[SOLVED] Ryzen 3000 overclock is useless and frequency is just a number

[Fiorezy]

I have the Ryzen 7 3700x, I used to run it using CTR at 4.45Ghz in sc and dc tasks, 4.4Ghz in 4c, 4.35Ghz in 6c, and 4.2Ghz all cores.

After that, I locked all cores at 4.1Ghz at 1.250V, and literally, I didn't notice any performance drop over the above frequencies in daily tasks and games, it runs much cooler with locked voltage as there were no more temperature spikes.

I will take this every time over a boost in frequencies that won't offer any performance benefit but results in a hot chip instead.

 

[logainofhades]

This is why I do not overclock Ryzen 3000 or 5000. PBO does a great job of handling it, for you.

 

[Fiorezy]

This is why I do not overclock Ryzen 3000 or 5000. PBO does a great job of handling it, for you.

Even PBO is a gimmick, it is just more heat and power delivery in exchange for a frequency number that won't boost the actual performance at all.

 

[drea.drechsler]

Even PBO is a gimmick, it is just more heat and power delivery in exchange for a frequency number that won't boost the actual performance at all.

I disagree.

Not that it doesn't result in greatly increased heat output (it does), but that it's just a gimmick on performance. With a tweaked PBO my 3700X performs like a 3800X (stock) in multi-threaded and single threaded tasks, when measured in the proper benchmark (CB20 and CB23). Without it holds up like a fairly mediocre 3700X in the same benchmarks. I just can't see the gimick in that.

But I do agree it needs fairly good cooling (I use a 240mm AIO) to maintain multi-thread performance for tasks like transcoding a long video. On the stock Wraith Prism I'm pretty sure it would throttle back to a run-of-the-mill 3700X, along with a screaming fan.

For a synthetic workload like Prime95 small FFT's, though, it will throttle back to what you've settled on...4125. So to be sure there's no gain at the absolute extreme but at least it's able to freely pull down 4400's and even 4425's on some cores during moderate to light , but realistic, work loads like gaming.

 

[Fiorezy]

I disagree.

Not that it doesn't result in greatly increased heat output (it does), but that it's just a gimmick on performance. With a tweaked PBO my 3700X performs like a 3800X (stock) in multi-threaded and single threaded tasks, when measured in the proper benchmark (CB20 and CB23). Without it holds up like a fairly mediocre 3700X in the same benchmarks.

But I do agree it needs fairly good cooling (I use a 240mm AIO) to maintain multi-thread performance for tasks like transcoding a long video. On the stock Wraith Prism I'm pretty sure it would throttle back to a run-of-the-mill 3700X, along with a screaming fan.

For something like Prime95, small FFT's, though it will throttle back to what you've settled on, 4125, so to be sure there's no gain at the absolute extreme. But at least it's able to freely pull down 4400's and even 4425's on some cores during moderate and light work loads like gaming.

Yes, but it is still just a number, as long that it is not a few hundred points difference in benchmarks you won't have an actual gain in daily tasks performance, or gaming and even in high load tasks like rendering.

Not sure if you've tried it before but CTR is the best way right now to maintain that juicy 4400 in gaming at reasonable voltages and power delivery, much much better than PBO. But as I mentioned before, it is just a number and you won't be able to notice any performance boost from 4200 or 4100. 1usmus himself already stated that with the way Ryzen has been designed, there won't be any performance gain past 4000.

Btw I have the Kraken x63.

 

[hotaru.hino]

The problem with modern day overclocking is two-fold:

1. What most people think is overclocking is simply raising the turbo boost ceiling. Most processors these days ship damn near their limits, hence why, at least without extreme measures, most people can't get add 0.1-0.2GHz on top of what they already get. This is exacerbated when they enable something like forcing all-core turbo boosting and they can't push the processor very far.
2. Overclocking still seems to have this mythical property that doing it, regardless of how much of it you do, gives you substantial performance improvements. i.e., it's essentially free performance. Again, processors these days are shipped to self-overclock near their limits. But if you sit down and do the math, you'll find that basically past 4.0GHz, adding another 0.1 or 0.2 GHz is a <5% improvement to clock speed. And while we'd like to believe that this translates to 1:1 increase in performance, it doesn't. However, physics is a cruel mistress so you still will get a proportionate increase in power dissipation.

