GRID 3.0使用M60 GPU在ESXI 6.0.2上成功安装但无法通过nvidia-smi进行验证

嗨，亚历克斯，
我担心自己不是VMware专家。
但我正在检查支持和产品团队的已知问题。
您有权获得M60和GRID 3.0的全面支持 - 您是否已提出支持案例？
最好的祝愿，
雷切尔



以下为原文

Hi Alex,

I'm afraid I'm not a VMware expert myself. But I'm checking for known issues with the support and product teams. You are entitled to full support with M60 and GRID 3.0 - have you raised a support case yet?

Best wishes,
Rachel

不，我没有提出支持案例。
我应该继续创建吗？



以下为原文

No I haven't raised a support case. Should I go ahead and create one?

是的，这将是好的，提出一个支持案例和pm（个人信息）我的数字，我会密切关注它。
我们的一位工程师已经在研究这个问题。
请将我的名字添加到故障单中，这样前线就不必追逐信息，我会把它们填入。
雷切尔



以下为原文

Yes that would be good, raise a support case and pm (personal message) me the number and I'll keep an eye on it. One of our engineers is already looking into this. Please add my name to the ticket so frontline don't have to chase info and I'll fil them in.

Rachel

在ESXi主机CLI上，请运行
lspci -n |
grep 10de
然后在这里发布结果。



以下为原文

At the ESXi host CLI please run

lspci –n | grep 10de

then post the result here.

嗨，杰森，这是结果
[root @ localhost：〜] lspci -n |
grep 10de
0000：04：00.0 0300：10de：13f2 [vmgfx6]
0000：05：00.0 0300：10de：13f2 [vmgfx7]
0000：08：00.0 0300：10de：13f2 [vmgfx4]
0000：09：00.0 0300：10de：13f2 [vmgfx5]
0000：83：00.0 0300：10de：13f2 [vmgfx2]
0000：84：00.0 0300：10de：13f2 [vmgfx3]
0000：87：00.0 0300：10de：13f2 [vmgfx0]
0000：88：00.0 0300：10de：13f2 [vmgfx1]



以下为原文

Hi Jason, here are the results

[root@localhost:~] lspci -n | grep 10de
0000:04:00.0 Class 0300: 10de:13f2 [vmgfx6]
0000:05:00.0 Class 0300: 10de:13f2 [vmgfx7]
0000:08:00.0 Class 0300: 10de:13f2 [vmgfx4]
0000:09:00.0 Class 0300: 10de:13f2 [vmgfx5]
0000:83:00.0 Class 0300: 10de:13f2 [vmgfx2]
0000:84:00.0 Class 0300: 10de:13f2 [vmgfx3]
0000:87:00.0 Class 0300: 10de:13f2 [vmgfx0]
0000:88:00.0 Class 0300: 10de:13f2 [vmgfx1]

是GPU的工厂吗？



以下为原文

were the GPU's factory fitted?

这些GPU是由NVIDIA直接提供的，因为我们是NVIDIA的合作伙伴，我们自己安装它们。
相同的GPU刚刚在此服务器上成功完成了NVQUAL。



以下为原文

These GPU's were provided from NVIDIA directly as we are an NVIDIA partner and we installed them ourselves. The same GPU's just successfully completed the NVQUAL on this server.

这表明M60 / M6 GPU已在图形模式下正确设置 - 任何遇到M60 / M6问题的人都可以轻松地仔细检查这一点，请遵循以下建议 http://nvidia.custhelp.com/app/answers/detail
/ a_id / 4106 / [/ url]。
我担心我对这个案子没有任何建议。

以上来自于谷歌翻译

以下为原文



This shows the M60 / M6 GPU is correctly set in graphics mode - anyone else experiencing M60 / M6 issues with can double-check this easily, following this advice [url]http://nvidia.custhelp.com/app/answers/detail/a_id/4106/. I'm afraid I haven't got any suggestions for this case though.

