Overclock.net banner
81 - 100 of 158 Posts

·
Registered
Joined
·
1,765 Posts
Didn't notice anything like this. What CO has to do with FCLK?
You can utilize the CPU more efficiently?
It effect core stability at least I can confirm.

Some things I run to benchmark like the higher FCLK for better performance. But also means I can't have as aggresive Curve Optimizer settings to be stable.

More FCLK the less Negative CO settings you can run and I need Positive offset to work stable on some cores.
 

·
Registered
Joined
·
804 Posts
You can utilize the CPU more efficiently?
Actually efficiency is a function of performance per power consumed.
So I don't fully understand what you mean here, since CO hardly affect that.
But also means I can't have as aggresive Curve Optimizer settings to be stable
Which is reflected in what, exactly? What measure of stability are we talking about?
 
  • Rep+
Reactions: TimeDrapery

·
Registered
Joined
·
1,765 Posts
Actually efficiency is a function of performance per power consumed.
So I don't fully understand what you mean here, since CO hardly affect that.

Which is reflected in what, exactly? What measure of stability are we talking about?
Stability in Prime95 & Y-cruncher was effected.
FLCK effect performance, more data/compute can be pushed through the cores overall, meaning better utilization, which means it's used more with a higher stress overall getting warmer & requiring more voltage to have stability with a higher load being put on the cores.

2100FCLK allowed me to go from +4 CO to -3 CO setting on the worst core that usually crashed first. There are other cores that were weak as well but haven't had time to test how much better CO can be used yet on each and every core that managed bad with 2366FCLK.

This could mean I might be able to get 2400+ FCLK stable if I use positive CO values in the +10 range for my CPU sample.
 
  • Rep+
Reactions: TimeDrapery

·
Registered
Joined
·
804 Posts
Stability in Prime95 & Y-cruncher was effected
Did you mean a single core or a usual allcore run when testing?
And what PBO2 settings turned out stable for you, if more conservative fabric OC used, lets say 1800-2000Mhz ?
 
  • Rep+
Reactions: TimeDrapery

·
Registered
Joined
·
1,765 Posts
Did you mean a single core or a usual allcore run when testing?
And what PBO2 settings turned out stable for you, if more conservative fabric OC used, lets say 1800-2000Mhz ?
Tested both single-core and all-core in varied tests.

Not gone through all cores but the following has been verified:
10X & +200Mhz

FCLK 2100
Core 0 | -3 OK [-4 BAD]
Core 1 | -11 OK [-13 BAD]
Core 2 | -17 OK
Core 3 | -16 OK
Core 4 | -11 OK [-13 BAD]
Core 5 | -16 OK? [-17 BAD]
Core 6 | -11 OK
Core 7 | -12 OK [-13 BAD]


FCLK 2366
Core 0 | +4 OK [+2 BAD]
Core 1 | -3 OK
Core 2 | -8 OK
Core 3 | -8 OK
Core 4 | -0 OK
Core 5 | -8 OK
Core 6 | -4 OK
Core 7 | -0 OK

Been busy with other stuff so not been going through each core that fast.
 

·
Registered
Joined
·
804 Posts
Tested both single-core and all-core in varied tests.
Well, at 2100 curve looks fine to me. A value fore the core 0 is a bit on a low side, but thats ok.
 
  • Rep+
Reactions: TimeDrapery

·
Registered
Joined
·
26 Posts
STAPM can be disabled and the "timer" is no longer in effect, you get infinite boost behaviour.
You can also alter the timing period in seconds on how long it will Boost if you want to keep it enabled.
After time is out, it stops the boost behaviour and limits down to Stock PPT/TDC/EDC limits.

By altering your PPT you alter the STAPM PPT limit. So you can go that route as well but if the timer is in effect it will ignore your custom PPT and use the Fused limits when your out of time aloted for STAPM Boost.

You need to disable STAPM either way if you don't want the throttle effect in length.
But be aware that your clocks will depend on your CPU temperatures.
Hi. I have a gigabyte board and wanted to just make sure that I actually disabled it correctly.

I found STAPM but it was named STAPM Control.
Here is how the menu is:
STAPM Control
  • STAPM Control
    • Auto/Manual
  • STAPM Boost
    • Auto/Disable/Enable
So, if I were to set the Control to Manual and Boost to Disabled, that should be it right?
 
  • Rep+
Reactions: TimeDrapery

·
Registered
Joined
·
1,765 Posts
Hi. I have a gigabyte board and wanted to just make sure that I actually disabled it correctly.

