Ryzen 3700x clockspeed issues. Prime 95 low clocks. 3.5 GHz, is this normal?

[sn2x]

X470 final ABB bios (to keep PCIe 4)
Hyper 212 EVO cooler

At stock in BIOS and idle voltages are very high, like 1.45
Drops with a load.

Prime 95 my clocks are dipping down to 3.5 GHz
Cinebench was around 4 all core and 4.35 GHz 1T.

At this point I'm thinking about just dialing in a 4.0 OC and calling it a day as I can easily get 4 stable in P95 with manual overclocking offset with a voltage I'm comfortable with. Yeah I know I'll be losing ST performance but 3.5 on P95 stock just seems too low to me, and the stock idle voltages are much higher than the 4.0 OC...

 

[Jism]

Yes it's normal. Its doing what it's supposed todo. Your all core clock is lowering because of the stress Prime95 puts through your CPU. You'll get a guaranteed base clock, unless you tinker around with PBO and it's settings, this all core clock could be raised as long as temp and VRM are good.

 

[eidairaman1]

Lock voltages down

 

[Jism]

Lock voltages down

Yes, and degrade the chip. Good advice.

Reports of Ryzen 3000 High Idle Voltage Exaggerated, a Case of the "Observer Effect"

With AMD's 3rd generation Ryzen processors finally falling into the hands of PC enthusiasts, many early-adopters are taking to tech communities such as ours, to share their experiences with others. A trend appears to be emerging of users reporting higher-than-usual voltages for these processors...

www.techpowerup.com

Strictly technical: Matisse (Not really)

07/08/2019 6:33 PM (GMT) - Update on the bios issue on Crosshair VIII Hero motherboard ("the thing"). Earlier today I received a response to my inquiries from ASUS. The response was rather technical and I cannot go into the specifics of what exactly it involved. However, it confirmed my...

www.overclock.net

https://www.reddit.com/r/Amd/comments/cbls9g

 

[tabascosauz]

X470 final ABB bios (to keep PCIe 4)
Hyper 212 EVO cooler

At stock in BIOS and idle voltages are very high, like 1.45
Drops with a load.

Prime 95 my clocks are dipping down to 3.5 GHz
Cinebench was around 4 all core and 4.35 GHz 1T.

At this point I'm thinking about just dialing in a 4.0 OC and calling it a day as I can easily get 4 stable in P95 with manual overclocking offset with a voltage I'm comfortable with. Yeah I know I'll be losing ST performance but 3.5 on P95 stock just seems too low to me, and the stock idle voltages are much higher than the 4.0 OC...

35x if it's in P95 Small, it's understandable, as Small runs a reduced multiplier and voltage to prevent melting 7nm Ryzens like reactor no.4. But still on the low side. Do you have PBO Enabled? If so, what are your PBO settings? You can try Smallest to see if your multipliers go up.

Be very, very cautious what you do if you test fixed frequencies and voltages. See what your chip needs to be stable at 40x, and make sure it's not more than 1.25v at that frequency at the very most.

1.4v-1.5v at idle is normal operating behaviour for Ryzen 3000; it's the Vcore you see under SVI2 TFN when all the cores are under 100% load in P95 or CB R20 that actually matters. No noteworthy amounts of current are going through the CPU at times when Vcore is that high.

Also, most importantly, you're shooting yourself in the foot by staying on the 1.0.0.3ABB BIOS. 1.0.0.3ABBA implemented major fixes for a calmer idle as well as boost speeds closer to rated frequency, and 1.0.0.4B even more so. You're not gaining any performance by staying on the old BIOS and betting on PCIe 4.0; chances are, going to the latest BIOS might solve your frequency problem.

 

[Zach_01]

What @tabascosauz said about AGESA versions is right...
1003-ABBA fixed the lower boosting issue and staying to ABB just for PCI-E 4.0 is not woth it.

Additionally ZEN2 is very temp sensitive regarding sustainable boosts (all core and single).
Really these chips are so complex and advanced that a static clock voltage is a killer deal (not in the good way). Do not confuse ZEN2 with ZEN/ZEN+ parts. They have similar features and behaviour up to a point. Further more ZEN2 is something entirely new in a lot of ways.

