Cannot get 3900x to Max Boost to 4.6Mhz.

[Tarvaln]

I've been doing testing with my 3900x and I can't get it to Boost to 4.6Mhz. I've done all setting through Ryzen Master and have tried Default, AOC, and PBO settings. I've been going through different forums and I see the same with other 3900x users. I know there have been investigations with other 3000x CPU's. I think this is an issue with just the 3900x.
I've created a consolidation post here https://community.amd.com/thread/242812 for testing results. If you have a 3900x, please check that topic and post your results as per the guidelines. I really want to see if anyone is getting the results as advertised.

Here is an example of my testing.

Cannot reach 4.6Mhz on Max Boost. Test with Auto Over Clock set with Ryzen Master. 

MB: Asus Prime x570 Pro BIOS: ver 1005 (AGESA 1.0.0.3 ABB), RAM: G.SKill 3200c14. Highest MAX Boost: 4.355

 

[Princess Luna]

You're aware that 4.6ghz is a single core boost and you'll never manage to actually clock all 12 cores at 4.6ghz running Cinebench right?

[Tarvaln]

If you look at the image, you will see I am running the single core test.

[Princess Luna]

1 minute ago, Tarvaln said:

If you look at the image, you will see I am running the single core test.

Yes, while running OS and background/side ground applications along side.

 

You have more than just one core in use already.

[Tarvaln]

Yes, Ryzen Master is running. Windows runs apps and other things all the time. Can you provide a screen shot of a 3900x hitting 4.6Mhz on any core through Ryzen Master? 

[Cyberdyne_]

I would try a different benchmark, I think Cinbench r20 uses instructions that could prevent a higher boost. I also think that top single core boost may only show itself under lower load, like a game. Maybe try r15 and/or a GPU limited game. Or even just try video playback in a media player.

Unfortunately you kinda need all the stars to align to get that top single core boost to show up. But I've seen it in reviews for games, pretty sure Far Cry 5 bench can do it for example.

[Tarvaln]

Cyberdyne_ : Thanks for the advice. I can tell you between myself and a few others we have tried Cinebench R15, R20, CPU-Z, Prime95: Blend, FFT Small and Smallest. I think there is another one I'm forgetting. That's part of the reason I made that consolidated topic https://community.amd.com/thread/242812 Test and results are all over different forums. I wanted a place where we could compare notes and try to solve this. I'll try the Farcry 5 bench but it would be odd that it's the only condition where it works. Also, most of the testing and reviews are for 3600x - 3800x. Those can hit their Max Boost depending on the MB and BIOS ver. The 3900x seems to be the exception across all MB's.

 

This is not a slight to you or anyone else. There seems to be this group think of "Well, under the right conditions..." My problem is no one can tell me what those conditions are. I've yet to see these conditions on any website or forum from AMD to MB vendors to user forums. So, unless these conditions can be duplicated, then I don't think they exists.

[Cyberdyne_]

I might have seen a different Zen 2 CPU in FC5.

I didn't realize the full scope of the mission here Tarvaln. Carry on soldier.

[DildorTheDecent]

45 minutes ago, Tarvaln said:

 

That's because no benchmark or program is truly "single-thread" anymore

 

If you want to see just one core being utilised then you should either:

• set the core affinity in task manager
• just enable one core in BIOS and benchmark it (serious suggestion, people actually do this)

 

32 minutes ago, Cyberdyne_ said:

I think Cinbench r20 uses instructions that could prevent a higher boost.

That's not it chief.

[Cyberdyne_]

His screenshot does show one core being used for r20, one active core at 4.2. Three other cores at 300 mhz, likely for Windows and such. 

 

You can't be seriously suggesting that AMD's boost rating is based on taskmanager affinity and disabling cores. I understand those things are useful for testing, but AMD can't be expecting people to do that.

[Tarvaln]

17 minutes ago, DildorTheDecent said:

That's because no benchmark or program is truly "single-thread" anymore

 

If you want to see just one core being utilised then you should either:

• set the core affinity in task manager
• just enable one core in BIOS and benchmark it (serious suggestion, people actually do this)

 

That's not it chief.

