• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.
  • The forums have been upgraded with support for dark mode. By default it will follow the setting on your system/browser. You may override it by scrolling to the end of the page and clicking the gears icon.

Ryzen Owners Zen Garden

I guess that I'm still used to the idea that 95c is like really, really hot.
It looks like you got some excellent advice about curve optimizer. In regards to 95c being hot or not, it is hot, but modern AMD chips more or less overclock themselves until they hit a power or thermal limit. For most people, that thermal limit comes first even with quality air cooling. If you improve cooling, like I did by adding custom water, the cpu will push itself further. This is a bit simplified for communication purposes. AMD set the thermal limit at 95c for mainstream CPU's so they must be fairly confident in it.

I'm still waiting on a 140mm fan, but from the constant weekly updates from Phanteks reddit rep it looks like the T30-140 will beat Noctua's 140mm to market. Unless something big happens, I won't be waiting for Noctua.
I am planning to change my whole desk, with new case, and new water cooling. I was planning to go upto a 140mm wide radiator but T30-140's weren't out. I can't wait.
Yeah, R23 is pretty definitive. If the core in question can't go any further then it just can't go any further. Perhaps you still have more CO headroom on that core, perhaps not.

Corecycler is fine as it's honestly still the most convenient way of getting an overview of all the cores (ie. run it as-is). It should at least get you to a point of not crashing under normal workloads. If you want to make sure you're 100% stable there are many options but they all will take exponentially more time and effort. Ycruncher is good but there's config involved as well. You have to figure out what tests are most effective for you.
Core Cycler/P95 is running into errors on the "preferred" cores at +-0 for CO settings. If I were to decrease my PBO override from 200, that would improve stability right? It would be better to have more consistent high clocks with a negative value for CO than a higher potential clock with a positive CO value. In the past I got the best benchmarks with +50 override but with new water cooling at lower CO values, I set a new benchmarking personal record with this 5950x with +200.
 
Core Cycler/P95 is running into errors on the "preferred" cores at +-0 for CO settings. If I were to decrease my PBO override from 200, that would improve stability right? It would be better to have more consistent high clocks with a negative value for CO than a higher potential clock with a positive CO value. In the past I got the best benchmarks with +50 override but with new water cooling at lower CO values, I set a new benchmarking personal record with this 5950x with +200.

I'm not sure if no boost override would change anything. Right now your preferred cores arent even breaking what, 4950 in R23? In actual effective clock.

No boost override for 5950X is 5050, still way off. Only if you are already bumping up against the stock Fmax, will boost override do anything for you.

If it's not actually crashing on you, you could just leave it alone. If I just went off default config corecycler I could run -5 -8, but on my long term test config it came out to -2 -5. But -5 -8 didn't actually cause me any real stability problems, just corecycler didn't like it.

Can run the cores in question individually through the whole ycruncher stress test suite, to get a second opinion.
 
I'm not sure if no boost override would change anything. Right now your preferred cores arent even breaking what, 4950 in R23? In actual effective clock.

No boost override for 5950X is 5050, still way off. Only if you are already bumping up against the stock Fmax, will boost override do anything for you.

If it's not actually crashing on you, you could just leave it alone. If I just went off default config corecycler I could run -5 -8, but on my long term test config it came out to -2 -5. But -5 -8 didn't actually cause me any real stability problems, just corecycler didn't like it.

Can run the cores in question individually through the whole ycruncher stress test suite, to get a second opinion.
I am hitting 5075 peaks but they are not sustained. 4950ish is about right. That brings up an interesting question. With other settings remaining the same, increasing the boost override increases my benchmark scores consistently.

I will have to try y cruncher. Previously you mentioned it takes some configuration. What more is there compared to core cycler?
 
Why does my mobo have AMD chipset drivers newer than what AMD's official website offers? same with the on-board VGA drivers.... but when I go to AMD official website, the drivers are still from December 8th... I am very confused.

february 13th released on the mobo website...

