vGPU配置的文档哪里可以找到

嗨M.C，没有这些设置的文档，它们不是用于用户修改。
唯一的例外是frame_rate_limiter，您将在当前的驱动程序发行说明中找到它。
-Andy



以下为原文

Hi M.C, there is no documentation for those settings, they are not intended for user modification. The only current exception to this is frame_rate_limiter which you will find documented in the current driver release notes.

-Andy

更多未记录的参数带有新的驱动程序版本（340.xx）。
对于GPU-Z，信息API（没有vbios的默认设置）似乎存在问题。
plugin0.cuda_enabled
plugin0.vbios_support
CUDA（6.5）即将推出Grid K1和K2 vgpu :-)快乐的黑客攻击。
M.C>
更新：Nvidia决定从Grid2.0驱动程序中拒绝CUDA对公众的支持（除了整张卡M * -8Q配置文件）。
CUDA：
CUDA + VBIOS：



以下为原文

More undocumented parameters came with new driver version (340.xx).
There seems to be problem with information API (default settings without vbios) for GPU-Z.

plugin0.cuda_enabled
plugin0.vbios_support

CUDA (6.5) is coming to Grid K1 and K2 vgpu very soon :-) Happy hacking.

M.C>

Update: Nvidia decided to decline CUDA support for public from Grid2.0 drivers (except whole card M*-8Q profiles).

CUDA:


CUDA+VBIOS:

新司机＆amp;
新参数（340.34 / 340.69）
plugin0.add_on_fb_tax
plugin0.enable_large_sys_mem
（还有一些被删除）
M.C>



以下为原文

New driver & new parameters (340.34/340.69)

plugin0.add_on_fb_tax
plugin0.enable_large_sys_mem

(and some removed)

M.C>

Grid 2.0（xen管理程序驱动程序352.46）
plugin0.address64
plugin0.allow_duplicate_client
plugin0.bindchanneldmacontext
plugin0.channel_reset_delay
plugin0.client_display_mode
plugin0.cuda_enabled
plugin0.debug
plugin0.dev_instances
plugin0.direct_gpu_timer_access
plugin0.disable_vnc
plugin0.disable_wddm1x_preemption
plugin0.displaywarningthreshold
plugin0.emulate_channels
plugin0.enable_guest_process_isolation
plugin0.enable_large_sys_mem
plugin0.enable_msi
plugin0.extra_fb_reservation
plugin0.fake_bar1_start_limit
plugin0.fb_bar1_strap_mb
plugin0.force_cache_snoop
plugin0.framebufferlength
plugin0.frame_copy_engine
plugin0.frame_copy_warning_interval
plugin0.frame_rate_limiter
plugin0.frl_config
plugin0.gpu-PCI-ID
plugin0.intervaltime
plugin0.is_simulation
plugin0.license_support
plugin0.loglevel
plugin0.mappable_cpu_host_aperture
plugin0.mappable_video_size
plugin0.max_fd
plugin0.max_instance
plugin0.max_resolution
plugin0.native_display
plugin0.no_version_check
plugin0.num_heads
plugin0.nvwatch_enabled
plugin0.pdev_id
plugin0.pipelined_pte_blit
plugin0.plugin_path
plugin0.refusebar1access
plugin0.remapper_optimization
plugin0.remappers_enabled
plugin0.reserved_fb
plugin0.sim_force_3d_device
plugin0.sim_thread.cpumask
plugin0.sim_use_vga_display
plugin0.sw_vsync_enabled
plugin0.track_guest_memory_map
plugin0.vbios_support
plugin0.vdev_id
plugin0.vgaintervaltime
plugin0.vgpu_type



以下为原文

Grid 2.0 (xen hypervisor driver 352.46)

