AMD announced its Linux-ready “Ryzen Embedded V1000” SoCs with up to 4x cores and 3.6 TFLOPS, designed for mid-range embedded devices, plus an up to 16-core, Xeon-like “Epyc Embedded 3000” for embedded servers.

A few weeks after information leaked about several new boards based on AMD’s new Ryzen Embedded V1000, AMD formally introduced the new SoC family, which is notable for its high-end Vega graphics and for offering twice the performance of its earlier R-Series. The chipmaker also launched the Epyc Embedded 3000 — a lower-end version of its Epyc 7000 server chip comparable with Intel’s Xeon-D. Both processor families aim to match the current Intel state of the art while offering significantly lower prices.

Both the Ryzen Embedded V1000 and Epyc Embedded 3000 are based on the same 14nm FinFET fabricated Zen cores that launched last year on AMD’s server-oriented Epyc 7000, as well as its desktop-focused Ryzen and gaming-centric Ryzen ThreadRipper processors. The V1000 borrows the Ryzen name due to its use of the same high-end Vega GPU core that drives the Ryzen. Last month at CES, AMD announced 2nd Gen models of the Ryzen and Threadripper, as well as a 7nm version of the Vega GPU.

AMD’s new Ryzen Embedded V1000 and Epyc Embedded 3000 SoCs
(click images to enlarge)

The dual- and quad-core Ryzen Embedded V1000 (“Great Horned Owl”) is the successor to the R-Series, which was last updated in 2015 to a “Merlin Falcon” release. In its V1000 announcement, AMD claims the SoC provides up to twice the performance of Merlin Falcon. AMD’s benchmarks compare the V1000 with dual- and quad-core Intel Core chips from Intel’s 7th Gen “Kaby Lake” family, as well as Intel’s latest, 14nm 8th Gen Kaby Lake Refresh chips, also known as “Coffee Lake.”

We saw no details about OS support for either the V1000 or Epyc Embedded 3000, but they would appear to support the usual x86 ready platforms. Ubuntu Linux was used for the V1000 benchmarks, and Microsoft Windows was tapped for the Epyc Embedded 3000 benchmarks. In addition, long time AMD partner Mentor announced that Mentor Embedded Linux would support both SoC families.

In addition to the shared Zen core, the Ryzen Embedded V1000 and the Epyc Embedded 3000 share security features including an on-chip secure processor, complemented by hardware validated boot capabilities. They both offer Secure Memory Encryption (SME) for deterring unauthorized physical memory access, as well as Secure Encrypted Virtualization (SEV), which encrypts virtual machine (VM) memory “without the need to make changes at the application level.”

AMD said nothing about the lower-end “Banded Kestrel” SoCs that were leaked two years ago by AMD along with the Great Horned Owl, and which are expected to compete more directly with Intel’s Atom SoCs. The 2016 leaked documents framed Banded Kestrel as the successor to the G-Series “Steppe Eagle”.

Banded Kestrel was described as a dual core, quad-thread SoC with 1MB and 4MB of L2 and L3, respectively. The SoC family was projected to have a TDP range of 4-15W and offer a similar Vega GPU an RAM support as the V1000.

In this report we’ll focus primarily on the V1000 which we imagine will appear far more frequently on LinuxGizmos than the Epyc Embedded 3000. (See farther below for a briefer summary of the Epyc.)

 
Ryzen Embedded V1000

AMD’s Ryzen Embedded V1000 is designed for embedded applications including “digital casino gaming, medical displays, thin clients, and industrial PCs,” says AMD. The SoC combines 2x or 4x 14nm fabricated Zen single or multi-threaded cores along with a Vega GPU with up to 11 GPU compute units on a single die. V1000 SoCs provide up to 2MB of shared L2 plus 4MB of shared L3.

The Ryzen Embedded V1000 offers up to 3.6 TFLOPS performance, or 2x the performance of the R-Series Merlin Falcon generation, claims AMD. The SoC is also claimed to deliver an up to 52 percent IPC boost at the CPU and up to a 107 percent improvement in GPU performance.

TDPs range from 12W to 54W, compared to 12W to 35W for the dual- or quad-core, single threaded Merlin Falcon. The SoC supports dual-channel 64-bit DDR4 RAM with performance up to 3200 MT/s, claims the chipmaker.

The Ryzen Embedded V1000 is also claimed to provide an up to 26 percent smaller footprint than Intel’s 7th Gen “Kaby Lake” based Intel Core i7-7700HQ, while delivering up to 3x the graphics performance. The quad-core, octa-threaded i7-7700HQ is clocked at 2.8GHz/3.8GHz with Intel HD Graphics 630 and a 45W TDP.

The V1000 is said to provide 46 percent more multi-threaded performance than the Intel Core i3 -7100U. This recently announced “Coffee Lake” entry provides 2x 14nm cores (4x threads) clocked to 2.20/3.40GHz, plus up to 1GHz Intel UHD Graphics 620 and a 15W TDP.

The Vega graphics GPU’s is the V1000’s major advantage. The GPU can drive four independent displays running in 4K, and can also deliver up to 5K graphics. It provides 10-bit decoding with VP9 and H.265, as well as 8-bit encoding, and supports High Dynamic Range (HDR) displays.

