News Survey: Only 5.6 Percent of Ryzen 9 3900X Hit Advertised Speeds, Most Other Models Suffer, Too

[tom1111]

My 9900K does 4800MHz on singlecore cinebench.

 

[germanium]

I have a Core I9 9900K running at 5GHz on all cores under liquid cooling. It may be possible to do with an air cooler such as Nochuas best cooler. I increase the allowed power dissipation to 245 watts for up to 127 seconds. Beyond that it will downclock.

My cooler is rated for 350 watts so no problem. For the most part I run without hyperthreading unless I find something that truly benefits from it. It causes huge increase in power far beyond what you would expect. Without hyperthreading power stays less than 180 watts but with hyperthreading can shoot up to 245 watts. Without hyperthreading it never downclock at all.

This is something that cannot be done with Ryzen processors at all. What good is an unlocked CPU that can't be overclocked even in the slightest even if it only meant being getting an all core turbo boost such as what I get. Granted Ryzen does have very good competitive performance especially in the IPC area as well as multithreading. But I find advertising them to be unlocked overclockable chips as disingenuous & false advertising when often they can't even reach stock boost speeds on one core let alone all cores.

By the way mine does Cinebench on all cores at 5GHz.

 

[bit_user]

And a ST run of Cinebench doesn't make your computer only use 1 thread, Windows needs threads too!

No. If your PC is otherwise idle, then you should see task manager show all-core CPU utilization at almost exactly 1/n, where n is the number of hyper threads supported by your CPU. Sure, there's the occasional background activity, in Windows, but the rest of the threads aren't active enough to make a difference.

Cinebench is a compute benchmark, meaning it spends basically all of its time doing userspace compute. In I/O benchmarks on fast SSDs, you'll see a lot of time spent in the kernel.

 

[bit_user]

Have they accounted for room temperatures and cooling solutions used? I guess you can't really expect the CPU to boost fully at 35+ degrees centigrade ambient and using the stock cooler.

Why do you think they did this in August?
; )

(Yeah, I know, because the CPU was launched in July. But, it's still a valid point that it is a warm month, where most Ryzen users live, and there's no control for ambient temps.)

 

[bit_user]

For the most part I run without hyperthreading unless I find something that truly benefits from it. It causes huge increase in power far beyond what you would expect. Without hyperthreading power stays less than 180 watts but with hyperthreading can shoot up to 245 watts. Without hyperthreading it never downclock at all.

Why is that so unexpected?

How do your multithreaded Cinebench (or other benchmark) scores compare with/without hyperthreading? With hyperthreading, the CPU is doing more work. It can draw from up to 2 instruction streams to keep all of its execution units busy. Without hyperthreading, more of the chip is idling due to cache misses and serial dependencies within the code. That's the point of it.

As long as the amount of additional performance provided by hyperthreading is roughly in line with the additional power, then using it is a good move. However, if you don't run highly-threaded workloads that exceed the capabilities of the CPU without HT, then go ahead and leave it off.

What good is an unlocked CPU that can't be overclocked even in the slightest even if it only meant being getting an all core turbo boost such as what I get.

The main benefit is that it's a higher-binned chip, so the achievable clocks are better than cheaper models. Unlocking is just a bonus, if you get lucky.

Part of the problem is that AMD's highest-binned chiplets are not going into the current high-end model. They're probably being saved for the 3950X and the upcoming ThreadRipper.

 

[jdwii]

I've built many Ryzen 3000 systems and all of them will hit the boost clock, while under low utilization, confirmed by a look at a recorded HWInfo max clock speed reading. Doesn't mean you'll get the boost clock under a full single threaded Cinebench run. Why is that conveniently the definition now?

My 8086k doesn't hit 5ghz under Cinebench ST runs either. Same story. The 9900k hits it because unlike the 8086k the boost is for 2 cores, not 1. And the 9900k will throttle itself under full load also, and not hit its all core boost speed either. A little perspective is necessary.

This is like only being mad at LG for 1ms lies/misleading claims, but not being against Asus or TN panel makers doing the same thing.

My 3700X will not hit 4.4ghz on a single core load with even a h150i

 

[jdwii]

Some people might try that, but I think the most likely outcome is for AMD to issue a statement characterizing the workloads that are used for Max Boost testing.


