The move to multicore processors is boosting system performance to levels that have never been seen outside of highend SMP systems. Better yet, this increase in processing power is happening without the huge increases in price and power requirements that moving to traditional SMP systems entails. For SQL Server systems, added processing power can mean increased levels of system performance as well as higher levels of scalability. Let's take a look at some of the latest developments in the dual-core and quad-core processors that Intel and AMD are bringing to market and see what this technology means for your SQL Server environment.
When manufacturers became unable to improve processing power simply by boosting processor speed, both Intel and AMD realized that the easiest path to more power was through parallelism.
The ever-shrinking size of processors made it possible to produce dual-core chips, which combine two processors on a single die. An added benefit of dual-core chips is that they nearly double the available CPU power while using the same power envelope (i.e., the same wattage requirements) as a single processor. Figure 1 shows the results of running the SAP Sales and Distribution (SD) Users benchmark on AMD Opteron systems that were identically configured except for the processor. The dual-core Opteron 875 provided a 74 percent increase in performance over the single-core Opteron 852.
In 2005, Intel became the first to enter the dual-core market with the release of the Pentium D processor, built using the Intel NetBurst microarchitecture. In January 2006, Intel switched to the Core microarchitecture, which uses a shorter instruction pipeline than does NetBurst, letting processors execute substantially more instructions per clock cycle and achieve higher levels of performance even though they run at a lower clock frequency than earlier Intel CPUs.
The shared front-side bus technology of Intel's dual-core design gives each processor half the bandwidth of the front-side bus. Memory and I/O access operations also share the bus, making the bus speed a critical factor in overall system performance. Intel's latest dual-core processor, the Core 2 Duo, is built using 65 nanometer (nm) technology and integrates both cores on a single die. Each core has 64KB of dedicated L1 cache—a 32KB instruction cache and a 32KB data cache—and both cores share a 4MB L2 cache. The Core 2 Duo has a new power-saving design and a 1066MHz front-side bus. It also supports Intel Extended Memory 64 Technology (EM64T), Intel's 64-bit memory extension, and Intel Virtualization Technology (Intel VT).
Following Intel's lead, AMD introduced the 64-bit dual-core Athlon 64 X2, and later the dual-core Opteron. In AMD's Direct Connect Architecture, each CPU has an integrated memory controller and the HyperTransport bus runs at 1GHz and allows an 8GBps direct connection between the CPUs, I/O, and memory. The AMD Opteron 875 dual-core processor has an L1 cache with 64KB for instructions and 64KB for data, plus a 1MB L2 cache. AMD manufactures its dual-core line using 90nm technology. In February 2007, AMD released new dual-core Opterons that run at clock rates up to 2.8GHz and provide greater power efficiency than earlier models.
The jump from dual-core to quad-core processors delivered another big performance boost. Figure 2 shows the results of some benchmark tests on the Quad-Core Intel Xeon processor X3220 and the Dual-Core Intel Xeon processor 3070. The SPECfp_rate_base2000 benchmark measures floating point performance, the SPECint_rate_base2000 measures integer performance, and the LINPAC measures billions of floating point operations per second. SPECjbb2005 is a Webbased Java benchmark that simulates an order entry system. Although the X3220 runs at a slightly slower clock rate, it outperforms the 3070 in all the benchmarks.
With the release of the Quad-Core Intel Xeon 5300 series in November 2006, Intel is the clear leader in the quad-core race. AMD won't have an entry until mid 2007, when it will introduce a quad-core chip code-named Barcelona. However, Intel's and AMD's quad-core designs are significantly different.
Intel's quad-core design puts two dualcore processors onto a single chip. In other words, instead of being a "native" quadcore processor, Intel's quad-core Xeon is actually a dual dual-core chip. Although this architecture enabled Intel to beat AMD to market, the design isn't optimal. When processors that are on separate cores exchange data, the data must be sent over the front-side bus and through the memory controller, which isn't the most efficient mechanism. In addition, as with previous Intel designs, this approach makes the overall system speed dependent on the speed of the front-side bus. Despite these drawbacks, the additional CPUs and improvements in the Intel Core microarchitecture make Intel's quad-core chips the fastest x64-compatible processors available today.
In contrast, AMD's upcoming Barcelona mounts four independent CPUs on one die. AMD's quad-core chip will utilize the Direct Connect Architecture. Barcelona will be built using a 65nm process technology and will have versions that utilize a 68-, 95-, or 120-watt power envelope. This model enables all four cores to act independently, leading to more efficient power consumption because each core can adjust its frequency according to the workload.
Among other important enhancements, the Barcelona design sports 128-bit floatingpoint processing and a new 2MB L3 cache that's shared by all the processors. Because each processor performs more work per clock cycle, an estimated 15 percent efficiency improvement per core results in an improvement in processor performance of about 40 percent. The AMD quad-design is socket-compatible with existing Socket F dual-core processors. Consequently, existing dual-core systems built with the AMD Socket F can be upgraded to quad-core by performing a CPU swap and then upgrading the BIOS. The scalability of the Barcelona should also be greater than that of Intel's quad-core CPU. Each core on Barcelona's quadcore die could theoretically be upgraded to a dual-core chip in the future, essentially enabling a design that incorporates four dual-core CPUs on one quad-core die.
SQL Server and Multicore
Because it's designed to take full advantage of multiprocessor SMP systems, SQL Server can utilize all of the cores in a multicore system; you don't need to make any system or configuration changes. In addition, Microsoft doesn't charge licensing premiums for multicore processors. The company charges for SQL Server (and all other Microsoft products that are licensed by CPU) according to the number of sockets rather than the number of cores. For example, if you have a 2-way system that's running single-core processors, you need to purchase a license for two processors. But if you later upgrade that 2way system to two dual-core processors, no change in licensing is required because the number of motherboard sockets doesn't change.
The best thing about hardware competition is the price and performance benefits it brings to customers. Intel and AMD's multicore duel brings those benefits in spades by delivering SMP power at single-CPU prices. Table 1 shows a few representative server offerings from HP, Dell, and IBM for dual-core systems, any of which would work well for running SQL Server.
The Future is Multicore
Intel announced its next line of multicore chips, code-named Penryn, last fall and expects to make those products available later this year. The Penryn line of processors will utilize a new 45nm manufacturing technology, enabling Intel to increase processing speed while simultaneously reducing power requirements and heat generation. The move to 45nm manufacturing will give Intel a temporary leg up on AMD in the game of processor leapfrog, but AMD plans to bound back with its own line of 45nm chips for 2008. Look for Intel's next big move in late 2008 with its rumored eight-core processor, code-named Dunnington.