- Meet Radeon RX Vega 56 and RX Vega 64
- Radeon RX Vega power profiles
- Radeon RX Vega: New tech features
- RX Vega Radeon Pack bundles
- Our test system
- Radeon RX Vega: Benchmarks galore
- Next page: Power, heat, noise, clock speeds
- Power, heat, noise, clock speeds
- Radeon RX Vega: The FreeSync variable
- Radeon RX Vega: Buying advice
Radeon RX Vega: New tech features
Few of Radeon RX Vega’s new underlying features come as a surprise, as AMD already pulled back the curtain on the key details during the Vega technical preview at CES in January. We’ll cover the highlights that will most likely make a difference to everyday gamers here, but hit that link for more details, or check out this white paper on Vega’s architecture.
Most interesting might be Vega’s revolutionary new high-bandwidth cache and high-bandwidth cache controller, which creates what Radeon boss Raja Koduri calls “the world’s most scalable GPU memory architecture.”
As we wrote in the technical preview, the high-bandwidth cache replaces the graphics card’s traditional frame buffer, while the cache controller provides fine-grained control over data and supports a whopping 512 terabytes—not gigabytes, terabytes—of virtual address space. Vega’s HBM design can expand graphics memory beyond onboard RAM to a more heterogeneous memory system capable of managing several memory sources at once.
That sounds like a feature best suited to data center workloads—and AMD is indeed rolling out Radeon Instinct cards for machine learning based around the Vega architecture—but a new Radeon Settings feature makes it useful for gamers, too. If you open Radeon Software’s Gaming tab and head to the Global Settings, you’ll see a new “HBCC Memory Segment” section.
That slider lets you allocate a portion of your system’s main RAM to gaming, combining with RX Vega’s 8GB of onboard HBM2 to create a larger memory pool. “The high-bandwidth cache controller will monitor the utilization of bits in local GPU memory and, if needed, move unused bits to the slower system memory space, effectively increasing the size of the GPU’s local memory,” AMD explains.
That sounds awfully intriguing, and it could theoretically prevent stuttering and slowdowns in extremely taxing games. Unfortunately, time constraints prevented us from testing the feature, and to be honest, using more memory than the native 8GB of HBM2 can handle might prove difficult. Color me excited though, especially because AMD says the HBCC can also improve overall memory utilization in games, raising minimum frame rates.
Vega also includes revamped “next-gen compute units” that can perform two 16-bit ops (aka FP16) simultaneously, which isn’t possible in previous AMD GPUs. AMD calls the feature “rapid packed math,” as you can see in the video above. Rapid-packed math wouldn’t work with all aspects of a game, but functions that can utilize it (like some lighting, procedural, and post-processing effects) can essentially be doubled, as demonstrated in February Vega demo where enabling RPM allowed a system to render 1,200,000 hair strands per second, as opposed to 550,000 with RPM disabled.
Far Cry 5 will support rapid packed math. In the video above, Steve McAuley, the game’s 3D technical lead, says the game will run faster, at a higher frame rate, and at a more stable frame rate as well. Wolfenstein II: The New Colossus will also support RPM in some way.
Vega has a couple of other features designed to improve performance by working smarter, not harder. Rendering a scene is a complex process. A new programmable geometry pipeline can use primitive shaders to identify polygons that aren’t visible to the player and cull them quickly, allowing the GPU to start rending the geometry you can actually see faster. Yay, efficiency!
Vega’s pixel engine introduces a “draw stream binning rasterizer” that also improves efficiency and, hence, performance. After the geometry engine performs its (already reduced amount of) work, Vega identifies overlapping pixels that won’t be seen by the user and thus don’t need to be rendered. The GPU then discards those pixels rather than wasting time rendering them. DSBR should also reduce the load on RX Vega’s memory. It’s similar to the tile-based rendering that helped give Nvidia’s GeForce graphics cards a tremendous boost in efficiency starting with the Maxwell architecture, though it remains to be seen whether AMD’s solution is as effective.
And get this: Vega is fully compliant with Microsoft’s DirectX 12 level 12.1 features.
Next page: Radeon Packs and our test system details
AMD Radeon RX Vega 56
AMD Radeon RX Vega 64 (Radeon Pack)
AMD Radeon RX Vega 64 (liquid-cooled)