Ryzen 3000/5000 Overclocks

[JD]

Since a bunch of us seem to have Ryzen 3000 chips now, wondering what you guys are running them at?

For now I've settled on:
CCD0/CCX0: 4450
CCD0/CCX1: 4400
CCD1/CCX0: 4400
CCD1/CCX1: 4300
Voltage: 1.3625
LLC: "Turbo"

My CB20 score at stock is ~7100, with the OC I can get ~7700.

CPU-Z @ Stock:


CPU-Z @ OC:


Which goes to show that it doesn't seem to hold it's boost at stock long enough to make an impact. Technically I'm down 100-150MHz on the fastest cores with the manual OC, but it still yields a slightly higher single thread score and obviously a much high multithread score since the stock all core usually tops out at 4.1GHz.

I've read posts saying the total opposite though - that by doing a manual OC, you'll hurt single threaded performance since it won't be able to boost as high as it might have been able to. From my standpoint though, it looks like it's still worthwhile to do things manually.

Gigabyte seems to release BIOS versions at a weekly pace, which I suppose is partially good, but also partially bad since that means there's a lot of bugs they have to keep resolving. I'm on the 3rd release of their 1.0.0.4 AGESA BIOS's, though I skipped the F10d (beta) that has CCX clocking in the BIOS supposedly. I'm on F10 now though, which supposedly non-letter ending versions are "final" in Gigabytes realm.

I'm still thinking there's more performance to squeeze out of this platform though, between AMD updates and BIOS updates to match, along with dialing in the OC a bit more...

 

[lowfat]

After upgrading to 1.0.0.3 ABBA I can't get the overclock I once had. I'm not sure if it was changes to the bios or the fact that I lost all my bios settings and can't get the tweaking right. Before I had CCX1 @ 4.4, CCX2 @ 4.375, CCX3 @ 4.325, and CCX4@4.275. 1.306V.

But now I can't get anywhere near that. It pretty much insta-reboots if I started rendering. So currently I'm @


I've ran about 4000 loops of Intel Burn Very High, so I think its pretty stable. I can't really run higher than 1.31V, else I'm pretty much bouncing off 95C under stressfull loads.

 

[JD]

After upgrading to 1.0.0.3 ABBA I can't get the overclock I once had. I'm not sure if it was changes to the bios or the fact that I lost all my bios settings and can't get the tweaking right. Before I had CCX1 @ 4.4, CCX2 @ 4.375, CCX3 @ 4.325, and CCX4@4.275. 1.306V.

Yeah I've seen a lot of that too, people getting lower clocks and less performance as these AGESA updates come out, when they're supposed to be increasing clocks.

Though I came across one guy that claimed to be able to run 4.525 on all cores at like 1.2V... he must have won the lottery.

 

[gingerbee]

Stock but since the last agesa update my chips does hit the 4.6 on some cores but most of the time it's maxing out at 4.5/4.4

 

[CMetaphor]

I'm considering getting a Ryzen system in the future but this seems pretty damn complicated compared to regular OCing. Is there a tutorial somewhere? Sorta looks like custom P States or something... Interesting though.

 

[gingerbee]

I don't think there is its just overlocking each CCX by them selfs is all just think that each CCX is a cpu and go from there. hell, I am not sure you can control each CCX voltage by itself lowfat/JD would better answer that ???. in this instant I think RM is your friend

 

[Marzipan]

AMD overclocking is no different than Intel...perhaps different terms used for the same concept / feature, but they're the same. well there is a difference, each have their own nuances to play and fiddle with.

simplest is just to boost the multiplier though.

 

[JD]

Best to clock each CCX individually as they vary in "quality", so some will clock better than others. Voltage is universal across the total chip, you are setting the "peak" voltage in Ryzen Master that you want to allow. At stock, for me it would hit 1.4V, so my OC is actually down 500mV (1.35V).

AMD overclocking is entirely different from Intel. There's no concept of CCX or CCD's in Intel products. Just "raising the multiplier" isn't a thing here.

 

[CMetaphor]

Ya, raising the Multi or base freq is what I'm used to. And while I know certain cores can be better than others, since I don't bother assigning specific processes to specific cores, I'd rather have all cores hit the same OC mark. Just simpler for me I guess? Haha. I'll be pretty happy if I can get all cores of a future 3700X/3800X to 4ghz or so, any more than probably isn't super realistic since the build is going to be ITX and have very limited space.

Thinking another way though: do you have a specific use case where getting certain cores way higher than others is very beneficial? Like say, get Core 0 (ccx?) To 4.4 and the rest to 4.0, but then force most single-threaded apps to use that specific core? Or is it mostly a benchmarking thing, getting the highest scores possible but getting the most out of every single core?

 

[JD]

With the latest Windows scheduler changes (in 1909) along with the Ryzen Power Plan, it should prefer the "faster" cores and that generally appears to be the case. Usually my CCD1 is sleeping (6 cores) in normal usage as they are the slower bunch.

So yes, having my CCX0 on CCD0 be able to run at 4.45GHz should mean most lightly threaded apps will go to those 3 "fast" cores.

Overclocking based on the lowest common denominator is the old way of thinking, and I would say that's mostly what you're doing with Intel today. I don't believe Intel offers any way to OC cores separately.