Today at Embedded World in Germany AMD’s embedded group is introducing three new product families as part of its G-Series of Embedded Solutions. First up we have the AMD G-Series I and J families which were codenamed Brown Falcon and Prairie Falcon. These two product families use a new die that has a very similar feature set to the Carrizo-based Merlin Falcon offerings which launched as the AMD R-Series SoCs near the end of last year. But they differ in two important ways: only a maximum of two Excavator cores and four GCN compute units are available on the die. AMD’s third product family is the G-Series LX SoCs which also use a new die and are best described as a cut down version of AMD’s older Kabini chip with only 2 cores and 1 GCN compute unit.
One of the big goals for AMD’s Embedded group is to offer customers the ability to scale their compute needs up and down AMD’s product stack when considering factors like price, performance, and power consumption. To that end all three of the product families that AMD is introducing today will we be pin compatible with one of AMD’s existing FT3 and FP4 BGA packages so that they can be implemented in existing customer’s products without a redesign.
The example that Scott Aylor, the VP and GM of AMD’s Embedded group, offered was of a company that already used AMD’s G-Series SoCs for its high-end and midrange offerings but opted for an embedded ARM chip for its lower end offerings because AMD’s existing offerings couldn’t get down to the power envelope that this customer needed. AMD’s new G-Series offerings are aimed at remedying that problem with new 6 Watt and sub 6 Watts SKUs in the FT3 packaged G-Series LX Family of SoCs and FP4 packaged G-Series J Family of SoCs.
These new embedded products from AMD are exciting for a variety of reasons. The biggest of which is that this is the first time AMD’s Embedded group is launching products with dies that been specifically designed for them rather than appropriating an existing die from one of AMD’s other product groups. When I noted this during my briefing with AMD they pointed out that they needed these new dies because they couldn’t get down to the power envelopes and cost levels they wanted using binned or castrated versions of AMD’s existing chips.
The other important thing this launch highlights is AMD’s level of commitment and willingness to invest in creating products for the embedded market. AMD’s leadership has been talking about investing in growth markets like embedded, enterprise, and semicustom for the better part of four years now. But while we’ve clearly and publicly seen the company invest into the enterprise market with the A1100 Series and the console chips; we’ve only seen AMD’s embedded group offering tailored versions of existing chips that were already launched in the mobile, enterprise, or desktop markets offered a few months later in a BGA package for the embedded market. With these new G-Series SoCs we now know that AMD is willing to treat the embedded market as a first class citizen in its corporate product portfolio by designing chips specifically for it.
At the end of last year AMD launched Merlin Falcon, an embedded version of Carrizo, as the AMD R-Series of embedded SoCs using the FP4 BGA package. Merlin Falcon is the highest performance offering in AMD’s FP4 packaged product stack. AMD’s new G-Series I Family of products codenamed Brown Falcon slot in below Merlin Falcon with 2 Excavator cores, 4 GCN compute units, and the rest of Merlin Falcon’s feature set. Below the I Family is the AMD G-Series J Family Codenamed Prairie Falcon. Brown Falcon and Prairie Falcon use the same die, but Prairie is a harvested version that offers only two GCN Compute Units unlike the four on Brown Falcon and loses the ECC memory support that both Merlin and Brown Falcon offer.
On the other hand, Brown Falcon gains DDR4/DDR3 1866 Memory support which is a step up from the DDR4/DDR3 1600 that Prairie Falcon musters. It also offers higher base and turbo CPU clock frequencies on its top SKU and the J family as a whole appears to address a wider range of power envelopes than the I Family getting down to 6 Watts while the I Family sits up at 15 Watts. One important thing to note is that are still SKUs in both of these families that AMD hasn’t finalized and that will be offered later in the quarter.
Again scalability is one of AMD’s big goals which is why Merlin Falcon, Brown Falcon, and Prairie Falcon all use the same FP4 BGA packing and are pin compatible. They also all support the same software and OS stack with the only compatibility difference present at the BIOS level where Merlin Falcon and Brown Falcon can use the same BIOS presumably because they both have dual channel DDR3/DDR4 controllers and ECC support while Prairie Falcon does not and requires a different BIOS.
