Zen2's 7nm Complications: Why not All Ryzen 3000 Cores Are Created Equal

Auer

[H]ard|Gawd
Joined
Nov 2, 2018
Messages
1,972
https://www.tomshardware.com/reviews/amd-ryzen-3000-turbo-boost-frequency-analysis,6253.html

The TLDR:
  • Only one core on our Ryzen 5 3600X processor will hit AMD's rated boost frequency. AMD confirmed some cores in Ryzen 3000-series processors are faster than others, which is denoted in Ryzen Master. That means that not all cores on can hit the single-core turbo frequencies. Instead, there are a mix of fast and slow cores.
  • We recorded up to a ~75 - 100MHz difference between the fastest and slowest cores, with only one core reaching the single-core boost frequency. AMD hasn't shared the definition of the frequency that is acceptable for slower cores (it is rational to expect this to be the base frequency). While we recorded relatively small frequency deltas in our sample, it impacts performance and denotes a shift in AMD's binning strategy compared to the per-core turbo performance of Zen/Zen+.
  • Previous-gen Ryzen processors can reach boost frequencies on all cores. Intel also says its processors can reach the turbo frequency on all cores.
  • Workloads running on Ryzen 3000's slower cores experience lower frequencies than the chip's rated boost speed, and thus lower performance.
  • The combination of Windows 10's new Ryzen-aware scheduler and AMD's chipset drivers allow the operating system to schedule single-threaded tasks into the fastest cores (thread pinning). AMD has previously disclosed the Windows 10 scheduler and the CPCC2 feature, but not that the combined features assign threads to the fastest cores. This functionality requires the latest version of Windows 10. This is somewhat similar to Intel's Turbo Boost Max 3.0 on its HEDT processors, but Intel doesn't set this as a requirement to reach the normal Turbo Boost 2.0 clock speeds.
  • Older versions of Windows cannot schedule threads into the fastest cores as efficiently, thus resulting in lower clock frequencies and performance for Ryzen 3000-series processors in some workloads, which may be at the root of many user complaints.
  • Most test utilities do not measure performance fast enough to catch bursty frequency boost activity. They also do not measure certain types of power states that could indicate higher boost activity.
  • We theorize the slower cores could be a contributing factor to low overclock ceilings with Ryzen 3000 processors. Ryzen 3000 series processors hit all-core overclocks 200-300MHz below the single-core boost frequency. Slower cores simply may not be able to achieve/sustain higher frequencies, thus serving as the weakest link in the chain.
 
Could explain the wild variation in performance on different workloads that we have been seeing, where it handily beats Intel at some and just keeps up in others.
 
Well, in reverse, if this was Intel they would have been dragged out behind the shed and shot.

So much this. People would be screaming that Intel is scamming everyone or doing fraud etc etc. But when someone points out the issues they are seeing with actual data on Ryzen CPU's all of a sudden it's picking on AMD?

When did this place become such a fanboy haven and not the want for information on whats really the best and it's limits?
 
So much this. People would be screaming that Intel is scamming everyone or doing fraud etc etc. But when someone points out the issues they are seeing with actual data on Ryzen CPU's all of a sudden it's picking on AMD?

When did this place become such a fanboy haven and not the want for information on whats really the best and it's limits?
Back in the day, going back to when the forum first opened, discussions got very heated any time there was a new release from the big 3. But ever since AMD has become the underdog it does seem that the attitude they can do no wrong has become more prevalent, especially here in the news section. It seems like Reddit is even starting to seep in now that the main site is closed.
 
Upgrade to win 10 1903?

PBO with AIO cooler?

Motherboard firmware is still very immature.. so that could help eventually

On another note, I think sustained boost after 45 min of heavy use is worth looking at more

After all if you have a few VMs in background for several days and load a game I wonder what your average boost will be
 
Last edited:
Not that this is something to cheer, but I do think that the past decade+ of Intel having huge resources compared to AMD yet making questionable decisions that could either be read at best as "resting on their laurels" to "actively exploiting/overcharging/benchmark and other content CPU specific benefits etc", it changes the balance. Contrast this with AMD seemingly being the customer friendly "good guy" more often than not (ie their new open source Linux video driver and general support of open source/Linux , FreeSync etc.. though these are more digs at Nvidia vs Intel - Intel has some Linux support though can be binary blob encumbered for drivers and whatnot) , and people understandably give AMD a little more of a break, especially after they created two massive successes in Zen and Zen+ .