The quote you posted is blank so, I don't know what you are trying to quote. However, the point of the tests is to see if any core can hit 4.6 Mhz as advertised. This should be able to be accomplished in normal operational parameters. No one should have to disable all cores but one to hit that Mhz. That's like saying PBO works but only in Juneau, Alaska. Yea, anyone could travel there and test it but, it seems rather unreasonable.
Do you have a link to any of these one core test?

[lansing]

I did a cinebench R15  on my 3900x a few days ago for single core, the highest I got is a 4475. I think the limit came from the motherboard.

[DildorTheDecent]

50 minutes ago, Cyberdyne_ said:

You can't be seriously suggesting that AMD's boost rating is based on taskmanager affinity and disabling cores. I understand those things are useful for testing, but AMD can't be expecting people to do that.

On the contrary you can't be seriously suggesting that Cinebench has "algorithms" to limit the boost clock of the processor.

 

Algorithms that determine the maximum boost clock for a variety of core loads are not defined by cinebench. They are defined by the processor and can be modified in the BIOS or via software, in this case it's Ryzen Master.

52 minutes ago, Tarvaln said:

The quote you posted is blank so, I don't know what you are trying to quote.

It's a common courtesy to snip large walls of test and images. Finding which post isn't hard as all you need to do is check the date and time of the post that I'm quoting. Alternatively, use the arrow in the top right of the quote box.

52 minutes ago, Tarvaln said:

However, the point of the tests is to see if any core can hit 4.6 Mhz as advertised. This should be able to be accomplished in normal operational parameters. No one should have to disable all cores but one to hit that Mhz. That's like saying PBO works but only in Juneau, Alaska. Yea, anyone could travel there and test it but, it seems rather unreasonable.

And how else are you going to achieve it? Run a single thread benchmark and watch the load dance around multiple cores because it's a scheduling issue or are you going to set strict parameters? And if you cannot achieve the result with the strict parameters in place and someone else can, then it's a setup issue.

 

As far as I'm concerned, affinity/disabling cores is the only sound solution for now. That's all that has ever worked for me during testing with various processors over the years. Feel free to give it a go yourself. Or don't, I'm not your boss.

52 minutes ago, Tarvaln said:

Do you have a link to any of these one core test?

https://hwbot.org/benchmarks/processor

 

Look out for the PI and x265 benchmarks in particular. Geekbench works too but don't take it too seriously.

 

As far as reviewers showing off boost frequency there's this by techpowerup.

[bimmerman]

Pretty sure you're likely to see pigs fly more frequently than any of the Zen 2 CPUs reaching their advertised max boosts in actual usage.

 

There is quite a bit of discontent on the internets about it. My 'must buy a 3950X' idea has become more hesitant until this gets sorted out, either by solution or statement from AMD on the matter. Beyond, y'know, their 'we explained boosts in this marketing copy, gfys' post.

 

My 3600 seems to be an exception, it's been doing all-core and single core 4.2GHz with just DOCP enabled. It won't exceed that, no matter what I try (PBO/AutoOC/pixie dust), but it routinely will hit its spec at least. I'd like more cores, but from what I've seen I won't necessarily get any better core performance from spending more, so.......shrug. I don't have enough use cases to really benefit from more cores at the same effective speed.

[Cyberdyne_]

Lol I didn't say algorithms! Why is that in quotations @DildorTheDecent ? I said instructions. R20 uses newer AVX instruction sets, which can change how a CPU boost. What a strange assumption you made here.

[Tarvaln]

3 hours ago, DildorTheDecent said:

 

As far as reviewers showing off boost frequency there's this by techpowerup.



Thanks for the link. Though, the review does show a Boost Graph, I can't find what test or tests if was achieved on. Also, I didn't see anything about disabling cores on that page to achieve it. 
 

Quote

The data below presents the minimum, maximum, and average clock frequency of a given core/thread-count combination for a typical heavy workload. We start with one thread and go all the way up to the CPU's maximum thread count while at the same time measuring the average clock frequency for these timed testing runs.

