更新日志
Update VPS Host v2016
- Disabled the fwupd package for Linux images as a workaround for an Ubuntu 24 issue under VPS environments.
- Removed rm_init_failed error from watchdog.
- Set VGA type to none for Windows templates to improve GPU passthrough support.
- Fixed incorrect running status for VPS created with errors (e.g. GPU cold3 state).
- Added audio device for Windows VPS.
- Added automatic scandisk with disk error fixing before VPS resume.
- Fixed script arrays handling in listIdGet for VPS.
- Fixed incorrect handling of resumed VPS being treated as config-only changes cause issue with resume under some conditions.
- Removed hv_ipi=on,hv_synic,hv_stimer from VPS params.
- Added aio=threads to scsi0 options.
Update Docker Host v1068
- Moved GPU purging for frozen jobs to occur before GPU reset.
- Improved detection of non-responsive Docker during veryfying docker instance working.
- Minor fixes in the local logging system.
- Removed rm_init_failed error from watchdog logic.
- Added automatic copying of mount point contents to Persistent Storage when enabled and initially empty.
Instance
- VPS: added Windows 11 image
Instance
- Docker: added "Add Persistence Volume" option in Explore Apps template details
- VPS: added ability to change SSH key in resume process
Update Docker Host v1065
- Updated statistics to use the new Docker API version.
- Restored GPU reset functionality after removing an instance.
- Disabled secure connection for simplepodai and pytorch-cuda Docker templates.
- Disabled secure connection for FileManager.
Update VPS Host v2011
- Improved the VPS shutdown procedure timing.
- Restored qm stop in addition to qm shutdown as a workaround when ACPI shutdown fails on frozen VPS instances.
- VPS instances are now reconfigured when the instanceId changes.
- Disabled KVM errors for users and limited them to smdebug only.
- Added log rotation with cleanup of old atop, pveupload, and other obsolete files on the main disk.
- Disabled KSM (Kernel Samepage Merging).
- Added several improvements for Windows VPS instances.
- Improved VPS creation speed.
- Fixed CPU affinity assignment for VPS instances using HugePages.
- Reassigned cut off memory blocks between NUMA nodes for VPS instances.
- Simplified the HugePage CPU counting procedure.
- Added a "VPS Resuming" status message in the user console.
- Added statLeak for debugging memory and process leaks.
- Added forced AES support to CPU options.
- Reduced the wait time for MAC address verification in cloud-init.
- Updated statVmi to treat the internal-error status as running (possible blue screen) instead of undefined.
- VPS instances with internal-error status are now restarted automatically.
Instance
- Docker: added Share button in Template Details at Explore Apps section
- many fixes
Update Docker Host v1062
- Added a watchdog for ShellInABox.
- Temporarily disabled the OOM check.
- Added bond+ support in the firewall and fixed gateway detection under certain bond-related conditions.
- Fixed an issue where MFS did not detect disks outside a datacenter under specific conditions.
- Updated MFS to version 4.58.2.
- Stats now use 100 percent of system RAM value from the API instead of 95 percent.
- Fixed killing CF tunnels inside Docker.
- Fixed chat websocket screen management.
Update VPS Host v2010
- Improved ping statistics
- Increased debug file history retention
- Plugin messages are suppressed in VPS console
- Simplified VPS shutdown procedure with log suppression
- sysStat now sends 100 percent of host physical memory size to the API letting it to memory managing
- VPS is recreated from scratch (except disk data) when a stopped state is detected instead of just rererun
- Fixed NODELOCK looping issue caused by grabbing VPS config while status is still running
- Fixed NODELOCK grabbing creating and refreshing logic
- Preparation to support HugePages with full NUMA GPU/Memory/CPUs redistribution for maximum performance
Update Docker Host v1061
- Watchdog: added “BAR not claimed” error (code h46)
- Disabled instance management during host shutdown and reboot
- Increased debug log rotation threshold
- Watchdog now resends the error code if the same error reoccurs and unlists the host again
- Added periodic autofix for crashed docker instance SSHD configurations
- Increased sleep duration to 5 minutes in forced reboot/shutdown scripts to ensure standard reboot procedures complete
- Temporarily disabled GPU reset
- Added GPU purge functionality for frozen jobs
- Improved GPU error handling in watchdog by validating PCI IDs against the GPU list
- Fixed SSH restarting inside Docker instances after configuration correction
Update VPS Host v2008
- Improved system process handling by resizing /run/lock
- Improved host reboot and shutdown procedures
- Improved disk performance on the host filesystem storage
- Enhanced GPU error handling in errParseDmesg.sh by validating PCI IDs against the GPU list
Update VPS Host v2007
- Improved and simplified VPS shutdown/stop operations
- Watchdog resend error to dashboard when error on host reoccured (and unlist host again)
Update Docker Host v1059
- Watchdog now reports specific AER errors instead of a generic “hardware error” for easier troubleshooting.
- Upgraded the MFS storage system to improve speed and overall performance.
- Added automatic disk optimization with fstrim running regularly in the background.
Update VPS Host v2005
- Changed "VPS Locked" message to more human
- SatGpu added new NV 3xxx definition
- Fixed removing coudinit and efi disks on mfs
- Added fstrim in crontab
- VPS start changed virtio-scsi-pci to virtio-scsi-single
- VPS start added iothread=1 option for disk
- Fixed not resuming instance when status=unknown (source host offline?)
Instance
- API: removed itemsPerPage and page query parameters in Find instances (/instances/market/list) from docs as they have no effect on returned data
- Added Brazil to countries list in user invoice details
Instance
- improved VPS disk speed by using different network storage system. From previous 130MB/s to 2200MB/s. All newly created VPS will be deployed on new storage. Old VPS will be migrated within few weeks.