You need AGESA 1.0.0.3_ABBA or 1.0.0.4_B for the CPU to give the best by AMD and still there is room for "improvement" = with lower Temp + PBO settings with out static values.
ZEN2 needs cooling. I saw a sustainable +25~30MHz (eff clock) increase in all core boost by a 5C max temp reduction (63 > 58) by changing TIM. And another +25~30MHz (eff clock) from manually setting PBO to custom for PPT/EDC and PBO scalar. This increase temp by 3C again to ~61C. Overall clock increase 50~75MHz (50~60MHz on effective clock).
And this benefits the CPU across boost range (from base clock up to single core) but the max is for all core, the min for single (slightly more sustained), and medium gains for medium loads like gaming.

My CPU is rated for 88Watt PPT (65Watt TDP) max draw (actual value = 87.5W PPT from HWiNFO64 sensors mode) like the 3700X. Now, after messing with PBO its on 92~93W PPT with +50~60MHz eff clock. If I could drop it to <50C I would gain another 50~75MHz all core. Rated boost comes with the right AGESA(under 75C) and PBO with additional cooling (max at 50C).
By using static OC and voltage you most likely (not sure) cancelling the FIT controller of the chip and that is not good (silicon FITness controller). This controls PrecisionBoost and PBO by regulating clock, voltage, power draw in conjunction with temp.

X470 final ABB bios (to keep PCIe 4)
Hyper 212 EVO cooler
1. At stock in BIOS and idle voltages are very high, like 1.45
Drops with a load.
2. Prime 95 my clocks are dipping down to 3.5 GHz
3. Cinebench was around 4 all core and 4.35 GHz 1T.
4. At this point I'm thinking about just dialing in a 4.0 OC and calling it a day as I can easily get 4 stable in P95 with manual overclocking offset with a voltage I'm comfortable with. Yeah I know I'll be losing ST performance but 3.5 on P95 stock just seems too low to me, and the stock idle voltages are much higher than the 4.0 OC...

1. Idle/low load voltage is perfectly fine at 1.45~1.50V depending the CPU type. Mine is at 1.475 max. High voltage on low current is not hurting the CPU. It is designed like this for reasons...
2. What was the temperature during this P95 run?
3. You need AGESA 1003ABBA/1004B to hit rated boost.
4. You want a static OC just for burn tests? You are hurting performance in 99% of cases on real life apps/games.

Can you please provide a screenshot of HWiNFO64 sensors mode(expanded) with:
1. During CB-R20
2. During P95
...like this below with no static OC. In order to make it easy for me to see some things just open both CB and that sensors window. Start CB run and then after it starts filling squares hit the clock button in the window below (down right), to reset values and time. Wait 45~50 sec and before it ends hit Ctrl+PrtScr while the HWiNFO window is selected... again like below.
Columns are multiply with blue arrows (down left).
What matters in that screenshot is the average values of: clock, average eff clock, core VIDs, CPU temps, PPT/TDC/EDC.



Additionally you can do the same with static OC.

 

[sn2x]

Installed the latest Chipset Drivers after I made this thread, not sure if that helped or not.

What @tabascosauz said about AGESA versions is right...
1003-ABBA fixed the lower boosting issue and staying to ABB just for PCI-E 4.0 is not woth it.

Additionally ZEN2 is very temp sensitive regarding sustainable boosts (all core and single).
Really these chips are so complex and advanced that a static clock voltage is a killer deal (not in the good way). Do not confuse ZEN2 with ZEN/ZEN+ parts. They have similar features and behaviour up to a point. Further more ZEN2 is something entirely new in a lot of ways.

You need AGESA 1.0.0.3_ABBA or 1.0.0.4_B for the CPU to give the best by AMD and still there is room for "improvement" = with lower Temp + PBO settings with out static values.
ZEN2 needs cooling. I saw a sustainable +25~30MHz (eff clock) increase in all core boost by a 5C max temp reduction (63 > 58) by changing TIM. And another +25~30MHz (eff clock) from manually setting PBO to custom for PPT/EDC and PBO scalar. This increase temp by 3C again to ~61C. Overall clock increase 50~75MHz (50~60MHz on effective clock).
And this benefits the CPU across boost range (from base clock up to single core) but the max is for all core, the min for single (slightly more sustained), and medium gains for medium loads like gaming.