Because they're effectively a beta driver that fixes specific issues to your hardware that not all devices needed

I guess that I'm still used to the idea that 95c is like really, really hot.
Same - but thats why i adjust the cTDP temp in the BIOS

If you're willing to lose that brief boost frequency, drop it to a max of 85c or 80c
 
Same - but thats why i adjust the cTDP temp in the BIOS

If you're willing to lose that brief boost frequency, drop it to a max of 85c or 80c
I just got to get into the mindset that even though I may think that it's hot, it's safe for it to run at those temperatures. Hell, even some Intel chips say that they can run safely at a positively mind-bendingly high 105c.
 
Core Cycler/P95 is running into errors on the "preferred" cores at +-0 for CO settings
If you're getting issues with different load types, it means theres an instability just not showing in the other tests - you've got a RAM config similar to mine, you could be having SoC/IF instability that shows in tests that use the RAM heavier or cycle from idle/load more often

You'd definitely need an SoC increase for 4x16GB, especially when you've got a board that's extremely similar to mine
(1.15v SoC is likely your minimum for full stability, i can stress test at 0.95v fine but error at idle/low loads. Also, knowing that board - replace the thermal pad on the chipset, they're garbage. I used a copper shim as i got tired of replacing it every year.)
 
I am hitting 5075 peaks but they are not sustained. 4950ish is about right. That brings up an interesting question. With other settings remaining the same, increasing the boost override increases my benchmark scores consistently.

I will have to try y cruncher. Previously you mentioned it takes some configuration. What more is there compared to core cycler?

Nothing "more", it's just a different workload. It has a range of different tests that can be useful for different areas of stability. Doesn't really take that much config, just enable all the tests and select the core(s) to run on.

If your 5075 "peaks" are just observed values under the Max column, they are borderline meaningless if they are far away from Effective clocks (>100MHz). If 5075 was sustained for any appreciable amount of time your score in R23 would be verging on if not over 1700.

Yes, you may observe a very slight increase in score just from opening up Fmax over 5050, but A) run-to-run variation makes one-off results less meaningful, and B) it's a negligible increase if your chip is low SP and incapable of actually sustaining 5050.

ycruncher.png

If you've owned an early Ryzen 3000 chip, you know exactly what is going on. A lot of late 2019 chips spectacularly failed to even approach their advertised "boost" clock, then around AGESA 1004 AMD changed it to be more optimistic in reporting Core Clock, which "solved" the problem by "boosting" higher. Except, actual performance and Effective Clock did not change at all, just AGESA picking rosier numbers to show the user. Obviously this is no longer an issue with AMD setting Fmax 150MHz higher now, but the way Core Clock is reported doesn't change.

If you enable Snapshot Polling, Core Clock will be a little more accurate but Effective Clock is still the one that tells the whole story here in R23.

At the end of the day, anything between 4.8-5.2GHz is more than good enough, because they'll all get steamrolled by X3D anyway :laugh: jk
 
The TL;DR of what saucy is saying is that when a CPU can boost for just 1ms of the 1000ms in a second, the values reported as max can be misleading (Hence, effective clocks and benchmark results being key)
 
I just got to get into the mindset that even though I may think that it's hot, it's safe for it to run at those temperatures. Hell, even some Intel chips say that they can run safely at a positively mind-bendingly high 105c.
I used to let my 3770K roast in the 90s under a Linpack load. It just laughed at me. Reboots usually between ~107-112c give or take a nut hair.
 
I've been around computers for nearly two and a half decades. Some irrational fears of overheating are tough to shake loose.

PCGamer had this to say...
In certain instances, a Ryzen 5000 series CPU may run at 90C or even 95C, which is "typical and by design," according to AMD. That said, if your idle temps are in the 90C range, or even close to that point, you have a cooling problem—maybe you forgot to apply thermal paste or have your home's thermostat set to 9th Circle of Hell.
I had to laugh at that.
 
There was a time chips didn’t like much more than 80 unless they were at stock.
 