This isn't to say this information isn't a bit disappointing and I hope there is a microcode or process fix (ie perhaps Threadripper Zen2 will be based exclusively upon top binned silicon as in previous generations. Curious what this means for the 3950X though) , but the PR appearance is a little different when it happens to AMD vs if the same thing happened to Intel while they were standing on their piles of money and yelling about how their product is the best.

Edit: Anyone know how Linux's scheduler handles the issue? According to what was noted about only the latest generation of Windows 10 properly pinning tasks to fastest cores etc... I wonder how Linux fares. It would seem that Linux should already have this capability at the kernel level unless this is different somehow.
 
Last edited:
I am still patiently awaiting the delivery of my 3900X, well I say patiently.... cause I mean its not like I have a choice on the matter, I am told they are expecting mine to shop towards the end of August which I supposes gives my EKWB order some time to arrive.
 
This was some good technical coverage, and it is very important to keep-this-level of technical journalism. This is what Tom's used to be WAY back in the day when I used to get the information about the core logics. The holy trinity at the time was Tom's, Ars Technica (before the split to Tech Report), and Anand. Them's were the days......
 
Edit: Anyone know how Linux's scheduler handles the issue? According to what was noted about only the latest generation of Windows 10 properly pinning tasks to fastest cores etc... I wonder how Linux fares. It would seem that Linux should already have this capability at the kernel level unless this is different somehow.
From what I've heard Linux does not have this issue.
https://www.anandtech.com/show/1385...eadripper-2-performance-and-windows-scheduler
 
People arguing over a 2% performance difference. Ir's amazing how bored you all are, now that overclocking is over :rolleyes:

Also, just because Intel leaves enough turbo in on all cores so you can MCE doesn't mean that it makes all that much of a difference. Turning on MCE gets you 2% higher performance than the stock turbos, and then only if you have your machine fully-loaded (usually not gaming).

You're arguing over a tiny difference, more than made up by the availability of HT on Ryzen 2 parts of the same price point.

3600x is faster than he 9600k fully-loaded, even when you turn on MCE.

3700x is faster than the 9700k, fully-loaded, even when you turn on MCE.

This is just anther excuse of an article for someone to whine but the Ryzen 2 because it's not clocked as fast as the Skylake cores. Boo hoo - THE PERFORMANCE RESULTS DON'T LIE.

Clocks are falling with Intel's 10nm process too. Get used to it.
 
Last edited:
So much this. People would be screaming that Intel is scamming everyone or doing fraud etc etc. But when someone points out the issues they are seeing with actual data on Ryzen CPU's all of a sudden it's picking on AMD?

When did this place become such a fanboy haven and not the want for information on whats really the best and it's limits?

Because people love to root for the underdog and hate the big guy. It is silly, but it is how a lot of humans are.
 
Lol, lets see how a few mhz difference turns into ryzen sucks../s
I think the pointy is more that it can't reach the frequency printed on the box. Who care about the brand, any company that prints false numbers (or best case scenario numbers) should be called out for it.

It's pretty obvious they wanted the numbers to look better on paper as that was the big concern for many people compared to Intel (the merit of that concern is another topic).
 
So much this. People would be screaming that Intel is scamming everyone or doing fraud etc etc. But when someone points out the issues they are seeing with actual data on Ryzen CPU's all of a sudden it's picking on AMD?

When did this place become such a fanboy haven and not the want for information on whats really the best and it's limits?

Intel has actually been doing fraud for the last decade. Go disable hyperthreading on your older gen Sandy/Ivy/Haswell CPU's to find out why.
 
Makes sense.

With silicon quality always being uneven across the wafer, and even across the same die, it makes perfect sense that some cores can be pushed further than others.

What is new here that this is actually being used in binning and boosting such that AMD is only boosting the capable core to the max performance.

There is nothing necessarily WRONG with this, and I expect we'll see more and more of it from more vendors as we get closer to the limits of silicon, but coming as it does without any advance notice, it is absolutely not what most users expect when they buy a CPU.

The expectation is that if a boost clock is listed as an official spec, that all cores (not at the same time) are guaranteed to be able to hit that clock clock with the stock cooler at normal room temperatures, and that all cores can do it at the same time given sufficient monstrous cooling and power delivery.

This explains why that is not the case for AMD.

So, they essentially figure out which is the most capable core on the package and use it for loads requiring the highest boost.

I wonder what happens with workloads that have certain core numbers statically assigned.

