Установки BIOS
Настройка BIOS вашей системы является важным шагом в конфигурировании кластера для предоставления наилучшей смеси производительности приложения и мощности эффективности. В данной главе мы определим установки, которые могут максимизировать производительность инфраструктуры Intel ® Omni-Path и приложения. Оптимально следует применить настройки аналогичные приводимым в процессе фаз подъёма кластера и его валидации чтобы продемонстрировать что ваша инфраструктура работает ожидаемым образом. В долгосрочной перспективе вы можете пожелать установить свои BIOS для предоставления более энергосберегающих настроек, даже если они могут снизить до некоторой степени общие производительности приложения и инфраструктуры.
Семейства Intel ® Xeon ® Processor E5 v3 и v4
В приводимой далее таблице приводятся установки BIOS в сервере с ЦПУ семейства процессоров Intel ® Xeon ® E5 V3 и V4, рекомендуемые для всего, что связано с производительностью инфраструктуры Intel ® Omni-Path:
Политика энергопотребления или производительности ЦПУ
Производительный или Сбалансированная производительность 1
Конфигурация рабочей нагрузки
Uncore Frequency Scaling (Масштабирование частоты вне ядра)
Расширенная технология Intel SpeedStep ®
Настраиваемый Intel TDP
Технология Intel ® Turbo Boost
Intel ® VT for Directed I/O (VT-d)
Energy Efficient Turbo
Технология Intel ® Hyper-Threading
не рекомендуется (проверьте в своей конфигурации чтобы проверить даёт ли она преимущества)
IOU Non-posted Prefetch
Запрещёно (по возможности) 2
Snoop Holdoff Count
1. Для получения более согласованной производительности режима Turbo для запрашиваемых рабочих нагрузок установите значение в "Производительный". Одно из двух указанных значений даст в результате хорошую производительность инфраструктуры Intel ® Omni-Path.
2. Доступно в версии BIOS R016 процессора Intel ® Xeon ® E5 v4.
3. В некоторых версиях BIOS также именуется как Memory.SocketInterleave=NUMA
4. Иногда установки PCIe* Max Payload Size и Max Read Req отсутствуют в BIOS. В этом случае могут быть сделаны описанные в Параметры модуля драйвера HFI1 установки параметра pcie_caps=0x51 для вашего драйвера Intel ® OPA (что подразумевает установку MaxPayLoad в 256B и MaxReadReq в 4096B).
5. В некоторых BIOS также именуется как Snooped Response Wait Time for Posted Prefetch .
Процессоры Intel ® Xeon ® Scalable
Для ЦПУ Intel ® Xeon ® Scalable рекомендуется следующее:
Установить самую последнюю версию BIOS доступную у вашего производителя со следующими или подобными настройками:
Политика энергопотребления или производительности ЦПУ
Производительный или Сбалансированная производительность a
Технология Intel ® Hyper-Threading
не рекомендуется (проверьте в своей конфигурации чтобы проверить даёт ли она преимущества)
Snoop Holdoff Count
Uncore Frequency Scaling (Масштабирование частоты вне ядра)
WFR Uncore GV Rate Reduction
Запрещено (по возможности) d
a. Для получения более согласованной производительности режима Turbo для запрашиваемых рабочих нагрузок установите значение в "Производительный". Одно из двух указанных значений даст в результате хорошую производительность инфраструктуры Intel ® Omni-Path.
b. Также в некоторых BIOS именуется как Cluster-on-Die .
c. В некоторых BIOS также именуется как Snooped Response Wait Time for Posted Prefetch .
d. Может иметь различные названия в разных BIOS.
Применяйте установленные по умолчанию установки, в том числе Intel ® Turbo Boost Technology= Enabled .
Для семейства процессоров Intel ® Xeon ® E5 V3 и V4 установки перечислены в Таблице 2.1. Intel рекомендует те же самые значения установить и для ЦПУ Intel ® Xeon ® Scalable, где такие установки возможны.
