AMD Dual Graphics: Hybrid CrossFire, Reloaded
Update (8/16/2013): In light of your requests in the feedback section, we added Catalyst 13.8 beta video results to page nine of this review.
What AMD now calls Dual Graphics was originally referred to as Hybrid CrossFire. From a technical standpoint, the older designation was probably more on-point, since the feature leverages AMD's multi-GPU technology as a means to scale the performance of an APU using discrete graphics.
Simply, that means you're able to take an APU-powered system and add a Radeon card, link them together, and harness the resources of both to push frame rates higher than you'd see from either the APU or add-in card on its own.
The Dual Graphics brand was introduced alongside AMD's Llano-based APUs in 2011, which I had the opportunity to review. While I acknowledged the appeal of Hybrid CrossFire, I experienced a few glitches with the original implementation. At the time, I hoped to revisit Dual Graphics once it matured. Over time, Llano gave way to Trinity, and Trinity was succeeded by Richland.
AMD had three generations to refine its hardware and software. So, we thought it high time to test Dual Graphics more thoroughly. After all, the company is making some pretty bold claims about the feature's potential gains. The slide below comes from the press deck that accompanied Richland's launch:
During the past two years, we've accumulated a handful of questions about Dual Graphics. For example, AMD recommends that you don't imbalance its APUs with anything more than a Radeon HD 6670. And yet, we've heard that the Radeon HD 7750 makes a good accompaniment in a Dual Graphics array. Is it even possible to mix a VLIW5-based APU and a GCN-based add-in card using this technology? If so, do cards faster than the Radeon HD 6670 yield worthwhile results for the extra money you're spending? Are there any Dual Graphics-related limitations you should know about? We're setting out to answer those unknowns.
I also wanted to incorporate our FCAT-based analysis, which measures the dropped and runt (too small to perceive) frames generated by a multi-GPU configuration using video capture. Unfortunately, we can’t get the tool to successfully process video generated by Dual Graphics, and AMD tells us that the issue we're encountering won't be fixed in the foreseeable future. If you remember back to AMD A10-6700 And A10-6800K Review: Richland Hits The Desktop, the problem was that bits of adjacent frames would show up where there weren't supposed to, like this:
See the tear in the image? We aren’t satisfied with this state of affairs, so we found another way to objectively evaluate the effectiveness of Dual Graphics: using the actual video capture we'd normally feed through FCAT to generate data. You're going to be astonished by the dramatic results (at least, we certainly were). We'll talk more about that video demonstration on the next page.