Kunpeng Hardware
How do I upgrade the Hi1822 NIC?
You can upgrade the Hi1822 NIC using Smart Provisioning or SmartKit.
What can I do if the BIOS cannot identify the RAID controller card?
If the BIOS of the TaiShan 2280 server cannot identify the RAID controller card, contact technical support.
How do I check the RAID controller card type?
- Log in to the iBMC WebUI.
- Choose to check the RAID controller card type.
What is the clock cycle of the Kunpeng 920 processor?
The clock cycle of the processor is the reciprocal of the CPU frequency. The CPU frequency of the Kunpeng 920 processor is 2.6 GHz, and the clock cycle is 0.38 ns.
How do I calculate the floating-point computing power of the Kunpeng 920 CPU? Is it the same as that of Intel CPUs?
It is the same as that of Intel CPUs.
Theoretical upper limit of CPUs: CPU frequency × Number of CPU FMA units × (Maximum bit width of the vector register/Data bit width) × 2 (multiply-add instructions indicate two operations)
The following uses the Kunpeng 920 as an example:
Half-precision: 2.6 × 2 × (128/16) × 2 = 83.2 GFLOPS
Single precision: 2.6 × 2 × (128/32) × 2 = 41.6 GFLOPS
Double-precision: 2.6 × 1 × (128/64) × 2 = 10.4 GFLOPS
How does the Kunpeng server assembly instruction obtain the system time?
The Kunpeng server reads the cntvct_el0 register, which is related to the CPU external frequency. The external frequency of the Kunpeng 920 CPU is 100 MHz.
Assembly instruction code:
__asm__ __volatile__ ("mrs %0, cntvct_el0" : "=r" (result));
Does the Kunpeng processor (Arm) support the AVX2 instruction set?
AVX2 is the vector instruction set for x86 processors. On the Kunpeng processor (Arm), the AVX2NEON tool is used to port or adapt the AVX2 instruction set to the NEON instruction set.
Does the Kunpeng server support NVIDIA cards?
Yes.
Which Arm instruction set versions are compatible with the Kunpeng processor?
The Kunpeng processor is fully compatible with the Armv8-A architecture and supports Armv8.1 and Armv8.2 extensions.
How do I obtain the server parts replacement tutorial video?
To obtain the tutorial video for replacing TaiShan server hardware parts, visit https://support.huawei.com/enterprise/en/doc/EDOC1100099787?section=j001.
How do I enable the performance mode?
On the BIOS screen of the server, choose , set Power Policy to Performance, and press F10 to save the BIOS configuration.
How do I query the root PCIe number of a NIC?
Run the lspci -tv command.
What are the performance differences between the 32-core and 64-core Kunpeng 920 processors?
The number of cores of the two processors differs by half, and the memory channels also differ. The Kunpeng 920 7260 processor has eight memory channels, and the Kunpeng 920 5220 processor has four memory channels.
Does the Kunpeng processor support virtualization?
Kunpeng processors support virtualization. You need to enable the virtualization capability in the BIOS.
- On the BIOS screen, choose and set Support Smmu to Enabled.
- On the BIOS screen, choose and set SRIOV to Enabled.
What is the URL for accessing the Compatibility Query tool?
For details about the OS/component compatibility of TaiShan servers, see Compatibility Query.
Does the QS server mainboard L21K-2041K have a circuit diagram?
The circuit diagram of the QS mainboard L21K-2041K is not available. The CPLD is mainly used for board startup and cannot be directly developed. The online version is an adjusted version.
How many Atlas 300 inference cards does a TaiShan 2280 server support?
Up to 8 Atlas 300 AI inference cards are supported.
What can I do if the SPE function cannot be enabled on the Kunpeng server whose firmware version is 1.7?
Download the latest firmware version from Huawei enterprise website and upgrade the firmware.
What can I do if the TaiShan 200 2280 SP525 HBA card indicators are off?
For details about how to check the server hardware, contact the Global Service Hotline by visiting Huawei enterprise website.
How do I obtain the TG225 B1 SNMP traps monitoring configuration document and MIB library?
- Contact Huakun Zhenyu's after-sales personnel to obtain the document and MIB library.
- Related documents are also available on Huawei enterprise website. If you have permission, you can download and use the documents directly.
