Precision Boost Overdrive How Much Over Max Boost You Have

55C under full load is ...no. Not happening unless your idea of room temp is 40 degrees F.

the vast majority here are running water cooled setups be them aio's or 400 dollar custom loops. All temps are relatively the same. with full load precision boost temps in the 70C range at best where the load is pegged with various micro-benchmarks designed to keep the cpu's busy.

edit: 55C is what you commonly see for single core load tests. Not full 24 thread/core load.

Darth Ender said:

55C under full load is ...no. Not happening unless your idea of room temp is 40 degrees F.

the vast majority here are running water cooled setups be them aio's or 400 dollar custom loops. All temps are relatively the same. with full load precision boost temps in the 70C range at best where the load is pegged with various micro-benchmarks designed to keep the cpu's busy.

edit: 55C is what you commonly see for single core load tests. Not full 24 thread/core load.

Click to expand...

My 3700X on custom water barely breaks 60C under full load. And that's only with certain loads. Typically, for tests like Cinebench 20, it sits around 55C.

I had 3900x on the brain.

Riccochet said:

My 3700X on custom water barely breaks 60C under full load. And that's only with certain loads. Typically, for tests like Cinebench 20, it sits around 55C.

Click to expand...

3900X gets a bit warmer with a second chiplet.

In a game like Far Cry 5, Rage 2 or Battlefront 2 I can get about 4250-4275ghz across all cores with a boost now and then to 4300-4400ghz on the 3900X. My temps are typically about 58c-60c using the AMD Wraith, ambient temps are 68F.

3900x will generally run up to 4300mhz across all cores if you're not undervolting and the variables it looks at are favorable. When it's running like that it's going to be in the 70 and 80c range. Unless you're doing some very specific micro benchmark that "loads" the cpu without having it actually doing any work.

typical single threaded loads will be in the 50c range.

That's with good water cooling.

That's going by my own tests and others i've seen. However, i doubt any of them was testing at < 70F. My own tests are about 10F hotter than that, ambient. 75-77F thru all the tests.

https://openbenchmarking.org/result/1908234-HV-1908220HV55

I went thru a bunch of pbo and undervolt settings to see what made a difference. A lot of trials didn't even make it into this because i stopped them after the first runs of the first test because they made no difference at all.

I'm not sure there are any games that can load up 12 cores, much less 24 threads.

I've made so many bios changes to digital power controls and amd's submenu for pbo etc that I'm already sick of cpu tweaking after 2 weeks of ownership.

Either the settings dont do what they say
or they dont expose a setting that is impacting precision boost

I have more than enough available current and voltage with my motherboard. Temps are well under thermal limits. By all signs, enabling pbo should make some difference compared to disabling it, but all it ever seems to do is make margin-of-error level of changes to performance and produce a ton more heat.

I played about 10 minutes of Rage 2 and this is where Im at, typically at 4250-4275ghz but boosting even higher to 4300ghz. The temp is 55c but will go up to 58c-60c

rage 2 may look like it's doing all core load. but the benchmarks show, it only really can feed 4 cores. The load it's giving to 24 amounts to the cpu's doing a lot of waiting. It is enough to keep them pegged at high frequency, but not making them actually work hard. And that's reflected in the temps.

run a benchmark that keeps all the cores fully fed with intensive operations... the most common one in windows seems to be cinebench.

in my benchmark link, a similar test like rage2 is ctx_clock. It does load all cores, but they're all really mostly doing a particular operation that doesn't really make the cpu work hard. It looks like 100 load though. You can see the temps reflect that compared to the other multi-thread tests. it looks much more like my single thread temps like the one right below it, sysbench.

So did you get the 3900X to run benchmarks or did you get it for gaming? I got mine for gaming and very pleased with the performance. You can run benchmarks like Cinebench or Prime 95 for stress testing but none of it translates to playing games. They are synthetic benchmarks for testing only, the real meat and potatoes is for real world gaming for most people.

kilroy67 said:

So did you get the 3900X to run benchmarks or did you get it for gaming? I got mine for gaming and very pleased with the performance. You can run benchmarks like Cinebench or Prime 95 for stress testing but none of it translates to playing games. They are synthetic benchmarks for testing only, the real meat and potatoes is for real world gaming for most people.

Click to expand...

Compiling mostly. I program. And i game. If i want to test how good my cooling is compared to the hardware settings i'm using, i use the most stressful test i can find to make sure that over the course of my actual use cases i'll never exceed what I expect or what the hardware can sustain. I wouldn't run a game benchmark that can only push the hardware to some fraction of what it is capable of and say "hey, it only hits 55C, it's great!" when there are real world actions that hardware will see that will utilize 100% of it's potential - and for extended periods of time.

When comparing to others, mostly a common factor has to be used. Loading some fraction makes that hard to compare to others. Loading 100% is easier.