Update VPS Host v2004
- Added MooseFS support as default storage
- Watchdog changed "Hardware Error" to "AER" detection string
- Added support for diskResize on host side
Instance
- Added visual charts to display balance changes in the billing section for easier tracking.
- Restricted login and registration access from Russia, Belarus, and Singapore due to unsupported activities in these regions.
- Applied various minor fixes and optimizations across the platform to enhance stability and performance.
Update Docker Host v1057
- back count GPU to simplest lspci due to issues with nvtool, nvidia-smi under overload
- added rriShowGpuReservations.sh script for debug purposes
- added MIG support
- statGpu gpuMemUtil now return max(100%) when invalid (not supported) value returned from driver
- rebootTwice deprecated code purged
- added missing part of log debugs in statusHost
- rriStartPortMissflag removed after send to API to not block host forever
- logRotate fixed checking if shared-disk rri contains mounted MFS resource
- improved GPU reset procedure (added killing binded/freezed apps before reset)
- increase log retention periods to 30 days for various log files and archives
- smdebug added mc config edit internal editor
- statsGpu PCIe gen reading only once
Update VPS Host v2003
- watchdogSystem added error handling h45 AER Hardware Error
- watchdogSystem disabled Reboot
- vmiManager fixed reading new assoc table of pveIds
- Prepared hosts for Hugepages support. With transparent option.
- Hide not matter error message vmiStat empty during VPS creating process
- "VPS stopped. VPS paused." messave changed to just one simple: "VPS paused"
- vmiStart added condition to catch status=stopped with errorCreating flag
- vmiStart removing EFI components moved before importdisk step
- vmiStart/Remove fixed NODELOCK
- vmiStart/Remove check lock (backup?) before doing job
Update VPS Host v2002
- added smdebug command
- better VPS stats errors handling
- vmiRemove using stop instead of shutdown during remove VPS (not pausing)
- statusHost added more debug code
- added ERROR unlisting: device mapping invalid (IOMMU?)
- added ERROR unlisting: vfio BAR not claimed (reset GPU failed?)
- vmiStart improved removing old rbd cloud-init file
- statVmi fixed checking subprocess existing
- statSys static sysHddSharedSize=40TB
- vmiStart new metod of veryfing if VPS is running
- vmiStart added removing rbd stale or locked files
- vmiStart added removing old/prev resources before adding them again
- vmiStart now not break (return) if some step failed
- vmiListIdsGet improved and used faster method
- vmiRemove added missing timeouts for two commands
- rriStart fixed typo in mfs mounting folder name
- rriRemove umount both msf/mfs names folders if present in shared-disk directory
- watchdog added error h45: Hardware Error string detection
- removed deprecated ifSys variable in code
- updated process name updVersion in watchdog code
- disable reboot in watchdog
- added dmesg new method reading based on journalctl with marks
- enabling watchdog script
Update v1056
- Fixed GPU usage percentage with abnormal value for non fully supported GPUs by NVIDIA driver. Value force to 100%
- Fixed docker GPU stats not displaying
- Added more debug for gpuStat and statusHost
- Make sure /shared-disk have proper privs
Update v1055
- Improved network speed, disk speed, stats
- Improved hard drive speed settings
- Fixed internal chat mechanizm doubled
- Fixed stale old rriMissGpuError flag causing keeping unlisted host
- Changed error "h36: GPU missing during starting new instance" from permanent to "once". Can be now cleared by user and user can list again the host
- Improved GPU resetting when GPU not fully purged by previous instance
- Fixed reading gpuCount from system by dropping all method to common lspci command
- Fixed zero sysNetSpeedJSON file causing looped speedtest on hosts
- Switched mfsManager install packages to batch mode (non-intercactive)
- Rclocal/rriStart clear GPU reservation error flagimproved include config.txt file
- Removed mapGpu2Rri. Now rriStart reads realtime GPU usage/reservation from docker daemon
- Improves GPU resetting (trying 3times)
- Changed error type to warning of rriStartGpuMiss temporary
- Fixed and improved syntax and speed of some jq queries
Instance
- Persistent Volumes Pricing Update
- New Volumes: Starting immediately, persistent storage is priced at $0.03/GB per month.
- Existing Volumes: Will remain free ($0.00/GB per month) until September 27, 2025. After this date, the new pricing of $0.03/GB per month will apply.
- Volume Resizing: Any resizing action will trigger the new pricing instantly, regardless of the original creation date.
Update v1050
- Fixed unlisting hosts because of querying NVIDIA GPU count (short time lower count of GPUs issue)
Instance
- Docker: Fixed empty ports data and other stat informations in some conditions
Instance
- Docker: Fixed display of tags in Docker templates
- VPS: Fixed price calculations in instance details
- VPS: Improved resume process for enhanced usability and reduced confusion
- API: added VPS renting methods and reorganized menu structure to support separate usage paths
Instance
- New Policy: If your account balance drops below 0, your VPS servers will be paused and you will be billed for disk usage.
- Upcoming Change: In the future, instances may be deleted if the balance remains negative for an extended period.
Instance
- Fixed: Issue with auto-resume functionality when pausing a server using RTX PRO 6000 Blackwell GPU.
Instance
- Fixed problem with resume process that involved other GPU type
立即开始您的GPU云旅程
SimplePod 让 GPU 租赁 变得简单。无论您的需求是什么，我们都为您提供利用 云端 AI 的工具。选择您的 GPU 服务器，设置软件，几分钟内启动项目。今天就加入我们，体验 云端 AI 的轻松。