The Dunia engine used in the Far Cry series - for instance - always appears to pin the third core. What if the most capable boost core doesn't line up with this statically assigned third core? Can AMD reassign workloads internally? Or do you just get worse than advertised performance?

If AMD was going to do this they should have given their customers a heads up so they knew what they were buying.

These really should have been sold as 4.1ghz boost CPU's.

It's a little bit disingenuous of them to launch these CPU's without letting customers know that they have fundamentally changed how boosting works in a negative way.
 
Last edited:
  • Like
Reactions: Meeho
like this
I think the pointy is more that it can't reach the frequency printed on the box. Who care about the brand, any company that prints false numbers (or best case scenario numbers) should be called out for it.

It's pretty obvious they wanted the numbers to look better on paper as that was the big concern for many people compared to Intel (the merit of that concern is another topic).
it can......just not on every core and it needs the correct chipset drivers and latest win 10 build to operate correctly! big deal lol...everbody already knew how these things reviewed
 
The article is misleading and really not 100% accurate for 3600X.I have a 3600X which can hit 4525Mhz on all cores and up to 3CCX or 6 cores same time.
Hell I did a short single thread run and 3600X Hitting 4500Mhz Cinebench 20 and over 4400Mhz Max boost on 3CCX in a minute video.
 
The article is misleading and really not 100% accurate for 3600X.I have a 3600X which can hit 4525Mhz on all cores and up to 3CCX or 6 cores same time.
Hell I did a short single thread run and 3600X Hitting 4500Mhz Cinebench 20 and over 4400Mhz Max boost on 3CCX in a minute video.


Yours can, but theyre not all the same.
 
Yours can, but theyre not all the same.
This may be true not all the same but I have a sample size of 2 x 3600X and both are capable of these clocks.BIOS are the real problem.
5 CCX at 4400Mhz
1 CCX at 4450Mhz and all cores boosted past Rated max boost printed on the box.Seems misleading to me,from personal experience.
  • Only one core on our Ryzen 5 3600X processor will hit AMD's rated boost frequency. AMD confirmed some cores in Ryzen 3000-series processors are faster than others, which is denoted in Ryzen Master. That means that not all cores on can hit the single-core turbo frequencies. Instead, there are a mix of fast and slow cores.
Desktop-Screenshot-2019-07-11-15-49-28-90-2.png
 
  • Like
Reactions: N4CR
like this
I agree with gerard that there are currently some bios issues that need working out. As I've said before, my 3700x runs ALL cores at 4275-4300mhz while gaming with PBO all maxed out +200, but zero of my cores will reach the rated 4400, or even 4350. I get a very rare (once a day?) boost to 4325 on one of my cores.

I've clean installed windows, defaulted bios, installed 1.0.0.3AB updated bios, installed todays new AMD chipset drivers, tried PBO off, on, tried every power plan, you name it. I don't believe that my chip is capable of doing 4300 on every single core but zero of the cores can go above 4325, and worse yet the box says 4.4ghz.

That said, there are clearly variations, always is, but the CPUs should at least be able to boost to what the box says as a baseline. Mine should say 4.3ghz if that's all it's ever going to reach. On the bright side, this CPU absolutely annihilates my old 4790k, so I'm still pretty happy.
 
Toms wins the title for bullshit titles. Complications? AMD frickin says right in ryzen master that best cores highlighted by a star.

The problem is not the chips the problem actually lies in agesa 1.0.0.3AB, with stock bios on my crosshair viii hero wifi I had better boost clocks. But it seems after updating the bios it never really picks the best core. If you see in ryzen master running one core benchmark it never runs the benchmark on the best rated core in CCX0. Hence my boost now gets limited to around 4300 on single core from 4600.

But here is the kicker, my 3900x is a beast. It does all core OC at 4400mhz with 1.328v under load.

Before the update my all core boost was around 4300 from the quick tests I ran but that came down to 4100+ ish. I did ask asus support to make the out of box bios available or email it to me. They promised support ticket is created and engineering will honor my request and email me the bios. But hey nothing came out of that. I must say asus has gone down hill, they have made it easy to reach customer support but man no one fuckin follows up it seems. My last asus board for me this gen.
 
I agree with gerard that there are currently some bios issues that need working out. As I've said before, my 3700x runs ALL cores at 4275-4300mhz while gaming with PBO all maxed out +200, but zero of my cores will reach the rated 4400, or even 4350. I get a very rare (once a day?) boost to 4325 on one of my cores.

