Overclock.net banner
421 - 440 of 470 Posts

·
Registered
Joined
·
43 Posts
For 5700g i need to lower EDC to 120-125A to do the same. Or, with EDC limit 300-350, most of the counts should have been upped by 3-5.
I think it was you who mentioned PPT doesn't change the power limit (changes STAPM); with that in mind, do you:
1. manually set PPT to 88 , and ~90 - ~125 for TDC/EDC?
2. Manually disable the STT/STAPM/Smartshift

If so, in cases where the limit is power draw, like HWbot x265 4k HEVC mode, does your chip remain stable at 88w draw with those current limits?

For some reason test each core w corecyler works with lower voltage but once I run prime95, y-cruncher all cores, it'd just crash so i decide to not using corecyler anymore.
I don't think the cstates was causing it to crash. In my experience, if corecycler reveals an error on a core with c states disabled, it will do so with them enabled too, more frequently. Corecycler is the best way to tune CO - but it does not account for temperature variations when all cores are working on and off in a real workload. I tuned every core to the edge of stability and wqs solid in corecycler for 30 hours, and when i ran y cruncher test 14 on all threads it crashed a couple cores in an hour. Not to mention, if you have a boost override and your core optimizer is stable in a house with higher ambient temperature, you can try testing with the A/C on or create an atmosphere where the temperature is much colder than usual, and you will find that corecyxler and test 14 will expose plenty of instability, because going down in temperaturw by 5 or 10 degrees will cause your cores to hit 25-100 mhz more on low loads which you hadnt seen before. I found this out testing ycruncher in my garage where the ambient temp is 18c, test 14 boosts to 4700... whereas my house which stays at 20-22c the cores would be around ~4660
 

·
Registered
Joined
·
1,324 Posts
do you:
1. manually set PPT to 88 , and ~90 - ~125 for TDC/EDC?
2. Manually disable the STT/STAPM/Smartshift
1. Not sure, what PPT was set to (it's ignored anyway), but of course, TDC/EDC is set manually.
2. Yep, stapm should be disabled when CPU is overclocked with PBO2. STT, iirc, is legacy power management from zen/zen+ apus. Smartshift is for sharing powerbudget in mobile.

If so, in cases where the limit is power draw, like HWbot x265 4k HEVC mode, does your chip remain stable at 88w draw with those current limits?
With the 5600/5700G power things are bit different. They are mobile by nature. For what we got here as G chips, AMD is just allowed to disable STAPM, thus basically removing power limit, and at the same time didn't even bothered to clean up bios from all mobile s**t for the desktop parts.

But going back to your question, as you could probably guess there's actually no power limit, so for example, in linpack xtreme SMT, it may consume up to 130-140W, given TDC limit is not reached.
At this point the only limiter left is FIT, i.e. the less CPU temperature, the more voltage margin can be utilized, and hence higher power consumed/dissipated.
And yes, the chip is stable, because it was tuned with such operating points in mind.

PS: I personally believe there's not much sense in finding stable counts irrespective of the type of workload and operating mode the CPU is mainly gonna be used at, i.e. just reducing voltage guardbands based on individual core stress workload, not taking into account the bunch of other factors which affecting stability in a normal operation. The power/system management is smart enough to keep the CPU from crashing if regular OS scheduling used (CPPC on), unless external power supply has some issues.
 
  • Rep+
Reactions: eighty20

·
Registered
Joined
·
51 Posts
and you will find that corecyxler and test 14 will expose plenty of instability, because going down in temperaturw by 5 or 10 degrees will cause your cores to hit 25-100 mhz more
Damn you're right. My PC crashed in timespy stress test when room temperature goes below 20. TIL AMD CPU is also behaving similar to GPU. Awesome, TY.
I have my room as cold as possible now and I also hook an extra portable air conditioner 1.0 HP right in front of the case. The temperature in the case right now is 16 Celsius. I'm gonna try to keep temperature around 16C since it rarely goes below 16 during a couple of recent years.
I have my RAM running at 4533C17 flat 2T GDM off now w 1.48V, subs are as tight as possible. And I'm able to stress 3dmark with iGPU 2300 without even need to reduce memory clock anymore. Dang, i always thought stock cpu is stable.
Edit: Now there's one ore more core can't not even set CO at -5. PBO is 130 85 130, 10x scalar, 200MHz core boost. Temperature in the case is 10C , seems like it can't go any lower.
 

