Are these good clocks for the 5900X?

[Deleted member 212040]

Recently upgraded my anemic 3900X to a 5900X, very happy. Somehow, despite all my beliefs on Zen 3, this CPU runs cooler. This confirms to me that the 3900X is simply made out of lava. There are also zero of the issues I had with the 3900X, no idle reboots, no bad clocks and high temps, no USB disconnects (didn't have them with the 3900X either), the CPU just works. It's also way snappier. I didn't expect such a difference from a one generation upgrade but AMD surprised me yet again.

Anyway.

I messed with PBO2 and Curve Optimizer in the BIOS, how are these max clocks?:

I know that one of the dies on a x900X CPU isn't binned as well as the first one, unlike on a x950X CPU which has both dies binned well, so these clocks look good yeah? All of the cores are able to go the max rated boost of 4.8. Something my 3900X could barely think about accomplishing.

Here's Ryzen Master:

I put negative 11 offset on my best cores, negative 15 on second best cores, and negative 17 on all the other cores. I enabled PBO and set PPT to 165, TDC to 115 and EDC to 150. I set the boost override to +200. This brought my single thread Passmark score from 3547 to around 3637. CPU-Z single thread and Cinebench single thread increased as well.

But multi core got the biggest benefit: I get 4.5 GHz all core in Cinebench R20, and ~4.9 GHz on single core R20 test. In games, it's usually 4900 - 4975 MHz, sometimes 5 GHz. On stock it was 4.15 GHz all core and 4.8 GHz on single core. Games capped out at around 4.9 GHz.

My memory latency is also 55-57ns. Is that good? Used to be 68-70ns before.

Info about my rig and cooling in my System Specs.

I am not going to ask about temps since they are irrelevant for Zen 3, and you shouldn't either. I learned the hard way not to be paranoid about temps. Especially on Ryzen.

Another question: Is PBO2 & Curve Optimizer safe to run? Are the CPU safety features (voltage fitness regulator) still in place? I avoid manual OC because it disables these features and it's not as good as the CPU's regular boosting, so it'd be nice to know if PBO2 & Curve Optimizer are safe or not.

Here's a loserbenchmark run: https://www.userbenchmark.com/UserRun/44531174

 

[GerKNG]

can you run core cycler without errors?

and to your question about PBO.
Core clock offsets, curve optimizer and powerlimits do not have any influence on the FIT values (V/F Curve behaviour)
Only the scalar itself influences it. (which is fine to set to 2x but i'd not go above 5X except for benching)

using the curve optimizer, powerlimits and clock offsets is totally fine, safe and still following the stock stats of the CPU.

 

[Deleted member 212040]

can you run core cycler without errors?

and to your question about PBO.
Core clock offsets, curve optimizer and powerlimits do not have any influence on the FIT values (V/F Curve behaviour)
Only the scalar itself influences it. (which is fine to set to 2x but i'd not go above 5X except for benching)

using the curve optimizer, powerlimits and clock offsets is totally fine, safe and still following the stock stats of the CPU.

Yeah it runs just fine. Temps in AIDA64, Prime95 and Cinebench are all way better than on my 3900X, with bigger all core boost.

Good to know it's safe, just free performance then.

I did not mess with the scalar.

What about the clocks themselves though? CPU boosting properly?

 

[mtcn77]

Yeah it runs just fine. Temps in AIDA64, Prime95 and Cinebench are all way better than on my 3900X, with bigger all core boost.

Good to know it's safe, just free performance then.

I did not mess with the scalar.

What about the clocks themselves though? CPU boosting properly?

Play around with the PPT&EDC settings in 5 unit increments without breaking the 1.1:1 ratio between them. If you run the tests back to back without cooldown, you'll gain much more heat load to elucidate the difference it makes. Start from the highest potential setting to the lowest to keep the heat pressure intact between runs.

 

[Deleted member 212040]

Play around with the PPT&EDC settings in 5 unit increments without breaking the 1.1:1 ratio between them. If you run the tests back to back without cooldown, you'll gain much more heat load to elucidate the difference it makes. Start from the highest potential setting to the lowest to keep the heat pressure intact between runs.

I've already done PPT, TDC & EDC testing and determined this is the best balance of temps and performance for my CPU.

Only question I have at this point is if the CPU clocks are fine and if the Loserbenchmark run doesn't show any obvious failures.

 

[freeagent]

Loserbenchmark run

It honestly made me laugh

It runs your stuff without crashing? Should be good

 

[Deleted member 212040]

It honestly made me laugh

It runs your stuff without crashing? Should be good

Yeah nothing is crashing, not even Prime95 Small FFT. All cores can easily reach and surpass the rated 4.8 boost.

Surprised at the 5 GHz. Clocks on stock are 4.9 - 4.95 max, but with PBO, +200 offset and curve optimizer I see it reaching over 5.1 GHz.

When I installed this CPU, everything on stock, the max boost in Ryzen Master for each CCX was interestingly 4900 MHz and not 4800 MHz as I would expect. My tweaks brought it to 5150.

There are no WHEAs, no instabilities, even passed memtest. AGESA 1.2.0.3 Patch A bios with latest chipset drivers on a fresh Windows 10 Pro 21H1 installation, literally just reinstalled OS last night.

 

[GerKNG]

What about the clocks themselves though? CPU boosting properly?

pretty decent.
my 5900X was not happy with 5Ghz boosts. that looks great.

 

[mtcn77]

I've already done PPT, TDC & EDC testing and determined this is the best balance of temps and performance for my CPU.

Only question I have at this point is if the CPU clocks are fine and if the Loserbenchmark run doesn't show any obvious failures.

The test has to match your real world usage TDP, so record what TDP level your daily usage maxes out and run the test simulating that heat load.

