@@SkatterBencher Okay, I've played with CO quite a bit since replacing my 3700X with a 5800X a few months ago and I've arrived here in the present with much confusion and a little frustration... Obviously the lack of any documentation on CO whatsoever (that most end users can access anyways) makes this all the more frustrating as I'm fairly certain SOMEONE at AMD wrote down how best to utilize this feature... 😈😈😈😈😈 I've come up with three or four different permutations of a "stable" curve via stability testing with CoreCycler and OCCT Pro that present with great boost behaviors and decent bench scores (some curves better than others) without full system crashes interrupting my usage or restarting the PC overnight... Yet I can't discern which of the curves meet the designing engineer's intent and/or best practices Some curves use higher positive curve offsets than I've seen posted around the internet by other users, some use higher negative curve offsets Some curves have relatively large disparities in the amount of curve offset relative to some other curves This would make sense were I comparing different CPUs (I'm not, this is all with the same 5800X) or different versions of mobo BIOS (only two and the discrepancies are present throughout, I flashed an older version of a stable BIOS to see what would be different but, alas, not much) but all of this was like to like with regards to the aberrant behaviors I'm seeing Any ideas? Anything to share that won't land you in hot water with the NDA enforcers over at AMD? It would be great to get some clarification here... 😂😂😂😂😂
The third video I see and I just subscribed. I own a 5600x and your videos about undervolt are well described and helped me to understand pbo 2.0. Great work.
phenomenally awesome video as usual. amazing how this cpu could reach such a high memory buss speed maximum vs say a 5800x that limits at 1800-1900. with video cards being so ridiculously priced this built in apu, cpu will be highly considered vs past considerations if becomes readily available
Thanks for the kind words! APUs have been known to be very easy-going on the Fclk. Cezanne's successor Renoir also had good Fclk overclocking results: www.reddit.com/r/Amd/comments/hhydnq/renior_eng_sample_2166mhz_fclk/ www.reddit.com/r/Amd/comments/hfhpog/amd_renoir_at_fclk_2100mhz/ wccftech.com/amd-ryzen-7-4700ge-renoir-apu-ddr4-4333-test/
Crazy amount of work here... I've been recently playing with a 4750G and have found it rather enjoyable to try and get every last little bit of performance out of.
Nice! I played around with Renoir a long time ago but ended up not making the video. The performance increase should be spectacular though ... it's the same integrated graphics, but much lower base clock than on Cezanne.
@@SkatterBencher I picked up a 5700G last week and ran some comps between my 4750G and 5700G, same components, and you might be surprised how little difference there is. For instance, Night Raid on a properly overlocked 4700G returned roughly 17800, on similar OC settings with the 5700G around 18100. I'm not using anywhere near as fast DDR4 as you are. FFXV bench shows zero difference and CS:GO shows about a 1 to 2 fps difference between the two configurations. I was surprised myself. I used your video as a tutorial for settings by the way and greatly appreciate your hard work!
Thanks for the kind words! Yea I'm not too surprised about the small difference in performance since Renoir to Cezanne only has the upgrade from Zen 2 to Zen 3 on the CPU side but no new GPU. Just a bit higher base frequency. Thanks for sharing your info!
great job Massman :) It was really hard work with Curve optimizer per core :) But I agree, it is best way for higher performance on Zen3 CPUs and daily coolings (like also VF curve on Intels). Old methods allcore frequecny are almost death today, usable only for extreme overlcocking :)
Thanks! Yea I agree on old vs new methods. It's all about hybrid overclocking nowadays. I hope Precision Boost Overdrive 3 will give us more tools to edge out the maximum even further.
5700G made in 2023 for china - 1.24v/4.6Ghz all cores full stable for avx, 5700G made in 2022 for global users - 1.32/4.6Ghz all cores. therefore, keep in mind that there is a HUGE variation in the quality of chips during overclocking. I was lucky that there were 2 identical processors and I just changed them and noticed such a difference
Wow. I've built 3 pcs with 5700Gs and none of mine were even stable at 4.5Ghz 1.325V with AVX2 manual all core. Recently built a home server with a 5700G manufactured 5th week 2023 (2305SUS) and that cannot run 4.6Ghz at 1.24V. That one needs 1.38V (yikes) to even run prime95 stable for 30 seconds. Package power is 150W at 87C. Whats the full date code on your processor? Where did you buy yours? I'd like to get a good one for once :) Thanks for your insight! Edit: My CPU is one of the newer china models (removal of taiwan text).
Great Video as always. You surely deserve more subscriptions man... I am currently also testing out an APU - 4650G in particular. Any experiences with iGPU Clock and safe voltages for these 4000 Series APUs? Which Voltages affect iGPU stability? I am on MSI so I do not have those GPU Boost settings like on an ASUS Board. I guess that Boost setting automatically adjusts related voltages as well.
Thank you for the thorough algo! Gonna use it for my 5600x after I OC the memory. There is a question though: did you encounter a low-current workload BSODs/crashes/restarts while running Curve Optimizer? I've seen multiple reports on reddit of that and haven't seen fixes yet.
Thanks mate im Hispanic, so thanks for literal speak good! Which of the Strategies did you recommend for gaiming? Setup B550M-PLUS tuf 3000 Corsair 2*8 Already has the strategy 1 with my cpu around 80C
Very informative and educational video! Great job! I just bought a Ryzen 7 5700G with a Patriot Viper 4400MHZ RAM. To get the maximum rated performance of the memory, do I just change the FCLK frequency to 2200Mhz? or do I need to overclock the CPU with one of the methods you showed. I didn't realize that 4400Mhz is way above the advertised CPU memory support of 3200Mhz. I'll greatly appreciate the help. Thanks.
