Overclock.net banner
381 - 400 of 410 Posts

·
Registered
Joined
·
98 Posts
for now dont know how use corecycler, the last bios get better perfomance with pbo mother limit,
dark hero dinamic oc 4775 ccx1 4625ccx2 1.275v
+100mhz scalar x2, co -30 all core
df state = disable
cppc=disable
cppc core=enable

llc 3
 

·
Robotic Chemist
Joined
·
3,886 Posts
Yeah, -30 all core isn't going to be stable.
 

·
Premium Member
Joined
·
6,260 Posts
for now dont know how use corecycler, the last bios get better perfomance with pbo mother limit,
dark hero dinamic oc 4775 ccx1 4625ccx2 1.275v
+100mhz scalar x2, co -30 all core
df state = disable
cppc=disable
cppc core=enable

llc 3
I'm pretty sure if you set the CCX 4775/4625, then your -30 all core is doing nothing. It's static.

You need to put the CCX on Auto for your -30 to really work, and a static CCX overclock performs better than a -30 for example.

And after you put the CCX on Auto, your -30 will likely not pass Core Cycler.

Edit: And I'm also really sure you need the CPU voltage on Auto, not static 1.275v for your -30.
 

·
Registered
Joined
·
220 Posts
I'm pretty sure if you set the CCX 4775/4625, then your -30 all core is doing nothing. It's static.

You need to put the CCX on Auto for your -30 to really work, and a static CCX overclock performs better than a -30 for example.

And after you put the CCX on Auto, your -30 will likely not pass Core Cycler.

Edit: And I'm also really sure you need the CPU voltage on Auto, not static 1.275v for your -30.
The Dark Hero can do both with dynamic overclock. It switches between PBO and per-ccx based on power load.

That said, 4775/4625 per-ccx would take some beefy cooling.
 

·
Premium Member
Joined
·
6,260 Posts
The Dark Hero can do both with dynamic overclock. It switches between PBO and per-ccx based on power load.

That said, 4775/4625 per-ccx would take some beefy cooling.
Oh, I see. Thanks for clearing that up.
 

·
Registered
Joined
·
98 Posts
I'm pretty sure if you set the CCX 4775/4625, then your -30 all core is doing nothing. It's static.

You need to put the CCX on Auto for your -30 to really work, and a static CCX overclock performs better than a -30 for example.

And after you put the CCX on Auto, your -30 will likely not pass Core Cycler.

Edit: And I'm also really sure you need the CPU voltage on Auto, not static 1.275v for your -30.
the mother change to all core cxx1 set and ccx2 set ando vcore set , when consume pas x amperes (45a recomend in bios) i set 75a, from this limit auto change to allcore static voltage, its similar to crt 2.1 , I'll leave corecycler running all night
75a - work pbo +100 and -30 all core work for now, need more test, single core not optimal performance, need more test, testing in ubuntu 21.04 now, better single core and worst multi in geekbench 5
cooler is masterliquid ml360r rgb , thermal paste mastermaker gel 11wk

 

·
Registered
Joined
·
267 Posts
I see a lot of users using Curve optmizer to undervolt the cores and get more boosts, but why are you adding +volts offset to vcore? Does this mean the threads which due to the undervolt from curve optimizer get their "voltage back" from the offset set on the vcore meaning they can boost higher?

I tried a positive offset on my vcore and the only result was higher temps, no better single thread performance, in fact... i tested a negative 0.50mv offset on the vcore and the result was lower temps and the same single thread + multi thread performance in cinebench r20.

So can anyone please explain to me why you are doing positive offsets for vcore, what is the supposed benefit of doing this? Also with a positive vcore voltage offset my max is over 1.5v , is this safe long term for a 5950x?