My CPU is rated for 88Watt PPT (65Watt TDP) max draw (actual value = 87.5W PPT from HWiNFO64 sensors mode) like the 3700X. Now, after messing with PBO its on 92~93W PPT with +50~60MHz eff clock. If I could drop it to <50C I would gain another 50~75MHz all core. Rated boost comes with the right AGESA(under 75C) and PBO with additional cooling (max at 50C).
By using static OC and voltage you most likely (not sure) cancelling the FIT controller of the chip and that is not good (silicon FITness controller). This controls PrecisionBoost and PBO by regulating clock, voltage, power draw in conjunction with temp.


1. Idle/low load voltage is perfectly fine at 1.45~1.50V depending the CPU type. Mine is at 1.475 max. High voltage on low current is not hurting the CPU. It is designed like this for reasons...
2. What was the temperature during this P95 run?
3. You need AGESA 1003ABBA/1004B to hit rated boost.
4. You want a static OC just for burn tests? You are hurting performance in 99% of cases on real life apps/games.

Can you please provide a screenshot of HWiNFO64 sensors mode(expanded) with:
1. During CB-R20
2. During P95
...like this below with no static OC. In order to make it easy for me to see some things just open both CB and that sensors window. Start CB run and then after it starts filling squares hit the clock button in the window below (down right), to reset values and time. Wait 45~50 sec and before it ends hit Ctrl+PrtScr while the HWiNFO window is selected... again like below.
Columns are multiply with blue arrows (down left).
What matters in that screenshot is the average values of: clock, average eff clock, core VIDs, CPU temps, PPT/TDC/EDC.

View attachment 141645

Additionally you can do the same with static OC.

I enabled PBO in BIOS, something I thought I did before but it's definitely working now. P95 boosts to ~4 but the power consumption is almost 50% more than the manual OC.

P95 SFFT
TOP: PBO (stopped the test after a minute because temps/voltage looked high)
BOTTOM: Manual OC

35x if it's in P95 Small, it's understandable, as Small runs a reduced multiplier and voltage to prevent melting 7nm Ryzens like reactor no.4. But still on the low side. Do you have PBO Enabled? If so, what are your PBO settings? You can try Smallest to see if your multipliers go up.

Be very, very cautious what you do if you test fixed frequencies and voltages. See what your chip needs to be stable at 40x, and make sure it's not more than 1.25v at that frequency at the very most.

1.4v-1.5v at idle is normal operating behaviour for Ryzen 3000; it's the Vcore you see under SVI2 TFN when all the cores are under 100% load in P95 or CB R20 that actually matters. No noteworthy amounts of current are going through the CPU at times when Vcore is that high.

Also, most importantly, you're shooting yourself in the foot by staying on the 1.0.0.3ABB BIOS. 1.0.0.3ABBA implemented major fixes for a calmer idle as well as boost speeds closer to rated frequency, and 1.0.0.4B even more so. You're not gaining any performance by staying on the old BIOS and betting on PCIe 4.0; chances are, going to the latest BIOS might solve your frequency problem.

Yeah that was the test, turns out PBO wasn't on for the first post. I thought I turned it on previously but I must not have saved or something. By default it was on [AUTO].

Seems stable at 1.21 volts, used the chip for less than a day and the longest stability test I did was P95 for about an hour with no errors.

 

[Zach_01]

Any CB-R20?
Your CPU is suffering from cooling. Even tho I cant see everything I wanted, I can tell this is the main issue. 4.0GHz, 1.05V and 75+C is not good. Thats why you hitting 90+ on 1.28V which is not too high voltage.
First, the version of HWinfo is old 6.12 (latest 6.20). Win version? Win Power plan?
Latest chipset drivers directly from AMD and not from vendor helps some things on Win10 1909, but for sure AGESA is also critical.

I see the "CPU Power Package" which is close to PPT as a value.
Lets make a compare and some math, tho it would be better with CB-R20 test for more direct comparison (same kind of load)

Your 3700X stock
8 cores

avg clock 3975MHz
avg voltage 1.28V
avg power 162W
avg current 69A

My 3600 stock (not on the pic below)
6 cores

avg clock 3975MHz
avg voltage 1.35V
avg power 87W
avg current 79A

3700X have 33.3% more cores than 3600
We can add to 87W another 29W (+33.3%) = 116W

So...
8cores 3975MHz 1.28V = 162W
8cores 3975MHz 1.35V = 116W

How is this possible? And the 3600 has the worst possible silicon quality requiring more voltage than any other ZEN2 SKU for the same speed (less efficient).
One thing that increases power draw is higher temp. Your CPU is dropping to 3.5GHz because Tj (throttle) temp is reached (95C). And why is hitting 160+ power draw when it should be capped at 88W.
I believe this is not stock settings.
What are your PBO settings on the first stock run? Manual? Any scalar change? (x2, x3?) Did you just loosen the limits of PPT/TDC/EDC. Doesnt work that way...

