Loongson 3A4000 processor real test: 28nm process performance can still be improved by more than 100%

Loongson is the Chinese Academy of Sciences under the computer research and development of independent property rights of domestic processors, has been commercialized by Loongson, and in a few days they will release a new generation of Loongson processor – Loongson 3A4000 series, which is an improved version of the existing Loongson 3A3000 series. According to the previous news, the Loongson 3A4000 processor is still 28nm process, the frequency from the Loongson 3A3000 1.5GHz to 2.0GHz, the architecture upgraded to GS464V, the matching chipset has also been upgraded to the Loongson 7A2000, 28nm process.

So how much performance improvement did the Loongson 3A4000 processor improve? In fact, Loongson official early also announced the performance level of the Loongson 3A4000, Loongson Chief Architect Dr. Hu Weiwu in the journal earlier this year introduced the performance test of The Loongson 3A4000 processor, the test is carried out on the FPGA simulation platform, simulation 3A4000 and 3A3000 main frequency are 1GHz, The memory controller frequencies are 500 MHz, with the following results:

Dragon core 3A4000 processor real test: 28nm process unchanged performance can still be improved by more than 100%

From the results, under the same main frequency, the fixed-point performance of the Loongson 3A4000 is 48.54% higher than that of the 3A3000, the floating-point performance is 43.77%, and the performance improvement in some tests exceeds 100%.

It is important to note that the above tests have not taken into account the 246-bit vector optimization supported by the Loongson 3A4000 processor, and based on performance testing on the X86 platform, the compiler’s automatic quantitative optimization can improve the fixed-point performance of the SPECCPU by about 5% and the floating-point performance by about 15%.

Dragon core 3A4000 processor real test: 28nm process unchanged performance can still be improved by more than 100%

What is the performance level advantage of the Loongson 3A3000? Dr. Hu Weiwu also compared the performance of AMD K10 processors at 1.5GHz, with the same frequency single-core, quad-core performance and K10 basically a level. In other words, the IPC performance of the Loongson 3A4000 processor has exceeded AMD’s K10 processor by about 50%, but the actual frequency of AMD K10 processor is very high, in this regard, there is still a large gap between domestic processors, process performance is the biggest bottleneck.

In addition, the above test is still compared at the same frequency 1.5GHz, in fact, the Loongson 3A4000 uses the 3A3000 28nm process, by optimizing the design to increase the main frequency from 1.5GHz to 2.0GHz. Under the same process conditions, the Loongson 3A4000 has twice as much performance as the 3A3000.

Dr. Hu Weiwu said that on this basis, if the use of more advanced technology to increase the main frequency to 2.5 to 3.0GHz, then the universal processing performance of the Loongson CPU can reach the international mainstream CPU level.

Dragon core 3A4000 processor real test: 28nm process unchanged performance can still be improved by more than 100%

According to the official road map of Loongson, Loongson 3A4000 will be followed by Loongson 3A5000 (the server version is Loongson 3B5000), is still 4 core or 16 core GS464V architecture, but the process upgrade to 12nm, The frequency increase to 2.5GHz will really reach the level available to mainstream desktop processors, which will be around 2020.

Add a Comment

Your email address will not be published. Required fields are marked *