLC actually overcompensates on voltage. My 7800X3D can do -39 with the LLC maxed, but runs ~8 degrees warmer in Cinibench, than - 20 with the regular LLC setting.
My cpu passed -40 on Ryzen master but wouldn't post. So i dropped it to -30 and ran cinebench got 18714 on multi core and 1828 on single core Didnt got past 75.3°C with a 360mm aio. Did i do it right?.. can i play call of duty like a pro now?