我尝试过的更多东西
我取出了4个GPU中的3个并禁用了SuperMicro BIOS中的“高于4G解码”。
（1）GPU和“4g以上解码”禁用并运行nvidia-smi以正确返回：
[root @ localhost：〜] nvidia-smi
2016年5月5日星期五19:18:01
+ -------------------------------------------------
----- +
|
NVIDIA-SMI 361.40驱动程序版本：361.40 |
| ------------------------------- + -----------------
----- + ---------------------- +
|
GPU名称持久性-M |
Bus-Id Disp.A |
挥发性的Uncorr。
ECC |
|
Fan Temp Perf Pwr：用法/上限|
内存使用|
GPU-Util Compute M. |
| =============================== + =================
===== + ====================== |
|
0特斯拉M60开|
0000：83：00.0关|
关|
|
N / A 38C P8 24W / 150W |
19MiB / 8191MiB |
0％默认值|
+ ------------------------------- + -----------------
----- + ---------------------- +
|
1特斯拉M60开|
0000：84：00.0关|
关|
|
N / A 33C P8 24W / 150W |
19MiB / 8191MiB |
0％默认值|
+ ------------------------------- + -----------------
----- + ---------------------- +
+ -------------------------------------------------
---------------------------- +
|
进程：GPU内存|
|
GPU PID类型进程名称用法|
| =================================================
============================ |
|
找不到正在运行的进程|
+ ---------------------------------------
我然后启用“4g以上解码”并离开（1）卡并运行nvidia-smi时得到：“无法初始化NVML：未知错误”
我安装了第二个GPU并禁用“4g以上解码”并启动并获得BIOS错误：“检测到的PCI资源不足”因此除非安装了超过（1）GPU的“4g以上解码”，否则它将无法启动。
所以我重新启动了“4g以上解码”并尝试了nvidia-smi并得到了相同的“无法初始化NVML：未知错误”
基本上它只用于安装（1）卡和禁用“4g以上解码”。



以下为原文

More things I’ve tried

I took out 3 of the 4 GPUs and disabled "Above 4G Decoding" in the SuperMicro BIOS.

(1) GPU and “Above 4g Decoding” disabled and ran nvidia-smi to properly return:

[root@localhost:~] nvidia-smi
Thu May 5 19:18:01 2016
+------------------------------------------------------+
| NVIDIA-SMI 361.40 Driver Version: 361.40 |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
|===============================+======================+======================|
| 0 Tesla M60 On | 0000:83:00.0 Off | Off |
| N/A 38C P8 24W / 150W | 19MiB / 8191MiB | 0% Default |
+-------------------------------+----------------------+----------------------+
| 1 Tesla M60 On | 0000:84:00.0 Off | Off |
| N/A 33C P8 24W / 150W | 19MiB / 8191MiB | 0% Default |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes: GPU Memory |
| GPU PID Type Process name Usage |
|=============================================================================|
| No running processes found |
+---------------------------------------

I then enabled “Above 4g Decoding” and left the (1) card and got the: “Failed to initialize NVML: Unknown Error” when running nvidia-smi

I installed a 2nd GPU and disabled “Above 4g Decoding” and booted and get the BIOS ERROR: “Insufficient PCI Resources Detected” So it won’t boot unless the “Above 4g Decoding” is enabled with more than (1) GPU installed.

So I rebooted with “Above 4g Decoding” enabled and tried the nvidia-smi and got the same ”Failed to initialize NVML: Unknown Error”

Basically it’s only working with only (1) card installed and “Above 4g Decoding” disabled.

https://gridforums.nvidia.com/default/topic/526/necessary-to-disable-quot-above-4g-decoding-quot-for-view-with-vgpu-/？
https://gridforums.nvidia.com/default/topic/546/nvidia-grid-vgpu/mmio-above-4-gb-esxi-6-0u1-vgpu/？
（http://www.supermicro.com/support/faqs/faq.cfm?faq=20016？）
NVQUAL：对于NVIDIA vGPU应用程序，GPU应映射到4GB地址空间（BAR1
JS：在ESXi中，您必须将MMIO设置为低于4G。
VMware文章是正确的。
尽管ESXi是64位虚拟机管理程序，但它仍然具有此限制。
这些陈述今天仍然有效吗？
你可以使用GPU作为passthrough（vDGA）并启用“4g以上解码”（https://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=2139299）。



以下为原文

https://gridforums.nvidia.com/default/topic/526/necessary-to-disable-quot-above-4g-decoding-quot-for-view-with-vgpu-/ ?
https://gridforums.nvidia.com/default/topic/546/nvidia-grid-vgpu/mmio-above-4-gb-esxi-6-0u1-vgpu/ ?
( http://www.supermicro.com/support/faqs/faq.cfm?faq=20016 ? )

NVQUAL: For NVIDIA vGPU application, the GPUs should be mapped below the 4GB address space (BAR1<32b).
JS: In ESXi you have to have MMIO set to below 4G. The VMware article is correct. Although ESXi is a 64bit hypervisor it still has this restriction.