Darth Ender said:

Compiling mostly. I program. And i game. If i want to test how good my cooling is compared to the hardware settings i'm using, i use the most stressful test i can find to make sure that over the course of my actual use cases i'll never exceed what I expect or what the hardware can sustain. I wouldn't run a game benchmark that can only push the hardware to some fraction of what it is capable of and say "hey, it only hits 55C, it's great!" when there are real world actions that hardware will see that will utilize 100% of it's potential - and for extended periods of time.

When comparing to others, mostly a common factor has to be used. Loading some fraction makes that hard to compare to others. Loading 100% is easier.

Click to expand...

Than that makes sense, many in here play games, and use those tools just to see if their system is running stable and within spec. I can run one pass of Cincebench R20 and stay at about 4ghz and 65c with air cooling, other than that gaming is my main area of focus, I dont even bother with Prime 95 anymore. Most people with these new AMD CPU's have found leaving everything on default auto and letting the CPU do its thing is pretty much the best approach. There is some speculation that PBO is not working or not working correctly.

I can't tell if pbo is not working or if it's overlapping with just the default settings for most motherboards.

I can say for certain that on the Asus pro x570 ws ace motherboard, that there doesn't exist a PBO setting for the 3900x that makes any significant / meaningful difference to performance for single or full multicore loads. All it does is significantly increase the heat output.

But what does PBO working even look like? because we get conflicting info all over. At least officially, PBO doesn't let you exceed the max frequency set in the chip. So 4.6Ghz more or less for the 3900x. So if you're hitting that with pbo disabled and just using pb2, then is pbo on your motherboard borked and the settings not doing what they say or is the default values for your motherboard already set to what the motherboard provides and not what AMD wants those values to be by default (the spec values) ?

up until pbo was a thing, i'm sure they were used to setting the values for what the vrm can provide current wise and wattage wise and voltage wise to what the motherboard actually can do and the cpu would only care about that if you manually told it to overclock. But now pb2 reads them and operates off them. PBO is just pb2 with those values supposedly set to what the motherboard can handle or manual...

so like edc/tdc etc values would be by default the amd spec for the given cpu.
There would be another set of these values that is called the motherboard - which has the actual motherboard capacities and values for them
then there would be a set that is user defined.

If the motherboard manufacturers just use the motherboard values for the default though, then pb2 out of the gate will behave like pbo, minus the additional flags that control the scaler values which controls the boost voltage offsets. So you get the limits of pbo, but with the boost voltage of normal pb2. Hence perhaps less max frequency depending on the silicone lottery.

At least that's my best guess considering the testing I did on my motherboard showing nearly no change enabling or disabling and a metric F ton of other tweaks to the manual pbo settings in and out of the amd submenus ... It all behaved basically like the values the cpu references for tdc and such were already the PBO values, and the only difference was the scaler.

But my sample size is 1 ... i dont have the opportunity to have a bunch of motherboards to back to back test. And at the end of it all, I'm not even sure if with an example of it working properly, how much of a difference will it even make compared to the heat it will add.

I have 2x 3600X.
I have 6 BIOS now.
1 BIOS boost Max boost
5 BIOS that do not
For me I only game and none of the BIOS whether boosting over max boost or not make a difference in gaming.
Plus Ryzen CPU's shut off cores when gaming ,even with an ALL core overclock.Just the way it is,sure is fun testing the chips but after 1 1/2 months I am almost done testing,fuck I am a good liar ,I will never be done trying to get 1 more FPS in KDC.

JCNiest5 said:

I'm at a loss of explanation...my 3900x on Gigabyte AORUS X570 Master, some cores can boost to 4298 when running AID64. Cinebench R20 can make all cores boost to 4198. Never got the 4525 or 4500 people are getting. A disappointment in term of OCing, but I do not run my system OCed at all, so I'm just happy running the system normally.

Click to expand...

I have same board. 3900x. Hit 4550 all the time.

Bios is f5I
Latest Aug 18th or whatever chipset drivers from AMD directly. Not the mobo maker.

Auto oc on
Pbo on enable

I maintain 4100 all core when doing heavy workloads

The proper question to ask is how much under max boost you're getting.

Well with pbo off, you are less limited by heat than you are by tdp.

I've run many tests and most normal benchmarks never fully utilize the CPU even when running 24 processes . This allows the frequency to go up.

Mprime, or prime95 in Windows, will fully load your cpu.

Normally without pbo or with, my full load all core frequency hits 4.1-4.2 ghz.

With mprime doing it's most stressful test, that went down to 3.7 ghz.

Temps were fine, low to mid 60c. The heat wasn't just focused in the processing parts of the cores but also the cache.

I'm going to test pbo enabled with 1x scaler but tdp pushed way up and see if temps stay normal but peak frequency is better off.
I'm referring to pb2 peak frequency.

