Not long ago, MC shared with you the detailed technical analysis of the third generation of AMD EPYC (Xiaolong) processor (hereinafter referred to as the third generation AMD EPYC processor), and it can be seen that the new Zen 3 architecture has made the floating-point performance, branch prediction ability, and security of this generation of processors have been significantly improved. Through architectural improvements, AMD claims that the IPC of the third-generation AMD EPYC processor is an increase of up to 19% over the previous generation. So how much has the third-generation AMD EPYC processor improved performance in real-world applications? With the strong support of AMD, we have conducted exclusive tests specifically for the 3rd generation of AMD EPYC processors.
<h1 class="pgc-h-arrow-right" data-track="3" > the reference processor specifications</h1>
MC this test object is AMD EPYC 7763, EPYC 7713, EPYC 75F3 these three processors, let's first look at their specifications. The EPYC 7763 is the most comprehensive performance of the third generation OF AMD EPYC processors, it features a 64-core 128-thread design with a reference frequency of 2.45GHz, a Boost frequency of up to 3.5GHz, and a TDP of 280W. EPYC 7713 also adopts a 64-core 128-thread design, and its frequency setting is 2.0GHz ~ 3.675GHz, compared to EPYC 7763, the base frequency of the former is slightly lower, and its Boost frequency is higher.
In addition, the EPYC 7713 has a TDP of 225W, which is lower than the EPYC 7763, so the former is more suitable for applications that require stronger single-core computing performance and lower power consumption. The third generation of AMD EPYC processors contains four products named with "F", the main purpose of this naming method is to highlight their higher single-core Boost frequency, of which the EPYC 75F3 has the strongest overall performance. It adopts a 32-core 64-thread design, the base frequency is 2.95GHz, the Boost frequency is as high as 4.0GHz, and the TDP is 280W, which is more suitable for applications in some scenarios with high processor frequency requirements.
▲MC selected three processors for this test, NAMED EPYC 7763, EPYC 7713, and EPYC 75F3 (the picture shows EPYC 7713 and EPYC 75F3)
Of course, the above three processors are built by the upgraded TSMC 7nm process and are based on AMD's latest Zen 3 architecture, they all have a capacity of up to 256MB L3 cache, but also support up to 8 channels of DDR4 3200 memory, memory bandwidth (per channel) up to 204.8 GB/s. If you want to learn more about the technical content of the 3rd generation AMD EPYC processor, you may wish to refer to MC's March 16 "Zen 3 Architecture Building, Up to 64 Cores!" AMD EPYC 7003 Series Processor Technology Architecture Analysis article.
<h1 class="pgc-h-arrow-right" data-track="3" > performance test</h1>
Let's take a look at the testbed. This test uniformly examines the performance of the dual-socket system, and uniformly uses amd DAYTONA_X motherboard with DDR4 3200 memory (32GB×16). In order to give everyone a more intuitive understanding of the performance of the third-generation AMD EPYC processor, MC will also compare the test results of the EPYC 7763, EPYC 7713, and EPYC 75F3 with the flagship product of the second-generation AMD EPYC processor, EPYC 7742. It should be noted that due to the update of the test methods of C-Ray 1.1 and Sysbench, the test results of EPYC 7742 are no longer comparable, so in these two tests, we will focus on the performance of EPYC 7763, EPYC 7713, EPYC 75F3 and other processors.
▲We are looking at the performance of the two-way system in this test
▲The test benches are using Micron DDR4 3200 ECC memory (32GB×16)
▲The test platforms are all using Samsung 860 PRO 256GB SSDs
3rd Generation AMD EPYC Processor Dual System Test Platform at a Glance
Processors: EPYC 7763×2, EPYC 7713×2, EPYC 75F3×2
Memory: DDR4 3200 ECC 32GB×16
Motherboard: AMD DAYTONA_X (BIOS version: RYM1001D)
Hard Drive: Samsung 860 PRO 256GB SSD×2
Operating system: Ubuntu 20.04
EPYC 7742 Dual System Test Platform at a Glance
Processor: EPYC 7742×2
Memory: Micron DDR4 3200 ECC 32GB×16
Motherboard: AMD DAYTONA_X (BIOS version: RDY1001C)
Hard Drive: Micron 9300 3840GB SSD×3
Os-X: Ubuntu 19.04 (for testing SPECrate 2017)
CentOS 7.6 (for testing additional items)
SPECrate 2017
SPECrate 2017 tests the number of instances running per unit of time, which is the main performance indicator when purchasing servers, so server manufacturers and processor manufacturers usually provide this test score. Specrate Integer and SPEKrate Floating Point are included in SPEKrate 2017, which tests integer concurrency performance and SPEKrate Floating Point.