Разрешите скорости Turbo, как это описано в Драйверы масштабирования частоты ЦПУ.
Процессоры Intel ® Xeon Phi™ x200
Для продуктов семейства Intel ® Xeon Phi™ x200 рекомендуется следующее:
Установить самую последнюю версию BIOS доступную у вашего производителя со следующими или подобными настройками:
Uncore setting: Cluster Mode
По выбору пользователя в зависимости от приложения a
Intel ® VT for Directed I/O (VT-d)
Snoop Holdoff Count
a. Зависит от приложения, однако режим "Quadrant" показвывает определённые преимущества в достижении пиковых значений скорости обмена сообщениями MPI.
b. Хотя виртуализация и не поддерживается в семействе продуктов Intel ® Xeon Phi™ x200, процессоры Intel ® Xeon Phi™ x200 работают в режиме X2APIC, который требует чтобы Intel ® VT for Directed I/O (VTd) оставался включённым. Если Intel ® VT for Directed I/O (VTd) обнаруживается в вашем BIOS, мы рекомендуем установить его в значение "Разрешено". При проверке настроек хоста через сценарий verifyhosts.sh или opaverifyhosts , пожалуйста, не выбирайте/ не включайте в состав тест vtd .
c. В некоторых BIOS также именуется как Snooped Response Wait Time for Posted Prefetch .
Применяйте установленные по умолчанию установки, в том числе Intel ® Turbo Boost Technology= Enabled .
Разрешите скорости Turbo, как это описано в Драйверы масштабирования частоты ЦПУ.
Summary
Dell PowerEdge servers provide a wide range of tunable parameters to allow customers to achieve top performance. The information in this paper outlines the tunable parameters available in the latest generation of PowerEdge servers (for example, R660, R760, MX760, and C6620) and provides recommended settings for different workloads.
Figure 1. PowerEdge R660
Figure 2. PowerEdge R760
The following tables provide the BIOS setting recommendations for the latest generation of PowerEdge servers.
Table 1. BIOS setting recommendations—System profile settings
System setup screen
Recommended setting for performance
for HPC and SPECcpu speed environments
Recommended setting for low latency, Stream, and MLC environments
Recommended
for general business/scientific throughput
(for example, SPECcpu2017)
System profile settings
Performance Per Watt [1]
First select Performance Optimized and then select Custom [1]
System profile settings
CPU Power Management
System profile settings
System profile settings
System profile settings
System profile settings
Autonomous or Disabled [6]
System profile settings
System profile settings
Memory Patrol Scrub
System profile settings
Memory Refresh Rate
System profile settings
System profile settings
Energy Efficient Policy
System profile settings
CPU Interconnect Bus Link Power Management
System profile settings
PCI ASPM L1 Link Power Management
[1] Depends on how system was ordered. Other System Profile defaults are driven by this choice and may be different than the examples listed. Select Performance Profile first, and then select Custom to load optimal profile defaults for further modification
[2] SST Turbo Boost Technology is substantially better than previous generations for latency-sensitive environments, but specific Turbo residency cannot be guaranteed under all workload conditions. Evaluate Turbo Boost Technology in your own environment to choose which setting is most appropriate for your workload, and consider the Dell Controlled Turbo option in parallel.
[3] Monitor/Mwait should only be disabled in parallel with disabling Logical Processor. This will prevent the Linux intel_idle driver from enforcing C-states.
[4] You can test your own environment to determine whether disabling Memory Patrol Scrub is helpful.
[5] Dynamic selection can provide more TDP headroom at the expense of dynamic uncore frequency. Optimal setting is workload dependent.