EDIT: And does it mean that if you give vcore a positive offset that your potential to go lower on the curve optimizer values for cores that may not have been stable before at default vcore with a negative value? Example my core 2 won't go any kower than -5, with a +.0.250mv vcore offset could i potentially reach -10 or more with this core in the curve optimizer, is that the idea?
I have been playing with my 5900x for 2 weeks now and indeed I can confirm that using positive offset helps.
Using Auto vcore, I have 2 cores in CCD1 that fails corecycler, I had to use lower negative CO -10 and -3 for Core0 and Core2 to pass. Overall score for single score benches wasn't great, cpu-z single is maxed to 670 pts with this settings.
I played with the vcore offset and used + 0.03 and that helped to stabilize my CO to -20 and -10 for Core0 and Core2. single score bench increased in cpu-z to 680.
Temp increased like 2~3c was around 73 - 76c during cinebench23 run.
Also the multicore scores increased , like 100 ~ 150 pts difference in cpuz 9850 - 10000
 

·
Registered
Joined
·
267 Posts
any one lately during that week received his 5900/5950 from Amazon?
How the new batches going? I read the new ones boost is limited compared to old batches
And when the stepping B2 should arrive?
 

·
Registered
Joined
·
220 Posts
any one lately during that week received his 5900/5950 from Amazon?
How the new batches going? I read the new ones boost is limited compared to old batches
And when the stepping B2 should arrive?
They adjusted the default curve formula. Newer batches are more stable out of the box, but also have slightly lower boost.

It's the same as having a bad stepping and adding positive offset to a core, they just did it for you.
 

·
Village Idiot
Joined
·
2,440 Posts
Cave man here checking in with his brand new 5900x (from Amazon this week) + X570 Aorus Master. Seems you need a Ph.D in Rocket Surgery to "overclock" these things. I thought Nvidia GPU Boost was a nightmare, but holy hell... GPU Boost seems like child's play in comparison to this thing.

Fumbling around I've got PBO enabled, curve optimizer at -30, and it runs 4625-4650mhz through Cinebench... but Ryzen Master reports average core voltage as an alarming 1.43v. All BIOS voltage are set to auto. Is this remotely OK? I remember reading about people frying their Zen 2 chips with 1.35v+, so this is slightly concerning to me.
 

·
Robotic Chemist
Joined
·
3,886 Posts
Fumbling around I've got PBO enabled, curve optimizer at -30, and it runs 4625-4650mhz through Cinebench... but Ryzen Master reports average core voltage as an alarming 1.43v. All BIOS voltage are set to auto. Is this remotely OK? I remember reading about people frying their Zen 2 chips with 1.35v+, so this is slightly concerning to me.
Yes, that is normal and fine. You will not be stable at -30 all core (run CoreCycler), but the issue is 1.35V at high loads. These CPUs will hit 1.5V with light loads at stock.

It still bothers me so I put in a -0.05V offset on the CPU voltage before tuning my curve offsets, the voltage offset also allows a slightly lower (larger negative) curve offset to be stable through CoreCycler. It should be fine and I am just being paranoid (and losing the top 25-50 MHz for very short very light loads, but I am fine with that). 1.5V is scary. :eek:

Edit: I see 4700-4675 MHz in CBr20 at 1.325V (1.318) at my settings (5950X, 71.3°C peak, PBO at motherboard limits, curve optimizer -13 -9 -11 -18 -15 -21 -23 -9 -21 -29 -27 -21 -25 -26 -28 -14, SMT disabled). I do see my max voltage at 1.45V, it will still boost one core to 5GHz for 10ms or something to open a new tab and stuff like that. With a voltage offset larger than -0.0625V I saw "clock stretching", or the CPU claiming to run at a higher MHz than it really does (Core Clock is noticeably higher than Core Effective Clock in HWiNFO while running CB, and CB scores are lower too).

Core 1 and 7 are my "favored" cores. If I lower any core's curve offset by one it gets errors in CoreCycler, but I never have errors or crashes anywhere with my fine tuned settings. I am impressed with how well CoreCycler worked for me. I could run -22 all core without any issues in CB or other benchmarks and games, but I would get a random restart once a week or so. Now it is rock solid.

Edit2: Oops, that was with SMT disabled, with it enabled I see 4.6GHz all core at 1.275V with a 76.0°C peak CCD temp. 1.43V does seem high to me while running CBr20, even accounting for my voltage offset. I don't trust Ryzen Master at all though, in fact I do not like having it installed.
 