I wanted to see these... but ok



Whats your board? Does it have any BIOS flashback button?
If it does you can flash the board with the latest BIOS... try it out to see if and what you will gain from latest AGESA (you will) and then if you want you can flash it with flashback to older version (it works, I have done it). I know the integrated UEFI utility may not accept older versions.

 

[sn2x]

Your CPU is dropping to 3.5GHz because Tj (throttle) temp is reached (95C).

Dropping to 3.5 isn't an issue for me any more.
It dropped to 3.5 because PBO wasn't enabled in the first test. With PBO disabled both the clocks and voltage were really low in P95 and the power draw and temps were low too.

 

[thesmokingman]

Yes, and degrade the chip. Good advice.

Reports of Ryzen 3000 High Idle Voltage Exaggerated, a Case of the "Observer Effect"

With AMD's 3rd generation Ryzen processors finally falling into the hands of PC enthusiasts, many early-adopters are taking to tech communities such as ours, to share their experiences with others. A trend appears to be emerging of users reporting higher-than-usual voltages for these processors...

www.techpowerup.com

Strictly technical: Matisse (Not really)

07/08/2019 6:33 PM (GMT) - Update on the bios issue on Crosshair VIII Hero motherboard ("the thing"). Earlier today I received a response to my inquiries from ASUS. The response was rather technical and I cannot go into the specifics of what exactly it involved. However, it confirmed my...

www.overclock.net

https://www.reddit.com/r/Amd/comments/cbls9g

That needs to be stickied on every forum so ppl who are ignorant of the process can stop complaining.

I'd also add that yall should stop using monitoring software that is poll based.

 

[sn2x]

Any CB-R20?

Here
TOP: PBO auto CPU volts
BOTTOM: Manual OC

PBO is again drawing 50% more power and is a bit louder than manual OC, and the end score is only 1% higher for PBO (4837 vs. 4777). For this thread my only problem in the first place was 3.5, which was solved with PBO, but after seeing PBO I feel better with manual OC.

 

[tabascosauz]

You're not reading what I wrote. The latest chipset drivers will not help you, because you don't have the latest AGESA in your dated BIOS. AGESA and firmware go hand-in-hand, same goes for the Ryzen Power Plans included in the driver package. At this point I can hardly recommend any additional steps for you to take because you're not even on 1.0.0.3ABBA. Like I said, these deprecated versions boost differently to the latest BIOSes.

To put it in perspective, there's no way a 40.8-41x multiplier results in a 4777pt CB score. I regularly hit those speeds or lower without PBO, at 4950pts. I stay at 41.3x with modified PBO at 4985pts. It all looks like your board vendor didn't work out the bugs for that deprecated BIOS version.

85w package power is nothing out of the ordinary. The stock chip is allowed to draw up to 88W PPT. You are missing quite a few sensors and settings in HWInfo as well.

Judging from the max temperature, you may also be thermally constrained. I top out at 70 without PBO and 76 with PBO, and the former number is with a -0.075v offset. All of which are taken on a significantly better cooler than the Wraith Prism.

I don't know what you think unofficial PCIe 4.0 is doing for you performance-wise; it's not doing anything. Just upgrade to a ABBA or newer BIOS.

 

[Cheeseball]

You gotta update to 1.0.0.4 if its offered for your X470 motherboard for better boost clocks.

1.0.0.4B (1405 BIOS - TUF GAMING X570-PLUS (WI-FI)) allows my 3800X to boost to 4.5 GHz properly. I get a nice 144 FPS with minimal drops in PUBG PC.

Prior to that I saw it only going to around 4.4 GHz.

 

[sn2x]

You're not reading what I wrote. The latest chipset drivers will not help you, because you don't have the latest AGESA in your dated BIOS. AGESA and firmware go hand-in-hand, same goes for the Ryzen Power Plans included in the driver package. At this point I can hardly recommend any additional steps for you to take because you're not even on 1.0.0.3ABBA. Like I said, these deprecated versions boost differently to the latest BIOSes.