Let me see 5950X, 5900X, 5800X3D, 5600X. I have also built 5600 based systems though. I can tell you that that the 5900X, 5950X or X3D all love to go high on voltage. The 5950X can run as high as 5.1 GHZ and the 5900X 5.0 GHZ (single core) but you need proper cooling and a high end board for that as some boards give more juice to the CPU than others. My Asus X570E does not get as high a boost clock as my MSI Ace Max but that board does that.
That clocks from 5GHz and above even for singlethread is achieved through oc (PBO) so it is normal that the voltage will be high. Thing is for normal power limits the voltage for multithread isn't as high. Singlethread boosts allow instant voltage boosts but only for a few miliseconds per core and the core used is changed very fast.
 
There was a time chips didn’t like much more than 80 unless they were at stock.
Hah, how optimistic
1676446704098.png


Even the intel CPU's of the time couldnt take 70C - I think the core 2 line went to 90/100C and things stayed higher after that point, but both brands went up and down like yoyos
1676446556215.png
 
Hah, how optimistic

Even the intel CPU's of the time couldnt take 70C - I think the core 2 line went to 90/100C and things stayed higher after that point, but both brands went up and down like yoyos

Didn't the AM3+ Bulldozers all have really weird temp sensors? Socket temp, package temp, core temp or something, too complicated for my monke brain. Glad AM4 did away with all that crap, and now we have on-die monitoring more advanced, complex and precise than Intel platforms

I've been around computers for nearly two and a half decades. Some irrational fears of overheating are tough to shake loose.

PCGamer had this to say...

I had to laugh at that.

Here's the thing.........at face value, the reasoning makes sense. It's just another paradigm shift in how these CPUs work, remember when Ryzen 3000 introduced CPPC2 and nobody could get used to the fact that the cores boost with a lightning fast reaction time. We're all used to it now.

However...

All it takes is some bigger CO offsets (-15 minimum) and lower limits: -30C temps, -30W power, same performance? Even those day 1 CPUs that could only do -10 or -15 all-core can still manage what, 10C reduction? "95C is the new normal" just looks like an excuse for history repeating itself...............the process needs extra voltage and power to get to where AMD wants the CPUs to be.

So to avoid a repeat of early Ryzen 3000 (not hitting advertised clocks) or early Ryzen 5000 (unstable at stock), this time they aggressively chase the 95C target. Whereas past generations had similar Tjunction but would start cutting clocks long before then.

7600/7700/7900 are pretty telling - both Zen 4 and N5 are clearly very capable, and 95C is not the new normal. I said then that AMD wanted to milk some extra $$, but just look at the minimal difference in boost freq vs. the huge difference in power and temps. I think N5 at the time was just not ready for the 7600/7700/7900 to launch (at least consistent enough to produce enough CPUs), back in September 2022.

The 7900 runs so cool that they could have simultaneously launched a 142W 12-core (a true 5900X successor), and it would still run completely fine and worlds cooler than the 7900X.
 
Last edited:
I've been around computers for nearly two and a half decades. Some irrational fears of overheating are tough to shake loose.
Same here. I remember the early days of those two and half decades when my mom would tell me to turn the computer off and play with something else because "it needs to cool down". :ohwell:
 
Didn't the AM3+ Bulldozers all have really weird temp sensors? Socket temp, package temp, core temp or something, too complicated for my monke brain. Glad AM4 did away with all that crap, and now we have on-die monitoring more advanced, complex and precise than Intel platforms



Here's the thing.........at face value, the reasoning makes sense. It's just another paradigm shift in how these CPUs work, remember when Ryzen 3000 introduced CPPC2 and nobody could get used to the fact that the cores boost with a lightning fast reaction time. We're all used to it now.

However...

All it takes is some bigger CO offsets (-15 minimum) and lower limits: -30C temps, -30W power, same performance? Even those day 1 CPUs that could only do -10 or -15 all-core can still manage what, 10C reduction? "95C is the new normal" just looks like an excuse for history repeating itself...............the process needs extra voltage and power to get to where AMD wants the CPUs to be.

So to avoid a repeat of early Ryzen 3000 (not hitting advertised clocks) or early Ryzen 5000 (unstable at stock), this time they aggressively chase the 95C target. Whereas past generations had similar Tjunction but would start cutting clocks long before then.