Well, you should also ask why Intel isn't being sued for several generations of their CPUs not hitting tubo boost speeds on AVX2 workloads. Relatively few people actually know about the reduced clocks for AVX2, yet there seems to be acceptance of that policy.

That further underscores that this might really be a matter of customer education, on AMD's part.

I tried every workload i possibly can the main reason why i buy newer chips is to test and compare them for fun. It took a lot of tweaking to rarely get my chip to hit 4.4ghz and it required PBO on with lower values then what is at default(yes lower then stock)

Running on a AIO 360 rad in a R6 case filled with Noctua fans.

This isn't good for Amd this can lead to more law suites. They shouldn't have said what the max turbo is if it never reaches that at stock. Intel does not do this and if they did i know lawyers would be happy.

 

[bit_user]

I tried every workload i possibly can the main reason why i buy newer chips is to test and compare them for fun.

So, you've gone beyond stress tests and tried some simple things - like an infinite loop in powershell?

I'm not saying you didn't get a bad CPU, but just checking whether there's really no way to see it reach the specified boost.

This isn't good for Amd this can lead to more law suites. They shouldn't have said what the max turbo is if it never reaches that at stock. Intel does not do this and if they did i know lawyers would be happy.

Huh?

It says right here:

Max Turbo Frequency: 5.00 GHz​

As others have said, many i9-9900K do not reach that on the Cinebench test.

 

[jdwii]

So, you've gone beyond stress tests and tried some simple things - like an infinite loop in powershell?

I'm not saying you didn't get a bad CPU, but just checking whether there's really no way to see it reach the specified boost.


Huh?

It says right here:

Max Turbo Frequency: 5.00 GHz​

As others have said, many i9-9900K do not reach that on the Cinebench test.

My CPU isn't bad at all in fact in AGESA 1.0.0.2 it worked fine except for memory support(which is perfect BTW on 1.0.03ABB)

I created a quick program that keeps trying to find prime numbers in python and that also barely hits 4375mhz typically i see 4300-4325mhz sustained on one core.


I used hwinfo64 at the polling rate of 500ms. I also tried Cinebench R15 and R20(takes forever) on just one core as well. I tried Dolphin Emulator on just one core and finally i played older games that use only one core(SimCity 2000).

Not sure what else i could do the temps are completely inline with what we are supposed to see in fact i typically hang around 55C or so in single core loads(this is very normal for Ryzen 3000).

Something to add in i'm currently using the latest bios, chipset driver and Windows 10 1903 with all the latest updates and i noticed that Ryzen Master claims core 4 is the best but in the task manager when single core loads hit windows puts it on core 2(really called core 1 in windows)

I never see core 4 get hit primary on single core loads isn't that what its supposed to do?

The last option that i thought about doing was redoing windows completely but i do notice that threads are scheduled on just one CCX if the program uses 4 cores or less so maybe its working somewhat right?

Amd should possibly try and look into this my last option which i did NOT try yet is installing Ubuntu or something and trying that.

Again in reality no one is going to notice a 100mhz difference except in benchmarks but that's not the point and this can be damaging to Amd's reputation and i'm sure lawyers will love this once again. I'm actually somewhat of a Amd fanboy lately with Ryzen so i have no idea why Amd did this unless they rushed the launch and later found out that if they kept the turbo at those levels the chips would degrade themselves.


Edit i guess i can try and force a program to use core 4 specially and see if it hits 4.4ghz this could be windows related as well

 

[jimmysmitty]

Do folks always forget that the first runs of new chips usually don't perform the best. Give the new process another 3 months and the boosts will hit far better I bet.

I never buy new chips for this reason. Always wait a while for the silicon to mature.

Not always. Intels 45nm chips came out and clocked well and overclocked very well.

I do think part of the issue is the process and that it's just not designed for big x86 cores.

Have they accounted for room temperatures and cooling solutions used? I guess you can't really expect the CPU to boost fully at 35+ degrees centigrade ambient and using the stock cooler.

Advertised boost speeds should be the max speed on a stock cooler if a stock cooler is included. Basically if AMD says max boost is 4.3GHz and includes a stock cooler that should be what it hits with said stock cooler. Same for Intel if they ever went back to including one.

The interesting thing is a lot of these people probably have better than the Wraith cooler for their setup.

