Exploring The Zen 5 SMT Performance With The AMD EPYC 9755 "Turin" CPU
([Processors] 4 Hours Ago
4 Comments)
- Reference: 0001498898
- News link: https://www.phoronix.com/review/amd-epyc-9755-smt
- Source link:
[1]
Continuing on with the testing around the [2]AMD EPYC 9005 series " [3]Turin " processors, today is a look at the Simultaneous Multi-Threading (SMT) performance impact for Turin while using the AMD EPYC 9755 as the highest-end "Turin Classic" processor with 128 cores / 256 threads. Similar SMT on/off tests for "Turin Dense" with the [4]EPYC 9965 192-core / 384-thread will also be coming in a future benchmarking comparison on Phoronix. These tests are mainly intended for reference purposes for those curious about the SMT benefits at such high core counts and what workloads may or may not still benefit from SMT especially when having so many threads while using 12-channel DDR5-6000 memory.
[5]
I ran more than 170 different benchmarks for this comparison of running the EPYC 9755 processor with its Zen 5 "classic" cores while in the default SMT-enabled mode of operation and then with SMT disabled. The 128-core AMD EPYC 9755 processor has a 2.7GHz base clock with 4.1GHz maximum boost clock, 512MB L3 cache, 500 Watt default TDP, and a 1KU list price of $12,984 USD.
[6]
Under Linux SMT can be easily disabled at boot time using the "nosmt" option or as most Phoronix readers are aware cores/threads can be offlined at run-time using the sysfs interface or in a more friendly manner via different utilities. The Linux run-time controls around core/thread offlining (and methods like CPU core pinning) are great for those wanting to modify the behavior at run-time depending upon the workload(s) being used for a given period of time. For making these tests straight-forward, SMT was toggled via the AMD Volcano system BIOS. All the tests were carried out for the EPYC 9755 with the default SMT enabled for 128 cores / 256 threads and then repeated with SMT disabled.
The EPYC 9755 was tested in a single socket configuration with 12 x 64GB Samsung ECC RDIMMs at DDR5-6000 speeds. Ubuntu 24.04 LTS was running on this AMD EPYC 5th Gen server with the default GCC 13.2 compiler. The CPU power consumption was also monitored for those curious about any Zen 5 CPU power consumption impact with SMT on/off.
[1] https://www.phoronix.com/image-viewer.php?id=amd-epyc-9755-smt&image=amd_epyc_smt_1_lrg
[2] https://www.phoronix.com/review/amd-epyc-9005
[3] https://www.phoronix.com/search/Turin
[4] https://www.phoronix.com/review/amd-epyc-9965-9755-benchmarks
[5] https://www.phoronix.com/image-viewer.php?id=amd-epyc-9755-smt&image=amd_epyc_smt_2_lrg
[6] https://www.phoronix.com/image-viewer.php?id=amd-epyc-9755-smt&image=amd_epyc_smt_3_lrg
Continuing on with the testing around the [2]AMD EPYC 9005 series " [3]Turin " processors, today is a look at the Simultaneous Multi-Threading (SMT) performance impact for Turin while using the AMD EPYC 9755 as the highest-end "Turin Classic" processor with 128 cores / 256 threads. Similar SMT on/off tests for "Turin Dense" with the [4]EPYC 9965 192-core / 384-thread will also be coming in a future benchmarking comparison on Phoronix. These tests are mainly intended for reference purposes for those curious about the SMT benefits at such high core counts and what workloads may or may not still benefit from SMT especially when having so many threads while using 12-channel DDR5-6000 memory.
[5]
I ran more than 170 different benchmarks for this comparison of running the EPYC 9755 processor with its Zen 5 "classic" cores while in the default SMT-enabled mode of operation and then with SMT disabled. The 128-core AMD EPYC 9755 processor has a 2.7GHz base clock with 4.1GHz maximum boost clock, 512MB L3 cache, 500 Watt default TDP, and a 1KU list price of $12,984 USD.
[6]
Under Linux SMT can be easily disabled at boot time using the "nosmt" option or as most Phoronix readers are aware cores/threads can be offlined at run-time using the sysfs interface or in a more friendly manner via different utilities. The Linux run-time controls around core/thread offlining (and methods like CPU core pinning) are great for those wanting to modify the behavior at run-time depending upon the workload(s) being used for a given period of time. For making these tests straight-forward, SMT was toggled via the AMD Volcano system BIOS. All the tests were carried out for the EPYC 9755 with the default SMT enabled for 128 cores / 256 threads and then repeated with SMT disabled.
The EPYC 9755 was tested in a single socket configuration with 12 x 64GB Samsung ECC RDIMMs at DDR5-6000 speeds. Ubuntu 24.04 LTS was running on this AMD EPYC 5th Gen server with the default GCC 13.2 compiler. The CPU power consumption was also monitored for those curious about any Zen 5 CPU power consumption impact with SMT on/off.
[1] https://www.phoronix.com/image-viewer.php?id=amd-epyc-9755-smt&image=amd_epyc_smt_1_lrg
[2] https://www.phoronix.com/review/amd-epyc-9005
[3] https://www.phoronix.com/search/Turin
[4] https://www.phoronix.com/review/amd-epyc-9965-9755-benchmarks
[5] https://www.phoronix.com/image-viewer.php?id=amd-epyc-9755-smt&image=amd_epyc_smt_2_lrg
[6] https://www.phoronix.com/image-viewer.php?id=amd-epyc-9755-smt&image=amd_epyc_smt_3_lrg