TLDR; Yes, to maximize the performance benefit from your DDR4-4400 memory it would be best to run FCLK at 2200MHz. Long answer: same as the TLDR but with the caveat that not all CPUs can run the FCLK at 2200 MHz, so you'll need to test if it works for your. I think my 5700G could boot 2200 MHz but was not stable. Also, most if not all motherboards have an auto-rule that will reduce the FCLK frequency when the memory frequency is over DDR4-3600. So, if you'd just enable XMP to get the DDR4-4400 the autorule would probably run the FCLK at less than 2200 MHz. In terms of overclocking strategies: I think OC Strategy #2 would be a good fit for your case. But you can always try the others (or even your own) to see what suits you best. In the end, what matters is that you're happy with your settings. Good luck! :)
I think stock should these days be done including XMP for AMD processors no ? and check what putting PBO and other dram settings bring to the table against cpu stock + xmp. Just a thought. what do you think oh and, thanks a lot for taking the time to make videos like this, helped a lot.
I understand your reasoning and it's a perfectly legitimate point of view. "Stock" performance is a bit of a grey area because (1) most motherboards enable PBO by default even though it's technically overclocking and (2) most motherboards will run DDR4-2133 out of the box instead of either the maximum guaranteed spec of DDR4-3200 or the memory XMP. (1) is obviously because motherboard vendors don't want to lose against their competitors in the benchmark charts. Personally, I always disable PBO for my stock performance because, well, PBO is not in-spec behavior. (2) is mostly for compatibility reasons. There are so many different DRAM ICs, PCBs, configurations, etc out there that it's virtually impossible to ensure all combinations with all CPUs will work out of the box. Even for DDR4-3200. My choice here is that I measure stock as what the system defaults are. For DDR4 that's usually DDR4-2133 and for DDR5 it'll be DDR4-4800. I agree there is an argument to be made "stock" should be with the maximum guaranteed specification for the memory. For Ryzen 5000 that's DDR4-3200. I don't feel there's a good case for using XMP because (a) XMP will differ between kits, so there's no universal performance baseline, and (b) XMP ... is actually an Intel technology :)
at last... youtube would send me all kind of unrelated videos for other cpus smh thanks! any way to rate the cpu like CRT? to see a measurement of how good my specific hardware is. Thanks!
When you say clock streching happens when effective freq. is significantly lower than normal freq. does 20-30mhz count or would it be more like half / in your example?
Just bought this system and Im going to swap it into my asus X470-F, let you know how it goes! going to pair it with my 2080 super, see how it does compared to my old 3600x 1900fclk with B-Die
Good video! Maybe you have input for my 5600X? -30 neg offset is stable up to pbo +50. Above it becomes unstable in certain single core scenarios. It seems the 2 worst core don't get utilized when doing SC work, but cycles between the 4 others. Would a lower neg offset on the 2 worst cores improve stability with pbo 100+?
Thanks for the kind words! So far I only thoroughly tried Curve Optimizer with this one CPU, so I'm not confident to say that my approach will work universally. However, from my observation Fmax itself doesn't really cause instability. It just lifts the ceiling of the maximum frequency. It doesn't increase the frequency by itself. However, in certain scenarios it's possible that PBO algorithm targets a certain voltage and frequency that isn't quite stable for a specific core. In order to test this, you'd have to check the stability of every core separately. The way I would approach this would be: - First, do the Curve Optimizer tuning for all-core OC like shown in the video from 15:37 - Then, load the same Prime95 and select 2 threads instead of 16 threads load. Then set alternate the CPU affinity in task manager and cycle through each core to see if there's any instability or clock stretching The point you brought up made me realize I missed this point of stability testing in my video. So, next time I play around with Curve Optimizer I will also test for single core stability. Great input!
@@SkatterBencher thanks for the reply! I fiddled around a bit before you replied and I increased the powerbudget at bit and that allowed for a bit higher voltage and so far it has been stable. MC holds 4650-4700MHz in CB20 now and voltage around 1.19V with moderate LLC. SC hold 4850MHz at about 1.3V. I can run unlimited as you as I'm on a budget B550 board with a small heatsink on vrm. My cooler (BQ Dark rock slim 180W) gets overwhelmed if consumption rises about 110W (4.8@1.32V oc does that) with thermal throttling so I must set max PPT to 100 if I want no thermal throttle and temps below 90C during stress. I will try your stability tips with P95 etc when I get back from work :)
@@SkatterBencherhey so I got the 5700g and the rx 6600. Along with a msi b550 edge, Just getting into pc building. Do I follow your settings exactly to maximize just the cpu without focusing any more power towards the integrated graphics?
Using the p95 method at 26:20 with -30 all core on my 5600x, I get zero difference beween core clock and effective clock in hwinfo while under load. Even though I know that -30 is NOT stable as it has crashed in the past. I usually run -10 all core. I wonder why its not showing clock stretching?
I tried this on my Ryzen 5600, most settings are Auto and I set -30 on all cores, when I stress test the effective clock seems to stay around 4,450Mhz on all cores, I guess.....that's good?....
This was the first time I included a "real" game in my videos (CS:GO). The most difficult part is to find an easy way to benchmark the games. I don't have the time to play through the games each time I run a test. Also, the primary purpose of showing the benchmark data points is to show the relative performance gain after overclocking. It's not really intended to be a close reflection of real-world performance. If there's an easy way to include popular esports titles, I'm all for it though!