·
Village Idiot
Joined
·
2,440 Posts
Yes, that is normal and fine. You will not be stable at -30 all core (run CoreCycler), but the issue is 1.35V at high loads. These CPUs will hit 1.5V with light loads at stock.


Edit2: Oops, that was with SMT disabled, with it enabled I see 4.6GHz all core at 1.275V with a 76.0°C peak CCD temp. 1.43V does seem high to me while running CBr20, even accounting for my voltage offset. I don't trust Ryzen Master at all though, in fact I do not like having it installed.
I know they'll run pretty high voltage for lightly threaded loads, but the all core full load voltage in CB seemed really high given what I remember reading about Zen 2. Knowing some X99 boards had a tendency to fry CPUs if left on "auto" voltage settings doesn't give me a warm fuzzy either.

Is Ryzen Master generally regarded as junk? What software can provide a real voltage reading? Running CB R23 on stock power limits I get three different voltage readings with three different software:

Ryzen Master = 1.20v average core voltage
HWiNFO64 = 1.139v core voltage / 1.214 VID
CPU-Z = 1.164v
 

·
Robotic Chemist
Joined
·
3,886 Posts
What software can provide a real voltage reading?
HWiNFO is the most reliable in my experience. CPU Core Voltage (SVI2 TFN) is the one I pay attention to.
 

·
Robotic Chemist
Joined
·
3,886 Posts
I know they'll run pretty high voltage for lightly threaded loads, but the all core full load voltage in CB seemed really high given what I remember reading about Zen 2. Knowing some X99 boards had a tendency to fry CPUs if left on "auto" voltage settings doesn't give me a warm fuzzy either.
PBO auto voltage control is very different, the board is opening up the power limits, but the CPU is still what is deciding what voltage to run at, based on load and temperature. The curve optimizer will change it too. CB is a pretty light load, if you run something heavier you will see much lower voltages and frequencies. It is Prime95 at 1.35V that really kills things.

But, yeah, I don't get warm fuzzies either. :p
 

·
Registered
Joined
·
19 Posts
Hi everyone, I just got my 5900x last Friday and since then I've been tinkering with it a lot, but I'm not that happy after all. It sits into a crosshair viii formula, I ran ctr 2.1 test and it has been seen as a silver sample. The thing is that I see veryone is scoring 600+ and 8.6k + in cb20 but I can't score more than 590 and something or 8200-8300 . I also got my ram overclocked from 3200 cl16 to 4000 cl16 ( dual rank ) and IF 2000 mhz in zentimings they are all at 1:1 ratio but I have my latency 65-66 ns in AIDA. I'm currently at 230 PPT 155 TDC and 170 EDC. Curve optimizer is set per core, -25 on the best cores, -21 on the secondary and -30 on the others. PBO scalar on auto ( 1x ) and +200 mhz my motherboard allows me up to 700 if I remember correctly. While running corecycler I see the VID and Core Voltage switching from 1.381 to 1.403v I'm on the latest bios During cb20 test the core speed is 4.575 reported by hwinfo ryzen master or my motherboard. In hwinfo I see that some of the cores go up to 5150 mhz and also in gaming for a fraction of second. Can somebody shed some light upon this situation?
 

·
Registered
Joined
·
406 Posts
Hi everyone, I just got my 5900x last Friday and since then I've been tinkering with it a lot, but I'm not that happy after all. It sits into a crosshair viii formula, I ran ctr 2.1 test and it has been seen as a silver sample. The thing is that I see veryone is scoring 600+ and 8.6k + in cb20 but I can't score more than 590 and something or 8200-8300 . I also got my ram overclocked from 3200 cl16 to 4000 cl16 ( dual rank ) and IF 2000 mhz in zentimings they are all at 1:1 ratio but I have my latency 65-66 ns in AIDA. I'm currently at 230 PPT 155 TDC and 170 EDC. Curve optimizer is set per core, -25 on the best cores, -21 on the secondary and -30 on the others. PBO scalar on auto ( 1x ) and +200 mhz my motherboard allows me up to 700 if I remember correctly. While running corecycler I see the VID and Core Voltage switching from 1.381 to 1.403v I'm on the latest bios During cb20 test the core speed is 4.575 reported by hwinfo ryzen master or my motherboard. In hwinfo I see that some of the cores go up to 5150 mhz and also in gaming for a fraction of second. Can somebody shed some light upon this situation?
Your high latency for 4000 CL16 indicates something is not set correctly. You might have instability and errors running at that speed. I recommend setting your memory to run at 3800 CL16, and see if your latency increases.