[6] Autonomous on Air Cooled system or Disabled on Liquid Cooled Systems
Table 2. BIOS setting recommendations—Memory, processor, and iDRAC settings
System setup screen
Recommended setting for performance
for HPC and SPECcpu speed environments
Recommended setting for low latency, Stream, and MLC environments
Recommended
for general business/scientific throughput
(for example, SPECcpu2017)
Memory Node Interleave
DIMM Self Healing
Correctable Error Logging
CPU Interconnect Speed
Maximum Data Rate
Maximum Data Rate
Maximum Data Rate
Maximum Data Rate
Adjacent Cache Line Prefetch
DCU Streamer Prefetcher
DCU IP Prefetcher
Sub NUMA Cluster
SNC 4 on XCC SNC 2 on MCC
SNC 4 on XCC SNC 2 on MCC
Dell Controlled Turbo
Dell Controlled Turbo Optimizer mode
DeadLine LLC Alloc
Dynamic SST Perf Profile
Operating Point 1
Operating Point 1
Operating Point ? [6]
Operating Point 1
[1] Use Optimizer Mode when Memory Bandwidth Sensitive, up to 33% BW reduction with Fault Resilient Mode.
[2] Only available when x4 DIMMS installed in the system.
[3] Logical Processor (Hyper Threading) tends to benefit throughput-oriented workloads such as SPEC CPU2017 INT and FP_RATE. Many HPC workloads disable this option. This only benefits SPEC FP_rate if the thread count scales to the total logical processor count.
[4] Dell Controlled Turbo helps to keep core frequency at the maximum all-cores Turbo frequency, which reduces jitter. Disable if Turbo disabled.
[5] Option is available on liquid cooled systems only.
[6] Depends on if your program is affected by Base and Turbo frequency. Will reduce CPU core count and give higher Base and Turbo frequencies.
Help me with an explanation of these BIOS options please
Can someone please post a brief description of what are these BIOS options for?
I have them in my laptop
Intel RC ACPI Settings
PTID Support
CPU Flex Ratio Override
Intel Trusted Execution Technology
FCLK Frequency (has values of 400MHZ, 800MHz, or 1GHz which is the default)
Race To Halt RTH
HDC Control
Energy Efficient P-State
Energy Efficient Turbo
VR Mailbox Command Options
CPU Lock Configuration
IMON Scaling Support
XTU Interface
vBoost Enable
Above 4GB MemoryMappedIO BIOS Assignment
X2 APIC Opt Out
- Add to quote
- ShareOnly show this user
Here is what I gathered so far.
Intel RC ACPI Settings
it's a power/monitoring chip so as an OEM you could get your own and feed in data to the BIOS/EC in order to control the system.
Allows to enable\disable proprietary temperature and voltage monitoring device PTID, which essentially drops one of the SSDT tables called PtidDevc from the BIOS if you set to disable this feature.
CPU Flex Ratio Override
Intel Trusted Execution Technology
FCLK = some sort of BUS between PCI-E and CPU, 1Ghz is the default when using premas bios, i asked him about that about 1 year ago, on the P775DM3 i never checked but i would guess that's 1ghz already too. Here is a good article about it, back then at release when 800mhz was still forced due to a bug.
Race To Halt RTH
Race to halt is the concept of boosting a workload quickly in order to sleep quickly (AKA race to sleep)
Energy Efficient P-State
Energy Efficient Turbo
VR Mailbox Command Options
CPU Lock Configuration
IMON Scaling Support
about RSR it's Residency State Regulation, don't count me on that but i think it's some sort of protection that the voltage doesn't bounce too dangerous levels when trying real high voltages, like for me when i tried the 4.9Ghz 5.0Ghz runs, the stock bios decided to limit it to max 1.38-1.4v so i never could get 5.0ghz validation stable, HWinfo then reported that RSR was the Throttling reason and my ratio got limit to max x41
Residency State Regulation (RSR)
Allows you to determine whether to automatically lower the CPU turbo ratio if the CPU voltage/temperature
is too high. source
Above 4GB MemoryMappedIO BIOS Assignment
X2 APIC Opt Out
- Add to quote
- ShareOnly show this user
I believe you might have found the answers since but here is a few from your list:
The list of UEFI feature(Workload Optimized 2)¶
UEFI settings of Workload optimized 2 is as follows. In addition, customers can not change the following settings.