How do I turn on NVMe SSD indicators in the operating system (OS) of a TaiShan 5280 server?
There is no IPMI command for turning on NVMe SSD indicators.
What can I do if the Kunpeng BMC displays a message indicating that the SMTP server cannot be connected?
Enable the SMTP service in the 163 mailbox and obtain the authorization password. After the configuration is complete, if the system displays a message indicating that the SMTP server cannot be connected, check whether the mailbox settings are correct or try another mailbox. If the fault persists, collect logs and contact the Global Service Hotline.
Can ECSs be live migrated across Kunpeng processors?
Currently, Kunpeng processors do not support live migration of ECSs across different models.
How do I obtain the parameter description of Kunpeng processors?
Visit the Kunpeng Community and search for keywords such as Kunpeng processor model or Kunpeng product model to obtain the parameter description of Kunpeng processors.
How do I configure the memory mirroring and tiered-reliability memory functions?
The Kunpeng 920 processor does not support memory mirroring.
For details about how to configure the tiered-reliability memory function, see Tiered-Reliability Memory User Guide.
What are the clock frequency and number of cores of the Kunpeng 920 5250 processor?
The clock frequency is 2.6 GHz, and the number of cores is 48.
How many floating-point operations does the Kunpeng 920 5250 processor perform per clock cycle?
Each core has two SIMD operation pipelines, and two floating-point instructions are executed per each clock cycle.
For the NEON instruction set of the Kunpeng 920 5250 processor, the operand width is 128 bits, allowing one instruction to process two FP64 elements, four FP32 elements or eight FP16 elements.
Number of floating-point operations per core per cycle = Number of instructions per cycle × Data count × 2 (An FMA instruction can perform a multiplication and an addition operations simultaneously.) Therefore, the number of FP64 floating-point operations per cycle is 1 × 2 × 2 = 4; the number of FP32 floating-point operations per cycle is 2 × 4 × 2 = 16; the number of FP16 floating-point operations per cycle is 2 × 8 × 2 = 32.
How do I obtain the information about the Kunpeng processor or general-purpose mainboard?
For details about the Kunpeng server mainboard, see https://support.huawei.com/enterprise/en/category/kunpeng-computing-pid-1548148188432?submodel=doc.
For details about the Kunpeng processor, see https://www.hikunpeng.com/en/compute/kunpeng920.
What do I do if some functions are suspended when Python is used for machine learning on the Kunpeng 920 processor?
When the Python machine learning library scikit-learn is used for machine learning on the Kunpeng 920 processor, the single-core running time is too long (more than 3 hours), which makes the system seem suspended. In this case, try to use the parallelization method to shorten the running time.
What instruction sets does the Kunpeng 920 5221K processor support?
The Armv8.2 instruction set is supported.
Why is the HTTP performance of the Kunpeng 920 processor with 96 cores lower than that of the processor with 24 cores?
A performance degradation occurs due to cross-NUMA-node access on the server. The highest performance is achieved when both the server-side process and the test program run on the same CPU socket. Performance can be restored to the expected level by binding both the server and client processes to the same NUMA node using the numactl command. Specifically, the numactl -C command can be used to bind cores during service startup.
For example, to bind a service to cores 0 to 15, run the umactl -C 0-15 command. Before binding, run the numactl -H command to check the NUMA distribution. You are advised to select the same NUMA node for core binding.
How do I reduce the number of cores on Kunpeng processors?
You can run the echo 0 > /sys/devices/system/cpu/cpuX/online command, where X indicates the CPU number. This command takes effect after being executed and becomes invalid after the system is restarted.
To make the configuration take effect for a long time, use either of the following methods:
- Compile a startup script to make the CPU offline during startup.
- Use the parameter maxcpus provided by the kernel. Add the parameter to the grub command line.
maxcpus= [SMP] Maximum number of processors that an SMP kernel will bring up during bootup. maxcpus=n : n >= 0 limits the kernel to bring up 'n' processors. Surely after bootup you can bring up the other plugged cpu by executing "echo 1 > /sys/devices/system/cpu/cpuX/online". So maxcpus only takes effect during system bootup. While n=0 is a special case, it is equivalent to "nosmp", which also disables the IO APIC.