@@SkatterBencher some games have in game benchmarks like shadow of the tomb raider and here is a link to games with in game benchmarks if you are curious www.google.com/amp/s/xsreviews.co.uk/news/games-with-built-in-benchmarks-how-to-benchmark-your-pc/%3famp . Though most games played on APUs like this are played on 720p to get closer to break over 30fps
Thx for so informative explanation. Wish there a one more guide but with Gigabyte mobo. Tried to go PBO approach, but realized that many of these settings i cant find on Gigabyte mobo, or if have - it uses different values (eg. Instead of Level 1-4, it uses High, mid, low, etc)
@@SkatterBencher Yay, nice one.. Waiting for this.. Coz I did not used Gigabyte boards before, still not figured out where they "hide" all the settings...
Tried 4th Strategy with manual Curve Optimizer per clock, got +20-25% performance. Main question if this is OK for daily gaming use. In some games, on heavy load I see consistent 1.4-1.45 CPU voltage, temperature is still pretty low though, 60-65C with far from the best air cooler. Is it some kind of a normal behavior for 5th gen Ryzen or I should worry about that high voltage?
1.4-1.45V is normal behavior for a Ryzen CPU even at default settings. As long as you have Precision Boost enabled, the algorithm should manage the voltage appropriately (within safety limits).
Hi! I did a SFF build with this in the inwin chopin pro (200w PSU). I used the viper steel 4400 2x8 kit. The best cooler I could find was the thermalright axp90 x47full cooler, noctua A9x14 and used arctic silver paste. Any chance you could do a SFF overclocking guide with the 200w chopin to maximize gaming performance? Trying to build the ultimate tiny gaming PC. A few things I noticed, the memory FLCK would target 2200 in the bios but would only ever be 2000 in the OS. Setting it higher than 2000 was causing instability. So I set the ram at DOCP II, 4266 19/39 1.35v. DOCP 1 was 1.45v 4400 19/39, and the ram was noticeably hot. Was also getting instability with the GPU boost at extreme (2300mhz) with the ram at DOCP I / 1.45v. With the ram at DOCP II 1.35v I was able to set it to extreme (2300 mhz) and be stable. Any manual setting I tried over extreme (2400, 2500 mhz) would cause serious instability, graphical failures / tears, and black screen shutdowns. My guess is that I was hitting the limit of a 200w PSU, the thermals all looked fine at under 70 deg and 300w minimum is specified by AMD. I left the CPU settings all Auto, including PBO (auto). Overall I'm happy with the gaming performance at 4000mhz ram / 2000 FLCK / 2300 mhz GPU boost, scored 1900 in Time Spy. But if I could get the gaming performance higher stably I would (looking at the GPU / FLCK / Ram speeds). Not really concerned about boosting the CPU since the GPU is heavily bottle necked.
@@PinchHarmonic69 It wasn't the PSU, the power draw was fine. The problem is silicone lottery. They give reviewers like this amazing chips that allow for settings most of us retail folks will never be able to copy. Most 5700g cannot clock over 2300 / 2400 and remain stable.
@@SkatterBencher thank you! I followed your steps and working so far. One thing I noticed. During prime95 the clock speed dropped to 4.0ghz. This only happened once when I ran it, is that normal?
Check the CPU operating temperature with HWiNFO. If the CPU temperature is at the limit (95C, I think), then the reduced frequency is the result of thermal throttling. This is very normal behavior in heavy workloads like Prime95.
If #3 worked and #4 did, then the likely reason why your system is unstable are the Curve Optimizer settings. Did you directly copy the Curve Optimizer settings from the video? The easy fix would be to manually tune the Curve Optimizer for your CPU. Start from -5 on all cores and check with an application like CoreCycler if each core is stable. If yes, change to -10. Etc.
Hi, I just got a 5700g, and it keeps crashing when doing 3DMark CPU tests when trying to overclock the RAM. I think it might need more SOC voltage. What is the safe limit for SOC voltage please?.. Some people are using 1.5v. Which is scary because with other CPUs the limit was around 1.2v I think?... Other people say 1.3v is OK because the SOC voltage is for the iGPU part of the chip as well, so it needs more than a normal CPU?... I have a MSI x470 Gaming Plus Max Crucial Ballistix 3600 cl16 - BL16G36C16U4RL.M16FE It doesn't crash with the XMP disabled, but that's no good, too slow...
It's always difficult to say what's safe or not. For example, the AMD Ryzen CPU boost algorithm has a maximum limit of 1.55V which it will reach on some CPUs (but only in special situations). For VccSOC I'd generally be comfortable with up to 1.35V. I would only set higher if I see a significant improvement in stability AND it doesn't impact temperature that much.
@@SkatterBencher OK thanks. I'm running it with a DGPU at the moment, to see if I can get the ram stable at 3800 cl16 1.37v. It passed 2028% Memtest - and is 9 hours 12 cycles into a Test Mem anta777 Extreme 500% - so far. I was trying to use the SOC v offset - 0.15v and it still crashed. Should I use offset or forced for upping the SOC v please?... And wouldn't running it forced just waste power?.. It won't auto adjust like with offset?... Should I put it under load with OCCT cpu test and Furmark together and watch "SoC Voltage (SVI2 TFN)" in HWmonitor and offset it until it gets to 1.35v?... If it crashes at that, I'll have to give up on the iGPU part...
@@SkatterBencher Been messing with it. No offset on the SOC will work. 1.3v Override works and seems stable, but I still don't feel comfortable using it. That is the max this MB will allow. I asked AMD, I wonder if they will tell me what is the safe upper limit for SOC... Other people are probably using XMP without touching the SOC; why can't I?.. I really don't want to degrade the chip...
