Discussion My New Ryzen 5 3600 Hits 4.5 Ghz @ 1.28v

[thor23]

The 3700x has a higher boost frequency than the 3600 requiring more single core voltage/temperature. Also when I was doing the stock power consumption test in my screenshot I was using a power plan switcher which changes the power plan on idle to 99% max frequency which reduces the idle temp a bit considering max vcore on idle is 0.975v.

 

[Shmee]

Interesting. After a bit of googling, it seems I may be best off with an all core OC and undervolt to keep temps down. I am trying 4.225 GHz at around 1.3V peak.

 

[Markfw]

Some reviewers do - TechPowerUp for instance.

WOW, the 3900x took 26.6% less energy to be 23.5% faster in cinebench than a 10900k

 

[Shmee]

I settled for now with 4.25 GHz @ 1.31V. Temps are idling over 10C cooler.

 

[Makaveli]

It would be nice to have an AMD thread without Intel fan boys all up in it.

There has been good info in here otherwise.

Insulting other users (such as using "fan boys")
is not allowed in tech areas.

AT Mod Usandthem

 

[JoeRambo]

I settled for now with 4.25 GHz @ 1.31V. Temps are idling over 10C cooler.

I would not go above 1.25V MT load volts on these 7nm CPUs, and their sweet spot is 100-150mhz lower, might require 1.15 for 4.1 and 1.3V for 4.25 ( both load volts).

 

[Shmee]

Ok thanks for the heads up. I may try to undervolt and underclock a bit further.

 

[IEC]

Ok thanks for the heads up. I may try to undervolt and underclock a bit further.

Also look for apps that take a lot of CPU cycles. For me I initially had high idle usage until I investigated. Once I disabled the offending app (iCloud Photos) my idle voltages and temps dropped through the floor.

This is with the latest chipset drivers (May 2020) + AGESA 1004-based BIOS/UEFI with the Ryzen Balanced power plan on Win 10 v1909.

Also worth experimenting with is ECO mode. I lose very little MT performance and have the same lightly-threaded performance. A small undervolt can offset the MT performance loss by increasing max MT clocks; you will have to experiment to find the sweet spot (generally between -50mV and -100mV in my experience)

 

[Shmee]

Hmm another thing to try for sure. I think it is looking pretty good right now with 4.2GHz and just under 1.25V. I may stop here for a while and do extended stability testing (games).

 

[DrMrLordX]

I will say that you guys have chips that run a lot cooler than mine. Is it possible that my launch 3700X is just a hotter chip, or is there something wrong with my old H110i GT do you think?

Idle temps seem around 50C in ryzen master. CPU vcore seems high though, around 1.4V peak. Using ryzen balanced power mode.

What are you using to measure vcore? 1.4v is too high, I wouldn't go above 1.325v in heavy-current workloads.

 

[cherullo]

That's some pretty impressive clocks. My 3600X is Prime95 stable with a -0.0875v undervolt at stock clocks. Tried to get 4.2Ghz@1.3v and got around 3750 points in CB20, but Prime95 (Small FFT, AVX) crashed instantly. Tried 1.35v too, but temperatures in Prime95 shot to 95C and it crashed too.

I envy even more @IEC 's idle power consumption, I can never get SOC Power below 12w.

 

[Shmee]

That was ryzenmaster or coretemp while stock. It was auto boosting to that.

 

[Kenmitch]

Hmm another thing to try for sure. I think it is looking pretty good right now with 4.2GHz and just under 1.25V. I may stop here for a while and do extended stability testing (games).

The XBOX for PC app is another one to look out for.

 

[DrMrLordX]

That was ryzenmaster or coretemp while stock. It was auto boosting to that.

In what workload?

edit: Ryzen Master is consistently wrong in its voltage reporting, since it never adjusts for vdroop or negative voltage offsets or anything else. At least on my rig. the only digital monitoring I trust on Matisse right now is CPU-z

edit edit: coretemp also seems to report high voltage. It's reporting VID which isn't necessarily the same thing as actual vcore . . . meanwhile, HWiNFO64 reports low core voltage.

 

[JoeRambo]

t least on my rig. the only digital monitoring I trust on Matisse right now is CPU-z

On my 3950x setup HWINFO64 CPU Core Voltage (SVI2 TFN) agrees with VR VOUT, that's what i believe. CPU-Z readings just don't make sense, showing stuck 1.2V even if CPU is under high load and should VDROOP.