How do I determine the CPU corresponding to each socket in the Kunpeng 920 processor NUMA architecture?
- Check the CPU architecture information.
lscpu
If the following information is displayed, each socket has 64 CPUs and corresponds to two NUMA nodes. Socket 0 corresponds to NUMA nodes 0 and 1, and socket 1 corresponds to node 2 and node 3.Architecture: aarch64 CPU op-mode(s): 64-bit Byte Order: Little Endian CPU(s): 128 On-line CPU(s) list: 0-127 Thread(s) per core: 1 Core(s) per socket: 64 Socket(s): 2 NUMA node(s): 4 Vendor ID: HiSilicon Model: 0 Model name: Kunpeng-920 Stepping: 0x1 BogoMIPS: 200.00 L1d cache: 8 MiB L1i cache: 8 MiB L2 cache: 64 MiB L3 cache: 128 MiB NUMA node0 CPU(s): 0-31 NUMA node1 CPU(s): 32-63 NUMA node2 CPU(s): 64-95 NUMA node3 CPU(s): 96-127 Vulnerability Itlb multihit: Not affected Vulnerability L1tf: Not affected Vulnerability Mds: Not affected Vulnerability Meltdown: Not affected Vulnerability Mmio stale data: Not affected Vulnerability Retbleed: Not affected Vulnerability Spec store bypass: Vulnerable Vulnerability Spectre v1: Mitigation; __user pointer sanitization Vulnerability Spectre v2: Not affected Vulnerability Srbds: Not affected Vulnerability Tsx async abort: Not affected Flags: fp asimd evtstrm aes pmull sha1 sha2 crc32 atomics fphp asimdhp cpuid asimdrdm jscvt fcma dcpop asimddp asimdfhm
- Verify the mapping between sockets, NUMA nodes, and CPUs.
lstopo -l
Package L#0 indicates the first physical CPU socket.
What is the power consumption of the Kunpeng 920 5225F processor?
The Kunpeng 920 5225F processor is a non-public model. Therefore, no information about the processor is available on the official website. For details, contact the customer manager.
What can I do if Accelerators Status is No Install in the BIOS?
If Accelerators Status is No Install in the BIOS, check whether the license is correctly installed. For details about how to install the license, see Huawei Server iBMC License User Guide.
Does the Kunpeng 920 5220 processor support CPU virtualization?
The Kunpeng 920 5220 processor supports CPU virtualization and Armv8 virtualization enhancement features such as I/O virtualization, two-level page tables, and interrupt virtualization. If the virtualization configuration does not take effect, contact the customer manager.
What can I do if a green screen is displayed when a container uses NVIDIA DeepStream under the Kunpeng 920 processor?
As a general-purpose server, Kunpeng provides standard PCIe interfaces. If a problem occurs when a third-party card is connected, you need to check the card driver, which is provided by the card vendor. The second possible cause is the OS, which provides the entire display solution. If a green screen is displayed when a container uses NVIDIA DeepStream, the board vendor and OS vendor need to cooperate in resolving the problem.
How many NUMA nodes does the Kunpeng 920 5220 processor have? How do I change the number of NUMA nodes?
A Kunpeng 920 5220 processor has only one NUMA node. If the server has two processors, there are two NUMA nodes. The number of NUMA nodes cannot be changed.
How do I obtain the frequency of the Kunpeng 920 high-performance processor (64 cores) in real time?
Run the cpufreq-info command or check the /sys/devices/system/cpu/cpu*/cpufreq/scaling_cur_freq file to obtain the real-time CPU frequency. If the frequency remains 2.7 GHz, set the BIOS to Efficiency mode and try again.
How do I obtain the technical documentation of the Kunpeng processor?
Kunpeng processors are compatible with the Armv8-A architecture. For details about the product, visit the Kunpeng Community.
What OSs are compatible with the Kunpeng 920 processor?
For details about the compatibility between the Kunpeng 920 processor and OSs, see Compatibility Query.
FAQs about NUMA node affinity
Q:
How do I allocate the array space applied at runtime to different NUMA nodes so that the core can access the memory of the NUMA node where the core is located?
A:
You can use either of the following two methods:
- Use the numactl command provided by the OS to set the memory allocation policy of a process.
numactl --membind=<node>;
- Use the numa_all_onnode(size_t size, int node) function in the NUMA library to set the memory allocation policy on the NUMA node.
