5900x boost question

Astrodave

[H]ard|Gawd
Joined
Apr 20, 2005
Messages
1,080
So I'm new to AMD and just built a 5900x system with a 3090. I can't seem to figure out boost though. I have hard set the bios multiplier to 47 and 1.35v and am stable at that setting with an all core boost at 47. When I do benchmarks in windows I see that all cores boost to 4.7ghz and it shows up in HWinfo showing the effective clock of all cores at 4.7ghz. When I game, however, the boost is rarely going above 4ghz. The only game I've seen that will cause a single core to boost to 4.7 is of course P3D which is known for single threaded reliance. WTF are the cores not boosting to 4.7 when I game? I am using a cooler master 360 AIO and my temps never go above 80c with extreme benchmarks. My 9900k machine boosts all cores to 5.1 whenever ANY load is on them. This processor seems to only apply boost when it thinks it needs to and games don't seem to be pushing it. I want to force this thing to boost to 4.7 on all cores ANY time a load is on the cores. Is this by design on AMD's part or is there something going on with my setup? I'm running an ASUS X570 TUF WIFI board. I have tried overclocking with Ryzen Master and direct through the BIOS and get the same results. I just installed the latest AMD drivers from AMD's site as well. I set the power plan in windows to high performance and it makes no difference. It's like the C state power settings are ignoring everything I have set. I'd appreciate some insight from someone who is familiar with these things.
 

alyjen

Weaksauce
Joined
Mar 24, 2011
Messages
102
same goes for single core benchmarks? does it boost single core correctly or also only 4GHz?

rule of thumb is that all core OC will work great for workload, video encoding and multithread benchmarks ;) you'll get better results with less power, but if you want gaming performance you have to stick to PBO & learn how to play around it so your CPU could boost even higher while gaming
 

thesmokingman

Supreme [H]ardness
Joined
Nov 22, 2008
Messages
6,617
You should not be randomly setting a manual voltage. You need to google silicon fitness monitor and learn the process to find your safe voltage for a manual overclock. Otherwise just leave it alone and let pbo or curve optimizer handle it.
 

Astrodave

[H]ard|Gawd
Joined
Apr 20, 2005
Messages
1,080
I enabled pbo in the bios but even with everything else at stock except for enabling pbo 4.2 is still the max it will go and in games it still doesn’t even boost to that reliably. You can definitely tell that it’s using multiple cores in game but not at max speed. I don’t get this amd stuff. Intel is set it and it works like you set. This amd boost does whatever it wants regardless of settings In the bios.
 

Astrodave

[H]ard|Gawd
Joined
Apr 20, 2005
Messages
1,080
And 1.35v is plenty safe for these chips and it runs reliably benchmarked at that voltage. let’s not overthink that part.
 

thebufenator

[H]ard|Gawd
Joined
Dec 8, 2004
Messages
1,305
And 1.35v is plenty safe for these chips and it runs reliably benchmarked at that voltage. let’s not overthink that part.

If you manually set the voltage you are stopping Precision Boost from doing what it needs to. It is trying to dynamically scale clocks and voltage, while keeping TDP and temps within a certain margin. When you set a manual voltage, the only think it can scale is clocks. Typically, these boost better when undervolted.
 

Astrodave

[H]ard|Gawd
Joined
Apr 20, 2005
Messages
1,080
I enabled pbo and set everything else to defaults and I still have weird behavior. CPUz says it’s running at 4.6 now but hwinfo says the effective speed is still 4.2. My r20 scores are also wayyy down compared to my manual 4.7. Is hwinfo showing wrong info on these new cpus? I need to test to see if it boosts in games or not but I’m betting it still will be running far below that like it has been. I need a walkthrough on overclocking these things on an asus x570 tuf...
 

Astrodave

[H]ard|Gawd
Joined
Apr 20, 2005
Messages
1,080
So CPUz and core temp both show boosting at 4.7-4.9 right now but hwinfo shows effective speed at 4.2 and judging by my cinebench score compared to 4.7 manual oc I think it’s running at 4.2. what are you guys using to monitor your speeds and voltages? Doesn’t seem like a lot of the tools I used for intel work on these new amd chips.
 

thebufenator

[H]ard|Gawd
Joined
Dec 8, 2004
Messages
1,305
hwinfo should show you the highest boost clocks each core is hitting. Manual OC you will likely end up with a higher multi core score on CB, but letting precision boost do its thing should give you a higher single core score. What temps are you hitting on CB multi-core?
 

thesmokingman

Supreme [H]ardness
Joined
Nov 22, 2008
Messages
6,617
Yeah so tell us how it works then because that’s what everyone else is doing when they oc for workloads...