 

[drea.drechsler]

Yes, but it is still just a number, as long that it is not a few hundred points difference in benchmarks you won't have an actual gain in daily tasks performance, or gaming and even in high load tasks like rendering.
...

I do see improvements in rendering, but the couple minutes of improvement on a 30 minute video transcode I use as a benchmark really isn't all that noticeable. So in that sense THAT much is arguably true. AMD's really pushed the envelope with their boosting algorithm and their bin sorting of chiplets.

It's that chiplet architecture that makes it possible to use them as buildling blocks in any number of different CPU's...all the way from basic, low-core count desktop up to HEDT and even massive server CPU's. That means the ones destined for lowly 3700X's really are being pushed even for that role, leaving very little for overclocking.

But my goal for overclocking always has been to buy a tier back in hopes of getting performance of the next tier product, un-overclocked. Which is what I'm seeing with my 3700X PBO and means I've met my goals, at least. But saying the realized gains aren't really all that noticeable does raise questions about whether 3800X's in stock performance are worthy of their price premium...a topic which has also seen much discussion.

Oh yes, and one other thing. A lot of people have been seeing greatly improved manual overclocking with newer, probably mature process, silicon. But even they seem to be back-tracking with time. I can imagine they've run high OC's with upwards of 1.45V, fixed, and thought it OK. If so, many could be finding their CPU's lose stability as they degrade.

 

[Unolocogringo]

The new chips are a lot better than the older ones for all core clocks.

And overclocking is hit or miss in the silicone lottery.But some times you get lucky.
My 3600 does 4.4ghz all core boost @1.28v with v droop as low as 1.269v.
PBO way over-volted the chip and made it run hot.
It runs cooler @4.4 that using PBO which dropped all core clocks to 3.9.
Everyone's mileage will vary. And the days of 30%-50% overclocks on air/water are gone forever.

 

[Crosslhs82x2]

Ya with these Ryzen cpu's overclocking
Expectations really needs to be tempered.

If my 3600x on a Asus Rog Strix F Gaming
Would boost as well as my son's 3600 on a Msi b450 Pro Carbon ac, I probably would not have a Oc on my 3600x.

His 3600 has 2 cores regularly boosting to 4.275g during minecraft and Detroit Human while the rest run steady at 4.225g with the ppt,tdc,edc limits raised, Pbo and a slight undervolt, so i feel on his leave it alone and be happy that it's able to boost over it's 4.2g turbo boost on it's own.

My 3600x 4.325g Oc @1.30???v
Gives me very close Mc cbr20 (within 15 pts) score as a 2700x so I won't say it wasn't worth it.
Anything more I really had to ask myself if it's worth the risk of Cpu Longivity having to push the voltage higher for very little more gain in performance.

 

[damric]

Yeah my 3000 and 5000 series already boost close to the maximum headroom so it's not as worth it to manually tune the CPU clock.

The last CPUs I was able to get respectable OCs were the Ryzen 1600AFs.

 

[drea.drechsler]

..
The last CPUs I was able to get respectable OCs were the Ryzen 1600AFs.

I think they use the improved Zen+ dies the 2600's use so you can get the performance margins they offer in addition to the uplift from original Zen.

I think most people are spoiled by expectations from earlier CPU's. I remember I could take my FX6300 from it's stock clocks of 3500-3600 to over 4500 for an amazing performance uplift. If I had a case that could handle a liquid cooler and a motherboard that could power the CPU they were good for upwards of 5Ghz. That was then, now we have 7nm and vastly better IPC instead.