Is this really how Tom's Hardware gets their 'News' stories now? Some first class nonsense... Watch a five minute Youtube video and write an article...This site hits a new all time low with every passing day...

They actually did testing of their own for this very story.

And not sure why you think one of the top overclockers in the world makes it invalid. His experience makes him a good source for this very topic.

 

[bigpinkdragon286]

So, with a software polling rate of 500 ms, people are expecting to get a clear view of what the clocks inside of a modern CPU are doing?

On the surface, I would expect boosting on the Ryzen 3000s to be similar to how other boosting algorithms have been demonstrated to work, but that's as much because I have no clear understanding of the intended behavior of the new chip boost algorithm yet (too distracted to read up on it.)

I would suspect, before one says with certainty the chips aren't rapidly clock cycling or prioritizing something else, such as power usage, you should measure with something a bit more fine grained than a software sampler running in a huge (for a CPU) time window of 500+ ms. Heck run HWMonitor and HWiNFO side by side and in a great many areas you'll see similar but different values, just due to when they are sampling. It's entirely possible the chips are boosting and un-boosting faster than the sampling rate and not maintaining the boost long enough to be sampled at it, or the systems are not actually configured to boost using the correct core.

 

[bit_user]

I created a quick program that keeps trying to find prime numbers in python and that also barely hits 4375mhz typically i see 4300-4325mhz sustained on one core.

That should be pretty good.

Computing a fractal is also very serial and iterative, if you implement it naively.

I used hwinfo64 at the polling rate of 500ms. I also tried Cinebench R15 and R20(takes forever) on just one core as well. I tried Dolphin Emulator on just one core and finally i played older games that use only one core(SimCity 2000).

If you want to tease out the highest clocks, then don't use stress loads. What you want is some relatively simple, serial, and scalar computation - like the python examples, above.

I never see core 4 get hit primary on single core loads isn't that what its supposed to do?

Yeah, it's a good point.

Amd should possibly try and look into this my last option which i did NOT try yet is installing Ubuntu or something and trying that.

Well... that probably won't have their latest patches. For the latest, use a rolling distro, like OpenSUSE Tumbleweed, Debian Unstable, or such. Here's a more complete list, but I cannot vouch for it - it's just one of the first search hits I got:

https://www.slant.co/topics/6289/~rolling-release-linux-distributions

Edit i guess i can try and force a program to use core 4 specially and see if it hits 4.4ghz

That's also a good idea.

 

[paul prochnow]

The Ryzen 3000 boost clock controversy deepens as a survey points out that only 5.6% of Ryzen 9 3900X are hitting their rated boost clocks, among other Ryzen CPUs that aren't matching the stated frequencies.

Survey: Only 5.6 Percent of Ryzen 9 3900X Hit Advertised Speeds, Most Other Models Suffer, Too : Read more

MSI X470 I seen 4600Mhz on rather a few cores in the measures I have like CPUZ HWMON....is that what you all are talking about?
Is 4.6GHz the turbo boost?

 

[bit_user]

So, with a software polling rate of 500 ms, people are expecting to get a clear view of what the clocks inside of a modern CPU are doing?

Well, under the assumption that turbo should boost for much longer than that. Except if turbo is mitigated by a high-stress load with lots of parallelism and vector arithmetic.

 

[jdwii]

So, with a software polling rate of 500 ms, people are expecting to get a clear view of what the clocks inside of a modern CPU are doing?

On the surface, I would expect boosting on the Ryzen 3000s to be similar to how other boosting algorithms have been demonstrated to work, but that's as much because I have no clear understanding of the intended behavior of the new chip boost algorithm yet (too distracted to read up on it.)

I would suspect, before one says with certainty the chips aren't rapidly clock cycling or prioritizing something else, such as power usage, you should measure with something a bit more fine grained than a software sampler running in a huge (for a CPU) time window of 500+ ms. Heck run HWMonitor and HWiNFO side by side and in a great many areas you'll see similar but different values, just due to when they are sampling. It's entirely possible the chips are boosting and un-boosting faster than the sampling rate and not maintaining the boost long enough to be sampled at it, or the systems are not actually configured to boost using the correct core.

If you watch the video(3:25 min into it) that is what Amd recommends during the survey to use 500ms as the polling rate which is why i used it.