Is the Kunpeng BMC compatible with other platforms except the Hi1711 chip?
Currently, the Kunpeng BMC supports the Hi1710 and Hi1711 chips only. The Kunpeng BMC system cannot run on other BMC chips.
Does the Kunpeng 920 processor support SM3/SM4 algorithm acceleration?
The Kunpeng 920 processor does not support SM3 and SM4 instructions, but integrates the SM3 and SM4 accelerators.
Does the Kunpeng 920 processor support Armv8_SM3 instruction acceleration?
Kunpeng 920 processors do not support SM3 instructions, but the integrated encryption and decryption accelerator contains the SM3 acceleration engine.
How do I obtain the documentation of the Kunpeng 920 processor?
You can search for them under the Products or Documentation in the Kunpeng Community.
For details about the Kunpeng processor, see https://www.hikunpeng.com/en/compute/kunpeng920.
Does the Kunpeng 3210 processor support 24 cores?
The Kunpeng 920 3210 processor is a 24-core CPU and supports 24 cores.
FAQs about the underlying configuration solution of the TaiShan 2280 server
Q1:
Does the Kunpeng 920 5220 processor have affinity differences among 32 cores when accessing the L3 cache and memory?
A1:
The Kunpeng 920 5220 processor is a single-chip processor with only one CPU die. For the L3 cache access affinity, each 4-core cluster access its corresponding L3D. The access latency to DDR is nearly uniform across all 32 cores.
Q2:
What is the connection mode between the two processors of the TaiShan 2280 server? What is the affinity of the two processors when accessing the cache and memory?
A2:
The two processors of the TaiShan 2280 server are connected through Huawei Cache Coherence System (HCCS). You can run the numactl -H command to view the affinity of the two processors when accessing the memory.
What are the differences between the Kunpeng 920 7265F processor and the Kunpeng 920 7260 processor, for example, between the 2.6 GHz CPU and the 3.0 GHz CPU?
The 3.0 GHz and 2.6 GHz CPUs differ in the frequency. Compared with the 2.6 GHz CPU, the 3.0 GHz CPU has 15.4% higher frequency, and the 3.0 GHz CPU performance is improved by about 15% in SPEC CPU and some services.
Does the number of physical threads of the Kunpeng 920 processor match the number of physical cores?
The number of physical threads of the Kunpeng 920 processor is the same as the number of physical cores. One physical core corresponds to one physical thread.
FAQs about hardware virtualization and AArch32 of Kunpeng 916 and Kunpeng 920 processors
Q1:
Do the Kunpeng 916 and Kunpeng 920 processors support KVM virtualization? If yes, how do they support the KVM?
A1:
The Kunpeng 916 and Kunpeng 920 processors support KVM virtualization. The community KVM runs stably on Kunpeng processors.
Q2:
Do the Kunpeng 916 and Kunpeng 920 processors support the AArch32 mode at EL0 and EL1?
A2:
The Kunpeng 916 processor supports the AArch32 mode at EL0 and EL1, but the Kunpeng 920 processor does not support the AArch32 mode at EL0 and EL1.
Q3:
Do Kunpeng 916 and Kunpeng 920 processors support running 32-bit Linux OSs using KVM virtualization?
A3:
The Kunpeng 916 processor supports running 32-bit Linux OSs using KVM virtualization in the AArch32 mode. The Kunpeng 920 processor does not support the AArch32 mode and cannot use the KVM virtualization to run 32-bit Linux OSs.
Does the Kunpeng 920 processor support hyper-threading? For example, can an 8-core CPU supports 16 threads with hyper-threading?
Some models of the Kunpeng 920 processor series support hyper-threading. With simultaneous multi-threading (SMT), one physical core can work like two logical cores.
Do Kunpeng 920 5230 and Kunpeng 920 5220 processors support hyper-threading?
No. One physical core corresponds to one physical thread.
What instruction sets does the Kunpeng 920 3210 processor support?
It supports the Armv8.2 instruction set.
Does the Kunpeng 920 processor support the RNDRRS instruction set?
The Kunpeng 920 processor is developed based on the Armv8 architecture and does not support the RNDRRS instruction set added in Armv8.5.