I found STAPM but it was named STAPM Control.
Here is how the menu is:
STAPM Control
  • STAPM Control
    • Auto/Manual
  • STAPM Boost
    • Auto/Disable/Enable
So, if I were to set the Control to Manual and Boost to Disabled, that should be it right?
Yeah, should be disabled like that.
You have to check and verify it's actually doing what it's supposed to do.

I've seen it bug-out a few times when you change the setting around.
For some reason I've not been able to "enable" it after managing to disable it the last few attempts.
There are some BIOS bugs with regard to Gigabyte & 5700G AMD_OVERCLOCKING settings I've noted.
Inconsistent application, where your settings aren't applied with F34 bios correctly.
 
  • Rep+
Reactions: TimeDrapery

·
Registered
Joined
·
26 Posts
Yeah, should be disabled like that.
You have to check and verify it's actually doing what it's supposed to do.

I've seen it bug-out a few times when you change the setting around.
For some reason I've not been able to "enable" it after managing to disable it the last few attempts.
There are some BIOS bugs with regard to Gigabyte & 5700G AMD_OVERCLOCKING settings I've noted.
Inconsistent application, where your settings aren't applied with F34 bios correctly.
Have you set SOC LLC? I am having issues with iGPU being stable as there is vdroop. What SOC LLC do you recommend?
 
  • Rep+
Reactions: TimeDrapery

·
Registered
Joined
·
1,765 Posts
Have you set SOC LLC? I am having issues with iGPU being stable as there is vdroop. What SOC LLC do you recommend?
I've just found the needed SoC voltage more depends on your FCLK if it's gonna be stable or not. Just check what FCLK requires it translates quite well to what the iGPU as well will require if not a little extra for the same.
The separate GFX Core voltage needs no adjustment on my end and will manage with the stock voltages.

The iGPU needs a stable FCLK to be stable. So adjust SoC accordingly.
 
  • Rep+
Reactions: TimeDrapery

·
Registered
Joined
·
36 Posts
Hey guys.

New build, work in progress ...

5700G @ stock, with CO per core (avarage about -20), but still testing the stability at Idle.
Gigabyte B550 Vision D, BIOS F13.
G.Skill TridentZ Neo DDR4 RGB 3600Mhz CL16-16-16-36 2x8GB, @4266 CL18-18-18-36 1.43v(BIOS sett), FCLK @2133.

I have adjusted the Curve Optimizer but the Precision Boost Overdrive is sett to disabled, for now. No overclocking yet, but it is so cool & quiet that I'm tempted to keep the PBO disabled.

First CB20 run, CPU @ stock with CO.
Font Screenshot Software Technology Rectangle
 

·
Registered
Joined
·
36 Posts
Higher FCLK has a effect on your Core voltage requirements so you might need to play around with Curve Optimizer adding positive offset to some cores.
I had some cores that became troublesome with higher FCLK needing help other than more VSOC to run fine.
I have noticed something similar too. The high FCLK doesn't like an aggressive CO. But a higher FCLK should require higher voltages for Uncore, including more VDDP too. What other voltages are you guys adjusting?

So we have to choose between a better CO or a higher FCLK. It's pretty time consuming to find an optimal CO, so it's a good idea to settle the FCLK first. ;) .

I'm still experimenting, but I have been getting much better almost-stable CO with my XMP-profle with FCLK @1800. Almost all my cores require 5-6 more stepts to become almost-stable with FCLK @2133, if I'm not mstaking. What is your sweet spot for FCLK?
 

·
Registered
Joined
·
804 Posts
So we have to choose between a better CO or a higher FCLK
As I said earlier, I see no reason why this should be the case and haven't heard convincing arguments yet. May be its sample specific, as in @Nighthog's tests and applies to FCLK set really high, i.e. 2300+ where the stability issues might come into effect.
 

·
Registered
Joined
·
1,765 Posts
I think it's just more demanding to run all in high frequency.

We know LN2 OC requires low FCLK to increase Core speeds without hassles. Might be that FCLK just adds so much more to core voltage requirements overall for stability purposes it's setting a hindrance.
So you might get more performance with a lower FCLK rather than trying to push FCLK higher but focusing on pure Core Frequency increases instead. Meaning you need less voltage for each step up in frequency. More efficient for performance gains.

We already notice FCLK eats a lot out of the CO adjustment possibilities when pushing it. Meaning you might loose too much frequency pushing FCLK from the cores headroom allowances.