Keep in mind the frequency can change every 2ms or so according to Amd

Edit

What is crazy is Ryzen Master actually makes Ryzen 3000 look worse and according to that program the peak frequency it hits is 4285 or something while hwinfo hits much more believable results

 

[germanium]

Why is that so unexpected?

Because the power draw is quite out of proportion compared to the improvement in performance as in much greater.l

 

[hannibal]

Yeah! You just need that chilled water cooler that keeps 3000 Ryzen at 0-10 celsius or lower
The term optimal situations is so vague...
I hope that marketing departments stop from giving hazy information. The cpu itself is super good, it just was marketeissa so that situation seems to be bad... and from marketing Sandpoint it is! If They would not have promised any boost speeds. There would be no any problem! And someone with stock cooler would get the basic + very Little. Someone with Nochtua 15 someone better. Someone with custom watercooler maybe more and someone with very big chiller something like They now promised. Everybody would be happy. You get something extra you cool more. Now you need in bad case that extreme cooling to get what some stupid pr Person says and people Are upset.

 

[closs.sebastien]

About the subject of the article: isn't it the fault of the standard heatspreader? does someone has a much bigger one and reach the boost without any problem?

 

[TJ Hooker]

This is during a Cinebench run. The boost isn't for a Cinebench run, that is obviously not how it is defined.... 🙄

As people have said, Intel's turbo boost doesn't apply for AVX, so you never see it with Intel either. And a ST run of Cinebench doesn't make your computer only use 1 thread, Windows needs threads too! Thus my 8086k NEVER runs at 5ghz. Marketing nonsense. It might run at 5ghz when doing nothing except clicking on something, just like with Ryzen. Intel lies for years no-one makes it an issue, Ryzen doesn't boost too, now it is an issue, the main reason being the boost speeds for Ryzen are higher than the all core overclock, while for Intel, the boost speeds are usually lower than you can hit with an all core overclock, making it so sometimes there is an advantage to relying on boost instead of setting your own overclock, for Ryzen.

According to Der8auer, AMD themselves recommended CB R15 1T to measure single core boosts. If this is "obviously not how it is defined", then how exactly is it defined? And where does this definition come from?

You mention you see your Ryzen 3000s hit advertised boost clocks under low utilization, but that seems like a catch 22 to me: if you're not loading the CPU then you don't really need it to boost in the first place.

 

[jdwii]

About the subject of the article: isn't it the fault of the standard heatspreader? does someone has a much bigger one and reach the boost without any problem?

I tried 3 different times to apply thermal compound and this time around 2 pea size amounts was ideal for my 3700X one if the middle and one around one chiplet.

However i never seen any big difference probably 3-5C at best.

Right now i turned off SMT to see if it helps allow my 3700X go to 4.4ghz under load.

This time around i ran each individual core for about 15min in my prime program and it required around 40% of one core. I did not notice any difference then letting W10 simply pick the core to run on.

SMT off i actually saw a 25mhz decrease in frequency in terms of boost which is so weird i contacted Asrock about the behavior something is still wrong.

SMT off
https://postimg.cc/bZD2D7V1

SMT on
https://postimg.cc/BjrPxmRk

 

[jdwii]

Great news to all this is why you keep complaining when their is issues Amd is such a great company!

https://pbs.twimg.com/media/EDjF1zmXYAABTYU?format=jpg&name=medium

Gonna get fixed in a few days

 

[NightHawkRMX]

Were gettings somewhere.

I wander how log it will be and what the change will actually improve.

 

[kinggremlin]

Actually those two cases have nothing in common. The first case was real, you could verify that the 500mb of ram was different from the rest. The second case is based on one flawed survey then parroted by this hack of a website that is circling the drain.

Would you care to explain why AMD has just announced they are releasing a BIOS update to fix a problem you claim isn't real? Surely AMD has better things to do than fix nonexistant problems.

 

[jdwii]

Would you care to explain why AMD has just announced they are releasing a BIOS update to fix a problem you claim isn't real? Surely AMD has better things to do than fix nonexistant problems.

Fanboys man these are the same people who defended the FX CPU's(the crappy ones not the good ones from back in the day)

SO HAPPY Amd is nothing like their fanbase

 

[NightHawkRMX]

Im not really sure how AMD will fix this issue. These chips rarily can hit rated boost on 1 core even when overclocked.