@@SkatterBencher I put the SOC back to auto and changed the GFX voltage to 1v, it sorts it out too, stops the crashes. HWm shows about 1.15v SOC and GPU voltage, it never goes lower when changing the GFX voltage; on auto it goes down to 0.75v. I guess having the voltage fixed, not auto adjusted helps. I want to have it auto adjust though. I shouldn't need to fix the GFX voltage to get stability.
Bah, I give up with this APU, it's too weird. There's some bug that AMD still hasn't fixed, making the mouse lag with USB 3.2 as well. Maybe I bought a lemon one, and other 5700gs are ok. Maybe the RAM is just not compatible - (there isn't even a RAM compatibility list on the MSI website for Ryzen Cezanne/Vermeer. Maybe they would fix the issue in a BIOS update later. Can't be bothered with it. Going back to the 3600xt.
The memory is set to DDR4-2133 for the stock benchmark ruclips.net/video/4gDfMwOBnoE/видео.html. This is what the system will set the memory frequency as when not adjusting any setting. What would be the right stock frequency for the memory according to you?
@@timeDrapery he's right, "stock" should be 3200, not 2133, since 3200 is the AMD's "stock" supported spec. And +60% performance gain is just hilarious and not true at all.
@@timeDrapery 3200mhz is officially supported memory speed that is guaranteed for amd's 5000 series cpu so as 2133. You can call both stock its just different spectrum. But using it as a base line in my opinion is relatively pointless and potentially misleading as anyone who watches these videos definitely using xmp/docp so 2133mhz numbers are pretty much irrelevant to vast majority of viewers here, but its good to have both tests completed.
@@chovekb ehh both can be considered stock but i agree with you that 60% performance gain is slightly misleading and mostly irrelevant. Video author should have included 2133mhz results but exclude them from being a base line, his "PBO + DOCP + GPU Turbo" at 36:25 should have been a baseline not that with lowest jedec speed ram.
OMG such a "huge" slow down from 80 fps to 75 fps LOOL And it still has a decent igpu and better memory controller doing 2200+ IF and 4400+ ram EZ, and also is cheaper.
@@Mr11ESSE111 your general assumptions without knowing in fact are garbage LOOL All AAA games are running with 30+ fps on a 5700g, especially with 4400 ram with tight timing LOLOL check the facts before talking garbage next time!
Extremely detailed and informative video about a piece of hardware that is very important right now but severely underreported. Thanks! Subscribed ;)
Glad you enjoyed it!
@@SkatterBencher
Okay, I've played with CO quite a bit since replacing my 3700X with a 5800X a few months ago and I've arrived here in the present with much confusion and a little frustration...
Obviously the lack of any documentation on CO whatsoever (that most end users can access anyways) makes this all the more frustrating as I'm fairly certain SOMEONE at AMD wrote down how best to utilize this feature... 😈😈😈😈😈
I've come up with three or four different permutations of a "stable" curve via stability testing with CoreCycler and OCCT Pro that present with great boost behaviors and decent bench scores (some curves better than others) without full system crashes interrupting my usage or restarting the PC overnight... Yet I can't discern which of the curves meet the designing engineer's intent and/or best practices
Some curves use higher positive curve offsets than I've seen posted around the internet by other users, some use higher negative curve offsets
Some curves have relatively large disparities in the amount of curve offset relative to some other curves
This would make sense were I comparing different CPUs (I'm not, this is all with the same 5800X) or different versions of mobo BIOS (only two and the discrepancies are present throughout, I flashed an older version of a stable BIOS to see what would be different but, alas, not much) but all of this was like to like with regards to the aberrant behaviors I'm seeing
Any ideas? Anything to share that won't land you in hot water with the NDA enforcers over at AMD? It would be great to get some clarification here... 😂😂😂😂😂
Great video, great work, actually a highly underrated channel, thanks for the grat content.
This video deserves more likes.
Extremely underrated.
I respect his efforts. ✨
Thanks for the kind words!
The third video I see and I just subscribed. I own a 5600x and your videos about undervolt are well described and helped me to understand pbo 2.0. Great work.
Thanks for the sub!
phenomenally awesome video as usual. amazing how this cpu could reach such a high memory buss speed maximum vs say a 5800x that limits at 1800-1900. with video cards being so ridiculously priced this built in apu, cpu will be highly considered vs past considerations if becomes readily available
Thanks for the kind words!
APUs have been known to be very easy-going on the Fclk. Cezanne's successor Renoir also had good Fclk overclocking results:
www.reddit.com/r/Amd/comments/hhydnq/renior_eng_sample_2166mhz_fclk/
www.reddit.com/r/Amd/comments/hfhpog/amd_renoir_at_fclk_2100mhz/
wccftech.com/amd-ryzen-7-4700ge-renoir-apu-ddr4-4333-test/
All that work is appreciated. Very good research report and tutorial 👍
Much appreciated!
1.35v on vsoc scared me 😹
Crazy amount of work here... I've been recently playing with a 4750G and have found it rather enjoyable to try and get every last little bit of performance out of.
Nice! I played around with Renoir a long time ago but ended up not making the video. The performance increase should be spectacular though ... it's the same integrated graphics, but much lower base clock than on Cezanne.
I would love to see this video! I just got a 4750 g today that I am trying to tweak as high as possible
@@SkatterBencher I picked up a 5700G last week and ran some comps between my 4750G and 5700G, same components, and you might be surprised how little difference there is. For instance, Night Raid on a properly overlocked 4700G returned roughly 17800, on similar OC settings with the 5700G around 18100. I'm not using anywhere near as fast DDR4 as you are. FFXV bench shows zero difference and CS:GO shows about a 1 to 2 fps difference between the two configurations. I was surprised myself. I used your video as a tutorial for settings by the way and greatly appreciate your hard work!