Maybe someone should just test lower FCLK and push more core frequency instead to compare if Core performance is more important than FCLK.
I expect that Core should be better than FCLK overall. But the question is how much frequency do we lose for each FCLK increase etc.
Where is the best balance between FCLK & Core speeds.
Do we only need 1200-1333FCLK and more Core frequency or should FCLK be ~1900-2133 range and then you see how much you get from the cores?
 

·
Registered
Joined
·
804 Posts
Maybe someone should just test lower FCLK and push more core frequency instead to compare if Core performance is more important than FCLK.
I think, there's no need to do it, since core performance contribution is obviously much bigger overall.
And what you say about CO might be equally applied to any CPU metrics, especially in case of even more power-gated APU's, still this doesn't change the fact, that a high FCLK only indirectly affects OC capability.
What would be more interesting to find out is the highest FCLK value, after which performance scaling is stopped due to the above reasons.
 

·
Registered
Joined
·
36 Posts
Do we only need 1200-1333FCLK and more Core frequency or should FCLK be ~1900-2133 range and then you see how much you get from the cores?
That's good question. I haven't tried ultra-high FLCK @2300+ range yet, But I've noticed that almost all Ryzen x-series CPUs are struggling with FCLK 1600+, and almost all of them hit a wall at 1800. So this range may be a good start point to research the answer of this question. These APU's can push the FCLK further, for sure, but it's affecting the CO, for some reason

I started with my XMP- profile (3800Mhz RAM and 1800-FCLK) to eliminate the RAM, Uncore and FCLK issues. I nailed an almost-stable CO with an average about -25 per core. But when I overclocked my RAM to 4266MHz (and FCLKC @2133), got many reboot at Idle/light loads. It crashes without BSOD on Idle/light loads. But it's passing RAM-tests, so I went back to CO. and tried some fast rounds with core.cycler. In my full surprise they need about 5-6 steps more for FCLK-2133. Now the average is about -20, and with this trend, I'm afraid that FCLK 2300+ will push some of my cores to + territory.



I'm stll experimenting, but my CO needs about 5-6 more stepts to get form 1800 to 2133 FCLK and stay almost stable.

Anybody has experimented with the effect of FCLK on CO between FCLK-1600 to 1800?
 

·
Registered
Joined
·
804 Posts
I started with my XMP- profile (3800Mhz RAM and 1800-FCLK)
Are you sure fclk was that and not 1900, cause 1800 is not sync'ed with MCLK and memory controller will work at half the frequency - 950Mhz in this case.
 

·
Registered
Joined
·
36 Posts
Are you sure fclk was that and not 1900, cause 1800 is not sync'ed with MCLK and memory controller will work at half the frequency - 950Mhz in this case.
I mean 3600Mhz and 1800FCLK. The FCLK ws right, anyways, but i mixed up The RAM speed o_O.

According to Zentmings, the 3600Mhz RAM runs with MCLK=FCLK=UCLK=1800, as this sreenshot. This is my XMP-profile, and this is what I have testet against FCLK 2133.
Rectangle Font Screenshot Parallel Technology


And this is the 2133-FCLK:
Rectangle Font Screenshot Parallel Technology
 

·
Registered
Joined
·
36 Posts
I found this interesting statement in "[Official] AMD Ryzen DDR4 24/7 Memory Stability Thread" page 645, from yesterday.

-25 all core has always worked for me. I did not want to try each core little by little, I start -10 all core and ran OCCT and when I passed I went to -15 and so on till I got to -25 and it passed, however, it does not pass now and the issue is my memory as it does pass when I set the memory ot DOCP or 3200 CL14
And he posted a screen-shot with his RAM running at 3800Mhz now.

These RAM-guys are discussing about CO degradtion from 3200Mhz to 3800Mhz RAM speed. This discussion is about X-series, but something similar should apply to our AUPs too, probably in a bit higher RAM/FCLK-speeds.

So it's a good idea to investigate it on these APUs. But this issue is more critical for us, becouse these APU's can hit much higher FCLK with much more degration of CO.
 

·
Registered
Joined
·
36 Posts
So back to the original poit, that I nade about the effect of FCLK on CO.

I think we can all agree and conclude that; It's important to settle the FCLK before using a lot of time on finding an optimal CO. Because changing the FCLK will require a new round of time consuming tests for finding a new optimal CO.

So, what's the sweet spot for FCLK on 5700G APUs? I hope you guys can share your findings.
 
81 - 100 of 158 Posts
Top