Does the Kunpeng 920 processor support the random number API (for key generation) Are development documents and instruction set description documents for the random number API available?
Kunpeng 920 supports random number APIs. The hardware random number generation (RNG) has passed the security certification of the State Cryptography Administration (SCA) and its performance can reach 1.6 Gbit/s.
The random number API can be the dev/random API provided by the OS or the random number API from OpenSSL. (If supported by the hardware, the Kunpeng 920 processor also includes the true random number API.) For details about how to use the API, search for the corresponding OS or OpenSSL documents.
How do I calculate the transactions per minute C (tpmC) of a single Kunpeng 920 processor?
tpmC varies according to the hardware configuration and needs to be tested.
tpmC is the measurement unit of the TPC-C benchmark test and must be obtained through standardized testing. The main steps are as follows:
- Configuration environment: Determine the CPU model/core quantity, memory capacity, storage performance, and network bandwidth.
- TPC-C test: Run the test to simulate the OLTP scenario (five types of transactions of the warehouse management system).
- Reference formula: tpmC = (Total_Transactions × 60)/Test_Duration(s).
How many cores does the Kunpeng 920 5220 processor have?
The Kunpeng 920 5220 processor has 32 cores.
How do I calculate the floating-point computing power of the Kunpeng 920 7270Z processor?
Double-precision computing power of a single CPU = Number of cores × Single-core computing power
Single-core computing power = CPU frequency × Number of floating-point operations per cycle
Number of floating-point operations per cycle = Number of floating-point units × (Vector register bit width/64) × 2
The frequency of the Kunpeng 7270Z processor is 2.6 GHz, the number of cores is 64, the number of floating-point units is 2, and the vector register bit width is 256.
Is Huawei certification required for using the trusted execution environment (TEE) technology of Kunpeng servers?
For details about how to use the TEE technology of Kunpeng servers, see TEE Kit on the Kunpeng developer website.
Which Kunpeng 920 processor models have the CPU frequency of 2.5 GHz?
The CPU frequency of models such as 7261R, 5251R, 5221R, and 3211R is 2.5 GHz. For details, see Kunpeng Community.
Does the Kunpeng 920 7260 processor support instruction sets such as svei8mm, svef32mm, svef64mm, svebf16, i8mm, and bf16?
The Kunpeng 920 7260 processor does not support the SVE instruction.
Does the Kunpeng 920 processor support the SM2 algorithm? If not, can it be implemented with technologies like TEE at the software layer?
Kunpeng 920 processor does not support the SM2 algorithm, which means that the Kunpeng 920 processor does not include an SM2 instruction or acceleration engine. The SM2 algorithm can be implemented through software programming, for example, using the open-source GmSSL software.
Do Kunpeng 920 5250Z and 7270Z processors support the Kylin OS?
Kunpeng 920 processors support the Kylin OS.
Does the Kunpeng 916 processor support the Armv8.2 instruction set?
The Kunpeng 916 processor supports some Armv8.2 instruction sets.
How do I query the number of threads of a Kunpeng 920 processor?
You can query the number of threads on the Kunpeng Community. The number of cores is the number of threads.
Does the 48-core Kunpeng 920 processor support dual threads?
The 48-core Kunpeng 920 processor does not support dual threads. One physical core corresponds to one physical thread.
Does the Kunpeng 920 processor support Windows XP?
Currently, the Kunpeng 920 processor does not support Windows XP.
Which Kunpeng 920 processor models support dual threads?
You can run the lscpu command to display the CPU architecture and related information in the Linux system. In the output of the lscpu command, the Thread(s) per core field indicates the number of threads per core. If its value is 2, dual threads are enabled.
Does the Kunpeng 920 Processor support the Armv8.2-A architecture?
The Kunpeng 920 processor supports the Armv8.2-A architecture.
Does the Kunpeng 920 processor support nested virtualization?
The Kunpeng 920 processor does not support nested virtualization.
How do I query the PID, BID, and CID of the Kunpeng 916 module?
PID (product ID): uniquely identifies a specific product model or version of the Kunpeng 916 module.
BID (board ID): identifies the mainboard on which the Kunpeng 916 module is installed.
CID (component ID): identifies each component of the Kunpeng 916 module.