I've clean installed windows, defaulted bios, installed 1.0.0.3AB updated bios, installed todays new AMD chipset drivers, tried PBO off, on, tried every power plan, you name it. I don't believe that my chip is capable of doing 4300 on every single core but zero of the cores can go above 4325, and worse yet the box says 4.4ghz.

That said, there are clearly variations, always is, but the CPUs should at least be able to boost to what the box says as a baseline. Mine should say 4.3ghz if that's all it's ever going to reach. On the bright side, this CPU absolutely annihilates my old 4790k, so I'm still pretty happy.

Did you test it out with older agesa? I have the same issue. Out of the box I actually had way better boost clocks, but after updating to 1.0.0.3AB I lost couple hundred off boost clock single and all core. Raising power limits does nothing or the +200. It almost feels like bios is messing up since out of the box I saw way better boost clocks. Plus my 3900x runs 4400mhz OC all cores at 1.328v, tested and tortured lol. So not even single core boost goes to 4400, hahaha. Something is clearly messed up with the newer bios. Plus sinlge core voltage is around 1.4v+ so I have no doubt I should be able to hit 4.6 on atleast one core. I did have 4.3ghz all core boost with original bios. THis went down to 4150ish and single core boost is the same as yours. 4300 and see some spikes to 4325.

Also I noticed in ryzen master that single core load is not assigned to best core, may be that is the issue with the newer agesa. Because it should be selecting the best single core labeled in ryzen master when running single core load and off load windows or other workloads to different cores.
 
Sounds like it will hopefully get better as they tweak things. Also need a way to set CPU affinity or at least preference to use the 'best' core.
 
Hallock having to put out such a detailed Community Update makes me wonder why they didn't just consider pushing the launch back a full month.
 
  • Like
Reactions: Auer
like this
Hmm, yeah, count me in as another one who's not getting the rated speed anywhere. Ryzen Master reports no cores ever going above 4.225; cpu-z shows the best one going up to 4.275 and briefly hitting 4.300 sometimes; the next best core in the same CCX gets 4.25. All core during a stress test is 4.100. Don't know if I got a bad one or if it's the motherboard/BIOS (I have the latest BIOS, which was dated 7/5; I didn't check the speeds before updating the BIOS.)

I'd be really tempted to take this back if I were still in the 15-day return period. I DO have a replacement plan so I could probably still do it but my motherboard's also outside the window and I didn't get a plan on that.
 
Hmm, yeah, count me in as another one who's not getting the rated speed anywhere. Ryzen Master reports no cores ever going above 4.225; cpu-z shows the best one going up to 4.275 and briefly hitting 4.300 sometimes; the next best core in the same CCX gets 4.25. All core during a stress test is 4.100. Don't know if I got a bad one or if it's the motherboard/BIOS (I have the latest BIOS, which was dated 7/5; I didn't check the speeds before updating the BIOS.)

I'd be really tempted to take this back if I were still in the 15-day return period. I DO have a replacement plan so I could probably still do it but my motherboard's also outside the window and I didn't get a plan on that.

Its agesa issue. They will eventually get ironed out and on the same page soon. I had reduced boost clocks after I updated.

Also what is your ram speed? Set it right below 3600 as well. Setting mine to 3533 increased my boost speed by hundred or so, there was a thread on Reddit. I had the original bios that boosted to 4.6, but after the new agesa that mobo makers put out on 7/5. I got capped at 4.3 for single core and 4150 for all core. Before it was 4.5+ all the time for single core and 4250-4300 all core.

But the next agesa 1.0.0.3ABB is suppose to have more permanent fixes.
 
That's an interesting point about the RAM speed. 66MHz on DDR4 for 100Mhz more core? That's a fair trade.
 
Did you test it out with older agesa? I have the same issue. Out of the box I actually had way better boost clocks, but after updating to 1.0.0.3AB I lost couple hundred off boost clock single and all core. Raising power limits does nothing or the +200. It almost feels like bios is messing up since out of the box I saw way better boost clocks. Plus my 3900x runs 4400mhz OC all cores at 1.328v, tested and tortured lol. So not even single core boost goes to 4400, hahaha. Something is clearly messed up with the newer bios. Plus sinlge core voltage is around 1.4v+ so I have no doubt I should be able to hit 4.6 on atleast one core. I did have 4.3ghz all core boost with original bios. THis went down to 4150ish and single core boost is the same as yours. 4300 and see some spikes to 4325.

