5800x and Asus Fmax Enhancer. Higher clockspeed but lower scores!?

[FireDemon]

Can anyone shed some light on why enabling the PBO Fmax enhancer on my Crosshair VIII Hero results in higher clockspeeds, but lower scores in Cinebench? Fmax off, I'm seeing 4850 max clocks with 15540+/- in Cinebench MT and 1590+/- ST. Fmax on, I'm seeing as high as 5050 clockspeed but 13880 MT and 1520 ST.

Average clocks seem to be higher as well. 4850 MT vs 4650 ST. What's up?

 

[jthomas]

Most likely it's because of a phenomena known as clock stretching. Essentially, the processor is throttling frequency down then back up faster than most monitoring software can recognize. The reason for lower scores is because your true average frequency ends up being lower over a given time period.

To confirm, use Hwinfo64 and compare reported core clocks against what is reported under "effective clocks", if there is a significant difference under load than you are experiencing clock stretching. It's a problem that seems to crop up a lot with the Fmax enhancer enabled.

 

[Jamie Marsala]

Turn off FMax enhancer. It does not work the same or at all on Zen3 chips.

 

[Digital Viper-X-]

You don't have to turn it off, just don't choke the cpu for power if you expect it to perform better.

 

[Jamie Marsala]

You don't have to turn it off, just don't choke the cpu for power if you expect it to perform better.

Maybe it is just the dual CCD units but when I enable FMax enhancer it cripples what my second CCD, peak drops by several hundred MHz from 4.6-4.7 to 4.4-4.5. It does nothing for my single core peaks which are 5.1Ghz on two cored on my 5900X. Power on mine is no where near crippled as my processor easily hits 200 Watts if it wants to.

 

[Digital Viper-X-]

Maybe it is just the dual CCD units but when I enable FMax enhancer it cripples what my second CCD, peak drops by several hundred MHz from 4.6-4.7 to 4.4-4.5. It does nothing for my single core peaks which are 5.1Ghz on two cored on my 5900X. Power on mine is no where near crippled as my processor easily hits 200 Watts if it wants to.

The reported Mhz is missleading. EG: If I move LLC to 2 or 3, my reported core speeds drops 100-200mhz but performance goes up.

 

[Jamie Marsala]

The reported Mhz is missleading. EG: If I move LLC to 2 or 3, my reported core speeds drops 100-200mhz but performance goes up.

My performance drops if I change LLC from auto to 2 or 3, or any setting really as I tried them all. But I am using Curve Optimizer so that may be why. That gives me the best multi core performance which is what I want. Just because it peaks at 5.1Ghz on two cores does not mean it can hold it there, it can hold around 4.9-4.95 on a few cores during synthetic benches. I only use the benches as references for increase/decrease in performance from changes I make tweaking it.

 

[Digital Viper-X-]

My performance drops if I change LLC from auto to 2 or 3, or any setting really as I tried them all. But I am using Curve Optimizer so that may be why. That gives me the best multi core performance which is what I want. Just because it peaks at 5.1Ghz on two cores does not mean it can hold it there, it can hold around 4.9-4.95 on a few cores during synthetic benches. I only use the benches as references for increase/decrease in performance from changes I make tweaking it.

Try these settings and tweak from there.

LLC 2 , CPU Current 110% , Curve optimizer off at first
Check performance there and see if it improves.

From that point go ahead and tweak the curve to drop temps until performance drops.

 

[Jamie Marsala]

Try these settings and tweak from there.

LLC 2 , CPU Current 110% , Curve optimizer off at first
Check performance there and see if it improves.

From that point go ahead and tweak the curve to drop temps until performance drops.

Oh with no Curve and even just PBO enabled my chip performs no where near what any tester has ever stated, except maybe in single core scores with a very specific setup. Not stock, not PBO enabled, not with board at max power settings and not with LLC in any of its settings. I currently run LLC=Auto and CPU current at 140%, CPU voltage on auto. Curve Optimizer is -20 on 4 of the 6 CCD1 cores and -25 on the other two. CCD2 is set at -10 on the two best and -8 on 2 more -5 on one and the worst core is at 0. This keeps the system stable 100% for days on end with no re-boots unless windows updates or I need to change something. With the following results.

I do not recall what I started with 8437 Multi and 616 Single on R20 with PBO Enabled. Those are just for comparison. If I set LLC to anything other than Auto I lose single core performance on the tests and even my TimeSpy scores drop compared to with it on Auto, multicore drops as well just not as much.

 

[FireDemon]

I'm going to side with the ones saying not to use the Fmax Enhancer on Zen3 chips.

"The Stilt" made a somewhat recent post regarding the feature and explaining in depth on how it works. He stated it was "exclusive" to 3000 series CPUs. He also went on to explicitly say that in the case of reduced CB MT scores, this is due to clock stretching and because of the nature of PBO, the LLC settings should NOT be changed manually, as it will cause the CPU to no longer be in control of the LLC. He states to add a positive offset to the CPU voltage, one step at time, until scores stabilize.

Doing this proved futile, as I was unable to obtain stock PBO MT scores with the Fmax enhancer until I reached a positive offset voltage of +.10000. Even with this, things seemed hit or miss as one run, the scores would be (at best) even with NON-Fmax PBO, and the next run they would be down by 200-300pts. Single core scores seemed largely unaffected if at all no matter what the voltage offset was (if any at all). So best case scenario, a feature designed to boost single core performance, was giving no increase at all in single core scores.

The only things I noticed that seemed to be guaranteed across the board were all detriments. Higher average temps, higher peak temps and higher power draw. Any increase in clockspeed seemed to be offset by poor performance. I suspect that it was mostly the result of clock stretching. 5050mhz is great for bragging rights but makes no sense when it draws more heat and power just to perform as good if not worse than 4900mhz.

A couple times I observed my CPU boosting to 5000+ with just normal PBO and no Fmax.

 

[Digital Viper-X-]

Oh with no Curve and even just PBO enabled my chip performs no where near what any tester has ever stated, except maybe in single core scores with a very specific setup. Not stock, not PBO enabled, not with board at max power settings and not with LLC in any of its settings. I currently run LLC=Auto and CPU current at 140%, CPU voltage on auto. Curve Optimizer is -20 on 4 of the 6 CCD1 cores and -25 on the other two. CCD2 is set at -10 on the two best and -8 on 2 more -5 on one and the worst core is at 0. This keeps the system stable 100% for days on end with no re-boots unless windows updates or I need to change something. With the following results.

I do not recall what I started with 8437 Multi and 616 Single on R20 with PBO Enabled. Those are just for comparison. If I set LLC to anything other than Auto I lose single core performance on the tests and even my TimeSpy scores drop compared to with it on Auto, multicore drops as well just not as much.

140% cpu current may be getting you into some throttling, are you monitoring the temps / edc etc?

 

[Jamie Marsala]

140% cpu current may be getting you into some throttling, are you monitoring the temps / edc etc?

Yes temps never go above 73 in the current setup but I can get them lower by lowering the EDC a hair and pretty much scoring the same for the comparison. But the loop keeps it cool so for normal operation and gaming. I am going to try to lower that cpu current to maybe 110 or 120 and see what it does. I went with suggestions from overclockers.net for using the Crosshair Hero and the Ryzen 3 chips. I am adding my video card to the loop tomorrow as I finally found a water block for it, it is not a reference card. So I will play more then to see what I can do.