I already explained what to do in my first post. And no one is randomly setting voltages on manual if they know what's appropriate for these chips. :rolleyes:
 

BrotherMichigan

Limp Gawd
Joined
Apr 8, 2016
Messages
343
So CPUz and core temp both show boosting at 4.7-4.9 right now but hwinfo shows effective speed at 4.2 and judging by my cinebench score compared to 4.7 manual oc I think it’s running at 4.2. what are you guys using to monitor your speeds and voltages? Doesn’t seem like a lot of the tools I used for intel work on these new amd chips.

"Effective speed" is a value calculated based on the actual C-State occupancy during the sampling period. Don't use that as an indication of the actual boost clocks being hit. If you're showing 4.7-4.9 GHz max clocks, everything is working as it should.
 

alyjen

Weaksauce
Joined
Mar 24, 2011
Messages
102
"Effective speed" is a value calculated based on the actual C-State occupancy during the sampling period. Don't use that as an indication of the actual boost clocks being hit. If you're showing 4.7-4.9 GHz max clocks, everything is working as it should.
that's part of the picture, but you may end up in clock stretch situation where you'll see high boost clocks reported yet effective clocks will never go near to these values, it's best to check all OC with some benchmark with good scaling, it's super easy to get cheated by this rather smart CPU pretending it's working faster but actually quite the opposite for the sake of stability
 

BrotherMichigan

Limp Gawd
Joined
Apr 8, 2016
Messages
343
There's no "pretending" or "cheating" going on for the sake of stability, it is simply the design of literally every modern CPU to enter low power states as quickly and as often as possible, especially for designs like Zen 2 which have intelligent boost algorithms that have strong dependencies on power draw and temperature. The CPU is actually running at the reported speeds when it is doing work during the sampling period, it is just that in the light load situations where you see maximum boost clocks the work is often much shorter than the duration of the sampling period, leading to the situations where the CPU will be spending much of that time in low power states. This is evident during something like normal desktop navigation, where my 3800X will report average core clocks of around 4 GHz on all cores and peaks of about 4.5 GHz, but an average effective clock of just around 300 MHz for that period.
 

alyjen

Weaksauce
Joined
Mar 24, 2011
Messages
102
I got what you say but it's not the complete picture. Check this text https://www.gamersnexus.net/guides/3494-amd-ryzen-3000-undervolting-offset-override they mention it somewhere on the way, and there were more reports of the same since then. Zen 3 is no different.
If you push it too high, while limiting voltage, it'll try to boost higher but fall back to lower clock to maintain stability, it's not C state thing.

You can easily check it by creating conditions in which your boost clock will be higher while your benchmark score and reported effective clocks will drop.
 

BrotherMichigan

Limp Gawd
Joined
Apr 8, 2016
Messages
343
That doesn't apply at all to stock operation and would still show up in the effective clockspeed reported by whatever monitoring tool you choose to monitor the speeds. So again, no "pretending" or "cheating."
 

amd7674

[H]ard|Gawd
Joined
Jan 1, 2007
Messages
1,772
I'm still waiting for my 5900x to finish my build. so I could be talking out of my ass. However I saw this posted on MSI forums:

CPU Frequency Cannot Go Turbo Frequency
Note. This issue should be common issue on all motherboard vendor

This issue happens when using Ryzen 5000 CPU + PBO enable + adjust CPU voltage
Symptom: CPU frequency locks at base frequency and will not boost to turbo
Solution: No ETA


I'm not sure if this is even related to OPs issue.
 

thesmokingman

Supreme [H]ardness
Joined
Nov 22, 2008
Messages
6,617
"Effective speed" is a value calculated based on the actual C-State occupancy during the sampling period. Don't use that as an indication of the actual boost clocks being hit. If you're showing 4.7-4.9 GHz max clocks, everything is working as it should.

This doesn't make any sense, at all.

It's effective clocks and they are in fact the only accurate reading and it is only available in hwinfo. The 4.7ghz he's seeing is discrete clocks and that is pretty much just a target but is in no way indicative of the real clock speed. That said the monitoring apps used here are incorrect as the only apps that show the right clocks are ryzen master and hwinfo.

https://www.hwinfo.com/forum/threads/effective-clock-vs-instant-discrete-clock.5958/
 

BrotherMichigan

Limp Gawd
Joined
Apr 8, 2016
Messages
343
This doesn't make any sense, at all.

It's effective clocks and they are in fact the only accurate reading and it is only available in hwinfo. The 4.7ghz he's seeing is discrete clocks and that is pretty much just a target but is in no way indicative of the real clock speed. That said the monitoring apps used here are incorrect as the only apps that show the right clocks are ryzen master and hwinfo.

https://www.hwinfo.com/forum/threads/effective-clock-vs-instant-discrete-clock.5958/

Your link actually agrees with what I said.
 