Are these statements still valid today ?

You can probably use GPU as passthrough (vDGA) with enabled “Above 4g Decoding” ( https://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=2139299 ).

是的ESXi仍需要低于4G。



以下为原文



Yes ESXi still requires below 4G.

现在我们到了某个地方。
是否安装了其他PCI设备？
您是否在机箱的最新BIOS修订版中？
您是否已检查VMware HCL以获取此配置？
您是否有办法尝试不限于以下4G解码限制的虚拟机管理程序？
（XenServer是最简单的）。



以下为原文



Now we're getting somewhere.

Are there any other PCI devices installed?

Are you at the latest BIOS revision for the chassis?

Have you checked the VMware HCL for this configuration?

Do you have the means to try a hypervisor that is not limited to the below 4G decoding limit? (XenServer is the easiest).

是的ESXi仍需要低于4G。
这个要求导致只能启用几张卡（例如，一张卡需要> 2x 128MB + 16MB + 32MB）。
512MB-1.5GB的mmio孔是标准配置，但我没有看到任何选项来配置BIOS中的mmio孔尺寸。
我的超微X9DR3-F有2GB mmio孔，我不能添加超过1-2张卡（在Xen dom0和禁用“4g以上解码”）。
我使用XenServer（> = 6.5）并且启用“4g以上解码”工作正常。
如果在ESXi中可用，请尝试分析此程序输出以获取内存PCI BAR映射：
#lspci -vv |
egrep“^ [a-f0-9] |记忆在”
#cat / proc / iomem
#dmesg |
grep“可用于PCI设备”
#dmesg |
grep“pci_bus”
#dmesg |
grep“ [mem”



以下为原文


Yes ESXi still requires below 4G.

This requirements leads to only few cards can be enabled (eg. one card needs >2x 128MB+16MB+32MB). The 512MB-1.5GB of mmio hole is standard but I do not see any options to configure mmio hole size in bios. My supermicro X9DR3-F has 2GB mmio hole and I cannot add more than 1-2 cards (under Xen dom0 and disabled "Above 4g Decoding"). I use XenServer (>=6.5) and it works fine with enabled "Above 4g Decoding".

Try to analyze this programs output for memory PCI BAR mappings if available in ESXi:
# lspci -vv | egrep "^[a-f0-9]|Memory at"
# cat /proc/iomem
# dmesg | grep "available for PCI devices"
# dmesg | grep "pci_bus"
# dmesg | grep "[mem "

在听完SuperMicro的回复并在这里进行一些调整后，我们现在有2个GPU在工作。
他们建议我们将MMIOHBase设置更改为“2T”。
该设置位于Advanced PCIe / PCI / PnP Configuration下。
我们还在BIOS中更改了以下内容：
https://dl.dropboxusercontent.com/u/4009063/PCIe_PCI_PnP_Part_2%20Changes.jpg



以下为原文

After hearing back from SuperMicro and doing some tweaking here we now have 2 GPUs working. They suggested we change the MMIOHBase setting to “2T”. The setting is located under Advanced PCIe/PCI/PnP Configuration. We also changed the following in the BIOS:
https://dl.dropboxusercontent.com/u/4009063/PCIe_PCI_PnP_Part_2%20Changes.jpg

None


以下为原文

The above BIOS settings work with 3 cards but when you add the 4th card then only 3 are recognized. I tested the one card itself and it works by itself with all the others removed. I just can't get all 4 to be seen via nvidia-smi and lspci -n | grep 10de

[root@localhost:~] lspci -n | grep 10de
0000:04:00.0 Class 0300: 10de:13f2 [vmgfx4]
0000:05:00.0 Class 0300: 10de:13f2 [vmgfx5]
0000:83:00.0 Class 0300: 10de:13f2 [vmgfx2]
0000:84:00.0 Class 0300: 10de:13f2 [vmgfx3]
0000:87:00.0 Class 0300: 10de:13f2 [vmgfx0]
0000:88:00.0 Class 0300: 10de:13f2 [vmgfx1]

More fun stuff to figure out.