You can query the BID and PID using OS commands or the one-click log collection function. CID information can be queried only on the iBMC. However, the currently released iBMC has limited functions and does not support CID information query.
You can query the BID and PID using the following methods.
Method 1: Run the ipmitool fru command in the OS CLI.
Method 2: Query the BID and PID by using the one-click log collection function by following the path: dump_info\RTOSDump\versioninfo\fruinfo. Then run the ipmitool fru command.
How do I enable CPU virtualization on the Kunpeng 920 processor?
For details about how to enable CPU virtualization, see Component Principles.
Does the 2-port 100GE NIC adapted to the Kunpeng 920 processor support DPDK?
The 2-port 100GE NIC adapted to the Kunpeng 920 processor support data plane development kit (DPDK). For details, see SP600 Standard NIC User Guide.
Does the Kunpeng 920 5250Z processor support the memory rated speed of 5600 MT/s?
The Kunpeng 920 5250Z processor supports the memory rated speed of up to 4800 MT/s, but not 5600 MT/s.
Does the Kunpeng processor support the reinstallation of the Arm-based Windows 11 system using a VM?
The Kunpeng processor does not support the reinstallation of the Arm-based Windows 11 system using a VM.
Can the CPU usage be reduced by adding an inference card if the openGauss database adopts 2 × Kunpeng 920 processors (256 GB memory) and is experiencing 100% CPU usage?
When CPU usage reaches 100%, you can use diagnosis tools such as top, perf, and flame graphs to check the attributes of the main CPU hotspots, and then take appropriate measures. If the hotspot is primarily in the database, you can optimize the database or shard the databases and tables. If the hotspot is mainly in AI inference and multiply–add operations, you can add an inference card.
Does the TaiShan 200 2280 server support the SVE1 instruction?
The TaiShan 200 2280 server does not support the SVE1 instruction. The TaiShan 200 2280 server is developed based on the Armv8 architecture (Armv8-A) and does not integrate the scalable vector extension 1 (SVE1) instruction set. Support for SVE1 requires the Armv9 architecture or later.
How many VMs can the Kunpeng 920 processor support?
The virtualization capability of the Kunpeng 920 processor depends on multiple factors, including the number of physical cores, memory capacity, virtualization technology, and OS and virtualization platform configurations. The Kunpeng 920 processor does not have a fixed specification specifying the number of VMs supported. The number of VMs supported is determined based on the actual deployment requirements, system load, and resource configuration. In actual applications, it can support dozens to hundreds of VMs in typical configurations.
Why does the PyTorch model run slowly on the Kunpeng 920 processor?
Performance optimization is a systematic project, which may involve BIOS configuration, OS configuration, application parameter configuration, and basic library performance. For details, see Kunpeng Performance Tuning.
FAQs about advanced host controller interface (AHCI) of the onboard drive controller
Q1:
For the Kunpeng 920 model, does the onboard drive controller AHCI pass-through mean that the drive backplane is directly connected to the mainboard CPU?
A1:
AHCI pass-through means that the SATA controller on the mainboard (usually the AHCI controller integrated in the CPU or chipset) directly manages SATA drives without passing through an additional RAID controller card or HBA. The AHCI controller of the CPU/chipset is connected to the SATA interface (on the mainboard or backplane) through the mainboard PCB trace, instead of being transferred through a third-party controller.
Q2:
Can the front pass-through backplane be connected to the mainboard through a SATA cable without a RAID controller card when using the onboard drive controller AHCI pass-through?
A2:
Yes, it works without a RAID controller card. The drive backplane connects to the AHCI controller on the mainboard through the SATA cable and communicates with the CPU through the PCIe bus.
FAQs about the Kunpeng server mainboard version
Q1:
How do I query the versions 3.0.0 and 1.2.0 mentioned in the EOS announcement of Huawei S920S00 3.0.0, S920S00K 3.0.0, and S920X00 1.2.0?
A1:
Query method: Go to the Support > Kunpeng Computing > Kunpeng Server Boards page, select the specific model, and access the software page. Then, you can obtain all versions of the product and find the EOS version.
The S920S00 3.0.0, S920S00K 3.0.0, and S920X00 1.2.0 described in the EOS announcement are C versions or basic versions. The actual released version may be the basic version itself or an iteration built upon it.