 

[Rigg]

So you sold one CPU, but you bought another one - which also loses value over time. You lose twice on opening the box.
Also, looking at your kit, it's not like you'd go bankrupt because of a few $.

And since 3900XT won't hit the shelves before mid July, you've willingly halved your multi-thread CPU performance for at least 1.5 month. Which probably means you don't need that much throughput anyway.

You like buying CPUs, just admit it. No shame in that. Why make up stories about losing resale value?

I certainly like buying CPU's. I won't deny that. I buy a lot of them. Especially Ryzen CPU's. I know the market trends on them well. I wasn't planning to upgrade until after Zen 3 released. Based on my experience (I've owned over 30 AM4 CPU's in the last 2 years) I'm betting that the value hit to this 3900x will be more than the price of this 3600 if I wait 6 months or more to sell it.

I don't anticipate needing the extra CPU power in that time frame so I'm not that worried about it. At this time I really don't need the 12 cores. I've been doing more gaming than video editing/audio mixing lately so I'm fine with riding it out on 6 cores for a while.

Considering I paid for my rig completely by trading in PC parts/builds over an 18 month period (and had a lot of fun doing it) I think it's safe to say I know what I'm doing. I haven't even sold the 3900x yet. I'm not committed to keep the 3600 until my return window closes in a week and a half. Assuming I do move forward with my current plan, I don't anticipate losing a dime selling it. I'll at least break even on the transaction by bundling it with other parts I bought at a discount (or doing a full build) and might even make a little money. All this despite paying full retail for it at launch. I don't need a lecture on flipping CPU parts. I have it down to a science.

 

[Rigg]

It's also hitting over 80*C under a custom water loop, which means this OC probably won't be accessible for most 3600 owners. It's a budget CPU after all.

See updated thermal data in the OP.

pretty much complete BS, a good value of money air cooler like a 212evo will handle it fine, i have a much smaller cooler on then that. Heat density is the problem for OCíng Zen not aggregate heat energy. If you had one you would know that.

I think this OC could be achieved with a 240 aio or large air cooler. A Deep Cool Caiptain 240 handled one of the launch 3600's I had at 4.3 @ 1.4v in real bench with mid 70's C temps. The temps do start skyrocketing on these when you bump frequency this high though.

 

[DrMrLordX]

On my 3950x setup HWINFO64 CPU Core Voltage (SVI2 TFN) agrees with VR VOUT, that's what i believe. CPU-Z readings just don't make sense, showing stuck 1.2V even if CPU is under high load and should VDROOP.

Odd, mine doesn't. It shows vdroop and it reacts to negative offsets. Ryzen Master doesn't show either one.

 

[darkswordsman17]

No. Running my 3900x (from July 2019) @ 4.4 GHz static is a substantial performance increase in everything except ST synthetics.

Any kind of modern game, MT benchmarks/productivity apps, etc. it's a performance gain.

Last year, people would have killed for a 4.5 GHz 3600! That's an insane clockspeed for a chip that used to struggle to hit 4250 MHz overclocked. Only 3800x chips could somewhat-reliably maintain all-core OCs that high.

So what exactly was going on? There were a fair amount of people reporting doing OCs that saw worse performance and so many were just leaving it to manage things itself. There was discussion about some things and a bit of a furor over clock speeds and how they were being reported, but I don't recall there being definitive "this is what you should do" which it seems like would be the case if things were that clear cut. And that definitely muddies things up when comparing testing from different sites as I absolutely guarantee there's not consistency in how they're managing things like that.

Now it seems like performance talk is all focused around memory.

Here's some data on how a 4.2ghz manual OC compares to a 3600 at stock. As you can see average core power consumption is about 30% lower with a manual OC.View attachment 22066View attachment 22067

What's the performance like though? I recall people OCing but then saying that they didn't see any improvement in actual performance (and some saying it actually was giving them worse). But others are saying that's not the case, but they're also not showing less power use and lower temps when OCing that you're indicating, so something isn't adding up.

 

[DrMrLordX]

So what exactly was going on? There were a fair amount of people reporting doing OCs that saw worse performance and so many were just leaving it to manage things itself.

Personally I never saw that (or don't remember that), though if you can link to such a discussion I might be able to sort out what was going on. I do remember people triggering clock stretching with undervolts.

 

[therealmongo]

Personally I never saw that (or don't remember that), though if you can link to such a discussion I might be able to sort out what was going on. I do remember people triggering clock stretching with undervolts.