我认为这些设置会导致从2TB开始激活“64位”PCI mmio BAR寻址（例如，低于4TB 32位地址限制）。
前几张卡很幸运，mmio BAR低于4TB 32bit限制，内核/驱动程序/ nvidia-smi可见，但第四张卡超出此限制且无法看见。
您可以尝试使用以下命令研究卡片的mmio BAR（10de：13f2）的内存分配吗？
#lspci -nvv |
egrep“^ [a-f0-9] |记忆在”



以下为原文

I think that the settings leads to activate "64bit" PCI mmio BAR addressing starting from 2TB (eg. bellow 4TB 32bit address limit). The first few cards get lucky and mmio BAR is under 4TB 32bit limit and visible to kernel/driver/nvidia-smi but 4th card is beyond this limit and invisible. Can you try and study memory assignments of mmio BAR of your cards (10de:13f2) with following command ?

# lspci -nvv | egrep "^[a-f0-9]|Memory at"

None


以下为原文

Here's what I get with the 4 cards installed:

[root@localhost:~] lspci -nvv | egrep "^[a-f0-9]|Memory at"
0000:00:00.0 Class 0600: 8086:2f00 [PCIe RP[0000:00:00.0]]
0000:00:01.0 Class 0604: 8086:2f02 [PCIe RP[0000:00:01.0]]
0000:00:02.0 Class 0604: 8086:2f04 [PCIe RP[0000:00:02.0]]
0000:00:03.0 Class 0604: 8086:2f08 [PCIe RP[0000:00:03.0]]
0000:00:04.0 Class 0880: 8086:2f20
0000:00:04.1 Class 0880: 8086:2f21
0000:00:04.2 Class 0880: 8086:2f22
0000:00:04.3 Class 0880: 8086:2f23
0000:00:04.4 Class 0880: 8086:2f24
0000:00:04.5 Class 0880: 8086:2f25
0000:00:04.6 Class 0880: 8086:2f26
0000:00:04.7 Class 0880: 8086:2f27
0000:00:05.0 Class 0880: 8086:2f28
0000:00:05.1 Class 0880: 8086:2f29
0000:00:05.2 Class 0880: 8086:2f2a
0000:00:05.4 Class 0800: 8086:2f2c
0000:00:11.0 Class ff00: 8086:8d7c
0000:00:11.4 Class 0106: 8086:8d62 [vmhba0]
0000:00:14.0 Class 0c03: 8086:8d31
0000:00:16.0 Class 0780: 8086:8d3a
0000:00:16.1 Class 0780: 8086:8d3b
0000:00:1a.0 Class 0c03: 8086:8d2d
0000:00:1c.0 Class 0604: 8086:8d10 [PCIe RP[0000:00:1c.0]]
0000:00:1c.4 Class 0604: 8086:8d18 [PCIe RP[0000:00:1c.4]]
0000:00:1d.0 Class 0c03: 8086:8d26
0000:00:1f.0 Class 0601: 8086:8d44
0000:00:1f.2 Class 0106: 8086:8d02 [vmhba1]
0000:00:1f.3 Class 0c05: 8086:8d22
0000:02:00.0 Class 0604: 10b5:8747
0000:03:08.0 Class 0604: 10b5:8747
0000:03:10.0 Class 0604: 10b5:8747
0000:04:00.0 Class 0300: 10de:13f2 [vmgfx4]
0000:05:00.0 Class 0300: 10de:13f2 [vmgfx5]
0000:07:00.0 Class 0200: 8086:1528 [vmnic0]
0000:07:00.1 Class 0200: 8086:1528 [vmnic1]
0000:08:00.0 Class 0604: 1a03:1150
0000:09:00.0 Class 0300: 1a03:2000
0000:7f:08.0 Class 0880: 8086:2f80
0000:7f:08.2 Class 1101: 8086:2f32
0000:7f:08.3 Class 0880: 8086:2f83
0000:7f:09.0 Class 0880: 8086:2f90
0000:7f:09.2 Class 1101: 8086:2f33
0000:7f:09.3 Class 0880: 8086:2f93
0000:7f:0b.0 Class 0880: 8086:2f81
0000:7f:0b.1 Class 1101: 8086:2f36
0000:7f:0b.2 Class 1101: 8086:2f37
0000:7f:0c.0 Class 0880: 8086:2fe0
0000:7f:0c.1 Class 0880: 8086:2fe1
0000:7f:0c.2 Class 0880: 8086:2fe2
0000:7f:0c.3 Class 0880: 8086:2fe3
0000:7f:0c.4 Class 0880: 8086:2fe4
0000:7f:0c.5 Class 0880: 8086:2fe5
0000:7f:0c.6 Class 0880: 8086:2fe6
0000:7f:0c.7 Class 0880: 8086:2fe7
0000:7f:0d.0 Class 0880: 8086:2fe8
0000:7f:0d.1 Class 0880: 8086:2fe9
0000:7f:0d.2 Class 0880: 8086:2fea
0000:7f:0d.3 Class 0880: 8086:2feb
0000:7f:0d.4 Class 0880: 8086:2fec
0000:7f:0d.5 Class 0880: 8086:2fed
0000:7f:0f.0 Class 0880: 8086:2ff8
0000:7f:0f.1 Class 0880: 8086:2ff9
0000:7f:0f.2 Class 0880: 8086:2ffa
0000:7f:0f.3 Class 0880: 8086:2ffb
0000:7f:0f.4 Class 0880: 8086:2ffc
0000:7f:0f.5 Class 0880: 8086:2ffd
0000:7f:0f.6 Class 0880: 8086:2ffe
0000:7f:10.0 Class 0880: 8086:2f1d
0000:7f:10.1 Class 1101: 8086:2f34
0000:7f:10.5 Class 0880: 8086:2f1e
0000:7f:10.6 Class 1101: 8086:2f7d
0000:7f:10.7 Class 0880: 8086:2f1f
0000:7f:12.0 Class 0880: 8086:2fa0
0000:7f:12.1 Class 1101: 8086:2f30
0000:7f:12.4 Class 0880: 8086:2f60
0000:7f:12.5 Class 1101: 8086:2f38
0000:7f:13.0 Class 0880: 8086:2fa8
0000:7f:13.1 Class 0880: 8086:2f71
0000:7f:13.2 Class 0880: 8086:2faa
0000:7f:13.3 Class 0880: 8086:2fab
0000:7f:13.6 Class 0880: 8086:2fae
0000:7f:13.7 Class 0880: 8086:2faf
0000:7f:14.0 Class 0880: 8086:2fb0
0000:7f:14.1 Class 0880: 8086:2fb1
0000:7f:14.2 Class 0880: 8086:2fb2
0000:7f:14.3 Class 0880: 8086:2fb3
0000:7f:14.4 Class 0880: 8086:2fbc
0000:7f:14.5 Class 0880: 8086:2fbd
0000:7f:14.6 Class 0880: 8086:2fbe
0000:7f:14.7 Class 0880: 8086:2fbf
0000:7f:16.0 Class 0880: 8086:2f68
0000:7f:16.1 Class 0880: 8086:2f79