To put it in perspective, there's no way a 40.8-41x multiplier results in a 4777pt CB score. I regularly hit those speeds or lower without PBO, at 4950pts. I stay at 41.3x with modified PBO at 4985pts. It all looks like your board vendor didn't work out the bugs for that deprecated BIOS version.

85w package power is nothing out of the ordinary. The stock chip is allowed to draw up to 88W PPT. You are missing quite a few sensors and settings in HWInfo as well.

Judging from the max temperature, you may also be thermally constrained. I top out at 70 without PBO and 76 with PBO, and the former number is with a -0.075v offset. All of which are taken on a significantly better cooler than the Wraith Prism.

I don't know what you think unofficial PCIe 4.0 is doing for you performance-wise; it's not doing anything. Just upgrade to a ABBA or newer BIOS.

4777 was for 4.0 all core Manual OC (4837 was PBO) with a bunch of programs open including hwinfo.

After closing programs All Core 4.0 is 4826, sounds like I could gain maybe 2.5-3% by updating BIOS and using PBO.

 

[tabascosauz]

4777 was for 4.0 all core Manual OC (4837 was PBO) with a bunch of programs open including hwinfo.

After closing programs All Core 4.0 is 4826, sounds like I could gain maybe 2.5-3% by updating BIOS and using PBO.

I get that an updated BIOS will improve PBO, but in the end from your numbers it looks like I'd gain maybe 2.5-3% and I bet the power draw, heat and noise would increase too, disproportionately compared to 4826 at (85w). Am I wrong in thinking this? Those are things I'm more concerned about at this point.

Wasn't your concern heat and noise? The newer BIOSes specifically address these issues at idle.

If you're not longer concerned about absolute scores, you can experiment with offsetting negative voltage or lock it to 4.0GHz at the lowest possible voltage. But again, without being on the latest BIOS makes all this pointless; on your current firmware, you get neither the idle improvements nor the performance improvements.

No, power doesn't scale that way, because the 3700X doesn't pull beyond 90W of package power in Cinebench unless you set a fixed clock and throw all the safeguards out the window...Ryzen boost algorithms are not that stupid.


Whatever floats your boat, I guess.

 

[sn2x]

Wasn't your concern heat and noise? The newer BIOSes specifically address these issues at idle.

If you're not longer concerned about absolute scores, you can experiment with offsetting negative voltage or lock it to 4.0GHz at the lowest possible voltage. But again, without being on the latest BIOS makes all this pointless; on your current firmware, you get neither the idle improvements nor the performance improvements.

No, power doesn't scale that way, because the 3700X doesn't pull beyond 90W of package power in Cinebench unless you set a fixed clock and throw all the safeguards out the window...Ryzen boost algorithms are not that stupid. And if you're worried about heat + noise, your first priority should be to get a better cooler. The Wraith Prism is adequate, and nothing more.

Old BIOS with manual OC has no issues with idle.

I was never concerned with absolute highest score. I'll easily sacrifice 2-3% for lower temps. This thread was only about 3.5, which has already been solved, I thought maybe my CPU was messed up but that was just P95 without PBO.

I'm not buying another cooler, I'm using 212 EVO which I've been using for years now. RIght now with manual OC the system is already quiet enough for my needs.

 

[Zach_01]

Here
TOP: PBO auto CPU volts
BOTTOM: Manual OC

PBO is again drawing 50% more power and is a bit louder than manual OC, and the end score is only 1% higher for PBO (4837 vs. 4777). For this thread my only problem in the first place was 3.5, which was solved with PBO, but after seeing PBO I feel better with manual OC.

PBO probably not working as it supposed to work and I blame AGESA 1.0.0.3 ABB. Something went wrong in this version, that was quickly fixed by 1.0.0.3 ABBA and 1.0.0.4 B. I have use the ABB and it was crap at boosting.

If you had the latest AGESA I could suggested some PBO settings and potential improve performance further but now it’s irrelevant.
Also 1usmus Power plan for Ryzen has some benefits for boosting and for windows scheduler for optimized core loading on low/medium loads. Cores on ZEN2 have a performance label/number and win10 scheduler without latest chipset (from AMD) + the proper power plan + win v1909 + latest BIOS/AGESA... does not know which cores to load and when (unaware of high perf cores).
Again all this is irrelevant with static OC.

Can I ask what is your usage/needs for PCI-E 4.0