The first version of the S920S00 3.0.0 series is S920S00 3.0.0.1.
The first version of the S920S00K 3.0.0 series is S920S00K 3.0.0.1.
The first version of the S920X00 1.2.0 series is S920X00 1.2.0.1.
This rule also applies to the replacement version mentioned in the announcement. For example, the replacement version of S920S00 3.0.0 is S920S00 3.1.0, and the first released version is S920S00 3.1.1.
Q2:
In Ubuntu, the dmidecode -t baseboard, dmidecode -t system, and dmidecode -t bios commands can be used to view the Version field. Which Version field displayed corresponds to the EOS version of the mainboard?
A2:
The EOS version information of the mainboard cannot be obtained by running the three commands.
The Version field in the dmidecode -t system command usually indicates the version of firmware or hardware of the system manufacturer, not the OS version.
The Version field in the dmidecode -t bios command indicates the BIOS firmware version.
The Version field in the dmidecode -t baseboard command indicates the mainboard version.
Query method: Go to the Support > Kunpeng Computing > Kunpeng Server Boards page, select the specific model, and access the software page. Then, you can obtain all versions of the product and find the EOS version.
Q3:
If the version information of the mainboard cannot be queried, how do I determine whether to upgrade the mainboard version?
A3:
You are advised to manually record the firmware version or compare the firmware versions. You can view the main firmware information of the mainboard in the current environment on the BMC WebUI.
How do I check whether the Kunpeng CPU frequency is reduced?
Kunpeng CPUs support Arm dynamic frequency adjustment (for example, the big.LITTLE scheduler in the ARMbig.LITTLE architecture). The system automatically adjusts the frequency based on the load.
You can run the cat /sys/devices/system/cpu/cpufreq/policy0/cpuinfo_cur_freq command to check the current CPU policy and determine whether the energy saving mode is enabled.
How do I query the three Intel third-generation CPUs corresponding to the 2.6 GHz 32-core, 48-core, and 64-core Kunpeng 920 processors?
For general-purpose CPUs, an overall comparison is performed based on the test data of the Standard Performance Evaluation Corporation (SPEC) CPU 2017. For different service applications, the comparison is performed based on the specific service benchmark test.
The SPEC CPU 2017 data for the Kunpeng 920 processor can be obtained through actual testing or by checking the official website. If the Kunpeng 920 performs similarly to Intel's Ice Lake, they can be considered corresponding products. Note that the SPEC CPU 2017 test data is closely related to the compiler version. You need to align the test data with the compiler version during comparison.
Does the Kunpeng 920 processor support signature verification?
Yes. Kunpeng 920 processors support the SM3/SM4 algorithm for hardware acceleration, and also support signature verification with open-source algorithm libraries such as OpenSSL and GMSSL.
How do I obtain the cryptographic algorithms and performance parameters supported by Kunpeng 920 series CPUs?
For details about the encryption and decryption algorithms supported by the CPU and their performance parameters, contact technical support.
Does the Kunpeng 916 processor support the 32-bit Arm instruction set?
The Kunpeng 916 processor supports the 32-bit Arm instruction set.
FAQs about configuring the PXE function of a NIC on the BIOS interface of the Kunpeng 920 processor
Q1:
Figure 1 shows the PXE configuration of a NIC on the BIOS. How do I check whether a NIC exists in the environment?
A1:
You can check the current NIC status in either of the following ways:
- View the list of available NICs on the BIOS > Advanced > Network Device List page.
- Check the product specifications or check whether the NIC device exists on the server. For details, see Kunpeng Computing Documentation and Computing Interactive Product Display based on the model.
Q2:
How do I set PXE for a NIC on the BIOS screen of the Kunpeng 920 processor?
A2:
For details about how to configure PXE for a NIC, see BIOS Parameter Reference.
How do I install the four DIMMs of the Kunpeng 920 processor when die interleaving is enabled?
Q1:
The Kunpeng 920 processor has two dies, and each die has four memory controllers. The 2 DPC configuration is supported. When inter-die interleaving is enabled, the two dies must have the same number of memory channels and the same memory size. Do the memory channels of the two dies correspond to each other in a one-to-one manner: channel 0 corresponds to channel 4, channel 1 corresponds to channel 5, channel 2 corresponds to channel 6, and channel 3 corresponds to channel 7? If there are only four DIMMs, which channels should the four DIMMs be installed in to meet the inter-die interleaving requirements?