0000:7f:16.2 Class 0880: 8086:2f6a
0000:7f:16.3 Class 0880: 8086:2f6b
0000:7f:16.6 Class 0880: 8086:2f6e
0000:7f:16.7 Class 0880: 8086:2f6f
0000:7f:17.0 Class 0880: 8086:2fd0
0000:7f:17.1 Class 0880: 8086:2fd1
0000:7f:17.2 Class 0880: 8086:2fd2
0000:7f:17.3 Class 0880: 8086:2fd3
0000:7f:17.4 Class 0880: 8086:2fb8
0000:7f:17.5 Class 0880: 8086:2fb9
0000:7f:17.6 Class 0880: 8086:2fba
0000:7f:17.7 Class 0880: 8086:2fbb
0000:7f:1e.0 Class 0880: 8086:2f98
0000:7f:1e.1 Class 0880: 8086:2f99
0000:7f:1e.2 Class 0880: 8086:2f9a
0000:7f:1e.3 Class 0880: 8086:2fc0
0000:7f:1e.4 Class 0880: 8086:2f9c
0000:7f:1f.0 Class 0880: 8086:2f88
0000:7f:1f.2 Class 0880: 8086:2f8a
0000:80:02.0 Class 0604: 8086:2f04 [PCIe RP[0000:80:02.0]]
0000:80:03.0 Class 0604: 8086:2f08 [PCIe RP[0000:80:03.0]]
0000:80:04.0 Class 0880: 8086:2f20
0000:80:04.1 Class 0880: 8086:2f21
0000:80:04.2 Class 0880: 8086:2f22
0000:80:04.3 Class 0880: 8086:2f23
0000:80:04.4 Class 0880: 8086:2f24
0000:80:04.5 Class 0880: 8086:2f25
0000:80:04.6 Class 0880: 8086:2f26
0000:80:04.7 Class 0880: 8086:2f27
0000:80:05.0 Class 0880: 8086:2f28
0000:80:05.1 Class 0880: 8086:2f29
0000:80:05.2 Class 0880: 8086:2f2a
0000:80:05.4 Class 0800: 8086:2f2c
0000:81:00.0 Class 0604: 10b5:8747
0000:82:08.0 Class 0604: 10b5:8747
0000:82:10.0 Class 0604: 10b5:8747
0000:83:00.0 Class 0300: 10de:13f2 [vmgfx2]
0000:84:00.0 Class 0300: 10de:13f2 [vmgfx3]
0000:85:00.0 Class 0604: 10b5:8747
0000:86:08.0 Class 0604: 10b5:8747
0000:86:10.0 Class 0604: 10b5:8747
0000:87:00.0 Class 0300: 10de:13f2 [vmgfx0]
0000:88:00.0 Class 0300: 10de:13f2 [vmgfx1]
0000:ff:08.0 Class 0880: 8086:2f80
0000:ff:08.2 Class 1101: 8086:2f32
0000:ff:08.3 Class 0880: 8086:2f83
0000:ff:09.0 Class 0880: 8086:2f90
0000:ff:09.2 Class 1101: 8086:2f33
0000:ff:09.3 Class 0880: 8086:2f93
0000:ff:0b.0 Class 0880: 8086:2f81
0000:ff:0b.1 Class 1101: 8086:2f36
0000:ff:0b.2 Class 1101: 8086:2f37
0000:ff:0c.0 Class 0880: 8086:2fe0
0000:ff:0c.1 Class 0880: 8086:2fe1
0000:ff:0c.2 Class 0880: 8086:2fe2
0000:ff:0c.3 Class 0880: 8086:2fe3
0000:ff:0c.4 Class 0880: 8086:2fe4
0000:ff:0c.5 Class 0880: 8086:2fe5
0000:ff:0c.6 Class 0880: 8086:2fe6
0000:ff:0c.7 Class 0880: 8086:2fe7
0000:ff:0d.0 Class 0880: 8086:2fe8
0000:ff:0d.1 Class 0880: 8086:2fe9
0000:ff:0d.2 Class 0880: 8086:2fea
0000:ff:0d.3 Class 0880: 8086:2feb
0000:ff:0d.4 Class 0880: 8086:2fec
0000:ff:0d.5 Class 0880: 8086:2fed
0000:ff:0f.0 Class 0880: 8086:2ff8
0000:ff:0f.1 Class 0880: 8086:2ff9
0000:ff:0f.2 Class 0880: 8086:2ffa
0000:ff:0f.3 Class 0880: 8086:2ffb
0000:ff:0f.4 Class 0880: 8086:2ffc
0000:ff:0f.5 Class 0880: 8086:2ffd
0000:ff:0f.6 Class 0880: 8086:2ffe
0000:ff:10.0 Class 0880: 8086:2f1d
0000:ff:10.1 Class 1101: 8086:2f34
0000:ff:10.5 Class 0880: 8086:2f1e
0000:ff:10.6 Class 1101: 8086:2f7d
0000:ff:10.7 Class 0880: 8086:2f1f
0000:ff:12.0 Class 0880: 8086:2fa0
0000:ff:12.1 Class 1101: 8086:2f30
0000:ff:12.4 Class 0880: 8086:2f60
0000:ff:12.5 Class 1101: 8086:2f38
0000:ff:13.0 Class 0880: 8086:2fa8
0000:ff:13.1 Class 0880: 8086:2f71
0000:ff:13.2 Class 0880: 8086:2faa
0000:ff:13.3 Class 0880: 8086:2fab
0000:ff:13.6 Class 0880: 8086:2fae
0000:ff:13.7 Class 0880: 8086:2faf
0000:ff:14.0 Class 0880: 8086:2fb0
0000:ff:14.1 Class 0880: 8086:2fb1
0000:ff:14.2 Class 0880: 8086:2fb2
0000:ff:14.3 Class 0880: 8086:2fb3
0000:ff:14.4 Class 0880: 8086:2fbc
0000:ff:14.5 Class 0880: 8086:2fbd
0000:ff:14.6 Class 0880: 8086:2fbe
0000:ff:14.7 Class 0880: 8086:2fbf
0000:ff:16.0 Class 0880: 8086:2f68
0000:ff:16.1 Class 0880: 8086:2f79
0000:ff:16.2 Class 0880: 8086:2f6a
0000:ff:16.3 Class 0880: 8086:2f6b
0000:ff:16.6 Class 0880: 8086:2f6e
0000:ff:16.7 Class 0880: 8086:2f6f
0000:ff:17.0 Class 0880: 8086:2fd0
0000:ff:17.1 Class 0880: 8086:2fd1
0000:ff:17.2 Class 0880: 8086:2fd2
0000:ff:17.3 Class 0880: 8086:2fd3
0000:ff:17.4 Class 0880: 8086:2fb8
0000:ff:17.5 Class 0880: 8086:2fb9
0000:ff:17.6 Class 0880: 8086:2fba
0000:ff:17.7 Class 0880: 8086:2fbb
0000:ff:1e.0 Class 0880: 8086:2f98
0000:ff:1e.1 Class 0880: 8086:2f99
0000:ff:1e.2 Class 0880: 8086:2f9a
0000:ff:1e.3 Class 0880: 8086:2fc0
0000:ff:1e.4 Class 0880: 8086:2f9c
0000:ff:1f.0 Class 0880: 8086:2f88
0000:ff:1f.2 Class 0880: 8086:2f8a