I contacted the author Wizard about his testing for Max Boost and I'm hoping he gets back to me soon. I really want to know what he used to observe these core Mhz,

 

4 hours ago, DildorTheDecent said:

As far as I'm concerned, affinity/disabling cores is the only sound solution for now. That's all that has ever worked for me during testing with various processors over the years. Feel free to give it a go yourself. Or don't, I'm not your boss.

Can you please share your results with this on a 3900x then?

[bignaz]

Mine has no problem hitting its boost. What are you cooling it with?

[Tarvaln]

bignaz: I'm using stock but under single core test, my temps don't go above 68C. That should be cool enough. 

Could you please post your findings in this topic https://community.amd.com/thread/242812 following the guidelines there. I want to replicate what you did. Also, please use Ryzen Master to view your cores.

 

A few of us are seeing different peak Mhz on cores between Ryzen Master and HWINFO. I'd like to keep results as uniform as possible so please use Ryzen Master.

 

Thanks!

[xg32]

3 hours ago, bignaz said:

.

zen 2 voltages always scares me lol.

[bignaz]

2 hours ago, xg32 said:

zen 2 voltages always scares me lol.

You should of seen when i put it on Ln2. Those were some nice voltages. But the cpu sucks on LN2 but is number one currently on water But my mobo reads .06v higher then what it actually is.  But im not scared i'll put 1.55 to this thing no problem. Just dont expect any all core full loads at that voltage lol. Voltage wont kill it. The current will. Thats why single core voltages are to the moon but all core voltages are around 1.35ish give or take.

 

 

To many unknowns about your system  I seen multiple 3900x hits 4.6 under a single thread load. But it seem most people dont understand what a single thread load is any more. So i just avoid it. They can chase after the pointless single thread clock why there performance is crap because they are ignoring the multi thread clocks...the ones that actually matter. And then im not going to compare my 3900x to yours. As of right now on water there are only 2 other 3900x's that can pull the same clocks. I'm still in the number one position but my cpu represents maybe .1% of them out there.  To hit that single core clock you might need more voltage. PBO loves the voltage and with that stock cooler it just wont happen. I mean it could be tons of things. 

 

But seriously focusing on the single thread is just a waste of time. I would focus on the multi core.

[G00fySmiley]

a lot of those things are for marketing. If you really want synthetics I am sure one of the cores will work to hit 4.6 just disable all the others in the bios. 

 

likely the 4.6 being seen IRL is going to mostly be cherry samples with water cooling and even then rarely be used. 

 

most max boost clocks are marketing and should come with mountains of fine print. My i7 8700k says it will max boost clock to 4.7 Ghz... but it never actually did that (pre overclocking and increasing voltage, now it does 4.9 just fine on air cooling with a honking noctua nh d15) 

 

what really matters is not synthetics but how well it handles your workloads. (unless you are a professional overclocker competing for highest clocks in synthetic of coarse i suppose) Now you could contact AMD and say it is not performing to expectations, maybe they will take an RMA and send you a better binned sample or just another chance at silicone lottery if it is bothering you that much. 

 

 

[Tarvaln]

This is not to throw shade at anyone on this or other forums. But this is something that keeps coming up.

 

So, a common trend I see, not only this forums but others is,:

Me: "I can't get get my Max Boost to hit 4.6Mhz. Can anyone else?"

Them: "Well if you adjust X,Y,Z, under certain conditions it should work."

Me: "Great, can you please show me what you did to achieve it?"

Them: "Well, getting Max Boost doesn't matter..."

 

Everyone seem to have advice or claims but, no one can seem to provide proof.

 

So, I don't have anything against AMD, Intel, Nvidia or any other company. I just want to see if anyone can hit 4.6Mhz Max Boost on a 3900x under normal conditions.

 

If you can, great! I'd love for you to follow the guidelines here https://community.amd.com/thread/242812 and post a pic of your results.

 

Thanks!