Our test results show a significant performance improvement in SPEKrate 2017 compared to the dual EPYC 7742. For example, in the SPECrate2017_int_base, the two-socket EPYC 7763 scored 777 points, while the dual-channel EPYC 7742 scored 682 points in this test, which improved the performance of the former by about 14%. In the SPECrate 2017_fp_base that examined the floating-point performance of the processor, the dual EPYC 7763 scored 610 points, which greatly led the dual EPYC 7742 with a performance advantage of 17%. It is worth mentioning that spec's official website data shows that the SPEDrate 2017_fp_base test score of the dual-channel EPYC 7763 has reached 629 points, compared with the dual-channel EPYC 7742, the performance advantage of the former is further expanded to 20%.
The TDP of the EPYC 7713 and EPYC 7742, although both 225W, is quite ahead of the dual EPYC 7742 in SPESRATE 2017 thanks to the more advanced Zen 3 architecture and higher Boost frequency. Among them, the SPECrate2017_int_base test results of the dual-channel EPYC 7713 are about 9% ahead of the dual-channel EPYC 7742, and in the SPEKrate 2017_fp_base, the performance advantage of the dual-channel EPYC 7713 is 14%.
As mentioned earlier, the EPYC 75F3 is a processor with 32 cores and 64 threads, but it brings a brilliant performance in SPEKrate 2017. First, in the SPEKrate 2017_int_base, the dual-socket EPYC 75F3 scored 556 points, which is only about 23% behind the dual-socket system composed of the previous generation flagship EPYC 7742. What is even more gratifying is that in the SPEKrate 2017_fp_base, the performance of the dual EPYC 75F3 has reached the level of the dual EPYC 7742.
Based on the test results of SPECrate 2017, it is not difficult to find that the performance improvement of the third generation OFD EPYC processor in floating-point computing is very gratifying, which is naturally due to the strengthening of the Zen 3 architecture in the floating-point computing part. At the same time, thanks to the larger execution window, as well as the blessing of dedicated branch prediction and address selector, the performance of the third generation OFD EPYC processor in integer operations has also been enhanced, which is also fully reflected in the test results of SPECrate 2017.
Stream-Traid
Stream is one of the most popular comprehensive real-world performance measurement tools for memory bandwidth. Unlike the theoretical maximum memory bandwidth provided by hardware manufacturers, Stream, written in two high-level and efficient languages, fortran and C, can give full play to the memory capabilities in testing. Stream contains a total of four operations, Copy, Scale, Add and Triad, of which Triad combines the first 3 operations, so its test results are more valuable.
Comparing the specifications of the participating processors, it is not difficult to see that the theoretical memory bandwidth of each channel of the third-generation AMD EPYC processor is consistent with the EPYC 7742. Judging from our test results, the three third-generation AMD EPYC processors in the test each composed of dual-socket systems, the actual memory bandwidth in this test is basically the same, and it is also slightly ahead of the dual-channel EPYC 7742.
NAMD
NAMD is a parallel molecular dynamics code developed by the Theoretical and Computational Biophysics Group at the Beckman Institute for Advanced Science and Technology at the University of Illinois at Urbana-Champaign, and it is primarily used for high-performance simulations of large biomolecular systems. In this test, we mainly use NAMD to examine the floating-point performance of the participating processors.
As can be seen from our test results, the dual EPYC 7763 outperforms the dual EPYC 7742 in NAMD with an advantage of about 3% thanks to the higher base frequency and the improvement in floating-point performance due to the higher base frequency and the increase in floating-point performance. In addition, test results show that the dual EPYC 7713 performs slightly inferior to the dual EPYC 7742, which we believe is due to the fact that the underlying frequency of the EPYC 7713 is slightly lower than that of the EPYC 7742.
OpenSSL
OpenSSL is widely used to secure communication between servers, which is an important protocol in many server stacks. The OpenSSL test mainly consists of generating signatures and verifying signatures, and we mainly conducted a signature generation test at this time.