Thanks for the kind words! Yea I'm not too surprised about the small difference in performance since Renoir to Cezanne only has the upgrade from Zen 2 to Zen 3 on the CPU side but no new GPU. Just a bit higher base frequency.
Thanks for sharing your info!
Nice video as usual :)
Thanks!!
This video explained everything! Dude you deserve more public!
Thanks!
I would love to see a fresh video with more details on all overclocking scenarios using Asus Dark Hero and 5950X, much appreciate it!
Great video! I would love to see FPS perpfmance in esports titles, stock vs. OC.
Dude, your vid is just amazing. Thank you man.
I'd like to the how memory sub-timings add to these improvements. That Trc is sooooooo high, I'd bet Trfc is really high too with XMP.
great job Massman :) It was really hard work with Curve optimizer per core :) But I agree, it is best way for higher performance on Zen3 CPUs and daily coolings (like also VF curve on Intels).
Old methods allcore frequecny are almost death today, usable only for extreme overlcocking :)
Thanks! Yea I agree on old vs new methods. It's all about hybrid overclocking nowadays.
I hope Precision Boost Overdrive 3 will give us more tools to edge out the maximum even further.
Hi i have a 4650g. I dont find anything about soc Voltage. Mine runn @ 1.225V@SoC. Should i try a higher SoC? Your explenation is awesome.
nice work, finest content
Thanks!
Hi any idea why i dont see my GPU boost?
Perfect! Very very good Job 👍🏻👍🏻
Bro este es el video perfecto para aquellos que armaron su equipo durante la crisis de las gpu (como yo) muchisimas gracias UwU
Gracias por el cumplido!
5700G made in 2023 for china - 1.24v/4.6Ghz all cores full stable for avx, 5700G made in 2022 for global users - 1.32/4.6Ghz all cores. therefore, keep in mind that there is a HUGE variation in the quality of chips during overclocking. I was lucky that there were 2 identical processors and I just changed them and noticed such a difference
Wow. I've built 3 pcs with 5700Gs and none of mine were even stable at 4.5Ghz 1.325V with AVX2 manual all core.
Recently built a home server with a 5700G manufactured 5th week 2023 (2305SUS) and that cannot run 4.6Ghz at 1.24V. That one needs 1.38V (yikes) to even run prime95 stable for 30 seconds. Package power is 150W at 87C.
Whats the full date code on your processor? Where did you buy yours? I'd like to get a good one for once :)
Thanks for your insight!
Edit: My CPU is one of the newer china models (removal of taiwan text).
very detailed and educational sharing indeed, I am gonna try to boost my pc's speed further.
Great Video as always. You surely deserve more subscriptions man...
I am currently also testing out an APU - 4650G in particular. Any experiences with iGPU Clock and safe voltages for these 4000 Series APUs? Which Voltages affect iGPU stability? I am on MSI so I do not have those GPU Boost settings like on an ASUS Board. I guess that Boost setting automatically adjusts related voltages as well.
I got really lucky with my testing for the PBO optimization: -30 crashed underload and on -25 all cores were stable.
so fkin detail and ofc confusing for me as a beginner. but ill save this video until I can understand everything. lol
I tried all your sellings and bricked my motherboard, I had to remove the CMOS and reset bios. I have rog x570f with 5700G
Thank you for the thorough algo! Gonna use it for my 5600x after I OC the memory.
There is a question though: did you encounter a low-current workload BSODs/crashes/restarts while running Curve Optimizer? I've seen multiple reports on reddit of that and haven't seen fixes yet.
I did not encounter any issue on my system, but I've heard of this before
@@SkatterBencher I wonder, are you going to cover Clock Tuner for Ryzen on your channel (to compare with your strategy #4 for example)?
Thanks mate im Hispanic, so thanks for literal speak good!
Which of the Strategies did you recommend for gaiming?
Setup
B550M-PLUS tuf
3000 Corsair 2*8
Already has the strategy 1 with my cpu around 80C
Very informative and educational video! Great job! I just bought a Ryzen 7 5700G with a Patriot Viper 4400MHZ RAM. To get the maximum rated performance of the memory, do I just change the FCLK frequency to 2200Mhz? or do I need to overclock the CPU with one of the methods you showed. I didn't realize that 4400Mhz is way above the advertised CPU memory support of 3200Mhz. I'll greatly appreciate the help. Thanks.
TLDR; Yes, to maximize the performance benefit from your DDR4-4400 memory it would be best to run FCLK at 2200MHz.
Long answer: same as the TLDR but with the caveat that not all CPUs can run the FCLK at 2200 MHz, so you'll need to test if it works for your. I think my 5700G could boot 2200 MHz but was not stable. Also, most if not all motherboards have an auto-rule that will reduce the FCLK frequency when the memory frequency is over DDR4-3600. So, if you'd just enable XMP to get the DDR4-4400 the autorule would probably run the FCLK at less than 2200 MHz.
In terms of overclocking strategies: I think OC Strategy #2 would be a good fit for your case. But you can always try the others (or even your own) to see what suits you best. In the end, what matters is that you're happy with your settings. Good luck! :)
All i can say is wow.This guys good.
Cheers, much appreciated!
If I can get 60% more of this apu it’s totally worth it over a 1050ti combo
Kkkkk
I think stock should these days be done including XMP for AMD processors no ? and check what putting PBO and other dram settings bring to the table against cpu stock + xmp. Just a thought. what do you think
oh and, thanks a lot for taking the time to make videos like this, helped a lot.