Ryzen Embedded V1000 block diagrams from AMD (left) and Congatec
(click images to enlarge)

In its announcement for its V1000 based Conga-TR4 COM Express Type 6 module, Congatec touted the SoC’s balance of relatively low TDP with high-end graphics power. “High performance 4k UHD system designs, which previously ran at 54W, now require less than half the TDP without loss of graphics performance,” stated Martin Danzer, Director of Product Management at Congatec. “As a result, active cooling solutions can be replaced by passive ones, with all the corresponding SWaP-C benefits. The new 15W mobile system designs can deliver truly impressive user experiences with brilliant 4k displays and outstanding 3D graphics quality.”

AMD is currently offering four V1000 models: the dual-core V1202B and three quad-core, octa-threaded parts. The quad-core V1807B and V1756B are very much alike, with 35-54W TDPs and 3.35/3.8GHz and 3.25/3.6GHz CPU clock rates, respectively. The V1807B has an 11 compute unit Vega GPU, however, compared to 8x CUs for the V1756B.

Ryzen Embedded V1000 models
(click image to enlarge)

The quad-core V1605B starts at a low 2.0GHz clock rate, but bursts to 3.6GHz, with 12-25W TDP. The dual-core, quad-thread V1202B (2.3GHz to 3.2GHz) has the same TDP range. It’s limited to a 3-CU GPU, however, and has only 1MB cache instead of 2MB on the quad models. While the quad-core parts support dual 10GbE ports, the V1202B supports only dual 1GbE ports.

In addition to the dual Ethernet capability, the V1000 is notable for supporting NVMe, as well as up to 16 PCIe lanes. I/O support is listed as:

• eMMC5.0, SD3, or LPC
• Up to 2x SATA ports
• NVMe support
• Up to 4x DisplayPort 1.46, HDMI 2.0b, or eDP 1.4
• 2x 10 Gigabit Ethernet (or 10GbE)
• Up to 4x USB 3.1 (10Gbps)
• 2x Type-C with ALT.DP power delivery capable
• 1x USB 3.1 (5Gbps)
• 1x USB 2.0
• 2x UART
• 4x I2C
• 2x SMBus
• SPI/eSPI, I2S/HDA/SW, GPIO
• Up to 16 lanes of PCIe Gen3 (8x GFX, 8x GPP) and 7-link max

Advantech SOM-5871

AMD listed the following products that integrate the Ryzen Embedded V1000:

• Advantech integrated casino gaming platform, multimedia gaming engine, mini-ITX board, and COM-Express module. (The COM Express is the SOM-5871 we covered on Feb. 15.)
• Esaote MyLab9 eXP ultrasound system for general medical imaging, women’s healthcare and cardiovascular diagnostics (Q3 availability).
• iBase MI988 Mini-ITX motherboard and SI-324 4x HDMI 2.0 digital signage player (both briefly covered in our Feb. 15 V1000 coverage
• Quixant QX-70 4K Ultra HD casino gaming platform (available today).

Other V1000 based products have been announced by Arrow Electronics Intelligent Systems, Congatec, QTechnology, Sintrones, Seco, and SMACH Z. We will be covering many of these in the coming weeks.

 
Epyc Embedded 3000

The edge server focused Epyc Embedded 3000 offers 4x to 16x cores available in single or multi-threading models. The multi-threaded models provide 8x to 32x threads, respectively. TDPs range from 30W to 100W. The headless chips are designed for lower-end server applications relative to the Epyc 7000, including network functions virtualization, software-defined networking, and networked storage devices.

Epyc Embedded 3000 models
(click image to enlarge)

Benchmarks compare the Epyc Embedded 3000 with the 5th Gen Broadwell based, 16-core Intel Xeon D1587. The Epyc Embedded 3000 is claimed to offer up to 2.7x more “performance-per dollar” than the Xeon, as well as 2x more “connectivity.” By connectivity, they mean that the high-end Epyc Embedded 3451 model supports up to 64 PCI Express high speed I/O lanes, 8x 10GbE ports, 16x SATA ports, and 4x memory channels via 32MB shared L3 cache. By comparison, the Xeon D-1587 supplies 32x PCIe lanes, 4x 10GbE, 6x SATA, and 2x memory channels.

The newer, Skylake based Xeon D-2191 would fare better in that comparison, but is far more expensive. According to HotHardware, AMD has pointed out the price comparison between the ($2,407 Xeon D-2191 and the Eypc 3451 ($880) and Eypc 3301 ($438), which “are not only cheaper, they offer more cores, cache, PCIe lanes, and memory channels, while matching boost clocks and exceeding base clocks.”

The Epyc Embedded 3000 is touted for its “unparalleled enterprise-grade” reliability, availability and serviceability (RAS). The processor provides data detection, correction, recovery and containment, “helping ensure that systems are continuously running even under the most stringent enterprise environments,” says AMD.

The only Epyc Embedded 3000 based product mentioned by AMD is the iBase FWA8800 network appliance.

 
Further information

The AMD Ryzen Embedded V1000 and AMD Epyc Embedded 3000 are available now and are already shipping in at least one product (V1000 based Quixant QX-70). More information may be found in AMD’s announcement, as well as the Ryzen Embedded V1000 and Epyc Embedded 3000 product pages.