I've never seen pbo break out into 4.7+ ghz no matter what setting or how hot the CPU gets on the golden cores

Quick and dirty checking in, 3600 (non X), PBO 'enabled', P95 Blend at 12 threads show all cores pinned at 4135. With only one thread running, one core is pinned at 4199.
Bios version 3.60 (8/6/19) of rig in sig.

Ruiner said:

Quick and dirty checking in, 3600 (non X), PBO 'enabled', P95 Blend at 12 threads show all cores pinned at 4135. With only one thread running, one core is pinned at 4199.
Bios version 3.60 (8/6/19) of rig in sig.

Click to expand...

Nice wqish I could hit that

The newer the BIOS the Lower the clocks which still does not effect performance.AMD to the rescue Sept 10 2019 with BIOS thru Motherboard manufactures at a later date.

For now do not worry about the speed.It is not all doom and gloom. Hell worst case here at 1920x1080

ironically, the thing that people are butt hurt the most about not reaching their single thread boost clocks at (most games) are also not cpu bound processes and spend a huge portion of time waiting on io and graphics bound processes to finish before being able to continue on in it's mostly single threaded task. So they do not benefit greatly from a cpu that's already spending time waiting for other non-cpu tasks to complete.

The tasks that are highly cpu bound tend to be also threaded. So they would be stuck with the lower clocks anyway, of which this bios fix would only marginally impact (unless it fixes pbo).

I suspect either the bios will simply correct the issue to meet legal obligations ...and so wont really make any difference performance wise, but be enough to shut the lawsuit happy talk down. Or it'll also fix pbo functionality, which will open up whatever little headroom there is to push the chips as hard as temps will allow instead of being hamstringed as they currently are.

Darth Ender said:

ironically, the thing that people are butt hurt the most about not reaching their single thread boost clocks at (most games) are also not cpu bound processes and spend a huge portion of time waiting on io and graphics bound processes to finish before being able to continue on in it's mostly single threaded task. So they do not benefit greatly from a cpu that's already spending time waiting for other non-cpu tasks to complete.

The tasks that are highly cpu bound tend to be also threaded. So they would be stuck with the lower clocks anyway, of which this bios fix would only marginally impact (unless it fixes pbo).

I suspect either the bios will simply correct the issue to meet legal obligations ...and so wont really make any difference performance wise, but be enough to shut the lawsuit happy talk down. Or it'll also fix pbo functionality, which will open up whatever little headroom there is to push the chips as hard as temps will allow instead of being hamstringed as they currently are.

Click to expand...

Its more on principal than anything. Lets just hope AMD labels boost clocks better in the future which are closer to what is realistic vs what is the very top end a user might see if all conditions are correct.

The thing is, there are no standards by which to measure and report these things. So if AMD takes a conservative approach, they run the risk of Intel marketing their stuff differently and looking better while in fact, not being.

Isn't that already happening with TDP? Intel reports some average TDP based on some simulated workload benchmark they created themselves but amd reports the tdp you would see at max full load (not overclocked via automatic or manual methods) ... So on paper, intel's cpu's seem much more energy efficient but those numbers dont reflect reality.

Or even cpu frequency in general from any company when used in mobile devices and laptops where laptop manufacturers will tout their cpu speeds but in reality, have no hope of ever running at them for any significant amount of time.
How long does it need to be able to be run at a given speed before it can be marketed as such? Forever? But in doing that, you do a disservice to what could be a much more common use case of only having sporadic high cpu load, which would benefit from being able to peak momentarily at speeds you can't sustain forever. How would companies convey that information in marketing ?

There's no such standards body to define rules for these things and how to convey them to consumers.

Car manufacturers do the same thing ... cars "top speeds" are set by essentially sampling 1 in the most perfect conditions. They dont take them all on a drive to ensure they can reach it. They dont shave some mph off to include a fudge factor.

Definitely there is room for improving how things are marketed when it comes to overclocked speeds and such. I'm just not sure there is one that will make everyone happy.

With PBO the max I see is 4.66ghz multiplier 46.75. Anything threaded there is zero difference or very little change.

The new update will most likely be a whole bunch of nothing unless more cores and threads are boosted more than now.

Ruiner said:

Quick and dirty checking in, 3600 (non X), PBO 'enabled', P95 Blend at 12 threads show all cores pinned at 4135. With only one thread running, one core is pinned at 4199.
Bios version 3.60 (8/6/19) of rig in sig.

Click to expand...

I should add the following data points:
running 12 threads, under aftermarket air (with fairly quiet fan settings) Ryzen Master shows 60C, 1.38V, PPT 2%, TDC 32%, EDC 46%
1 thread: 56C, 1.37V, 1%, 10%, 8% respectively

I'll install the new bios version when it drops but don't expect any increases in boost if I'm already getting this.