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

[Rigg]

In anticipation of the upcoming zen 3 release and Zen 2 XT refresh, I decided to sell my 3900x now before resale plummets in the near future. After hearing it mentioned that the more recent 3600's were clocking much higher than older silicon I decided to go ahead and pick up an R5 3600 last week as a place holder. The 3600 CPU was manufactured the 7th week of 2020. I have a full custom loop so I figured I'd OC the snot out of it and share what I found.

PCPartPicker Part List

Yes, I fully realize this a ridiculous build to pair with a 3600. Still I think impressive results could be had with a good mid-range air cooler and B series board.

I ended up achieving stability with CCX0 @ 4.55 Ghz and CCX1 @4.5 Ghz. This took around 1.28V as reported by HWI64 on SVI2 under heavy load. I'm running low (level 2) LLC so voltage is set to a fixed value in BIOS of 1.3125. I was able to pass multiple runs of Intel Burn Test Standard, over an hour of Prime 95 with AVX disabled, and a full 8 hour Real Bench stress test. As I found with my 3900x the temps get toasty in a hurry when you start pushing the clocks. Even with fairly sane voltage. Here are some results to check out:





EDIT:

I have updated thermal data after playing around with my fan curve. An extra 500 RPM dropped water temps by 5c after reaching steady state. I'm including a data with %100 fan speed for reference. While I originally had the fan profile set for absolute silence I think the new data is probably more realistic. While the fans are now audible under heavy load they are what I would consider completely tolerable.

 

[beginner99]

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.

The issue with manual OC is that it runs at that clock constantly right? since with avergae PC use, you will use much more power as the cpu will mostly idle without manual OC.

 

[DrMrLordX]

The issue with manual OC is that it runs at that clock constantly right? since with avergae PC use, you will use much more power as the cpu will mostly idle without manual OC.

Matisse doesn't actually do that, at least not if you use Ryzen Master. I think it may do that if you OC through the UEFI.

 

[lightmanek]

Matisse doesn't actually do that, at least not if you use Ryzen Master. I think it may do that if you OC through the UEFI.

Even 1st gen Ryzen didn't do that. I tested my static OC idle power consumption against Default settings extensively, and the difference in idle on Ryzen 1700 was 10W at the wall, which came mostly from running memory at 3200MHz and increased vSOC voltage required for that.
In manual OC mode, Zen cores are still aggressively parking unused cores and clock stretch to save power.

 

[darkswordsman17]

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.

Was that all core or single core? And you were seeing that big of a difference from that small of v change? You're seeing less power even with the higher v on the OC?

 

[thor23]

The issue with manual OC is that it runs at that clock constantly right? since with avergae PC use, you will use much more power as the cpu will mostly idle without manual OC.

The screenshot clearly shows max and average power consumption is much less with the manual OC even though manual OC forces the frequency to run at max all the time. Average temps are even 5c lower than stock with a manual OC. Modern cpu's spend most of their idle time in a deep c state anyway so it doesn't really matter if your cpu clocks down at idle or not.

 

[thor23]

Was that all core or single core? And you were seeing that big of a difference from that small of v change? You're seeing less power even with the higher v on the OC?

At stock I wasn't running any vcore offset, with the manual overclock I was using a -0.025v offset for an effective voltage of 1.075v@4.2ghz allcore.

 

[eek2121]

To the OP, your chip is clock stretching. It isn’t really running at 4.5 GHz. My 3900X with a CCD disabled gets 577 point in CB single core.

EDIT: that is the average EFFECTIVE clock. Effective clocks matter when overclocking. My chip will reach 4.9 GHz if a screw around with it, but effective clocks end up being 1.5 GHz.

 

[Rigg]

To the OP, your chip is clock stretching. It isn’t really running at 4.5 GHz. My 3900X with a CCD disabled gets 577 point in CB single core.

EDIT: that is the average EFFECTIVE clock. Effective clocks matter when overclocking. My chip will reach 4.9 GHz if a screw around with it, but effective clocks end up being 1.5 GHz.

As far as I'm aware clock stretching doesn't happen with a fixed clock multiplier and v core voltage. Instead of clock stretching it crashes. My average effective clock is 4.525 in a full all core load. If your CCD disabled 3900x is scoring 577 its probably holding higher average single core boost than 4.55 Ghz. The clocks and voltage change on the millisecond level when not set to fixed values. No software monitoring I know of polls fast enough to give an accurate reading.

 

[DrMrLordX]

I think @Rigg 's MT results indicate a pretty high clockspeed. My 3900x @ 4375 MHz scores around 7900 in CBR20 MT. He has half the cores but he's scoring more than half as much as I. Amdahl's Law doesn't really bite you hard on benchmarks like CBR20.

Also, at least according to Anandtech Bench, a stock 3600 scores ~3500 in CBR20 MT:

CPU 2019 Benchmarks - Compare Products on AnandTech

CPU 2019 benchmarks: Compare two products side-by-side or see a cascading list of product ratings along with our annotations.

www.anandtech.com

He's got a big lead over that.

 

[therealmongo]

Agree with the OP and DrMrLordX above,

I also have a newer batch Ryzen 3600 and while my stable overclock is a "measly" 4400 mhz @ 1.337v (prime95 blend load), a CB20 I did @ 4.5 Ghz scored 4141 pts.

This score is not indicative of any clock stretching and as the OP said, if the voltage is not high enough either CB20 will crash or the PC will reboot....

 

[VirtualLarry]

I also have a newer batch Ryzen 3600 and while my stable overclock is a "measly" 4400 mhz @ 1.337v (prime95 blend load), a CB20 I did @ 4.5 Ghz scored 4141 pts.

Now I really wish I had taken a shot of the heatspreader. Sounds like when I bought a recent new-batch 3600 CPU, that perhaps, I didn't really get a new enough batch. I think that the first part of the date-code was "19" and not "20", TBH.

I can't even get Windows 10 to boot on mine at 4.4Ghz all-core, even with a good bump in voltage. I'm at 4.20Ghz @ 1.3500V, something like that, that was required to be "PrimeGrid stable".

Edit: Or possibly, I need to step up to something with beefier VRMs than my Asus B450-F ROG STRIX Gaming ATX mobo, like an MSI B550 Tomahawk (my next board I'm eyeing).