How AMD Bulldozer Performance

"Bulldozer" is the core of AMD's thorough redesign and will be AMD's next-generation high-performance processor technology for the client and server segments, which will increase the core, approximately 50%, performance compared to the Opteron 6100 Series.

As a new generation of processor architecture, AMD "bulldozer" will use 32nm SOI technology, which allows "bulldozer" compared to "Magny-Cours" Opteron processors can increase the core number of 33% without increasing power consumption, Increase the throughput by 50%.

Unlike all previous AMD processors, the "bulldozer" uses a "modular" design. Each "module" contains two processor cores, which is somewhat like a single-core processor with SMT enabled. Each core has its own integer scheduler and four proprietary pipelines, and both cores share a floating-point scheduler and two 128-bit FMAC multiply accumulators.

Turbo Core Full Core Acceleration Technology Turbo Core technology mainly refers to speeding up the clock speed for some workloads that are not fully consumed to the maximum extent. On a variety of different workloads, the use of Turbo Core can increase performance by a maximum of 500 MHz. The most important point is that Turbo Core Acceleration refers to the acceleration of all nuclear power. It is obviously different from some nuclear acceleration technologies. Previous nuclear acceleration technologies may need to close some cores and accelerate only some of them. With Turbo Core technology, it is possible to increase all nuclear power by 500 megahertz. If you shut down some nuclear power plants again, the acceleration will exceed 500 megahertz. At the same time, we have further optimized the memory controller to increase the memory throughput.

In addition to each core exclusive 4 integer computation pipelines, on the floating-point operations, "bulldozer" uses "FlexFP" technology, two cores share a floating-point scheduler and two 128-bit FMAC multiplication accumulators, can be combined Two 64-bit double-precision calculations or four 32-bit single-precision calculations can be completed every clock cycle. If one core does not perform floating-point operations, the other core can use the two 128-bit FMACs to complete 4 double-precision operations or 8 single-precision calculations in one clock cycle, and AMD names it AVX mode. This technology guarantees the "bulldozer" floating-point computing capability, and does not sacrifice performance for "sharing" in high-performance computing.

The new interface and new process bulldozer processor will use Socket AM3+ interface, 942 pins, different from the current 938-pin Socket AM3 interface, its advantage is that it can support DDR3-1600 memory and advanced energy-saving technologies, and AM3+ will be AM3+ will be AMD's last-generation pin-grid array (PGA) package, which will later use a contact grid array (LGA), and will use the LGA AF1 new interface when the Fusion Fusion processor comes, with up to 1591 contacts. DisplayPort 1.2 standard, PCI-E 3.0 specification (32 channels), four channels of memory.

Enhanced Memory Controller AMD first introduced integrated memory controllers eight years ago. Based on AMD's experience in this field and very good technology, it also improved the performance of memory controllers in this generation of products. First of all, the memory controller has been redesigned and improved in terms of efficiency, thus achieving a 30% improvement in memory performance. On the basis of 30% improvement in performance, let the memory support 1600MHz frequency, you can get an extra 20% of performance. Adding two items together can improve the 50% throughput of the memory controller.

Supporting both AVX and SSE instructions FLEX FP is AMD's most innovative floating-point computing technology to date, and each module has a FLEX FP for floating-point operations. If you use traditional 128-bit encoding, it means that each core will have a separate floating-point unit. Compared with other affiliates, the number of implementations performed by AMD is twice as high as the 128-bit encoding. If it is 256-bit AVX encoding, Bulldozer can put two floating-point units together. Therefore, in the 256-bit code execution mode, the number of executions is the same as that of the other company. But Bulldozer has a very big advantage, that is, it can execute 256-bit AVX instructions and SSE instructions at the same time. Friends can not do this, they can only choose one of AVX or SSE, this advantage can make Bulldozer have higher performance in high-performance computing, media codec and some technical operations.

The more advanced power management technology requires only 12% of the total core area for the second integer core in each module, which will only add 5% to the entire core at the chip level. More core, less space, which is obviously beneficial to improve the unit power consumption, unit cost performance.

The amount of power consumption is determined by the number of clocks that are powered on. It depends on how many transistors need to be powered on to perform a common instruction (operation). With the percentage of the maximum clock supply, the Bulldozer has a very good energy performance in both normal and idle states. At the same time, it has been optimized on various energy consumption units and can be turned off in various units. The high energy consumption of high-performance computing is mainly due to floating-point operations, and general application operations are mainly consumed in the execution unit. At the same time, in the idle state, AMD's technology can completely turn off the power supply for those cores that are completely useless. Last year, AMD's products had a major transformation. AMD introduced new slots. Bulldozers launched in 2011 can use 2010 slots. Friends of the business to launch a new platform, while introducing a new slot, which also makes AMD more dominant.