[root@localhost:~] nvidia-smi
Fri May 6 20:03:28 2016
+------------------------------------------------------+
| NVIDIA-SMI 361.40 Driver Version: 361.40 |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
|===============================+======================+======================|
| 0 Tesla M60 On | 0000:04:00.0 Off | Off |
| N/A 38C P8 24W / 150W | 19MiB / 8191MiB | 0% Default |
+-------------------------------+----------------------+----------------------+
| 1 Tesla M60 On | 0000:05:00.0 Off | Off |
| N/A 34C P8 23W / 150W | 19MiB / 8191MiB | 0% Default |
+-------------------------------+----------------------+----------------------+
| 2 Tesla M60 On | 0000:83:00.0 Off | Off |
| N/A 33C P8 24W / 150W | 19MiB / 8191MiB | 0% Default |
+-------------------------------+----------------------+----------------------+
| 3 Tesla M60 On | 0000:84:00.0 Off | Off |
| N/A 30C P8 23W / 150W | 19MiB / 8191MiB | 0% Default |
+-------------------------------+----------------------+----------------------+
| 4 Tesla M60 On | 0000:87:00.0 Off | Off |
| N/A 33C P8 25W / 150W | 19MiB / 8191MiB | 0% Default |
+-------------------------------+----------------------+----------------------+
| 5 Tesla M60 On | 0000:88:00.0 Off | Off |
| N/A 30C P8 23W / 150W | 19MiB / 8191MiB | 0% Default |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes: GPU Memory |
| GPU PID Type Process name Usage |
|=============================================================================|
| No running processes found |

