Eric Demers and GCN, Part II



Company: AMD
Author: James Prior
Date: December 22nd, 2011

ZeroCore Power

R3D: The low power states are very impressive, I think it's going to be really, really good to have the zero power and idle timeout.

Eric: I think my idle timeout at home is set to 4 minutes, basically if I just run out of the room I don’t want it kicking in; otherwise there's no time [to resume] you just have to hit the keyboard. I think being able to go to a couple of watts is a huge win. I'll be honest with you, Windows is not the best at the whole power management thing, like going into an S3 sleep which is what you really should do if you're gone for 10 minutes. We're taking the GPU out of the equation, we think that's awesome. For a lot of people it's inconsequential, who look at peak performance, but real world it's actually tangible and measurable that it will reduced the heat, it will reduce the noise, it will reduce the power consumption. I think it's great technology.

Mobility technology in desktop GPU's

Eric: It's an example of technology we brought from the notebook segment into the desktop; we take a lot of things the other way, right? This is something we brought up from that side, power domains and those ideas.

R3D: It's based on the Bus Alive, Chip Off premise in the mobility chips?

Tom: That's very much it, it's exactly that.

Eric: [laughing] It's exactly it, instead of restricting it to parts that would only be using mobile why not put it everywhere.

Tom: We initially put it into parts that customers had the highest value in that feature, and it's very successful so we made it a general feature.

Eric: Mobile is a little bit different, there are really three power modes for this guy [Tahiti]; it's off/long idle [<3W], light use which is teen to twenty watt kinda range; and heavy use. Mobile has a lot more variation there, because there is idle with display but you're not doing rendering so you want to be turned down; it's got a lot more depth, beyond the flexibility of this guy. I think we leveraged the things that make sense from that realm back into the higher end GPU.

Eric:I read on Rage3D, there was one guy asking 'when will Crossfire support zero power'..? I never commented on that,

R3D: I know! Watching the presentation I was immediately suspicious! Who is this guy?

Eric: Not me!

Tom:[laughing]

R3D: That's a plant, that's gotta be![laughing]

Eric: He brought up 'why aren't they turning off the second GPU' and actually it does go to idle range [ULPS], so I think it's about 20W right now [Cypress/Cayman], which is pretty low.

R3D: I thought I was being pretty smart, because I responded 'they would need to use BACO, like in the mobility parts'

Eric: And now we'll do that, it did require a hardware change to get that working, we needed voltage islands and a lot of software changes.

R3D: Was it a fairly involved process putting it in there?

Eric: Oh Gosh, Yes! Honestly, so much work on hardware and software. Voltage islands and getting everything right. Surprisingly difficult, even though it's a simple concept. I mean nowhere near desiging a new shader, but surprisingly difficult these kinds of system level changes.

The testing continues

If you made it through all that, then congratulations! We'll follow up with our performance analysis and review shortly, a few last minute problems prevented us from getting it finished for the launch day itself. Thanks to all at AMD who provided us with the resources and access for covering their new generation of Graphics boards.