In addition, your curve optimizer settings are likely to be not suitable for your CPU. You need to set it based on stability testing of each core using CoreCycler. If you give it too much negative curve offset, you'll lose performance. You mentioned that you ran Corecycler, what settings did you run it on?

I run my 5900X at only 4.7 GHz @ 1.300 V, using memory at 3600 CL14-15-14-24. Even with this conservative settings I'm already getting 9050 points in CB R20. And my Aida64 latency is 57.3 ns. Yours should be much quicker and lower latency that my system, so I believe you have some settings set in a way that is not matching to what your CPU can do.

2517853


2517854
 

·
Registered
Joined
·
19 Posts
Your high latency for 4000 CL16 indicates something is not set correctly. You might have instability and errors running at that speed. I recommend setting your memory to run at 3800 CL16, and see if your latency increases.

In addition, your curve optimizer settings are likely to be not suitable for your CPU. You need to set it based on stability testing of each core using CoreCycler. If you give it too much negative curve offset, you'll lose performance. You mentioned that you ran Corecycler, what settings did you run it on?

I run my 5900X at only 4.7 GHz @ 1.300 V, using memory at 3600 CL14-15-14-24. Even with this conservative settings I'm already getting 9050 points in CB R20. And my Aida64 latency is 57.3 ns. Yours should be much quicker and lower latency that my system, so I believe you have some settings set in a way that is not matching to what your CPU can do.

View attachment 2517853

View attachment 2517854
hi, thank you so much for your quick reply!
. You need to set it based on stability testing of each core using CoreCycler. If you give it too much negative curve offset, you'll lose performance. You mentioned that you ran Corecycler, what settings did you run it on? /////// what do you mean by settings please?
2517860
2517861
 

·
Premium Member
Joined
·
4,391 Posts
So I built this rig as a workstation for video editing and photogrammetry processing, and was going to leave it at stock because of all the stability issues I've heard about, but couldn't resist the urge to tinker a little. I'm new to AMD so it took some reading up, but using PBO is ultimately far easier than old school overclocking. I'm pretty happy with the results as well. Having done a lot of reading, I think too many people are getting hung up on clocks (it is called overclocking, so I get it) and ignoring actual performance, especially performance in real-world use cases and not just synthetic benchmarks. For real-world use, transient/moderate load clocks and how long those moderate boost clocks are being held matters a whole lot more than HWM is showing as your max clock hit.

Here's what I've got:

5950x in a Crosshair VIII Hero
Dark Rock Pro 4 cooler
48GB (mismatched 16gb/32gb kits) of GSkill 3200CL16
960 Evo 500GB for OS drive
Asus RTX3070 OC running at stock OC settings

DOCP has no problem with four sticks on 3200mhz and matching Infinity Fabric. That was the first thing I tested because I was concerned about four sticks and mismatched kits no less.
PBO limits are set at 225/140/180 - I'm completely baffled by people running higher TDC than EDC, completely goes against what those two are supposed to represent.
CO is set to -5 on four best cores, -10 on everything else.
LLC minimum/off
Scalar, boost increase, voltage offset, etc. everything else is untouched and at stock for best CPU longevity and stability.

Clocks achieved are... doesn't matter. Haven't even looked other than to make sure there's no clock stretching.

Completely unoptimized benches (i.e. still a million background processes running):

CR23: 29073/1608
CPUZ: 13,439/680
Geek5: 16779/1710
RB2.56: 313,833

Maxes at around 82C in CR23 in 75F room.