lspci @ ESXi可能有不同的输出格式化尝试修改regexp。
我的输出（XenServer 6.5，不是完整输出，但包括K1，K2，K2200，启用“4g以上解码”，64位mmio BAR，不强制启动@ 2GB（MMIOHBase））：
06：00.0 0300：10de：0ff2（rev a1）（prog-if 00 [VGA控制器]）
区域0：内存为dd000000（32位，不可预取）[size = 16M]
区域1：内存为380ff0000000（64位，可预取）[size = 128M]
区域3：内存为380ff8000000（64位，可预取）[size = 32M]
07：00.0 0300：10de：0ff2（rev a1）（prog-if 00 [VGA控制器]）
区域0：db000000处的内存（32位，不可预取）[size = 16M]
区域1：内存为380fe0000000（64位，可预取）[size = 128M]
区域3：内存为380fe8000000（64位，可预取）[size = 32M]
08：00.0 0300：10de：0ff2（rev a1）（prog-if 00 [VGA控制器]）
区域0：内存为d9000000（32位，不可预取）[size = 16M]
区域1：内存为380fd0000000（64位，可预取）[size = 128M]
区域3：内存为380fd8000000（64位，可预取）[size = 32M]
09：00.0 0300：10de：0ff2（rev a1）（prog-if 00 [VGA控制器]）
区域0：内存为d7000000（32位，非预取）[size = 16M]
区域1：内存为380fc0000000（64位，可预取）[size = 128M]
区域3：内存为380fc8000000（64位，可预取）[size = 32M]
82：00.0 0300：10de：13ba（rev a2）（prog-if 00 [VGA控制器]）
区域0：内存为fa000000（32位，非预取）[size = 16M]
区域1：内存为381fc0000000（64位，可预取）[size = 256M]
区域3：内存为381fd0000000（64位，可预取）[size = 32M]
85：00.0 0302：10de：11bf（rev a1）
区域0：内存为f8000000（32位，不可预取）[size = 16M]
区域1：内存为381fe8000000（64位，可预取）[size = 128M]
区域3：内存为381ff0000000（64位，可预取）[size = 32M]
86：00.0 0302：10de：11bf（rev a1）
区域0：内存为f6000000（32位，不可预取）[size = 16M]
区域1：内存为381fd8000000（64位，可预取）[size = 128M]
区域3：内存为381fe0000000（64位，可预取）[size = 32M]
预计你的3张卡的mmio BAR映射在4GB（32位边界）下，最后一张超过4GB。



