Ryzen 3950X disappointment

Only 4.3 GHz all core WATER REQUIRED


Why not 5.0 GHz All Core from factory for a change?
not-sure-if-i-used-this-meme-correctly_o_2101849.jpg
 
To me the 4.4ghz all core they were able to overclock all 16 cores to is actually pretty good. The more cores you have the less likely you are to get them all to clock well.

With the 3950X being supposedly higher binned chips, I would hope the advertized 4.7ghz boost speeds may actually be possible.

Though even 4.5-4.6ghz sustained boost would be really nice.

Also, with 16 cores to choose from, you may actually find 2, 4, wishfully maybe 6 of them that are actually capable of hitting 4.7ghz locked with an OC.

I can't remeber if Ryzen Master allows for per core specific individual OCs, but if so that would allow you to find those golden cores and have them at 4.7ghz locked and have the worse performing cores at 4.4ghz or wherever they are stable.

If even a few cores are actually capable of 4.7ghz, I'd love to see if they could be pushed to 4.8 or further if possible (under 1.5 vcore). Such single core OCs could really improve those single threaded scenarios where ryzen has historically struggled.
 
Last edited:
The idea that you think that the 3950x would reach 5Ghz I'd retarded. How did you expect a 16 core to manage 5 GHz when all the lower core CPUs can't even come close under LN overclocking?
 
Based on a mix of funny expectations, 64 core running at all-core 5 GHz would actually start to cool air around it. A Beowulf cluster of such CPUs could cool off the planet and slow down global warming until they start cramming in more cores with 4 or even 8 way SMT and throw off the equation back into regular physics.
 
what's funny is that zen2 3800x matches coffee's 5ghz allcore 9900K in most cases, so you're getting 5ghz allcore intel performance from these and in games within short distance.
appearantly the gap is gonna become somewhat smaller soon with new microcode (No miracles!) but it's already something you can't notice.
 
In full SMP tasks, the 3900X is still kicking ass and taking names, especially for the price.
Adding 4 more cores will only help, assuming memory bandwidth being split to 4 more cores doesn't hamper certain workloads.
 
  • Like
Reactions: erek
like this
Looks like you have been living under a rock. No disrespect, but where were you when 3900x came out? Did you really expect a 16 core part to OC to 5ghz when 12 core part couldn't? ROFL.

he's too busy living in the past with his retro stuff that he forgot it's October 2019 and we already know zen 2 can't do 5Ghz..
 
16c 32t @ 4GHz for $749 part. Not bad, not bad at all - - except the price. I know price 'makes' sense, but i think it should be $600 ($599);
Zen2 should be cheaper per core basis, and it should reflect on our wallet too, not only amd's.

// Note: intel doesn't have anything within performance, and price range of this desktop part.
 
16c 32t @ 4GHz for $749 part. Not bad, not bad at all - - except the price. I know price 'makes' sense, but i think it should be $600 ($599);
Zen2 should be cheaper per core basis, and it should reflect on our wallet too, not only amd's.

// Note: intel doesn't have anything within performance, and price range of this desktop part.

I'm cool with paying AMD more (as its still far less than Intel.) Because that gives them more R&D funds to make better products in the future.
 
Isn't it not a good idea to do all core overclocks on these anyway? Just let it do its' own boosting, and you get 4.5ish on a core or 2 for the busiest threads.

And that's what you get for watching AdoredTV, that guy is just an AMD shill.
 
i do all core 4.3 instead of precision boost. Precision boost on the 3900x still doesn't work correctly even after the last bios updates. I still see it adhering to the tdp in many cases when the system is not throttled by temps or power supply. So I'd rather run it with all cores at 4.3ghz since I know my system can handle it than get intermittent and random clocks that sometimes only hit anywhere between 4.0 and 4.3Ghz all core. Single core concerns are BS on ryzen. Your single core process has a random chance of being allocated to the golden cores that can actually reach 4.5+ ghz and even when they are assigned to them, the difference in overall performance between 4.3 and 4.5-4.7 ghz is not going to be significant, especially when it's not going to peg that speed for a significant amount of time, nor will the process scheduler keep the process on that cpu indefinitely without manually assigning it.

OS schedulers need to be coded to be aware of these heterogeneous cores in linux and windows so that they can keep them available and pegged to high cpu intensive tasks. Until then, they're opportunistic and randomly assigned and all core reliability in clock speed becomes far more beneficial than the extra 200 mhz single core potential performance by far.
 
Last edited:
Isn't it not a good idea to do all core overclocks on these anyway? Just let it do its' own boosting, and you get 4.5ish on a core or 2 for the busiest threads.

And that's what you get for watching AdoredTV, that guy is just an AMD shill.
Anyone remember when he created an account here to respond to criticism because a news thread about one of his Youtube videos got a lot of traction? I 'member. That was quite entertaining.

https://hardforum.com/members/adoredtv.310714/
 
Only 4.3 GHz all core WATER REQUIRED


Why not 5.0 GHz All Core from factory for a change?