For instance, unoptimised games are great for this. Some used to reflect instability better than prime 95, so it was a waste of time trying to program for a benchmark when real world use failed at that setting.

Just find your most difficult power consuming use case and luckily try to meet the same demand in the benchmark without tripping temperature limit. If the clocks do not devolve with the necessary PPT & EDC enhancement that indicates not those pbo settings, but that TDP level is fine for your cpu. Don't fixate on individual pbo setting levels.

 

[freeagent]

Not sure why they say 4850, at bone stock it shows the limit at 4950.. +200 right? So looks like someone at HQ missed that memo. I find the limits I set don't matter, it will take more or way more if it wants. So we aren't really controlling anything, just manipulating.

 

[Dux]

I have this weird issue with my MSI Gaming Plus X570 motherboard and 5900X CPU where it will run at maximum of 1750MHz in games and everything else if i don't manually set the multiplier in BIOS. Multiple BIOS versions and Windows reinstalls had no effect.

 

[bissag]

Looks good
What is your cpuz benchmark score?

 

[Deleted member 212040]

Looks good
What is your cpuz benchmark score?

680 on single thread, 9950-10050 on multicore

 

[tabascosauz]

@Emily If you set any Curve Optimizer offsets, Corecycler is probably the best way to test stability, but not really at default config.ini settings. The config file explains in depth about a lot of its settings.

Corecycler uses SSE Prime95 but running one core at a time, ideal because SSE gets the max possible clocks the boost algorithm allows on any given core. But the default config only runs a core for about 6min on Large FFT which is pretty woefully inadequate; change up the config a bit and you might soon see errors on what default Corecycler said was stable.

I spent a lot of time running a single FFT range (eg. Small or Large) but found that the safest stability test would be to set 68min per iteration, running the All FFT setting to cover the entire possible FFT range once per iteration. Then testing for about 4-5 iterations per core overnight, testing 2 cores per day so it wouldn't take forever. It took a lot of trial and error over a few weeks but the curve is 100% stable now.

I regularly see a lot of misinformation about P95 being bad for clocks - that much is true if you set the script to run AVX, but I can tell you right now that the way Corecycler runs P95 SSE, on any FFT length, will hit equal or higher boost clocks to any unoptimized single-threaded program or game. Just close as many background processes as you can when you run Corecycler, any bit of unnecessary load will drag down clocks.

If you're easily hitting 5.0-5.1 Effective Clock in HWInfo, that's already a pretty good chip. Mine tops out at about 4.95 even with +200MHz. You *probably* won't have visible daily instability on something that passes default 6min Corecycler, but if you want to be absolutely sure, the config file has all you need.

 

[Chomiq]

I'm just here to say - new user with 5 posts and has properly filled in system specs from day 1. Awesome!

 

[Deleted member 212040]

I'm just here to say - new user with 5 posts and has properly filled in system specs from day 1. Awesome!

I was here before. In the Discord as well.

If you're easily hitting 5.0-5.1 Effective Clock in HWInfo, that's already a pretty good chip. Mine tops out at about 4.95 even with +200MHz. You *probably* won't have visible daily instability on something that passes default 6min Corecycler, but if you want to be absolutely sure, the config file has all you need.

Passes default, don't feel like putting in all that work. Everything I do is rock solid stable with what I assume are good clocks. If it shows instability I'll tweak and do more extensive testing

Woke up today to more 5 GHz on more cores:

 

[tabascosauz]

I was here before. In the Discord as well.


Passes default, don't feel like putting in all that work. Everything I do is rock solid stable with what I assume are good clocks. If it shows instability I'll tweak and do more extensive testing

Woke up today to more 5 GHz on more cores:

Had a feeling that you were on TPU before, but didn't want to spoil it until you said it first.

That's a good way to go, as long as you can pass say iterations per core of default (when on the edge, might pass say 2 iterations out of 3), should be fine for everything you can throw at it.

Use HWInfo and watch Effective Clocks though, "Core Clock" isn't worth a damn on Ryzen 5000. Even on mine I can say my chip boosts up to 5050 all the time if I just look at Core Clock or HWMonitor metrics, but that's far removed from reality.

 

[Deleted member 212040]

Effective clock is either bugged on mine, or either I have the best CPU in the world because it shows 34 GHz.

It beats 11900K in single core benchmarks, I'm happy.

 

[Space Lynx]

these 5900x scores are very nice indeed.

Amazon still has the 5900x in stock in USA... really surprised it hasn't sold out yet.

 

[xtreemchaos]

id be well happy with those clocks bud. you have gave me a reason for a upgrade .

 

[Chomiq]

id be well happy with those clocks bud. you have gave me a reason for a upgrade .

3D v-cache is just around the corner.

 

[Deleted member 212040]

id be well happy with those clocks bud. you have gave me a reason for a upgrade .

Not gonna lie, I'm not sure what was wrong with the 3900X (probably a lot considering the idle reboots), but this feels so much snappier. I expected just better single core but what I got was a better overall experience.

I would've waited for 3D V-cache but I'm not sure if that Zen 3 refresh will even be a thing, found this CPU in stock at a good price so I just went for it. It'll do nicely until I upgrade again in a few years.

 

[xtreemchaos]

im very tempted to get one, ill have a look on scan to see what deals thay have and think on a way of getting it past the misses .

 

[Space Lynx]

3D v-cache is just around the corner.

I mean 3d vcache was already benchmarked by lisa su you are talking 8-9 fps increase due to it... and that was a highly selective title to test it on too... I'm not convinced it will matter honestly. at 1440p and 4k it doesn't even matter at all for most games, but we will see when they come out.

 

[Deleted member 212040]

There was also a weird delay when moving the CPU-Z window around with the 3900X, but no such thing with the 5900X.