Boot Feature¶
Setting items (key)
Value set as (Default)
Instead of the startup the logo display screen, it is set to display the POST screen.
In order to support an OS other than UEFI-compatible OS, it is the corresponding set to the traditional BIOS using the CSM.
Set the display mode of the option ROM. Force BIOS is selected to use the Option ROM display mode that has been set by the system BIOS.
Set the power-up state of the NUMLOCK key.
If an error occurs, and then force the system to select [Enable] to wait until the key <F1> is pressed.
In the setting to perform the notification of the interrupt signal, it is set to take effect if the option ROM is installed on more than one expansion card.
After a failed initial boot, it is set to be able to re-boot the system from the boot device.
This setting is to enable the watchdog timer to restart the system.
The system will control how to shut down when the power button is pressed.
This setting is for the power state at the time of recovery after a power failure. Currently it is set to be able to resume the last of the power state before power loss.
CPU Configuration¶
Setting items (key)
Value set as (Default)
It is the monitoring settings of BIOS to reduce the level of electromagnetic interference.
It is the support setting for the Intel Hyper-Threading technology to improve the performance of the CPU.
Valid setting of the CPU core. In the default setting, it is set “0” in order to enable all of the CPU cores in the system.
It is the setting for the processor that can specify the area and non-area capable of running the application code in memory.
Setting the inventory number (PPIN) control of the protected processor in the system.
It is a setting for read-ahead data and instructions flow from the main memory to the L2 cache and to improve the performance of the CPU.
This setting is for prefetching 128 bytes of cache line as the CPU configuration.
It is a prefetcher setting of the DCU streamer. It prefetches the data stream from the cache memory for the access data and processing the DCU(Data Cache Unit) to speed in order to improve the performance of the CPU.
In order to improve the performance of the network connection and the system, it is a setting for IP prefetcher in the DCU (data cache unit) to prefetch the IP address.
In order to improve the efficiency of the transfer and data access, it is a setting for the Intel DCA (direct cache access) technology.
Based on Intel‘s hyper — threading architecture, each logical processor (thread) is assigned an ID (sensor assistance) of 256 APIC with an 8 — bit bandwidth. The APIC ID extends from 8 bits to 16 bits to provide 512 sensor assistance for each thread to improve CPU performance.
Based on the Intel VT-d, to enable or disable the X2APIC_OPT_OUT Flag on the system.
It is a setting to ensure the security of data utilizing Intel’s AES (Advanced Encryption Standard).
It is a setting in order to use support of Intel Virtualization Technology for support of direct I/O VT-d by reporting I/O device assignment (virtual machine monitor) to VMM via DMAR ACPI table.
Setting items (key)
Value set as (Default)
It is a setting to support the power-saving mode. Customize the power settings of the system is selected as a default setting.
Configure the settings of the power performance bias adjustment by the BIOS or OS.
It s a setting for the control method or how actively is used the particular hardware-based power management options.
Configure the settings of the power-management features.
Setting items (key)
Value set as (Default)
It is a setting that enables a system to adjust the voltage and core frequency of the processor to take advantage of EIST (Enhanced Intel SpeedStep Technology) and automatically reduce power consumption and heat dissipation.
The setting is used in turbo mode in order to enhance the system performance.
This function is to change the P-state Coordination (power-performance state) adjustment type. P — state Coordination is known as SpeedStep for Intel processors. “HW_ALL” to change the P-state Coordination type of the hardware component is selected.
Setting items (key)
Value set as (Default)
This setting is to limit the C-state package registration.
It is a setting for the BIOS to enable the report of the CPU C3 state (ACPI C2) to the OS.