·
Registered
Joined
·
1,324 Posts
Not to mention, if you have a boost override and your core optimizer is stable in a house with higher ambient temperature, you can try testing with the A/C on or create an atmosphere where the temperature is much colder than usual, and you will find that corecyxler and test 14 will expose plenty of instability, because going down in temperaturw by 5 or 10 degrees will cause your cores to hit 25-100 mhz more on low loads which you hadnt seen before.
This can happen mainly in heavy workloads, but the reason of such instabilities may be not the frequency, cause the target core voltage derived from both frequency and temp. Actually, the global frequency is calculated first, and the target core VID is defined based on fuse VFT curves and guardbands.
Imho, all crashes caused by ambient temp variance, results from either not accurate enough BTC (boot time calibration) or too agressive CO counts.
There are plenty of various thresholds in SMU to fine tune, such as core Cac weights, C-state boost margins, etc. If properly tuned, I think even modest quality silicon sample should not fail in these scenarios.
 

·
Registered
Joined
·
43 Posts
@eighty20 once you change the curve optimizer to something other than 0 you are not at stock

And if you run high ram frequencies you are arguably not at stock

And if your motherboard manufacturer says your board can do ram freqs up to 4400 and you go above that you are not on a stock system any more
 

·
Registered
Joined
·
1,324 Posts

·
Registered
Joined
·
1,324 Posts
I just wanna make sure it won't be degraded after 6 months.
There's no reason to worry if you not gonna manually OC it. Regarding your question about 4400CL18, it really depend on your DDR modules.
I don't think I can help you with anything other than Samsung b-die chips, because I've never dealt with other samsung dice, microns, hynix, etc, sorry.
 
  • Rep+
Reactions: eighty20

·
Registered
Joined
·
51 Posts
There's no reason to worry if you not gonna manually OC it. Regarding your question about 4400CL18, it really depend on your DDR modules.
I don't think I can help you with anything other than Samsung b-die chips, because I've never dealt with other samsung dice, microns, hynix, etc, sorry.
Thanks. Looks like AMD heading toward no OC for CPU and pbo is doing great so far for me.

My memory kit is samsung B-die. I confirmed w a few people, they all said that Tforce xtreem >4133 ( non rgb) use samsung b-die only. And tforce taiwan also claimed it samsung B-Die. Well, it's just the original B-Die, not the special one that they sold for a limited number a couple years ago.

Anyway, i gave up trying to get gdm off 1T at 4400, so much less headache. I'm having them run at 4600C16 GDM on now, benchmark numbers are quite impressive above 4600.
 

·
Registered
Joined
·
1,324 Posts

·
Registered
Joined
·
51 Posts
Yeah... I was looking at your old screenshots with crucial and gskills [email protected], and didn't see Tforce. In this case, 4400 CL18 1T is definitely possible.
Azure Rectangle Font Screenshot Software


This is my current memory setting, overclock mode is manual. 4200C16 and 4400C18 giving either error 1 or 6.

@Conradical I'm still trying to get CPU stable, corecycler isn't foolproof solution so my method for now is running a lot of y-cruncher 2.5B, blender, aida64 memory stress test, a bunch of 3dmarks benchmark looped and stress testing. I have my CO -19 all except -13 core 5 320ppt 160tdc&edc, 10x scalar + 200MHz. These are with C-states, dfstates auto. I'm not sure if it's just me but whenever I disable C-states, the same memory tested stable start giving weird errors so I set them all on auto.
 