plugin0.address64
plugin0.allow_duplicate_client
plugin0.bindchanneldmacontext
plugin0.channel_reset_delay
plugin0.client_display_mode
plugin0.cuda_enabled
plugin0.debug
plugin0.dev_instances
plugin0.direct_gpu_timer_access
plugin0.disable_vnc
plugin0.disable_wddm1x_preemption
plugin0.displaywarningthreshold
plugin0.emulate_channels
plugin0.enable_guest_process_isolation
plugin0.enable_large_sys_mem
plugin0.enable_msi
plugin0.extra_fb_reservation
plugin0.fake_bar1_start_limit
plugin0.fb_bar1_strap_mb
plugin0.force_cache_snoop
plugin0.framebufferlength
plugin0.frame_copy_engine
plugin0.frame_copy_warning_interval
plugin0.frame_rate_limiter
plugin0.frl_config
plugin0.gpu-pci-id
plugin0.intervaltime
plugin0.is_simulation
plugin0.license_support
plugin0.loglevel
plugin0.mappable_cpu_host_aperture
plugin0.mappable_video_size
plugin0.max_fd
plugin0.max_instance
plugin0.max_resolution
plugin0.native_display
plugin0.no_version_check
plugin0.num_heads
plugin0.nvwatch_enabled
plugin0.pdev_id
plugin0.pipelined_pte_blit
plugin0.plugin_path
plugin0.refusebar1access
plugin0.remapper_optimization
plugin0.remappers_enabled
plugin0.reserved_fb
plugin0.sim_force_3d_device
plugin0.sim_thread.cpumask
plugin0.sim_use_vga_display
plugin0.sw_vsync_enabled
plugin0.track_guest_memory_map
plugin0.vbios_support
plugin0.vdev_id
plugin0.vgaintervaltime
plugin0.vgpu_type

2015年8月之后发布的所有驱动程序（Grid2.0）都通过/ usr / share / nvidia / vgpu / vgpuConfig中的数字签名（例如，不可更改）参数集忽略/usr/share/nvidia/vgx/*.conf中的参数。
xml（由新的nvidia-vgpud命令（init脚本）上传）。
...
...
...
...
...
...
...
您不能通过降低参数来限制（降低）vGPU性能（最大显示大小和/或frl（帧速率限制））。
这可能是由于M60 / M6的新NVidia GRID2.0许可（没有NVidia许可证，视频被迫降至1280x1024和3 FPS）。
从Grid4.x更新：即使没有许可证，也需要以完整功能运行许可的vGPU。
但是，在Windows上，在获取许可证之前，每次vGPU无法获得许可时都会向用户发出警告。
获得许可后，这些警告即告终止。
从Grid5.x更新：未经许可的限制又回来了！
- 视频被迫降至1280x1024和3 FPS。
NVidia可以重新考虑这个配置（例如，不阻止K1 / K2的这个参数和/或允许从允许的最大值降低参数，是的，这很简单，因为使用算术运算符“
M.C>
PS：没关系。
绕过这个限制相对容易，但不要问我怎么做。



以下为原文

All drivers released after 8/2015 (Grid2.0) are ignoring parameters from /usr/share/nvidia/vgx/*.conf by digitally signed (eg. unchangeable) parameter set from /usr/share/nvidia/vgpu/vgpuConfig.xml (uploaded by new nvidia-vgpud command (init script)).


...
...

...
...
...
...
...

You cannot throttle (lower) vGPU performance from now (max display size and/or frl (frame rate limit) by lowering that parameters). This is probably due to new NVidia GRID2.0 licensing for M60/M6 (without NVidia license the video is forced to drop to 1280x1024 and 3 FPS).

Update from Grid4.x: vGPUs that require licensing run at full capability even without a license. However, on Windows, until a license is acquired, users are warned each time a vGPU fails to get alicense. These warnings cease after a license is acquired.

Update from Grid5.x: Unlicensed restriction is back ! - video is forced to drop to 1280x1024 and 3 FPS.

Can NVidia rethink this config (eg. not blocking this parameters for K1/K2 and/or allow lowering parameters from allowed maximum values, yes, that is simple as use arithmetic operator "<" (less than) in NVidia code) ?

M.C>

PS: Never mind. It is relatively easy to bypass this restriction but do not ask me how.

从vgpuConfig.xml-352.83 ... 367.92（Grid4.2）中修复的最大值不支持降低分辨率和帧速率限制...



以下为原文

There is not support to lower resolution and frame rate limit from maximum that is fixed in vgpuConfig.xml - 352.83 ... 367.92 (Grid4.2) ...