The advantage of this level of compatibility for OEMs is that, similar to what we’ve seen with AMDs socket compatible Carrizo and Carrizo-L in the mobile market, with a single chassis and motherboard they can address the same array of price, performance, and power envelopes that would have required multiple chassis and motherboard designs in prior years; or worse yet for AMD, the use of a competitor’s product to meet a given set of design requirements. Thus the advantage of a converged infrastructure for AMD is twofold in lowering the cost of designing products with their chips and in keeping their customers inside of their product stack.
Another aspect of AMD’s vision for these embedded products is to help the company outflank Intel. AMD believes they have an advantage in that they can offer performance levels that are competitive with Intel’s embedded products, as they proved with Merlin Falcon last year, while scaling below them with lower power and lower cost products with the Brown and Prairie Falcon and above them with higher power and consequently higher performance products like Merlin Falcon which is configurable from 12 to 35 Watts. In addition to attacking from above and below AMD is also trying to move its x86 products into markets where Intel doesn’t compete and x86 hasn’t been an option in recent memory with it iTemp SKUs.
As I covered when Merlin Falcon launched, AMD announced that they’ve be offering the first iTemp SKU in Q1 2016. These chips are aimed at aerospace, military, and scientific applications where extreme and quickly changing ambient conditions are the norm and have a maximum temperature operating range of -40 to 105 degrees Celsius. These iTemp chips are a class of product that Intel does not offer. In this market AMD’s modern, HSA compliant, x86-based chips will be competing against ARM and MIPS-based products that are rarely updated and have long life cycles.
Thus iTemp-style offerings are an important part of AMD’s vision for its embedded product portfolio. To that end AMD is preparing a second Brown Falcon-based iTemp SKU in addition to the Merlin Falcon-based version they announced last year. When AMD’s iTemp SKUs finally launch sometime later this quarter AMD will have two options: one with four cores and another with two cores. It’ll be interesting to see what partner designs that AMD showcases with its iTemp SKUs in the coming year.
Also launching today is AMD’s G-Series LX family of SoCs which use a new die that’s basically a Kabini chip cut in half: complete with 2 cores, one GCN compute unit, and a single MB of shared L2 cache. The G-Series LX family rounds out the list of choices that come in AMD’s FT3 package which first launched back in 2013 with AMD 1st Generation of G-Series SoCs which were based on Kabini and was updated in 2014 with a version Beema tailored for the embedded market codenamed Steppe Eagle. AMD’s FT3 packaged products offer one big thing that its FP4 products can’t and that’s lower cost due to the smaller die sizes of Jaguar cored products like Kabini than Excavator cored products like Carrizo which consumes more than double the silicon area of the former.
AMD’s G-Series LX family of embedded SoCs can’t offer much in the way of compute but they also don’t consume much power with a ~6 to 15 Watt TDP range and they do enable access to the x86 software and OS stack which is complemented by a decent selection of I/O options like 4 PCI-E lanes, SPI, USB 2 and 3, SATA 3, 2 display outputs, and ECC Memory. If you want the cheapest x86 cores with decent I/O in an embedded package and a 10-year availability period, then AMD G-Series LX family is your only choice. Unless of course you’re willing to compromising on some of those features and go with one of Intel’s IoT products like Galileo or Edison.
What AMD’s done today with this launch is demonstrate its commitment to the embedded market. From building out the scalability of its high-end product stack from 35 to 6 Watts with the I and J families to updating its existing G-Series line up with lower cost and lower power consumption offerings thanks to the G-Series LX chips AMD’s desire to be a one stop shop for embedded compute is clear.S|A
Latest posts by Thomas Ryan (see all)
- AMD Launches its Radeon Instinct Server Accelerators - Jun 22, 2017
- A Look at AMD’s EPYC Line-up - Jun 21, 2017
- AMD Details EPYC’s Power Efficiency - Jun 20, 2017
- Intel’s Skylake-X and Kabylake-X Meet Our Scale - Jun 19, 2017
- MIPS for Self-Driving Cars with the I6500-F CPU - Jun 19, 2017