7600/7700/7900 are pretty telling - both Zen 4 and N5 are clearly very capable, and 95C is not the new normal. I said then that AMD wanted to milk some extra $$, but just look at the minimal difference in boost freq vs. the huge difference in power and temps. I think N5 at the time was just not ready for the 7600/7700/7900 to launch (at least consistent enough to produce enough CPUs), back in September 2022.

The 7900 runs so cool that they could have simultaneously launched a 142W 12-core (a true 5900X successor), and it would still run completely fine and worlds cooler than the 7900X.
a LOT of cpus back then relied on motherboard socket temperatures instead of something on the actual core(s)

It's why they were really inaccurate, then intel came along with "TJmax" and "distance to Tjmax" got flipped and gave people estimated CPU temperatures they took as gospel, and now we get people freaking over temperatures that arent crazily different until now when we've gotten 100c+ as stock


If i was to summarise my views on the state of the CPU market, it'd be to not buy any X CPU's from AMD - and on the intel side, dont buy anything with E-cores

The overall performance difference for anything but true workstation/server/rendering workloads is too damn small for the energy and cooling requirements
 
I've been around computers for nearly two and a half decades. Some irrational fears of overheating are tough to shake loose.

PCGamer had this to say...

I had to laugh at that.

Still working on it with the 7700X. It's not uncommon to see me staring at HWiNFO while emulating PS3 games mumbling "it's normal, it's normal" like a mantra. Old fears and habits do die hard.
 
Still working on it with the 7700X. It's not uncommon to see me staring at HWiNFO while emulating PS3 games mumbling "it's normal, it's normal" like a mantra. Old fears and habits do die hard.
This is the reason why I took the 7700 non-X my Noctua NH/D15 chromax black keep it around 45c with only the middle fan installed using Aida64 stress test so I am happy.

With the small performance difference there really is in gaming and such I enjoy the lower TDP of the non-x version and if I need the boost I can do PBO as reviewers has shown.

I am really rocking the Raphael onboard graphics works like a charm even with the newest drivers and since gotten my system I haven't had my RX 6800 XT Red Devil installed once o_O

It would be awesome to see a AMD Ryzen 7 7700 or 7800 non-x 3D that would be a mess but awesome :rolleyes:
 
a LOT of cpus back then relied on motherboard socket temperatures instead of something on the actual core(s)

It's why they were really inaccurate, then intel came along with "TJmax" and "distance to Tjmax" got flipped and gave people estimated CPU temperatures they took as gospel, and now we get people freaking over temperatures that arent crazily different until now when we've gotten 100c+ as stock


If i was to summarise my views on the state of the CPU market, it'd be to not buy any X CPU's from AMD - and on the intel side, dont buy anything with E-cores

The overall performance difference for anything but true workstation/server/rendering workloads is too damn small for the energy and cooling requirements
As a 7700X owner, I agree with you. The non-X is nearly as fast as the X, while offering huge savings on power, heat, and buying cost.

E-cores don't interest me as long as they require you to install Windows 11 to work properly, which I'm not going to do.
 
As a 7700X owner, I agree with you. The non-X is nearly as fast as the X, while offering huge savings on power, heat, and buying cost.

E-cores don't interest me as long as they require you to install Windows 11 to work properly, which I'm not going to do.

I doubt AMD has Win 10 in mind either, moving forward. 7900X3D and 7950X3D's asymmetrical CCD config requires significant new optimizations and so far it looks like Win 11 is the focus, if not the only one to receive them:


If AMD wants those CPUs to run properly, it needs to do what Intel's Thread Director does but in AGESA and Windows. It's like having E-cores without having E-cores.

7800X3D might still be fine.
 
I doubt AMD has Win 10 in mind either, moving forward. 7900X3D and 7950X3D's asymmetrical CCD config requires significant new optimizations and so far it looks like Win 11 is the focus, if not the only one to receive them:


If AMD wants those CPUs to run properly, it needs to do what Intel's Thread Director does but in AGESA and Windows. It's like having E-cores without having E-cores.