A1:
Inter-die interleaving requires symmetrical DIMM installation methods for the two dies in a CPU, that is, the number of 1 DPC (one DIMM per channel) and 2 DPC (two DIMMs per channel) must be the same. If the four DIMMs are inserted into Physical 0 of channels 0, 1, 4, and 5, the inter-die interleaving requirements can be met.
Q2:
Figure 3 shows the memory installation rules of dual-socket Kunpeng 920 processors. When eight DIMMs are installed, four DIMMs are configured for each CPU. The DIMMs of CPU 1 are distributed in Physical 0 of channels 1, 3, 4, and 6, which is different from Physical 0 of channels 0, 1, 4, and 5 in A1. When eight DIMMs are installed according to the DIMM installation rules, the DIMMs of each CPU are installed in Physical 0 of channels 1, 3, 4, and 6. Does this configuration not meet the prerequisites for die interleaving? Are there any problems with the DIMM installation rules?
A2:
The configuration of "Physical 0 of channels 0, 1, 4, and 5" in A1 is a common configuration method that meets the inter-die interleaving requirements. The configuration method provided in the DIMM installation rules usually ensures verified and good DIMM performance. Whether the DIMM installation method meets the inter-die interleaving requirements depends on the specific product model.
Does the Kunpeng server support the encryption module?
Kunpeng 920 processors are designed based on the Armv8 architecture and support the encryption instruction set. Encryption algorithms such as SM3 and SM4 can be implemented on the Kunpeng 920 server through software instructions without dedicated hardware-level acceleration. You can use the Kunpeng Accelerator Engine (KAE) to enable hardware-level acceleration (for example, in high-performance scenarios).
How do I apply for a license and import the license to the Kunpeng server to activate the KAE function?
You can log in to the ESDP platform to apply for a KAE license. For details about how to import a license to activate the KAE function, see User Guide (KAE).
How do I calculate the floating-point computing power of the Kunpeng 920 processor?
- Number of floating-point operations per cycle of a single physical core = Number of CPU FMA units × (Maximum bit width of the vector register/Data bit width) × 2
- Floating-point computing power of a single physical core = Number of floating-point operations per cycle of a single physical core × Frequency
- Floating-point computing power of the entire system = Floating-point computing power of a single physical core × Number of cores
(Note: 2 in the formula indicates that the multiply-add instruction performs two operations.)
Take the Kunpeng 920 7260/5250/5220 processor as an example. For details about the parameter values, see Table 1. The following can be calculated:
Half-precision floating-point computing power of the 64-core Kunpeng 920 7260/5250/5220 processor = 2 × (128/16) × 2 × 2.6 × 64 = 5324.8 GFLOPS
Single-precision floating-point computing power of the 64-core Kunpeng 920 7260/5250/5220 processor = 2 × (128/32) × 2 × 2.6 × 64 = 2662.4 GFLOPS
Double-precision floating-point computing power of the 64-core Kunpeng 920 7260/5250/5220 processor = 1 × (128/64) × 2 × 2.6 × 64 = 665.6 GFLOPS
|
CPU Model |
Number of CPU FMA Units |
Maximum Bit Width of the Vector Register |
Data Bit Width |
Frequency |
Number of Cores |
|---|---|---|---|---|---|
|
Kunpeng 920 7260 |
Half-precision: 2 Single-precision: 2 Double-precision: 1 |
128 |
Half-precision: 16 Single-precision: 32 Double-precision: 64 |
2.6 GHz |
64 |
|
Kunpeng 920 5250 |
48 |
||||
|
Kunpeng 920 5220 |
32 |
Do the password modules built in Kunpeng 920 series servers support virtualization and provide independent key management for VMs?
The password modules of Kunpeng 920 series servers support virtualization deployment and provide independent key management for VMs. This function must be used with related software. For details, see Kunpeng Documentation > Commercial Cryptography Application > Technical White Paper.
Do Kunpeng 920 series servers support the memory sparing technology?
Yes. In addition, Kunpeng 920 series servers support memory protection functions including ECC, SEC/DED, SDDC, and patrol scrubbing.