Yeah, i would imagine he is referring to that (clock stretching).

As long as the cooling is adequate, performance scales as expected.

RAM overclock got focused on afterwards as great gains could be seen, especially in gaming when reducing the latency ....

 

[thor23]

So what exactly was going on? There were a fair amount of people reporting doing OCs that saw worse performance and so many were just leaving it to manage things itself. There was discussion about some things and a bit of a furor over clock speeds and how they were being reported, but I don't recall there being definitive "this is what you should do" which it seems like would be the case if things were that clear cut. And that definitely muddies things up when comparing testing from different sites as I absolutely guarantee there's not consistency in how they're managing things like that.

Now it seems like performance talk is all focused around memory.



What's the performance like though? I recall people OCing but then saying that they didn't see any improvement in actual performance (and some saying it actually was giving them worse). But others are saying that's not the case, but they're also not showing less power use and lower temps when OCing that you're indicating, so something isn't adding up.

Well my manual OC was 4.2ghz and stock single core boost is only 4.2ghz so the manual OC was slightly better. My 24/7 OC is 4.4ghz@1.2v, I was just using 4.2ghz to test against stock.

 

[darkswordsman17]

Personally I never saw that (or don't remember that), though if you can link to such a discussion I might be able to sort out what was going on. I do remember people triggering clock stretching with undervolts.

I'll have to see if I can find that, its been months back (I didn't keep up much as I didn't have a Zen 2 system and wasn't sure when/if I'd be building one, and so kinda stopped paying a lot of attention to tweaking, so I didn't keep up with things, and then it seemed like next thing I know all the performance talk was focused around memory). I'm guessing there'd be some in the builders thread. I recall discussions about PBO (?) or not, and some other things as well, so perhaps it was just related to certain ways of doing it.

I don't recall people talking about undervolting along with it, if anything I think I recall people upping voltage to see what was stable (which perhaps that was the issue is upping the voltage caused thermal constraints to kick in more?). I don't think it was drastic, but people were basically seeing that manual overclocking didn't seem to bring any real benefit (but I think they were using synthetic benchmarks to test performance) so several said they just let the CPU handle things itself. Its possible it was in certain situations (i.e. stock HSF, where it'd likely thermally constrain more than the high end air coolers and water cooling with better TIM). This might've even been back when there was issues with the firmware and BIOSes (I searched to see if I'd posted anything around that time, but it was just remarking about people having various issues related to BIOS/AGESA).

Ah, found one:

Discussion - Ryzen 3000 CPUs overclocking: exercise caution

Buildzoid has admitted that the static overclock of his AMD Ryzen 7 3700x has degraded the processor. More on it here: https://www.youtube.com/post/UgwIDprhUtrJzFuhW0N4AaABCQ

forums.anandtech.com

This post in particular:

Discussion - Ryzen 3000 CPUs overclocking: exercise caution

Buildzoid has admitted that the static overclock of his AMD Ryzen 7 3700x has degraded the processor. More on it here: https://www.youtube.com/post/UgwIDprhUtrJzFuhW0N4AaABCQ

forums.anandtech.com

I recall others talking about it though.

So, not sure that helps, as they don't offer much specific. Weirdly markfw first agrees then disagrees later saying he both overclocked and undervolted.

Well my manual OC was 4.2ghz and stock single core boost is only 4.2ghz so the manual OC was slightly better. My 24/7 OC is 4.4ghz@1.2v, I was just using 4.2ghz to test against stock.

Ah, so you undervolted too then?

Did I completely miss Zen 2 being another AMD excessive stock voltage chip deal?

 

[thor23]

I was using a -0.025v vcore offset when I ran 4.2ghz, Normally I run a +0.1v offset on my 4.4ghz OC.

 

[therealmongo]

adjective
adjective: excessive

1. more than is necessary, normal, or desirable; immoderate.

Dont think thats the word you were wanting to use as that would be bias.

Voltages can be knocked back if you are running a manual overclock and you know the type of workload you will be running. In this way you can tune your voltage to the application and thermal headroom you have available to you.

Now with default settings aka PBO, the PBO algorithm has to take into account each and every possibilty and will ramp up/down voltage accordingly.

The only "excessive" I came across was in the early days with early BIOS and the misunderstadning of the uneducated who didnt understand Ryzen architecture as it is different to your classical CPU, where as the CPU is now acting more like a GPU in that it will boost according to application and thermal head room.

Thats my quick analysis, hope it helps