It is a setting for the BIOS to enable the report of the CPU C6 state (ACPI C3) to the OS.
Setting items (key)
Value set as (Default)
it is a setting to carry out the reduction of CPU power consumption, in order to reduce the clock cycle and voltage of the CPU when it’s in a significantly stopped state.
Chipset Configuration¶
Setting items (key)
Value set as (Default)
It is a setting to always be cleared during the electrical tuning EV_DFX Lock bits that are disposed on the processor.
This item is only displayed.
Setting items (key)
Value set as (Default)
It is a setting to support the Intel I/OAT (I/O Acceleration Technology).
It is a setting that does not support the snoop mode for each CB equipment.
It is possible to violate the strict ordering rules on the PCI bus for transactions when a particular transaction is completed before other transactions that have already been enqueued. It is a setting to enable support for Relaxed Ordering.
Setting items (key)
Value set as (Default)
It is an Intel virtualization technology setting. It performs Direct I/O VT-d support by reporting I/O device assignment to VMM (virtual machine monitor) via DMAR ACPI table.
It is a setting to enable the access control service.
It is a setting to enable the interrupt remapping in order to improve the performance of the system.
Setting items (key)
Value set as (Default)
It is a setting to choose the frequency for the QPI link connection.
It is a setting to support the Link L0p in order to reduce the power consumption.
It is a setting to support the Link L1 in order to reduce the power consumption.
It is a setting to enable the Cluster-On-Die support, in order to improve the system performance in cloud computing,
It is a setting is to enable the Early Snoop in order to improve the system performance.
Run the Directory mode, to enable the in memory snoop directory.
It is a setting to enable the Isochronous support in order to meet the QoS requirements (Quality of Service).
Setting items (key)
Value set as (Default)
It is a setting for applying the POR restrictions on the DDR4 frequency and voltage programming.
It is a setting for the maximum memory frequency of on-board memory module.
It is a setting to enhance the integrity of the system performance and data.
It is a setting to set the base line of the power limit at the time of execution of the DRAM module.
it is a setting to set via the automatic voltage control during an idle state of the CPU, thereby reducing power consumption, in order to improve the reliability of PU.
The 4G address space or more memory that is divided between the 2 sockets is set to be enabled.
It is a setting to support the A7 (addressing) mode in order to improve the memory performance.
Setting items (key)
Value set as (Default)
It is a setting to enable the RAS support.
It is a setting to enable support for memory sparing for rank, in order to improve the memory performance.
It is a setting that allows to enable / disable the rank sparing mode.
Patrol scrub is the process that allows the CPU to correct the correctable memory error detected in the memory module and send the correction to the requester (the original source). If this item is enabled, the IO hub is read and if there is no delay due to internal processing, write back 16 K cycles per cache line. With this method, approximately 64 gigabytes of memory behind the IO hub is scrubbed daily.
In this feature, it sets the waiting time of the system before the next patrol scrub is executed.
Demand scrub is a process that enables the CPU to correct correctable memory errors detected in memory modules. If the CPU or I/O issues a demand read command and it is found that the data read from the memory have a correctable error, it corrects the error and sends it to the requester (original source). The memory is updated as well.
It is a setting to support a device tagging.
Setting items (key)
Value set as (Default)
It is a setting to support the legacy USB devices on board.
It is a work-around solution setting for the OS that does not support the XHCI handoff.
It is for operating systems that do not support EHCI handoff. If this item is enabled, EHCI ownership change will be requested by the EHCI driver. Settings are enabled and disabled.
Provides full legacy USB keyboard support for operating systems that do not support legacy USB devices. This setting enables I / O port 60h / 64h emulation support.
It is a setting to enable the USB 3.0 support.
It is a setting to enable the EHCI (Enhanced Host Controller Interface) support of the USB 2.0 connector.
It is a setting to enable the EHCI (Enhanced Host Controller Interface) support of the USB 2.0 connector.