Thread Rating:
  • 0 Vote(s) - 0 Average
  • 1
  • 2
  • 3
  • 4
  • 5
Ryzen APU Thread
Let's kick it off:
Quote:While "Summit Ridge" is the combination of two "Zen" CCX (quad-core CPU complex) units making up an 8-core CPU die that lacks integrated graphics, the "Raven Ridge" silicon combines one "Zen" CCX with an integrated graphics core based on the "Vega" architecture. AMD's new Infinity Fabric interconnect ferries data between the CCX and the iGPU, and not an internal PCIe link. The CCX houses four "Zen" CPU cores with 64 KB of L1I cache, 32 KB of L1D cache, 512 KB of dedicated L2 cache, and 4 MB of L3 cache shared between the four cores.

The integrated graphics core is a different beast. It features similar (albeit scaled-down) front-end and back-ends from the "Vega 10" silicon, a similar video engine, and an SIMD area with 10 "Vega" next-gen compute units (NGCUs). This works out to a stream processor count of 640. Other key specifications include 40 TMUs, and 16 ROPs.

The video engine is now extremely capable, supporting hardware-accelerated decoding of CODECs such as VP9 10-bpc and HEVC 10-bpc at frame-rates of up to 240 for 1080p, and 60 for 4K UHD. It can also encode H.265 8-bpc at frame-rates of up to 120 at 1080p, and 30 at 4K UHD. You finally get to use the display connectors on your socket AM4 motherboards, as the iGPU supports DisplayPort 1.4 and HDMI 2.0b, with resolutions of up to 3840 x 2160 @ 60 Hz with HDR, 1440p @ 144 Hz, and 1080p @ 240 Hz.
AMD seems so have increased the amount of power-gating on its silicon. Disabled or idling components triggered by lower power-states, are now power-gated (their power-supply cut off), and not clock-gated (their clock cadence cut-off). The chip is peppered with multiple LDO (low-dropout regulator) regions for the CCX, iGPU, and uncore regions, with a common VDD package rail for both the off-chip (on motherboard) and on-chip voltage controllers.
Valve hater, Nintendo hater, Microsoft defender, AMD hater, Google Fiber hater, 4K lover.
Quote:One chip that isn't on the road map that MoePC got ahold of is an as-yet-unknown AMD "Fenghuang" APU with "15FF" graphics. This product found its way into the SiSoft results database recently. The 15FF IGP apparently has 28 Radeon compute units for a total of 1792 shader processors, accompanied by 2 GB of an unknown type of VRAM. Other data that made its way into the database would appear to be spurious, like the 555-MHz clock speed, 16kB of L2 cache, and a supposed 32-bit path to memory. There isn't much in the way of details about this chip, but it definitely appears to be a prototype. Still, if it is real, Fenghuang's IGP would represent a considerable increase in graphics resources over today's Vega 8 and Vega 10 IGPs. The power needed to support that much graphics horsepower could peg this chip as a desktop part, but we won't know either way until AMD offers more details, if it ever does.
Valve hater, Nintendo hater, Microsoft defender, AMD hater, Google Fiber hater, 4K lover.
They've been announced:
Valve hater, Nintendo hater, Microsoft defender, AMD hater, Google Fiber hater, 4K lover.

Forum Jump:

Users browsing this thread: 1 Guest(s)