Continuing its expansion of bringing modern processor and graphics designs to as many of its targeted market segments as possible, AMD announced today two new families that address the embedded processor space. The company has already seen double-digital YoY sequential growth in revenue from embedded markets, but the release of the Epyc Embedded 3000 and Ryzen Embedded V1000 family create significant additional opportunity for the company.
Embedded markets are unique from traditional consumer and enterprise channels as they address areas from military and aerospace applications to networking hardware and storage devices to retail compute and even casino and arcade gaming. These markets tend to be consistent and stable without the frequent or dramatic swings in architectural preference or market share that we often witness in consumer PCs. As AMD continues to grow and look for stable sources of adjacent income, embedded processors are a critical avenue and one that I believe AMD has distinct advantages in.
Research firm IDC estimates the market size that AMD can address with this pair of chip families exceeds $14-15B annually. The largest portion of that ($11-12B) includes storage and networking infrastructure systems that the Epyc 3000 line will target. The remaining amount includes IoT gateways, medical systems, and casino gaming hardware and is the purview of the Ryzen V1000.
Competitors in this space include Intel (with its Xeon D-series and Core family of chips) and many Arm-based designs that focus on low power integration. Intel has the most potential for immediate negative impact with AMD’s expansion in the embedded markets as the shared architecture and compatibility mean customers can more easily move between platforms. AMD is positioning both parts directly against Intel with proposed advantages in value and performance, hoping to move embedded customers to the combined AMD solution.
The Ryzen V1000 family combines the company’s recent processor and graphics architectures on a single chip, similar in function to the consumer Ryzen design that was released for notebook and desktop PCs. For the embedded customers and devices being targeted, this marks a completely new class of product with two key benefits over competing solutions. First, it allows for smaller and cooler system designs (critical for the cramped working environments of the embedded space) while increasing maximum performance.
Second, the V1000 allows integrators to downscale from using a combination of an Intel processor and a separate, discrete graphics chip to a single chip design. This both raises the ASP (average selling price) for AMD, increasing revenue and potential margin, while lowering the price that customers pay in total for system components.
While AMD struggles to find ways to promote the value of higher performance graphics on its new processors, where it has a significant advantage over Intel, for the consumer and business space, in the embedded markets that additional performance value is well understood. Casino gaming often utilizes multiple high-resolution displays for a single device with demand for high-quality rendered 3D graphics, of which the V1000 can now provide in a single chip design. The same is seen with medical imaging hardware, including ultrasound machines for women’s healthcare and cardiovascular diagnostics.
The Epyc Embedded 3000 family does not include integrated graphics on-chip and instead offers higher core performance and performance per dollar compared to competing Intel solutions. AMD believes that the Epyc 3000 will double the total addressable market for the company when it comes to networking and storage infrastructure.
AMD previously has disclosed its partnership with Cisco that included AMD-built processor options for some families of switches and other networking gear. As the demand for edge computing grows (systems that will exist near the consumer or enterprise side of a network to aid in computational needs of high speed networks), AMD is offering a compelling solution to counter the Intel Xeon family of processors.
Both the Epyc 3000 and Ryzen V1000 chips represent the first time AMD has targeted embedded customers with specific features and capabilities at the hardware level. During the design phase of its Zen CPU and Vega graphics architecture, business unit leaders included capabilities like multiple 10-gigabit network integration, support of four 4K display outputs, ECC memory (error correction capability for mission-critical applications), and unique embedded-based interfaces for external connectivity.
While these were not needed for the consumer segments of the market, and weren’t exposed in those hardware launches, they provide crucial benefits for AMD customers when selecting a chip for embedded markets.
This is very different from
This is very different from what normally goes into the embedded category.
Compare the I/O to something like the i.MX series.
Curious as to what package
Curious as to what package the Epyc 3000 chips are in. That looks like only 2 zeppelin die.
That’s not a real die shot
That’s not a real die shot but will some variant of the Epyc socket SP3, but AMD could have made a 16 core and 4 CCX unit on a single die variant or just crammed 2 Zeppelin dies closer on a redesigned MCM(Most probable) for the embedded Epyc market. AMD has now had the time to do up many variants while they where/are also working on zen+ and Zen2(design already frozen/certifing engineering processes ongoing) with Zen3 in active development.
Maybe some other website has more of the press kit information released but AMD has plenty of options going forward once it has Navi and some modular Navi GPU DIEs that are the same as those Zen Zeppelin Dies that can be scaled up and “glued” up via the Infinity Fabrric to make increasingly more powerful computing(graphics also once Navi arrives) processors/platforms without having to use a brand new tapeout to get more processor performance. Navi is more about getting a Navi Scalable GPU die design/s more so than any radical Navi GPU micro-arch changes over the Vega GPU micro-arch.
The reason that the Vega 10 base die tapeout was not so good for gaming was not because of any Vega GPU micro-arch shortfalls it was because of that Vega 10 base die tapout had to be designed to do double duty for professional compute/AI in addition to gaming. And AMD lacked the funds to quickly workup a GTX1080Ti competitor. So once Navi arrives with a Navi mdular GPU die design/design variants AMD will be able to scale up GPU’s Like AMD scaled up Zen/Zeppelin without requiring any time intensive and costly new die tapouts. The miners sure love Vega(Polaris also) and the only reason that Nvidia’s GPU became wanted for mining also is because the supplies of the Vega desktop GPU variants dried up(even drier) like Zevulon the Great.
Leela: Zevulon the Great, He’s teriyaki-style!
Yes, it’s a 2-die MCM on a
Yes, it’s a 2-die MCM on a much smaller package than server EPYC.
They’re unlikely to be
They’re unlikely to be socketable, embedded systems normally use BGA.
super informative write up on
super informative write up on embedded processors and where they are used and very exciting news for amd
thanks ryan