Also I noticed in ryzen master that single core load is not assigned to best core, may be that is the issue with the newer agesa. Because it should be selecting the best single core labeled in ryzen master when running single core load and off load windows or other workloads to different cores.

Honestly I did not try the bios it shipped with (aorus pro wifi x570, bios F3). I upgraded it to F4f using 1.0.0.3AB right away. Gigabyte rep on another forum said AMD has a new AGESA already in their hands for a new bios update I should see in a few days...

Yeah something is clearly up with your system too.. and damn @ 4.4ghz @ 1.328... I can't get 4.3ghz at that voltage or close to it to survive cinebench r20 on 3700x, though my RAM is heavily oced and tightened and I didn't bother trying to default it for CPU overclocking..

C 05 is marked as my best core, so far in testing games it's never the one that's getting hit with big load. Far Cry 5 seems to nail 3 ( best core on ccx 0) and occasionally c07 (second fastest on ccx 1).

Seriously though, try switching to AMD Power Saving Plan and see if your all core jumps like mine.. I get an extra 50mhz that way for some weird reason, even on the new chipset driver.
 
Its agesa issue. They will eventually get ironed out and on the same page soon. I had reduced boost clocks after I updated.

I hope so--like I said I never checked the out-of-box bios, and of course it's not available on Asus' website.

Also what is your ram speed?

It's only 3000MHz.

But the next agesa 1.0.0.3ABB is suppose to have more permanent fixes.

I hope so. I do see that Asus has a newer bios for my older mobo, an x470, although the notes don't say what AGESA level it is.
 
For anyone interested-some of the chipset drivers do not have AMD Ryzen™ Power Saver.How to in spoiler
AMD_Chipset_Drivers_v1.07.26.0551-Beta Drivers -July 25 2019 With AMD Ryzen™ Power Saver Plan which does not come with this chip set ,so here is a copy for anyone who wants this power plan.

Google drive AMD Ryzen™ Power Saver Plan and How to import power plan from desktop.

Open Command Prompt as(Admin)- powercfg -import “Full path of .pow file”
So from desktop command will look like - powercfg -import "C:\Users\YourName\AMD Ryzen™ Power Saver.pow"


https://drive.google.com/file/d/1nXzg4CgLUoA-141QiQ0TdKL3WZ01putS/view?usp=sharing
 
Huh. I just checked and I'm using AMD Ryzen Balanced. Well, maybe there's my problem? I'll try the other Ryzen plans too.

Just an anecdote from my army of 1 3700x, but I get 4,300 mhz all core on AMD Power Saver. If I change to AMD high perf or Balanced, I watch all cores drop right down to 4,242 and they never go higher. Note that these are peak all core clocks, under stress they still drop, but I can play games like Far Cry @ 4300mhz all night.

I don't know if this has happened to anybody else or if anybody has noticed besides me at this point..
 
Upgrade to win 10 1903?

PBO with AIO cooler?

Motherboard firmware is still very immature.. so that could help eventually

On another note, I think sustained boost after 45 min of heavy use is worth looking at more

After all if you have a few VMs in background for several days and load a game I wonder what your average boost will be

PBO isn't the answer. I and other reviewers have often found it does nothing at all, or does very little. Your really better off using a good AIO and PB2 than using PBO.
 
I guess this explains why the same core seems to always be used for single threaded activity. It much know that one is the best and can clock the highest.
 
I guess this explains why the same core seems to always be used for single threaded activity. It much know that one is the best and can clock the highest.

It does. Ryzen Master can show you which cores are the better ones. They are marked with gold starts. Silver stars, then dots mark the next ones in order. No dot, no star = worst cores.
 
Ok, none of the other plans did anything meaningful. Well, it's hard to tell. Ryzen Master never shows any core going over 4.225. Switching to CPU-Z, and switching the power plans, my best core sometimes goes to 4.275 and I can get 1-3 more to hit 4.250, but that's hardly any better.

I'm using the CPU-Z "Bench CPU" button on the Bench tab. Also used Cinebench R15--all-core stress tests gives me mostly an all-core 4.100, with some cores bouncing between that and 4.075. Temps approach 80C. With a 360mm Corsair H150. I still need, I guess, to check how the TIM spread itself, because in another thread someone mentioned checking how solidly that's attached. (BTW, that's a pretty nice AIO, but the AM4 mounting method is shit. You bolt two square metal pieces with holes in them to the pump; they have holes in them that fit over the horns on the stock AM4 mounting thing. Just horrible to try to mount if your motherboard's already in a case.
 
Back
Top