I understand your reasoning and it's a perfectly legitimate point of view.
"Stock" performance is a bit of a grey area because (1) most motherboards enable PBO by default even though it's technically overclocking and (2) most motherboards will run DDR4-2133 out of the box instead of either the maximum guaranteed spec of DDR4-3200 or the memory XMP.
(1) is obviously because motherboard vendors don't want to lose against their competitors in the benchmark charts. Personally, I always disable PBO for my stock performance because, well, PBO is not in-spec behavior.
(2) is mostly for compatibility reasons. There are so many different DRAM ICs, PCBs, configurations, etc out there that it's virtually impossible to ensure all combinations with all CPUs will work out of the box. Even for DDR4-3200. My choice here is that I measure stock as what the system defaults are. For DDR4 that's usually DDR4-2133 and for DDR5 it'll be DDR4-4800.
I agree there is an argument to be made "stock" should be with the maximum guaranteed specification for the memory. For Ryzen 5000 that's DDR4-3200. I don't feel there's a good case for using XMP because (a) XMP will differ between kits, so there's no universal performance baseline, and (b) XMP ... is actually an Intel technology :)
with a good AIO will the cpu degrade faster at those voltages?
at last... youtube would send me all kind of unrelated videos for other cpus smh
thanks!
any way to rate the cpu like CRT? to see a measurement of how good my specific hardware is. Thanks!
When you say clock streching happens when effective freq. is significantly lower than normal freq. does 20-30mhz count or would it be more like half / in your example?
Next time try to use CoreCycler or thread switcher to test Curve Optimizer, they are great
Oh awesome! I've been looking for something like this for months now. Thank you very much for sharing!!
@@SkatterBencher you're welcome. At least I prefer to use CoreCycler, because it is easy to customize.
Good stuff.
Glad you enjoyed it
Just bought this system and Im going to swap it into my asus X470-F, let you know how it goes! going to pair it with my 2080 super, see how it does compared to my old 3600x 1900fclk with B-Die
Good luck! Looking forward to hearing how it goes!
Good video! Maybe you have input for my 5600X? -30 neg offset is stable up to pbo +50. Above it becomes unstable in certain single core scenarios. It seems the 2 worst core don't get utilized when doing SC work, but cycles between the 4 others. Would a lower neg offset on the 2 worst cores improve stability with pbo 100+?
Thanks for the kind words!
So far I only thoroughly tried Curve Optimizer with this one CPU, so I'm not confident to say that my approach will work universally. However, from my observation Fmax itself doesn't really cause instability. It just lifts the ceiling of the maximum frequency. It doesn't increase the frequency by itself.
However, in certain scenarios it's possible that PBO algorithm targets a certain voltage and frequency that isn't quite stable for a specific core. In order to test this, you'd have to check the stability of every core separately. The way I would approach this would be:
- First, do the Curve Optimizer tuning for all-core OC like shown in the video from 15:37
- Then, load the same Prime95 and select 2 threads instead of 16 threads load. Then set alternate the CPU affinity in task manager and cycle through each core to see if there's any instability or clock stretching
The point you brought up made me realize I missed this point of stability testing in my video. So, next time I play around with Curve Optimizer I will also test for single core stability. Great input!
@@SkatterBencher thanks for the reply! I fiddled around a bit before you replied and I increased the powerbudget at bit and that allowed for a bit higher voltage and so far it has been stable. MC holds 4650-4700MHz in CB20 now and voltage around 1.19V with moderate LLC. SC hold 4850MHz at about 1.3V. I can run unlimited as you as I'm on a budget B550 board with a small heatsink on vrm. My cooler (BQ Dark rock slim 180W) gets overwhelmed if consumption rises about 110W (4.8@1.32V oc does that) with thermal throttling so I must set max PPT to 100 if I want no thermal throttle and temps below 90C during stress.
I will try your stability tips with P95 etc when I get back from work :)
with the price of this build, it'd be more economical to get a dedicated GPU :))) great detail on the settings though ;))))
I was so preoccupied with whether or not I could that I didn't stop to think if I should
@@SkatterBencherhey so I got the 5700g and the rx 6600. Along with a msi b550 edge, Just getting into pc building. Do I follow your settings exactly to maximize just the cpu without focusing any more power towards the integrated graphics?
Was planning on a asus rog strix b550-i with a 5600g but Asus says it is not supported. You just proved them wrong using a 5000g series on a b550-i?
Oops! :)
@@SkatterBencher already have everything in order so no backing down now. Did you have to update the bios or did it just worked out of the box
is this safe for daily use for gaming and streaming?
Using the p95 method at 26:20 with -30 all core on my 5600x, I get zero difference beween core clock and effective clock in hwinfo while under load. Even though I know that -30 is NOT stable as it has crashed in the past. I usually run -10 all core. I wonder why its not showing clock stretching?
I tried this on my Ryzen 5600, most settings are Auto and I set -30 on all cores, when I stress test the effective clock seems to stay around 4,450Mhz on all cores, I guess.....that's good?....
i get 15300 freom r23 with 5700g but my igpu seems like attachted a power limit? any way to unlock the igpu power limits without OC
just set ratio to 41.50 and offset negative .5 with pbo auto, is this ok for daily driver running 3600mhz docp? my soc is 1.34v
How do we know which are best cores because Ryzen Master does not work with this APU.
Pull up your bootstraps and roll up your sleeves, it's time for the manual grind :)
Are u sure? I oced my 5700g using rm
@@andia968 if you disable iGPU, you cant use Ryzen Master from my experience. But it works if iGPU is enabled.