·
Registered
Joined
·
1,324 Posts
This is my current memory setting, overclock mode is manual. 4200C16 and 4400C18 giving either error 1 or 6.
Idk what settings were used at 4200 and 4400, but for me your tRFC 140ns, tRRDL 4, RTP/WR 5/10 seems too tight.
As well as RDRD and WRWR should be 5 and 7 for the single ranks.
ProcODT for the 5700G UMC is kinda low. I found 43Ohms to be the most stable, but you should find your optimal impendance value yourself.
 
  • Rep+
Reactions: eighty20

·
Registered
Joined
·
526 Posts
View attachment 2556526

This is my current memory setting, overclock mode is manual. 4200C16 and 4400C18 giving either error 1 or 6.

@Conradical I'm still trying to get CPU stable, corecycler isn't foolproof solution so my method for now is running a lot of y-cruncher 2.5B, blender, aida64 memory stress test, a bunch of 3dmarks benchmark looped and stress testing. I have my CO -19 all except -13 core 5 320ppt 160tdc&edc, 10x scalar + 200MHz. These are with C-states, dfstates auto. I'm not sure if it's just me but whenever I disable C-states, the same memory tested stable start giving weird errors so I set them all on auto.
This is with C-states disabled

Colorfulness Purple Rectangle Font Violet
 

·
Registered
Joined
·
40 Posts
That's nice but i have 4533C17 flat 2T 1.48V (same subtimings at 4000) passed absolut 36 cycles with or without C-states enable awhile ago.

@Conradical I have my ram stable at 4000C15 true 1T, I think it's rock solid after passed a lot of absolut cycles, karhu and other ram stress testing. And I tested C-states enable and disable today, went thru all stress test such as blender, y-cruncher ( benchmark, fft,n32,n64), corona benchmark, OCCT small fft mode extreme, bunch of games and 3dmarks and some rendering. It's funny that my CO didn't crash in both states. I have -19 all cores except one -13, not sure if VCORE offset -0.8V did something since i want voltage stay below 1.3V. Anyway, with c-states disabled, i got a bit higher scores in cinebench, but enable helps lower power a lot lower when idling.

Edit: i just realized i logging w different email on Brave browser. I usually use firefox. So i have created 2 different accounts here.
 

·
Registered
Joined
·
43 Posts
View attachment 2556526

This is my current memory setting, overclock mode is manual. 4200C16 and 4400C18 giving either error 1 or 6.

@Conradical I'm still trying to get CPU stable, corecycler isn't foolproof solution so my method for now is running a lot of y-cruncher 2.5B, blender, aida64 memory stress test, a bunch of 3dmarks benchmark looped and stress testing. I have my CO -19 all except -13 core 5 320ppt 160tdc&edc, 10x scalar + 200MHz. These are with C-states, dfstates auto. I'm not sure if it's just me but whenever I disable C-states, the same memory tested stable start giving weird errors so I set them all on auto.
Corecycler isnt foolproof but damn is it cool software. One time i planned on running it over night but I just completely forgot about it and it ran for 3 days lol. It exposed a couple cores during that run, lol.

Also dude, make it easier on yourself. You need to be running like 18 cycles of testmem5 anta extreme. (idk about absolut, extreme1 is just better to me)

I can tell you spend a lot of time doing this, as do I, so do yourself a favor and run 18 cycles of extreme1 with your case fans on low to really torture it. Then you can know for certain that its not your memory when youre doing all these tests.

Matter of fact, you need to be running extreme1 tm5 with all those tests anyways, because if your curve optimizer is not stable, extreme1 WILL show it. If you run it for enoigh cycles. It does such rapid swapping of memory loads and your cpu clocks stay pretty much maxed. Then youre gonna be in the rabbit hole of not knowing if its your curve or memory, *****

Sounds like you got a good chip though.
 

·
Registered
Joined
·
43 Posts
That's nice but i have 4533C17 flat 2T 1.48V (same subtimings at 4000) passed absolut 36 cycles with or without C-states enable awhile ago.