用户指南/发行说明中正式记录了很少的参数（由“vgpu_extra_args”激活）：
disable_vnc
frame_rate_limiter
enable_large_sys_mem
extra_fb_reservation



以下为原文

Few parameters are officially documented in user guide / release notes (activated by "vgpu_extra_args"):

disable_vnc
frame_rate_limiter
enable_large_sys_mem
extra_fb_reservation

Grid 4.2的vGPU插件参数（xen hypervisor driver 367.92）：
address64
allow_duplicate_client
channel_reset_delay
cuda_enabled
调试
dev_instances
direct_gpu_timer_access
disable_vnc
disable_wddm1x_preemption
emulate_channels
enable_guest_process_isolation
enable_large_sys_mem
enable_mfn_caching
enable_msi
enable_win10_dual_heads_for_small_vgpu
extra_fb_reservation
fake_bar1_start_limit
fb_bar1_strap_mb
fb_scrubbing_enabled
force_cache_snoop
framebufferlength
frame_copy_engine
frame_copy_warning_interval
frame_rate_limiter
frl_config
GPU-PCI-ID
guest_managed_hw_alloc
intervaltime
linux_interrupt_optimization
日志等级
mappable_cpu_host_aperture
mappable_video_size
max_fd
max_instance
max_resolution
no_version_check
num_heads
nvwatch_enabled
pdev_id
pipelined_pte_blit
plugin_path
reclaim_bar2
remapper_optimization
remappers_enabled
reserved_fb
sw_vsync_enabled
track_guest_memory_map
use_per_vm_heap_memory
use_subheap_at_beginning
vbios_support
vdev_id
vgaintervaltime
vgpu_type
请参阅https://docs.nvidia.com/grid/latest/grid-vgpu-release-notes-citrix-xenserver/index.html#win-10-max-heads-under-1-gb-frame-buffer



以下为原文

vGPU plugin parameters for Grid 4.2 (xen hypervisor driver 367.92):

address64
allow_duplicate_client
channel_reset_delay
cuda_enabled
debug
dev_instances
direct_gpu_timer_access
disable_vnc
disable_wddm1x_preemption
emulate_channels
enable_guest_process_isolation
enable_large_sys_mem
enable_mfn_caching
enable_msi
enable_win10_dual_heads_for_small_vgpu
extra_fb_reservation
fake_bar1_start_limit
fb_bar1_strap_mb
fb_scrubbing_enabled
force_cache_snoop
framebufferlength
frame_copy_engine
frame_copy_warning_interval
frame_rate_limiter
frl_config
gpu-pci-id
guest_managed_hw_alloc
intervaltime
linux_interrupt_optimization
loglevel
mappable_cpu_host_aperture
mappable_video_size
max_fd
max_instance
max_resolution
no_version_check
num_heads
nvwatch_enabled
pdev_id
pipelined_pte_blit
plugin_path
reclaim_bar2
remapper_optimization
remappers_enabled
reserved_fb
sw_vsync_enabled
track_guest_memory_map
use_per_vm_heap_memory
use_subheap_at_beginning
vbios_support
vdev_id
vgaintervaltime
vgpu_type
see https://docs.nvidia.com/grid/latest/grid-vgpu-release-notes-citrix-xenserver/index.html#win-10-max-heads-under-1-gb-frame-buffer

这些设置是否也可以在vSphere中运行？



以下为原文

Do these settings also work in vSphere somewhere?

搜索“适用于VMware vSphere的GRID虚拟GPU发行说明”，例如http://docs.nvidia.com/grid/4.0/pdf/grid-vgpu-release-notes-vmware-vsphere.pdf并搜索“frame_rate_limiter”。
..



以下为原文

Search for "GRID Virtual GPU for VMware vSphere Release Notes" for example http://docs.nvidia.com/grid/4.0/pdf/grid-vgpu-release-notes-vmware-vsphere.pdf and search for "frame_rate_limiter" ...

我的观点：NVidia对此没有帮助。
好的，NVidia的方法迫使我接受挑战来解决这个问题。
4小时的实验，现在可能所有参数都可以修改，而不需要触摸我的thinVDI虚拟化堆栈中的任何NVidia文件，也可能在XenSever 7.x中。
我不会发表如何 - 挑战自己。
很简单就有可能。
例如 -  CUDA再次工作（可能在任何地方）。
应谨慎使用Pascal之前的芯片CUDA  - 较旧的芯片不支持抢先式多任务处理。
我根据我的真实编程实验仅描述了我的观点，并且我不试图干涉NVidia的Visions。



以下为原文


My opinion:  NVidia was not helpful with this. Ok, NVidia's approach forced me to accept challenge to resolve this problem. 4 hours of experiments and now probably all parameters are modifiable without touch of any NVidia file in my thinVDI virtualization stack and probably in XenSever 7.x. I will not publish how - challenge yourself. Simple it is possible. For example - CUDA is working again (probably everywhere). CUDA with chips before Pascal should be used carefully - the older chips do not support preemptive multitasking.