7800X3D might still be fine.
To be honest, I don't understand why you even need Win 11 for an asymmetrical core config. Windows 10 has worked with both AMD's and Intel's preferred cores for ages. You could just assign the Intel P-cores, or AMD's 3D cache cores to be the preferred ones. Easy.

It must be an artificial, unwarranted limitation imposed by Microsoft, like not having DirectX 10 on Windows XP back in the days.
 
To be honest, I don't understand why you even need Win 11 for an asymmetrical core config. Windows 10 has worked with both AMD's and Intel's preferred cores for ages. You could just assign the Intel P-cores, or AMD's 3D cache cores to be the preferred ones. Easy.

That won't work, because the Vcache CCD will not be preferred for anything that requires clocks (rendering, application workloads, literally anything that isn't gaming). The CCD that clocks to the max boost advertised will be the non-Vcache CCD, because it's not thermally constrained (possibly *still* Vcore constrained??). You can clearly see from the 7800X3D vs 7900X3D/7950X3D Fmax that Vcache is still hampering clocks.

Just speculating now but they might want to either split the core hierarchy in two and maintain two lists (1 for each CCD and therefore workload type) then assign based on workload like Thread Director. Or just maintain the CPPC list as is, and identify gaming loads in particular to move to the Vcache CCD, and set Preferred Cores on the regular CCD for everything else (since gaming does not need clock).

CCD2 has always been a lower quality die tacked on for MT purposes, so maybe they'll just stack the cache there, migrate gaming loads and confine them on CCD2, and call it a day.

In a sense it might be even more complex than Intel's challenges - there's no real need to maintain a performance hierarchy amongst the E-cores.

AMD/Intel and Microsoft need to push changes on their respective ends to introduce any of these optimizations. I doubt there's anything stopping them from implementing on 10, but you know MS won't. 11 is their baby.
 
That won't work, because the Vcache CCD will not be preferred for anything that requires clocks (rendering, application workloads). The CCD that clocks to the max boost advertised will be the non-Vcache CCD, because it's not thermally constrained (possibly *still* Vcore constrained??). You can clearly see from the 7800X3D vs 7900X3D/7950X3D Fmax that Vcache is still hampering clocks.

Just speculating now but they might want to either split the core hierarchy in two and maintain two lists (1 for each CCD and therefore workload type) then assign based on workload like Thread Director. Or just maintain the CPPC list as is, and identify gaming loads in particular to move to the Vcache CCD, and set Preferred Cores on the regular CCD for everything else (since gaming does not need clock).

In a sense it might be even more complex than Intel's challenges - there's no real need to maintain a performance hierarchy amongst the E-cores.

AMD/Intel and Microsoft need to push changes on their respective ends to introduce any of these optimizations. I doubt there's anything stopping them from implementing on 10, but you know MS won't.
I see what you mean. There's still no logical reason why these changes couldn't be implemented in Windows 10 other than Microsoft trying their best to push 11.

11 is their baby.
Yep, that's the thing. They need all that sweet telemetry data, I suppose. :(
 
I see what you mean. There's still no logical reason why these changes couldn't be implemented in Windows 10 other than Microsoft trying their best to push 11.

I respect that not everyone likes Win 11, but 11 in 2023 is very different from the absolute animal that launched in fall 2021. If you're willing to be a bit of a power user there are tools like NTLite that you can use to bend 11 to your liking.

At the end of the day I don't see not having the X3D optimizations as a big deal.

7800X3D is the everyman's SKU, and requires none of these novel scheduling tricks - it's just a new 5800X3D.

The hardships that come with 7900X3D/7950X3D are the result of wanting to have your cake and eat it too (Vcache + more cores). Not wanting to lose to Intel in any metric at the top end.

For people benching all-core, the 7950X is probably still going to walk all over those two, 16 cores that aren't hobbled by a silicon insulator. And for pushing DDR5 clock, it's still the 1CCD parts that will lead. And if the scheduling isn't perfect, then the 7800X3D will come out on top anyway (see 7950X problems).

I hope the AGESA team gets their shit together but I don't see the 2CCD Vcache parts being anything other than a hassle.
 
Back
Top