AMD Comments on Ryzen Core Boosting Controversy and Lack of Manual OC Headroom

[Tsing]

AMD's Ryzen 3000 chips have received a very positive response from the tech community since their launch on Sunday, but their core boosting behavior has generated a bit of controversy. Some reviewers found that their Zen 2-based CPUs were not hitting the boost speeds AMD had advertised.

Senior Technical Marketing Manager Robert Hallock has chimed in on this, stating the chips perform as promised. Hallock believes it's user error, and he may be right: some reviewers (e.g., AnandTech) used outdated firmware for their initial tests and saw superior results after updating their boards with a newer BIOS.

It's going to vary with workloads, but it should be absolutely possible for the CPU to boost up to 4.6 on the 3900X. We have seen some reviewers say their CPU never got there, and I'm confident that this is a configuration issue that can be resolved.

Others have apparently misunderstood AMD's claims for max boost clock. The specification does not imply that all cores will hit that number.

If you're asking whether or not all cores will hit the max boost clock: no. It will not do that, nor have we ever promised or implied that. We've been very clear for 1.5 years that the Precision Boost 2 behavior is a "curve" that tries to get the loaded cores to the highest possible frequency with respect to the aforementioned limits. Even with all cores loaded, the CPU can maintain frequencies that are hundreds of MHz higher than base.

Hallock also mentioned there was little point in manually overclocking the new Ryzen chips because of Precision Boost, which does a lot of the work automatically. The Ryzen 7 3700X might be worth tweaking due to its lower (65W) TDP, however.

The other goal of our engineering effort is to absolutely maximize the performance of the product out of the box. //EDIT: By designing algorithms that extract the maximum silicon performance automatically (e.g. Precision Boost 2) without asking the user to tinker or risk their warranty. So, no, you're not going to see a whole lot of manual OC headroom. That's just performance an average person--who doesn't know how to OC--can't access. Why would we do that? It is not our intent to leave anything on the table.

It's more beneficial to enable PBO, overclock the fabric, overclock the memory. But that's true of Ryzen 2000 Series, too.

 

[Riccochet]

Steve at GN didn't see any, or very little, improvement when comparing different BIOS versions.

 

[Dan_D]

Steve at GN didn't see any, or very little, improvement when comparing different BIOS versions.

You won't. I've actually done the testing and I'll be posting more about that, but the short version is that you won't see any real change in the results between BIOS versions.

 

[Zarathustra]

You won't. I've actually done the testing and I'll be posting more about that, but the short version is that you won't see any real change in the results between BIOS versions.

Which suggests one of two things.

1.) The Agesa is still faulty and needs to be fixed with yet another version
2.) Robert Hallock is full of ****, and AMD are shipping CPU's with binning that won't let them hit advertised clocks.

My expectation is that I should be able to open a retail package, slap a Ryzen 3000 CPU on any officially supported motherboard using the stock cooler and in single threaded tests be able to hit the advertised boost clock at typical room temps. This is the bare minimum expectation.

if I can't, something is wrong.

I'd also LIKE to if I have greater than stock cooling, be able to hit even higher clocks than advertised, but that is not a guarantee, and never has been.

 

[alxlwson]

Which suggests one of two things.

1.) The Agesa is still faulty and needs to be fixed with yet another version
2.) Robert Hallock is full of ****, and AMD are shipping CPU's with binning that won't let them hit advertised clocks.

My expectation is that I should be able to open a retail package, slap a Ryzen 3000 CPU on any officially supported motherboard using the stock cooler and in single threaded tests be able to hit the advertised boost clock at typical room temps. This is the bare minimum expectation.

if I can't, something is wrong.

I'd also LIKE to if I have greater than stock cooling, be able to hit even higher clocks than advertised, but that is not a guarantee, and never has been.

I find this to be extremely reasonable.