@Conradical I have my ram stable at 4000C15 true 1T, I think it's rock solid after passed a lot of absolut cycles, karhu and other ram stress testing. And I tested C-states enable and disable today, went thru all stress test such as blender, y-cruncher ( benchmark, fft,n32,n64), corona benchmark, OCCT small fft mode extreme, bunch of games and 3dmarks and some rendering. It's funny that my CO didn't crash in both states. I have -19 all cores except one -13, not sure if VCORE offset -0.8V did something since i want voltage stay below 1.3V. Anyway, with c-states disabled, i got a bit higher scores in cinebench, but enable helps lower power a lot lower when idling.

Edit: i just realized i logging w different email on Brave browser. I usually use firefox. So i have created 2 different accounts here.
Did you even watch the video? None of the benchmarks you listed besides Y-cruncher are even remotely good at singling out core optimizer settings. In fact, some of them likely never will, because the cores that are unstable will be throttled by the cores that need higher voltage. I even gave you an EXACT config file to use that will demonstrate proof to you yourself within 5 minutes.
 

·
Registered
Joined
·
51 Posts
Did you even watch the video? None of the benchmarks you listed besides Y-cruncher are even remotely good at singling out core optimizer settings. In fact, some of them likely never will, because the cores that are unstable will be throttled by the cores that need higher voltage. I even gave you an EXACT config file to use that will demonstrate proof to you yourself within 5 minutes.
My bad, I made it a bit confusing. So snorlaxgangs was another account I created here awhile ago but forgot then I accidentally logged using gmail without noticing that I was using different account.

Anyway, when I disabled C-states, I found that Corecycler did not work, I can set -30 for every single core, passed all kind of test mode but it would fail right away in all core stress test. Same thing happened w Ryzen Master new auto CO feature, if you disable C-states it would just run -30 all cores and then crash at the last stage and failed to apply the setting.

So during the time I disabled c-states, IDK what else should I do in order to test all CO stability, hence I threw all the benchmarks,stress tests I could find it. Curve went from -25, then -19 then -16 but still some instability occurred once awhile and then I saw some people using auto CO so I tried it out and found that c-states disable would cause the crash eventually.

In the end I enable C-states, start testing w 8 physical cores first bc I'm using it for fps game. Turned out my CPU is kinda **** haha, can only do negative 8 8 7 8 8 12 12 10 , ppt/edc/tdc: mb, +200 boost, - 0.75 Vcore for. It runs without any problem so far, I'm not sure if my setting is correct tho.


Some benchmark from benchmaestro and bios setting in attached files. Btw, I'm using b550-i strix bios mod 2603i from hardwareluxx.de
 

Attachments

·
Registered
Joined
·
43 Posts
My bad, I made it a bit confusing. So snorlaxgangs was another account I created here awhile ago but forgot then I accidentally logged using gmail without noticing that I was using different account.

Anyway, when I disabled C-states, I found that Corecycler did not work, I can set -30 for every single core, passed all kind of test mode but it would fail right away in all core stress test. Same thing happened w Ryzen Master new auto CO feature, if you disable C-states it would just run -30 all cores and then crash at the last stage and failed to apply the setting.

So during the time I disabled c-states, IDK what else should I do in order to test all CO stability, hence I threw all the benchmarks,stress tests I could find it. Curve went from -25, then -19 then -16 but still some instability occurred once awhile and then I saw some people using auto CO so I tried it out and found that c-states disable would cause the crash eventually.

In the end I enable C-states, start testing w 8 physical cores first since bc I'm using it for fps game. Turned out my CPU is kinda **** haha, can only do negative 8 8 7 8 8 12 12 10 , ppt/edc/tdc: mb, +200 boost, - 0.75 Vcore for. It runs without any problem so far, I'm not sure if my setting is correct tho.


Some benchmark from benchmaestro and bios setting in attached files. Btw, I'm using b550-i strix bios mod 2603i from hardwareluxx.de
What made you decide to use the -.075 vcore offset?
 
421 - 440 of 470 Posts
Top