I described only my point of view based on my real programming experiments and I do not try to interfere with NVidia's Visions.

Grid 6.0的vGPU插件参数（管理程序驱动程序390.42）：
address64
allow_duplicate_client
channel_reset_delay
调试
dev_instances
direct_gpu_timer_access
disable_vnc
disable_wddm1x_preemption
emulate_channels
emulate_usermode
enable_large_sys_mem
enable_mfn_caching
enable_msi
enable_nvenc_for_small_vgpu
enable_win10_dual_heads_for_small_vgpu
extra_fb_reservation
fake_bar1_start_limit
fb_bar1_strap_mb
fb_scrubbing_enabled
force_cache_snoop
framebufferlength
frame_copy_engine
frame_copy_warning_interval
frame_rate_limiter
frl_config
GPU-PCI-ID
guest_managed_hw_alloc
intervaltime
linux_interrupt_optimization
日志等级
mappable_cpu_host_aperture
mappable_video_size
max_fd
max_instance
max_resolution
migration_feature
no_version_check
num_heads
nvwatch_enabled
pdev_id
pipelined_pte_blit
plugin_path
pte_blit_enabled
pte_blit_src_buff_size
reclaim_bar2
reserved_fb
same_host_guest_bar1_offset
save_pfn_mapping
sw_vsync_enabled
track_guest_memory_map
use_per_vm_heap_memory
use_subheap_at_beginning
vdev_id
vgaintervaltime
vgpu_device_caps
vgpu_type
vmiop_vmm_migration_supported
请参阅https://docs.nvidia.com/grid/latest/grid-vgpu-release-notes-citrix-xenserver/index.html#bug-200130864-512-mb-frame-buffer-memory-exhaustion



以下为原文

vGPU plugin parameters for Grid 6.0 (hypervisor driver 390.42):

address64
allow_duplicate_client
channel_reset_delay
debug
dev_instances
direct_gpu_timer_access
disable_vnc
disable_wddm1x_preemption
emulate_channels
emulate_usermode
enable_large_sys_mem
enable_mfn_caching
enable_msi
enable_nvenc_for_small_vgpu
enable_win10_dual_heads_for_small_vgpu
extra_fb_reservation
fake_bar1_start_limit
fb_bar1_strap_mb
fb_scrubbing_enabled
force_cache_snoop
framebufferlength
frame_copy_engine
frame_copy_warning_interval
frame_rate_limiter
frl_config
gpu-pci-id
guest_managed_hw_alloc
intervaltime
linux_interrupt_optimization
loglevel
mappable_cpu_host_aperture
mappable_video_size
max_fd
max_instance
max_resolution
migration_feature
no_version_check
num_heads
nvwatch_enabled
pdev_id
pipelined_pte_blit
plugin_path
pte_blit_enabled
pte_blit_src_buff_size
reclaim_bar2
reserved_fb
same_host_guest_bar1_offset
save_pfn_mapping
sw_vsync_enabled
track_guest_memory_map
use_per_vm_heap_memory
use_subheap_at_beginning
vdev_id
vgaintervaltime
vgpu_device_caps
vgpu_type
vmiop_vmm_migration_supported
see https://docs.nvidia.com/grid/latest/grid-vgpu-release-notes-citrix-xenserver/index.html#bug-200130864-512-mb-frame-buffer-memory-exhaustion

Grid6.0中Grid 6.2（管理程序驱动程序390.42）的新vGPU插件参数：
cilp_disabled_on_wddm
pte_blit_cmd_buff_size
请参阅https://docs.nvidia.com/grid/latest/grid-vgpu-release-notes-citrix-xenserver/index.html#bug-200333574-cilp-not-working-in-windows-vms



以下为原文

New vGPU plugin parameters for Grid 6.2 (hypervisor driver 390.42) from Grid6.0:
cilp_disabled_on_wddm
pte_blit_cmd_buff_size
see https://docs.nvidia.com/grid/latest/grid-vgpu-release-notes-citrix-xenserver/index.html#bug-200333574-cilp-not-working-in-windows-vms