Initial attempts at overclocking 3700X

[porina]

Stability test/criteria I'm using is Prime95 29.8b5, running 8x 128k FFTs. The small size pushes the cores and not IMC/ram. Asus Prime X370 Pro with current bios 5008, which has AGESA Combo-AM4 1.0.0.2. Ram is running SPD 2400 so shouldn't be a factor. CPU cooling is 240mm AIO, I think it is a CM MasterLiquid Pro.

Using Ryzen Master I started at 1.20v set, observed indicated load voltage, and incremented clock in 25 MHz step on all cores until P95 showed an error. Increase set voltage by 0.05 and repeat.

1.20v set, 1.150v load, 4025 ok? 4050 error. top temp 68c
1.25v set, 1.194v load, 4100 ok? 4125 error. top temp 72c
1.30v set, 1.238v load, 4175 ok? 4200 error. top temp 78c
1.35v set, 1.281v load, 4200 error. top temp 84c

The "ok?" part is because I didn't run long, maybe tens of seconds at each step before increasing. This was the rough set. I think temps are excessive at 1.35v and current cooling so had to stop there since I still have some voltage headroom. On the longer test at 1.30v set, I had to drop the clock to 4125 to actually run P95 any amount of time without errors, in the end I tested clear for 12 minutes at 4125.

Remembering I had NZXT digital PSU attached, I installed the software to see what that was reporting as power on the ATX EPS power connector.
CPU stock: 16W idle, 86W load
CPU 1.30v 4125: 13W idle, 114W load

I note the system seems to give up to 1.5v to the CPU at idle, presumably this is needed for the 4.4 GHz single core boost. This might explain why I had lower idle power at 1.3v set in OC.

When AMD said they didn't need AVX offset ratios like Intel due to power, I don't think they were exactly giving the whole story there. Based on observations, FMA code is still an above average stress on these CPUs, but rather than having a separate multiplier just for that, they manage the whole chip power envelope. So it is still running lower clock than lesser loads, just there isn't a separate knob to set it. Their approach works for normal people, but isn't going to help overclockers unless we move to trying to manage through power and currents more than clock at fixed voltage directly.

 

Follow up with Cinebench R15 testing, results as follows.
Vset Vload Clock Score Max temp
1.30 1.250 4250 2198 65c
1.30 1.250 4275 2209 66c
1.35 1.300 4300 2221 69c
1.35 1.300 4325 2241 70c
1.35 1.300 4350 2246 70c
1.40 1.344 4375 2260 75c
Couldn't stabilise 4400 at set 1.425v and stopped.

Cinebench R15 could run at a higher clock at a given voltage than Prime95. Using the Intel example, I could use roughly some "AVX offset" to control this difference. What I didn't do through my testing was observe the CPU power and currents. If I had, I could try setting those to limit the CPU boost clock to remain with the stability envelope.

 

I also did a suicide run at 1.50v set to see how high I can turn up the clock before it died. Got 4590 MHz and the next step was an instant crash. Assume no stability at this clock, in these conditions.

[WereCat]

If you see 1.5V on idle then you're either experiencing iCUE or new RyzenMaster bug. 

 

Assuming you're measuring with CPUZ or HWInfo. 

[Jurrunio]

I wonder if turning up BCLK until system crashes single core workload, back a bit and then start pushing PBO a viable tuning tactic? Of course otherwise Zen 2 is all about overclocking the memory

[porina]

1 hour ago, WereCat said:

If you see 1.5V on idle then you're either experiencing iCUE or new RyzenMaster bug. 

 

Assuming you're measuring with CPUZ or HWInfo. 

Got more info on that "bug"? Using hwinfo64 primarily, Ryzen master also reports 1.46-something as stock voltage.

1 hour ago, Jurrunio said:

I wonder if turning up BCLK until system crashes single core workload, back a bit and then start pushing PBO a viable tuning tactic? Of course otherwise Zen 2 is all about overclocking the memory

I've yet to explore ram speeds. I have Kingston Predator 4000 B-die kit. Didn't want to work on 3600+B450 reduced down to 3400 where I gave up. Same kit works at XMP3600 on the X370. Haven't tried XMP4000 yet. One step at a time  I want to do 4000 at some point, but the timings are much tighter at 3600 and it also is supposed to be sweet spot for Zen 2.

[WereCat]

30 minutes ago, porina said:

Got more info on that "bug"? Using hwinfo64 primarily, Ryzen master also reports 1.46-something as stock voltage.

I've yet to explore ram speeds. I have Kingston Predator 4000 B-die kit. Didn't want to work on 3600+B450 reduced down to 3400 where I gave up. Same kit works at XMP3600 on the X370. Haven't tried XMP4000 yet. One step at a time  I want to do 4000 at some point, but the timings are much tighter at 3600 and it also is supposed to be sweet spot for Zen 2.

 

[porina]

31 minutes ago, WereCat said:

 

Thanks, but on closer inspection that isn't what I'm seeing. For starters, I don't have iCue on that system. I got it mixed up with NZXT CAM. Also it isn't "stuck" at the higher voltage, I only see that if single core loaded or idle. The voltage is lower when most cores are loaded. I think what I'm seeing is basically AMD overclocking their CPU to try and reach a higher single core clock.

[Whalelicker]

Im stable at 4,350Mhz with 1,4V. Dark rock air cooler.

[Bravo1cc]

There is not point in an all core OC, its make no difference in performance, if your cooling solution is good enough it will get there on its own.

[porina]

47 minutes ago, Whalelicker said:

Im stable at 4,350Mhz with 1,4V. Dark rock air cooler.

Define stable?

39 minutes ago, Bravo1cc said:

There is not point in an all core OC, its make no difference in performance, if your cooling solution is good enough it will get there on its own.

To me it is only of interest for competitive overclocking. I normally run CPUs stock. It is good for the non-overclocker that Ryzen runs closer to its limits than Intel for example, but for enthusiasts the amount of "free" extra performance isn't there...

[Bravo1cc]

3 minutes ago, porina said:

Define stable?

To me it is only of interest for competitive overclocking. I normally run CPUs stock. It is good for the non-overclocker that Ryzen runs closer to its limits than Intel for example, but for enthusiasts the amount of "free" extra performance isn't there...

True but not sure how much you will get, Ryzen 2 parts are mostly maxed out out stock. But i mush agree free performance is always good. I will keep an eye on this thread

[Whalelicker]

1 hour ago, porina said:

Define stable?

To me it is only of interest for competitive overclocking. I normally run CPUs stock. It is good for the non-overclocker that Ryzen runs closer to its limits than Intel for example, but for enthusiasts the amount of "free" extra performance isn't there...

Well, stable for benchmarking, multiple runs of cinebench etc, but 1.4V is definitely not daily voltage.

[Tiago8v]

Hey there... I'm kinda late but...

I was able to OC my 3700x allcore at 4300mhz with 1.3v.

So far, it's super stable on C15, and gaming.

 

My temps are also fine, on idle im around 42ºC with a Cooler Master ML240L RGB.