BrotherMichigan

Limp Gawd
Joined
Apr 8, 2016
Messages
343
Huh? You are confused dude. Read your post I quoted and read it again. You wrote to ignore effective clocks and to watch discrete clocks.
Oh, yeah, but just to the point of verifying that everything is working correctly after he reverts his changes, which is fine. It's much harder to verify max boost clocks by placing the CPU under a load and watching the effective clock because under the sort of loads where you're going to hit max boost, the effective clock will almost always be lower because of the nature of those workloads and how effective clock is measured.
 

thesmokingman

Supreme [H]ardness
Joined
Nov 22, 2008
Messages
6,617
Oh, yeah, but just to the point of verifying that everything is working correctly after he reverts his changes, which is fine. It's much harder to verify max boost clocks by placing the CPU under a load and watching the effective clock because under the sort of loads where you're going to hit max boost, the effective clock will almost always be lower because of the nature of those workloads and how effective clock is measured.

You're still not getting it I'm afraid. Discrete clocks do not equal real clocks.
 

BrotherMichigan

Limp Gawd
Joined
Apr 8, 2016
Messages
343
Actually, it is you that isn't getting it. Discrete clocks ARE real clocks at a particular instant in time. Effective clocks are an average of the speed over some sampling period.

From the link you yourself posted:
Hence a new approach needs to be used called the Effective clock. This method relies on hardware's capability to sample the actual clock state (all its levels) across a certain interval, including sleeping (halted) states. The software then queries the counter over a specific polling period, which provides the average value of all clock states that occurred in the given interval.
The Effective frequency does not represent a particular real clock, but the average clock value where sleeping states do not contribute to clock.
So for example when a core is running: 800 MHz, 0 (sleep), 0 (sleep), 0 (sleep)
the average value (effective clock) is: (800 + 0 + 0 + 0) / 4 = 200 MHz

I can feel FrgMstr eyeing this thread right now and readying his big green text, so I will leave it at this. My recommendation is that the OP leave everything at stock. They can verify their max boost clocks and compare his sustained clocks under particular loads to those achieved by places like review outlets to make sure everything it working correctly.
 
Last edited:

thesmokingman

Supreme [H]ardness
Joined
Nov 22, 2008
Messages
6,617
Actually, it is you that isn't getting it. Discrete clocks ARE real clocks at a particular instant in time. Effective clocks are an average of the speed over some sampling period.

From the link you yourself posted:



I can feel FrgMstr eyeing this thread right now and readying his big green text, so I will leave it at this. My recommendation is that the OP leave everything at stock. They can verify their max boost clocks and compare his sustained clocks under particular loads to those achieved by places like review outlets to make sure everything it working correctly.

Sorry, you are really missing it. Discrete clocks are only applicable at the time the clocks are polled, that one time. Telling ppl to watch discrete clocks is very misinformed. The only meaningful clocks to watch are the effective because they take into account the varying states. Don't get all caught up on semantic of real clock wording.
 

BrotherMichigan

Limp Gawd
Joined
Apr 8, 2016
Messages
343
I think you're having a hard time reading what I wrote, but that's fine. The OP will be fine if he just sets everything back to stock and does some research of their own. Just don't watch GN.
 

thesmokingman

Supreme [H]ardness
Joined
Nov 22, 2008
Messages
6,617
I think you're having a hard time reading what I wrote, but that's fine. The OP will be fine if he just sets everything back to stock and does some research of their own. Just don't watch GN.

While I agree and have written that op should google and learn about the voltage, you are just wrong and now playing semantics.

The point I made is encapsulated in the opening paragraph of the hwinfo link.

It has become a common practice for several years to report instant (discrete) clock values for CPUs. This method is based on knowledge of the actual bus clock (BCLK) and sampling of core ratios at specific time points. The resulting clock is then a simple result of ratio * BCLK. Such approach worked quite well in the past, but is not longer sufficient. Over the years CPUs have become very dynamic components that can change their operating parameters hundreds of times per second depending on several factors including workload amount, temperature limits, thermal/VR current and power limits, turbo ratios, dynamic TDPs, etc. While this method still represents actual clock values and ratios reported match defined P-States, it has become insufficient to provide a good overview of CPU dynamics especially when parameters are fluctuating with a much higher frequency than any software is able to capture. Another disadvantage is that cores in modern CPUs that have no workload are being suspended (lower C-States). In such case when software attempts to poll their status, it will wake them up briefly and thus the clock obtained doesn't respect the sleeping state.
 

THUMPer

2[H]4U
Joined
May 6, 2008
Messages
3,511
Manual OC for the normal user is out. Leave CPU stock or play with Curve Optimizer for best results.
 

MavericK

Zero Cool
Joined
Sep 2, 2004
Messages
30,528
For my chip, I have seen individual cores get up to 5050 Mhz, but on a full-bore multi-core workload like Cinebench R20, they seem to settle around 4.3 Ghz. Only using PBO and max offset, haven't done any manual overclocking.
 
Top