The test results show that the dual EPYC 7763 can generate 27163.2 signatures per second in our test, which is about 2% faster than the dual EPYC 7742. In addition, even though the number of cores and threads in the EPYC 75F3 is only half that of the EPYC 7742, the "high frequency" of the EPYC 75F3 processor makes it perform well in OpenSSL – the dual EPYC 75F3 can generate signatures in this test with about 66% of the efficiency of the dual EPYC 7742.
UnixBench Dhrystone 2 and Whetstone tested
This project is primarily used to test Unix system performance, including two test projects, Dhrystone 2 using register variables for string processing and Double-Precision Whetstone for testing the speed and efficiency of floating-point operations. In addition, single-threaded or multi-threaded tests can be used in both test projects, and this test we are using multi-threaded.
The powerful floating-point performance of the third-generation AMD EPYC processors was fully demonstrated in this part of the test. For example, in the Double-Precision Whetstone test project, the test score of the dual-channel EPYC 7763 far exceeded the two-socket EPYC 7742 of the previous generation positioning flagship, leading by up to 40%, while the dual-channel EPYC 7713 also led the dual-channel EPYC 7742 by about 34% in this test. What's even more gratifying is that the dual EPYC 75F3 can perform up to 89% of the dual EPYC 7742 in this part of the test.
In the Dhrystone 2 using register variables test project, the third generation of AMD EPYC processors also performed well. For example, compared with the dual EPYC 7742, the dual EPYC 7763 and the dual EPYC 7713 are about 17% and 3% ahead of the test scores, respectively, and the performance of the dual EPYC 75F3 can also reach about 78% of the dual EPYC 7742.
C-ray 1.1
C-ray is a commonly used ray-traced benchmark that shows differences in processors under multithreaded workloads, with shorter time indicating stronger system performance. In this test, we used both 4K and 8K resolutions to compare the performance of the participating systems under different loads.
Thanks to the high core frequency, the dual EPYC 75F3 performs well in C-ray. Although the number of cores and threads is only half that of the dual EPYC 7763, thanks to the higher core frequency, the dual EPYC 75F3 takes only about 1 second to complete the test task at 4K resolution than the dual EPYC 7763, and even at 8K resolution, the dual EPYC 75F3 takes only about 6 seconds more to complete the test task than the dual EPYC 7763.
Sysbench CPU test
Sysbench is a widely used Linux benchmark, it can perform performance tests on the CPU, in the test is mainly through the CPU to add prime numbers, the prime number limit used in this test is 10,000.
Our test results show that the dual EPYC 7763 can complete 487771 rounds of computation per second, while the dual EPYC 7713 performs about 92% of the performance of the dual EPYC 7763. It is worth praising that the dual-channel EPYC 75F3 can also complete 304232 round of operations per second, and its performance is about 62% of that of the dual-channel EPYC 7763. While the number of cores and threads in THE EPYC 75F3 is only half that of the EPYC 7763, we consider the higher base frequency to be a "plus" for the former.
<h1 class="pgc-h-arrow-right" data-track="3" > goal: to continue to strengthen the market share of the server space</h1>
Looking at the results of this test, it is not difficult to see that the Zen 3 architecture has greatly enhanced the comprehensiveness of the third generation of AMD EPYC processors. For example, while maintaining the same number of cores and threads, the dual EPYC 7763 has a leapfrog performance improvement over the dual EPYC 7742 in floating-point operations. At the same time, even though the number of cores and threads is only half that of the dual EPYC 7742, the floating-point performance of the dual EPYC 75F3 can be compared.
Not only that, thanks to the larger execution window, as well as the blessing of dedicated branch prediction and address selector, the performance of the third generation OF AMD EPYC processor in integer operations has also been enhanced, which is also fully reflected in this test. This, combined with the 3rd generation of AMD EPYC processors, increases flexibility in L3 cache applications and memory configurations, as well as significantly enhanced security. Not only that, the third-generation AMD EPYC processor has also launched a number of sub-product lines including "F" series products for different customers and different scenarios, a total of 19 products, and the entire product layout is very perfect.
At the same time, AMD is still focusing on building its EPYC processor ecosystem. AMD also announced that by the end of 2021, AMD EPYC processors are expected to work hand-in-hand with a number of partners, including Amazon, Cisco, Lenovo, Tencent Cloud, etc., to bring more than 400 cloud instances and more than 100 new OEM platforms to its ecosystem. Therefore, we believe that the third generation of AMD EPYC processors will help enterprise users meet complex IT challenges and become a new generation of strategic products for AMD to gain more data center market share.