The Lowest Power-per-Core¹ for Enterprise-class Embedded Systems – Meeting the Standards
Telecom providers are facing a real challenge – they have to continue to enhance their infrastructure, moving from 3G to 4G, to meet demand for higher bandwidth mobile phones with all-time access to pervasive apps like Pandora, Hulu and Facebook. They have to improve bandwidth and capacity efficiently. One way they are meeting this challenge is by choosing platform based on standards like the Advanced Telecommunications Computing Architecture (ATCA). By going with ATCA-compliant platforms, such as those based on AMD Opteron™ 4100 Series embedded processors, telecom providers are looking to improve time-to-market, lower CAPEX costs, and provide improved performance over older systems.
Of course, going with a standards-based approach can have its challenges. For example, ATCA specifies the mechanicals (14 8U high blades) and power (blades at 150W or 200W); meeting those specifications can require some help. For that reason, a new ATCA-compliant blade based on the six-core AMD Opteron 4100 Series processor, at 40W or 65W, provides a very compelling solution that can provide up to 88% improvement in performance-per-watt², up to 66% improvement in memory bandwidth³, and up to 33% improvement in I/O bandwidth⁴ over previous generation embedded AMD processors in the EE power band.
Another key market that can benefit from AMD Embedded Solutions and an industry standard form factor is enterprise storage. There is a tremendous increase in digital content, and need for continuous data protection and disaster recovery. One cost effective standards-based solution is Storage Bridge Bay. The 1U small form factor canister requires low power – 100W or 150W total. A solution based on the new AMD Opteron 4000 Series platform can provide high performance network connectivity and disk throughput. The AMD Opteron 4100 Series processor has two 16-bit lanes of HyperTransport™ 3.0 Technology links for up to 25GB/s peak bandwidth per link between processors and I/O. This is a 33% improvement in I/O bandwidth over the previous embedded AMD processor implementations and helps enterprise storage systems move data quicker. And of course, there’s the reliability of the AMD platform, which is of utmost importance for storage systems.
No matter the market, embedded systems can significantly benefit from multiple AMD 5600 Series chipsets for tremendous I/O connectivity to PCIe® Gen 2 interfaces such as 10 Gigabit Ethernet, Infiniband, FCoE. This data throughput can meet the I/O performance required by today’s systems and be ready for tomorrow’s demands.
And stay tuned, because the AMD Opteron 4000 Series platform’s performance and low power will be compatible with the planned next-generation “Bulldozer” processor and this will be a very exciting product for embedded systems.
Dave Jessel is the manager of AMD’s Embedded Enterprise market development. His postings are his own opinions and may not represent AMD’s positions, strategies or opinions. Links to third party sites are provided for convenience and unless explicitly stated, AMD is not responsible for the contents of such linked sites and no endorsement is implied.
¹ 1 As of June 8, 2010, AMD Opteron™ processor Models 4162 EE and 4164 EE have the lowest known power per core of any server processor, at 5.83W (35W/6 = 5.83W/core). Intel’s L5609 is 10W/core (40W/4 cores). See http://www.intel.com/p/en_US/products/server/processor/xeon5000/specifications.
²Internal testing of AMD Opteron 4100 Model 41GLEE (6c 40W 1.8GHz) versus published results of AMD Opteron 2300 Model 23KS (4c 50W 2GHz)
³AMD Opteron 4100 Processor 2channel DDR3-1333 vs AMD Opteron 2300 series 2 channel DDR2-800
⁴AMD Opteron 4100 Processor HT3 6400MT/s (25.6GB/s) vs AMD Opteron 2300 HT3 4800MT/s (19.2GB/s)
POSTED IN: AMD Opteron, Embedded Systems
TAGS: AMD, AMD Opteron 4000 Series, ATCA, embedded systems
SHARE:
COMMENTS: 1
The 4100 Opterons seem to deliver great performance/power values and are very interesting for small servers in data centers where “Green IT” is one of the most important paradigms nowadays. I am really impressed.
What I personally would really also like to see are six-core Opterons that reach higher clock speeds (a kind of “SE” edition, e.g. something like an Opteron 4186, 4188 or 4190). I think such series could be very interesting for high-end workstations and could be seen as a link between the single-thread performance of high-end desktop processors (like the Phenom II X6 1090T) on the one side and server processors with their well-known extensibility (more than one processor per board, more RAM due to the registered architecture) on the other side.
Obviously, the trend goes to multi-core and, no doubt, a 2P or even 4P socket G34 based Opteron 6100 system would make quite a nice base for a workstation. But there are still tasks which cannot be parallelized very well – even in the scientific or engineering area, either due to the lack of multi-core support in (commercial) applications one needs to work with or due to the fact that there are problems for which parallelization is just impossible.
At least in my working environment, I have very often a mixture between single- and multi-threaded applications on my development machine, so a highly-clocked 12-core system (two six core processors) would be ideal. Therefore I had really appreciated to see a socket C32 six-core processor that comes with the features or at least the same clock speeds that Phenom II X6 1090T offers. (Does anybody know what happend to the Opteron 4186 that was announced originally but doesn’t seem to be on the list anymore? Will it be released later?)
Undoubtedly, “Green IT” is important, especially when it goes about scaling over several thousands of machines on a limited amount of space and with expensive cooling requirements. But there are applications where raw performance is still more important and sometimes I get the impression that the high-end workstation market could be supported a bit more. But maybe this market has gotten just too small in the meantime as the typical desktop machines offer more and more performance (but still have some bottlenecks like 16 GB of RAM as a maximum).
Best Regards
Steffen