以下为原文

lspci @ ESXi has probably different output formatting try to modify regexp. My output (XenServer 6.5, not full output, but include K1,K2,K2200, with enabled "Above 4g Decoding", with 64bit mmio BAR, not forced to start @ 2GB (MMIOHBase)):

06:00.0 0300: 10de:0ff2 (rev a1) (prog-if 00 [VGA controller])
Region 0: Memory at dd000000 (32-bit, non-prefetchable) [size=16M]
Region 1: Memory at 380ff0000000 (64-bit, prefetchable) [size=128M]
Region 3: Memory at 380ff8000000 (64-bit, prefetchable) [size=32M]
07:00.0 0300: 10de:0ff2 (rev a1) (prog-if 00 [VGA controller])
Region 0: Memory at db000000 (32-bit, non-prefetchable) [size=16M]
Region 1: Memory at 380fe0000000 (64-bit, prefetchable) [size=128M]
Region 3: Memory at 380fe8000000 (64-bit, prefetchable) [size=32M]
08:00.0 0300: 10de:0ff2 (rev a1) (prog-if 00 [VGA controller])
Region 0: Memory at d9000000 (32-bit, non-prefetchable) [size=16M]
Region 1: Memory at 380fd0000000 (64-bit, prefetchable) [size=128M]
Region 3: Memory at 380fd8000000 (64-bit, prefetchable) [size=32M]
09:00.0 0300: 10de:0ff2 (rev a1) (prog-if 00 [VGA controller])
Region 0: Memory at d7000000 (32-bit, non-prefetchable) [size=16M]
Region 1: Memory at 380fc0000000 (64-bit, prefetchable) [size=128M]
Region 3: Memory at 380fc8000000 (64-bit, prefetchable) [size=32M]
82:00.0 0300: 10de:13ba (rev a2) (prog-if 00 [VGA controller])
Region 0: Memory at fa000000 (32-bit, non-prefetchable) [size=16M]
Region 1: Memory at 381fc0000000 (64-bit, prefetchable) [size=256M]
Region 3: Memory at 381fd0000000 (64-bit, prefetchable) [size=32M]
85:00.0 0302: 10de:11bf (rev a1)
Region 0: Memory at f8000000 (32-bit, non-prefetchable) [size=16M]
Region 1: Memory at 381fe8000000 (64-bit, prefetchable) [size=128M]
Region 3: Memory at 381ff0000000 (64-bit, prefetchable) [size=32M]
86:00.0 0302: 10de:11bf (rev a1)
Region 0: Memory at f6000000 (32-bit, non-prefetchable) [size=16M]
Region 1: Memory at 381fd8000000 (64-bit, prefetchable) [size=128M]
Region 3: Memory at 381fe0000000 (64-bit, prefetchable) [size=32M]

It is expected that your 3 cards has mmio BAR mapped under 4GB (32bit boundary) and the last one over 4GB.

感谢您抽出宝贵时间向每个人介绍supermicro的建议。
我们的支持组织将针对虚拟机管理程序改进有关MMIO的BIOS需求的文档，以便您的体验和时间用于改善其他人的体验。
当支持写这个时我会更新线程。
谢谢，
雷切尔



以下为原文

Thank you for taking the time to update everyone on supermicro's recommendations. Our support org will look to improve the documentation around BIOS needs on MMIO for hypervisors so that your experience and time is used to improve the experience for others.

I'll update the thread when support write this up.

Thank you,
Rachel