Core technology of Turbo Frequency: It mainly refers to speeding up the clock speed for some workloads that are not fully consumed to the maximum extent. In various workloads, using TurboCore can improve performance by up to 500 MHz. Most importantly, TurboCore acceleration refers to the acceleration of all cores, which is obviously different from some nuclear acceleration technologies. In the past, some cores may need to be shut down and only some cores need to be accelerated. Using TurboCore technology, all cores can accelerate by 500 MHz at most, and if some cores are turned off, the acceleration will exceed 500 MHz. At the same time, we further optimized the memory controller and improved the memory throughput.
Except that each core has four integer calculation pipelines, in floating-point operation, "Bulldozer" adopts "FlexFP" technology, and two cores * * * share a floating-point scheduler and two 128-bit FMAC multiplication accumulators, which can be used in combination. Each clock cycle can complete two 64-bit double-precision calculations or four 32-bit single-precision calculations. If one core does not perform floating-point operations, the other core can occupy these two 128-bit FMAC and complete four double-precision operations or eight single-precision calculations in one clock cycle, which AMD named AVX mode. This technology ensures the floating-point computing ability of bulldozer, and will not sacrifice performance because of "* * *" in high-performance computing.
Second, new interface technology.
Bulldozer processor will adopt 94 1 pin SocketAM3+ interface, which is different from 938-pin SocketAM3 interface. Its advantage is that it can support DDR3- 1866 memory and advanced energy-saving technology, and AM3+ will be the last generation of pin grid array (PGA) package of AMD, and then it will switch to contact grid array (LGA). Fusion processor will use the new interface of LGAAF 1, with as many contacts as 159 1, support DisplayPort 1.2 standard, PCI-E 3.0 specification (32 channels) and four channels of memory.
Third, the enhanced memory controller.
AMD takes the lead in introducing integrated memory controller. According to AMD's experience and excellent technology in this field, the performance of memory controller has been comprehensively improved in this generation of products. First, the efficiency of the memory controller has been redesigned and improved, so the memory performance has been improved by 30%. On the basis of improving the performance by 30%, the memory can support the frequency of 1600MHz, and an additional 20% performance can be obtained. Together, these two items can increase the throughput of the memory controller by 50%.
4. Support AVX instruction and Shanghai Stock Exchange instruction.
FLEXFP is AMD's most innovative floating-point computing technology so far, and each module has a FLEX FP for floating-point computing. If the traditional 128 bit encoding is used, this means that each core will have an independent floating-point arithmetic unit. Compared with Shang You, under the premise of 128 bit encoding, AMD's performance will be doubled. If it is 256-bit AVX coding, bulldozers can execute two floating-point arithmetic units together. Therefore, in the 256-bit encoding execution mode, the number of executions is the same as that of friends. But bulldozer has a great advantage, that is, it can execute 256-bit AVX instruction and SSE instruction at the same time. Friends can't do this. They can only choose between AVX and Shanghai Stock Exchange. This advantage can make bulldozers perform better in high-performance computing, media coding and decoding and some technical operations.
Five, more advanced power management technology
The circuit required by the second integer core in each module only accounts for 12% of the total core area, which will only increase the circuit of the whole core by 5% from the chip level. More cores and less space are obviously beneficial to improve the performance of unit power consumption and unit cost.
The energy consumption is determined by the number of clocks that are energized and depends on how many transistors need to be energized to execute a common instruction (operation). Bulldozers have very good energy consumption performance in the maximum clock power supply percentage, normal application state and idle state. At the same time, each energy consumption unit is optimized, and each unit can turn off the power supply. The high energy consumption of high-performance operations is mainly due to floating-point operations, while general application operations are mainly consumed in execution units. At the same time, in the idle state, AMD technology can completely turn off the power of those cores that are completely unnecessary. AMD products have undergone a major transformation. AMD has introduced a new slot, and the bulldozer introduced by 20 1 1 can use the slot of 20 10. In order to launch a new platform, Youshang also launched a new slot, which also gives AMD an advantage.
Summary: The key improvements of bulldozer include power consumption/core area efficiency, new ISA support, core expansibility and modular design concept, which is expected to significantly improve the performance per unit power consumption and area.