Some thought would be required to see why that is a foolish notion. You do remember the Bulldozer days wheremore than a very few enthusiasts were able to crank up 5.0 GHZ clock speeds. AMD is already slightly ahead of Intel on IPC . My Ryzen 9 3900X is within 2 to 3% of the I9 9900K in performance in gaming and kicks its ass in productivity , streaming, and ALL other tasks. That is at 4.3 GHZ so why do I need 5GHZ, so you can satisfy your roid rage???
PLEASE grow up.
 
And doing so isn't even worth while outside of very specific benchmarks or use cases. I've done this type of testing and you are better off leaving it alone and letting Precision Boost 2 do its thing. If you set all the cores to 4.3GHz, you lose single-threaded performance which hurts you more than you might think. These CPU's still achieve 4.0GHz+ all core, so 4.3GHz isn't much of a boost outside of specific scenarios most desktop users will never benefit from.

Actually overclocking CPU's to 4.3GHz if they can do it is quite easy. It's only two or three settings that you have to manipulate to make this happen in most cases.
I do it manually without a problem. Precision Boost is BS. All that fiddling to achieve what a manual overlcok can do easier and better.
 
I'm cool with paying AMD more (as its still far less than Intel.) Because that gives them more R&D funds to make better products in the future.
Your logic is sound, but they are charging us a lot more on those just because intel doesn't have anything better. Intel was doing this since inception, but doesn't mean we cannot hold our Ayymd to better standards. I only hope the shintel fans shilled out their pentium dollars on zen2 instead.
(I assume they are the ones crying here, now that it doesn't hit 5GHz)
 
I do it manually without a problem. Precision Boost is BS. All that fiddling to achieve what a manual overlcok can do easier and better.

Except, the data shows otherwise. If you manually lock the CPU's clocks to 4.3GHz or lower, you will NEVER see boost clocks over 4.3GHz. That means, those one or two cores that can hit 4.7GHz or close to it never will. That's a fair bit of single-threaded performance to give up in those cases, and the numbers back that up. There are some heavily multi-threaded applications where the manual overclock will work better. I've said as much, but for general performance, that's not the case.
 
Except, the data shows otherwise. If you manually lock the CPU's clocks to 4.3GHz or lower, you will NEVER see boost clocks over 4.3GHz. That means, those one or two cores that can hit 4.7GHz or close to it never will. That's a fair bit of single-threaded performance to give up in those cases, and the numbers back that up. There are some heavily multi-threaded applications where the manual overclock will work better. I've said as much, but for general performance, that's not the case.

you have 1/12 (1/8 for the 3950 if 2 cores out of 16 are golden) chance to momentarily get scheduled on the cores that can do > 4.3Ghz. If the process takes a significant amount of time (something that would make a frequency increase of 10% noticable) then it's very likely to get preempted off that core during the course of it's runtime, and get put on a core that can't hit 4.7.


So no, I do not buy the argument that you are hurting single core performance to any statistically measurable amount outside of controlled benchmarking situations. That goes for windows and linux (but especially linux since linux is much more likely to be configured to be highly preemptable than the much more voluntary options - which windows tends to behave more like)

edit: and just to be clear, you dont need special highly scalable multi-threaded programs to cause the cpu to limit access to those > 4.3Ghz frequencies. If you run a virus scanner along with an application it can frequently peak the cpu it's running on and trigger the golden cores to drop to multi-core speeds as well as any number of other background tasks that can wake up or that you happen to be running on purpose. You dont need to load all the cores to eliminate those single core frequencies, further reducing the likelihood of them contributing any benefit to real world use.
 
Last edited:
Didn't AMD recently specifically say they were updating the scheduler to help with that? And there is the upcoming microcode updates. I could swear they said their scheduling was getting an update in Windows to more intelligently use the fast cores.
 
Come to think of it.. I'm curious how bclk overclocking works with AMD's boost technologies. I wonder if that would work with their dynamic overclocking methods to get you an extra 100mHz or so. Perhaps it would still boost to the same point or worst case it would just cause instability.

Perhaps BCLK overclocking would give some of the benefits of all core overclocks while preserving the advantages of the single core boosts of AMD's boost tech
 
Come to think of it.. I'm curious how bclk overclocking works with AMD's boost technologies. I wonder if that would work with their dynamic overclocking methods to get you an extra 100mHz or so. Perhaps it would still boost to the same point or worst case it would just cause instability.

A slight increase in the base clock will increase your boost clocks. However, it doesn't take but a few MHz before you'll start seeing problems. The MSI MEG X570 GODLIKE even warns you that going past a certain speed results in it dropping the SATA controller and other things. I've only seen very small frequency increases using these method. Once you get past 50MHz or so, it causes problems.
 
Back
Top