Judging by other results I've seen, this is quite good for minimal added CPU wear/tear and rock solid stability, especially considering the slower RAM speeds. Seems like pushing harder is extraordinarily more effort, wear, and questionable stability for very minimal gains. I would like to keep tweaking with a better mixed/real world benchmark though. RB is hugely, hugely inconsistent and seems to need to be run 10x times to settle on a score average (it did this to me on Intel as well). The encoding sections takes anywhere from 1 to 10 minutes, which is absurd. Geek5 works pretty well, but the stuff it's doing is fairly irrelevant to what I'm actually doing.
 

·
Robotic Chemist
Joined
·
3,886 Posts
Scalar, boost increase, voltage offset, etc. everything else is untouched and at stock for best CPU longevity and stability.
I think you must not be aware of how aggressive AMD is with voltage at "stock" PBO? And you have a curve offset too... you not stock anymore, voltage or otherwise.

PBO limits are set at 225/140/180 - I'm completely baffled by people running higher TDC than EDC, completely goes against what those two are supposed to represent.
But why do we need a lower power limit long term compared to short term? My cooling is more than sufficient, I have no need to lower performance for workloads that run long. Setting a higher TDC disables that unneeded downclock.

Judging by other results I've seen, this is quite good for minimal added CPU wear/tear and rock solid stability, especially considering the slower RAM speeds.
I think of myself as an aggressive overclocker and I set a negative voltage offset on these CPUs. 1.5V is too much for me. :cautious:

This also allows higher performance during the mid-clock range because the CPU uses less power and has lower temperatures, so it boosts a little higher.

How have you judged "rock solid stability"? Have you run CoreCycler or anything? A -5/-10 curve offset is a pretty good PBO overclock, I don't think you are as close to "stock" as you think you are. :)
 

·
Premium Member
Joined
·
4,391 Posts
I think you must not be aware of how aggressive AMD is with voltage at "stock" PBO? And you have a curve offset too... you not stock anymore, voltage or otherwise.
I didn't say it was completely stock. I said scalar, boost max, and CPU voltage were stock. We have people running around with 10x scalar, positive vcore offset, and maxed LLC - that's going to be way more wearing on the CPU than leaving those things alone.


But why do we need a lower power limit long term compared to short term? My cooling is more than sufficient, I have no need to lower performance for workloads that run long. Setting a higher TDC disables that unneeded downclock.
Because that's how power delivery works. Those are motherboard delivery specs to let the CPU know what's available. Surely the VRMs are not able to deliver the same continuous current as they are able to deliver peak/spike current. So if someone is just running with these numbers at maxed values so as to be "uncapped", then that's fine and dandy. But running TDC higher than EDC makes no sense whatsoever - which is exactly what I said in my original post.


I think of myself as an aggressive overclocker and I set a negative voltage offset on these CPUs. 1.5V is too much for me. :cautious:
Why? It doesn't appear to be too much for AMD at bone stock settings, so why do you feel it's too much for you?

This also allows higher performance during the mid-clock range because the CPU uses less power and has lower temperatures, so it boosts a little higher.
But the problem I've noticed is that's everyone is all about what boost speed they're getting and judging their OC on that, rather than about what performance numbers they're achieving. I can easily hit higher boost clocks and higher average clocks... all with decrease scores in every benchmark. Every YouTube video and how-to post I've seen is about maximizing negative offset in CO under the assumption that lower is better, just like how everyone seems to run PBO numbers as high as cooling will allow. In my experience, neither has straight correlation with performance gains, even when a given setting results in higher average clocks.

How have you judged "rock solid stability"? Have you run CoreCycler or anything? A -5/-10 curve offset is a pretty good PBO overclock, I don't think you are as close to "stock" as you think you are. :)
Yes I've run CoreCycler for hours, days sitting idle on desktop or light apps, etc. Stability matters a lot to me. Most productivity software I'm using on a daily basis will crash a bad overclock faster than any benchmark or stress test I've used.

Again, I never said it's very close to stock. I said the things that have been tweaked are things that aren't particularly far out of spec for the CPU and don't actually alter AMD's voltage/boost behavior like scalar does and doesn't take the CPU out of prescribed voltage limits like running vcore offset does. Meaningful extra degradation is unlikely and yet the performance gains are reasonably useful.
 
381 - 400 of 410 Posts
Top