Can u test some games preferably those esports titles on this?Hard to find some legit numbers on this
This was the first time I included a "real" game in my videos (CS:GO). The most difficult part is to find an easy way to benchmark the games. I don't have the time to play through the games each time I run a test. Also, the primary purpose of showing the benchmark data points is to show the relative performance gain after overclocking. It's not really intended to be a close reflection of real-world performance.
If there's an easy way to include popular esports titles, I'm all for it though!
@@SkatterBencher some games have in game benchmarks like shadow of the tomb raider and here is a link to games with in game benchmarks if you are curious www.google.com/amp/s/xsreviews.co.uk/news/games-with-built-in-benchmarks-how-to-benchmark-your-pc/%3famp . Though most games played on APUs like this are played on 720p to get closer to break over 30fps
Thanks Travis!
able to overclock 5600g?
The Information is legit 👍🔥
the goals to get the best Bios setting and get the max performance are achived 👍👍
Glad you think so!
Thx for so informative explanation. Wish there a one more guide but with Gigabyte mobo. Tried to go PBO approach, but realized that many of these settings i cant find on Gigabyte mobo, or if have - it uses different values (eg. Instead of Level 1-4, it uses High, mid, low, etc)
Oh you're in luck! I am just putting the finishing touches on a video with the 5300GE and B550 Aorus Master! :)
@@SkatterBencher Yay, nice one.. Waiting for this.. Coz I did not used Gigabyte boards before, still not figured out where they "hide" all the settings...
Where can I buy this cpu?
Currently the Cezanne APU is only available to AMD OEM partners. AMD said they should appear in retail channels too ... but not sure when
How about optimizing memory timings/frequency?
Tried 4th Strategy with manual Curve Optimizer per clock, got +20-25% performance. Main question if this is OK for daily gaming use. In some games, on heavy load I see consistent 1.4-1.45 CPU voltage, temperature is still pretty low though, 60-65C with far from the best air cooler. Is it some kind of a normal behavior for 5th gen Ryzen or I should worry about that high voltage?
1.4-1.45V is normal behavior for a Ryzen CPU even at default settings. As long as you have Precision Boost enabled, the algorithm should manage the voltage appropriately (within safety limits).
@@SkatterBencher Thank u for ur answer! And ur videos ;)
If you have asus rog crosshair viii can you still do this????
Yes, the overclocking process will be very similar although the final frequency and settings may differ depending on your CPU quality
Qual o ganho em fps comparado ao stock?
Trying to pull off the manual overclock on a hyper 212 cooler, can't quite get there but 2400 MHz is good.
Hi! I did a SFF build with this in the inwin chopin pro (200w PSU). I used the viper steel 4400 2x8 kit. The best cooler I could find was the thermalright axp90 x47full cooler, noctua A9x14 and used arctic silver paste. Any chance you could do a SFF overclocking guide with the 200w chopin to maximize gaming performance? Trying to build the ultimate tiny gaming PC. A few things I noticed, the memory FLCK would target 2200 in the bios but would only ever be 2000 in the OS. Setting it higher than 2000 was causing instability. So I set the ram at DOCP II, 4266 19/39 1.35v. DOCP 1 was 1.45v 4400 19/39, and the ram was noticeably hot. Was also getting instability with the GPU boost at extreme (2300mhz) with the ram at DOCP I / 1.45v. With the ram at DOCP II 1.35v I was able to set it to extreme (2300 mhz) and be stable. Any manual setting I tried over extreme (2400, 2500 mhz) would cause serious instability, graphical failures / tears, and black screen shutdowns. My guess is that I was hitting the limit of a 200w PSU, the thermals all looked fine at under 70 deg and 300w minimum is specified by AMD. I left the CPU settings all Auto, including PBO (auto). Overall I'm happy with the gaming performance at 4000mhz ram / 2000 FLCK / 2300 mhz GPU boost, scored 1900 in Time Spy. But if I could get the gaming performance higher stably I would (looking at the GPU / FLCK / Ram speeds). Not really concerned about boosting the CPU since the GPU is heavily bottle necked.
How much more performance (%) do you think you could squeeze out of it with a 300w psu?
@@PinchHarmonic69 It wasn't the PSU, the power draw was fine. The problem is silicone lottery. They give reviewers like this amazing chips that allow for settings most of us retail folks will never be able to copy. Most 5700g cannot clock over 2300 / 2400 and remain stable.
only thing that makes me cry is that im using a gigabyte mainboard >.> takes ages to find the settings :D
Hi on my gigabyte motherboard when setting max CPU boost is it negative or positive 200mhz?
You should have the option to choose between -1000 and +200
Just doing curve optimizer and +100 single core on my 5600g is less than 10% from my 5900x.
Thank you for the video and the guide! However, this has the look and tone of a HOSTAGE video. 😲
Haha, I promise you I wasn't held captive! :)
1.35 soc seems too high ?
I NEED to do this with mine
Step 3 I am to hot... 😭.. Need a cooling solution
Is this stable for daily usage?
Considering the system is stable throughout all the benchmarks I used, it should be stable for daily usage yes.
If we’re using XMP for the ram, can we skip the OC ram steps and just work on the processor?
Also can we still use these steps even if we’re using a dedicated GPU?
Yes (to both questions)
@@SkatterBencher thank you! I followed your steps and working so far. One thing I noticed. During prime95 the clock speed dropped to 4.0ghz. This only happened once when I ran it, is that normal?
Check the CPU operating temperature with HWiNFO. If the CPU temperature is at the limit (95C, I think), then the reduced frequency is the result of thermal throttling. This is very normal behavior in heavy workloads like Prime95.
After method 4 it keeps restarting out of nowhere. Any fixes?
If #3 worked and #4 did, then the likely reason why your system is unstable are the Curve Optimizer settings. Did you directly copy the Curve Optimizer settings from the video?
The easy fix would be to manually tune the Curve Optimizer for your CPU. Start from -5 on all cores and check with an application like CoreCycler if each core is stable. If yes, change to -10. Etc.
So AMD puts an automatic overclocker that they created to be completely efficient yet it voids warranty. Sounds like a dick move
Hi,
I just got a 5700g, and it keeps crashing when doing 3DMark CPU tests when trying to overclock the RAM. I think it might need more SOC voltage. What is the safe limit for SOC voltage please?.. Some people are using 1.5v. Which is scary because with other CPUs the limit was around 1.2v I think?... Other people say 1.3v is OK because the SOC voltage is for the iGPU part of the chip as well, so it needs more than a normal CPU?...
I have a MSI x470 Gaming Plus Max
Crucial Ballistix 3600 cl16 - BL16G36C16U4RL.M16FE
It doesn't crash with the XMP disabled, but that's no good, too slow...
It's always difficult to say what's safe or not. For example, the AMD Ryzen CPU boost algorithm has a maximum limit of 1.55V which it will reach on some CPUs (but only in special situations).
For VccSOC I'd generally be comfortable with up to 1.35V. I would only set higher if I see a significant improvement in stability AND it doesn't impact temperature that much.
@@SkatterBencher OK thanks. I'm running it with a DGPU at the moment, to see if I can get the ram stable at 3800 cl16 1.37v. It passed 2028% Memtest - and is 9 hours 12 cycles into a Test Mem anta777 Extreme 500% - so far. I was trying to use the SOC v offset - 0.15v and it still crashed. Should I use offset or forced for upping the SOC v please?... And wouldn't running it forced just waste power?.. It won't auto adjust like with offset?...
Should I put it under load with OCCT cpu test and Furmark together and watch "SoC Voltage (SVI2 TFN)" in HWmonitor and offset it until it gets to 1.35v?... If it crashes at that, I'll have to give up on the iGPU part...
@@SkatterBencher Been messing with it. No offset on the SOC will work. 1.3v Override works and seems stable, but I still don't feel comfortable using it. That is the max this MB will allow. I asked AMD, I wonder if they will tell me what is the safe upper limit for SOC... Other people are probably using XMP without touching the SOC; why can't I?.. I really don't want to degrade the chip...
@@SkatterBencher I put the SOC back to auto and changed the GFX voltage to 1v, it sorts it out too, stops the crashes. HWm shows about 1.15v SOC and GPU voltage, it never goes lower when changing the GFX voltage; on auto it goes down to 0.75v. I guess having the voltage fixed, not auto adjusted helps. I want to have it auto adjust though. I shouldn't need to fix the GFX voltage to get stability.
Bah, I give up with this APU, it's too weird. There's some bug that AMD still hasn't fixed, making the mouse lag with USB 3.2 as well. Maybe I bought a lemon one, and other 5700gs are ok. Maybe the RAM is just not compatible - (there isn't even a RAM compatibility list on the MSI website for Ryzen Cezanne/Vermeer. Maybe they would fix the issue in a BIOS update later. Can't be bothered with it. Going back to the 3600xt.
is it safe for long term use
running memory at 2100mhz as "stock" is like cheating, that is not stock .
The memory is set to DDR4-2133 for the stock benchmark ruclips.net/video/4gDfMwOBnoE/видео.html. This is what the system will set the memory frequency as when not adjusting any setting.
What would be the right stock frequency for the memory according to you?
@@SkatterBencher What an absolutely polite response and invitation to continue conversing in an entirely productive manner... Too bad dude is so silly
@@timeDrapery he's right, "stock" should be 3200, not 2133, since 3200 is the AMD's "stock" supported spec. And +60% performance gain is just hilarious and not true at all.
@@timeDrapery 3200mhz is officially supported memory speed that is guaranteed for amd's 5000 series cpu so as 2133. You can call both stock its just different spectrum. But using it as a base line in my opinion is relatively pointless and potentially misleading as anyone who watches these videos definitely using xmp/docp so 2133mhz numbers are pretty much irrelevant to vast majority of viewers here, but its good to have both tests completed.
@@chovekb ehh both can be considered stock but i agree with you that 60% performance gain is slightly misleading and mostly irrelevant. Video author should have included 2133mhz results but exclude them from being a base line, his "PBO + DOCP + GPU Turbo" at 36:25 should have been a baseline not that with lowest jedec speed ram.
1
🥳
who the hell says centigrade... it's Celsius. also, 85°C with a cooler is damaging to the CPU at long periods of time
you can oclocked as much you want 5700G it will be still around 15% slower in games or sometimes more then 5800x/5600x !! AMD cripled alot own apu-s
OMG such a "huge" slow down from 80 fps to 75 fps LOOL And it still has a decent igpu and better memory controller doing 2200+ IF and 4400+ ram EZ, and also is cheaper.
@@chovekb igpu-s are garbage anyway and basically only for getting video on screen and nothing more
@@Mr11ESSE111 your general assumptions without knowing in fact are garbage LOOL All AAA games are running with 30+ fps on a 5700g, especially with 4400 ram with tight timing LOLOL check the facts before talking garbage next time!
Man, this was an amazing and outstanding job explaining everything, respect. 🫡
What a wonderful explanation video. Great work!!
Glad you liked it!