Author Topic: [MOVED] Hacking NVidia Cards into their Professional Counterparts  (Read 1645181 times)

0 Members and 5 Guests are viewing this topic.

Offline vsgan

  • Newbie
  • Posts: 4
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #350 on: May 22, 2013, 05:31:43 pm »
Small update:

I know most of you are trying to get professional cards out of consumer, I just want a card for passthrough so I could game "on linux". In every test I have run, this GTX680 modded to a Grid K2 runs exactly the same as the GTX680 except with Physx. Physx doesn't work with the "professional" drivers so I forced the consumer drivers to install. They work and run fine and now Physx will say it is enabled, but the card will not do any of the work. It offloads everything to the CPU while still reporting it is working to the program. Its a bit strange and annoying. Going to mod it back to confirm my results and trying a Telsa mod just in case that will support it (probably not).

Nvidia, if you are reading this, I just want virtualization FOR PLAYING GAMES. Seriously, AMD actually worked with the community on this one, why can't you just enable it? It clearly works fine.

How did you force the driver to install?
 

Offline gordan

  • Frequent Contributor
  • **
  • Posts: 277
  • Country: 00
    • Tech Articles
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #351 on: May 22, 2013, 06:56:12 pm »
Small update:

I know most of you are trying to get professional cards out of consumer, I just want a card for passthrough so I could game "on linux". In every test I have run, this GTX680 modded to a Grid K2 runs exactly the same as the GTX680 except with Physx. Physx doesn't work with the "professional" drivers so I forced the consumer drivers to install. They work and run fine and now Physx will say it is enabled, but the card will not do any of the work. It offloads everything to the CPU while still reporting it is working to the program. Its a bit strange and annoying. Going to mod it back to confirm my results and trying a Telsa mod just in case that will support it (probably not).

Nvidia, if you are reading this, I just want virtualization FOR PLAYING GAMES. Seriously, AMD actually worked with the community on this one, why can't you just enable it? It clearly works fine.

+1

I actually had more success with PhysX. My Win7 64-bit VM with a Quadro 2000 worked fine with PhysX once I installed the PhysX package in addition to the Quadro drivers. I haven't checked whether it was offloading onto the CPU or not, though, but games and GPU-Z all reported PhysX capability.

On XP64, however, that didn't work - no PhysX on the Quadro 2000, with or without the PhysX software installed.

In the end, I gave up on Nvidia and decided to save myself money, modding effort and time and just got an ATI card instead, because I, too, only wanted a decent GPU to game with in a VM without having to dual boot machine. At least until Steam has a better selection of Linux capable games and get their client software working without requiring bleeding edge glibc.
 

Offline gamezr2ez

  • Contributor
  • Posts: 30
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #352 on: May 22, 2013, 07:42:33 pm »
Does the card actually work as NVIDIA GRID VGX after modification? e.g. vmware vsphere vgsa?

I tried a GTX 680 > Grid K2 and it did not work. I was sad.

How did you force the driver to install?

The Quadro K5000 is supported by the "consumer" graphics but it will lockup on VM boot with those drivers.

The Grid K2 is not supported, but since my card is for all intents and purposes a GTX 680 the drivers should work fine. I just did some inf edits to add the new strings for the card. The driver will not be a signed driver for that PCI ID though. Doesnt affect performance, but it will throw warnings.
« Last Edit: May 22, 2013, 07:44:25 pm by gamezr2ez »
 

Offline gamezr2ez

  • Contributor
  • Posts: 30
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #353 on: May 22, 2013, 07:50:27 pm »
I actually had more success with PhysX. My Win7 64-bit VM with a Quadro 2000 worked fine with PhysX once I installed the PhysX package in addition to the Quadro drivers. I haven't checked whether it was offloading onto the CPU or not, though, but games and GPU-Z all reported PhysX capability.

I may retract my CPU offloading comment later. After modding to a k5000 again the physx score is the same. Both grid and quadro report supporting phsyx, but they both only give the same performance as my aging cpu.

Setting up a new windows install to test the card types to verify everything.

In the end, I gave up on Nvidia and decided to save myself money, modding effort and time and just got an ATI card instead, because I, too, only wanted a decent GPU to game with in a VM without having to dual boot machine. At least until Steam has a better selection of Linux capable games and get their client software working without requiring bleeding edge glibc.

I use a chroot for steam on linux so I dont break the rest of my system or have it all bleeding edge. Space isnt an issue now a days and there is no less performance. Annoying, but workable.
« Last Edit: May 22, 2013, 11:01:18 pm by gamezr2ez »
 

Offline vsgan

  • Newbie
  • Posts: 4
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #354 on: May 23, 2013, 04:05:50 am »
I just did some inf edits to add the new strings for the card. The driver will not be a signed driver for that PCI ID though. Doesnt affect performance, but it will throw warnings.

Could you share the method for modifying the driver? I tried to add some strings for my hardware id but the driver still cannot install correctly. Thanks.
 

Offline verybigbadboy

  • Contributor
  • Posts: 38
  • Country: ru
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #355 on: May 23, 2013, 07:40:48 am »
Does the card actually work as NVIDIA GRID VGX after modification? e.g. vmware vsphere vgsa?
I tested vga passthrough with xen only. It works fine.
6'7''
 

Offline gamezr2ez

  • Contributor
  • Posts: 30
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #356 on: May 23, 2013, 02:55:04 pm »
Does the card actually work as NVIDIA GRID VGX after modification? e.g. vmware vsphere vgsa?
I tested vga passthrough with xen only. It works fine.

That is different than the vGPU offered by VGX. vGPU in VGX did not work. I may play around with it more later to figure it out if it is possible.

Update on Physx offloading to cpu:
Turns out this is a well known issue. nVidia disables a bunch of stuff when it is not the primary graphics card in the OS. This is the case with typical vga passthrough and xen. I am trying to get primary passthrough to work now.
 

Offline gordan

  • Frequent Contributor
  • **
  • Posts: 277
  • Country: 00
    • Tech Articles
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #357 on: May 23, 2013, 09:07:33 pm »

In the end, I gave up on Nvidia and decided to save myself money, modding effort and time and just got an ATI card instead, because I, too, only wanted a decent GPU to game with in a VM without having to dual boot machine. At least until Steam has a better selection of Linux capable games and get their client software working without requiring bleeding edge glibc.

I use a chroot for steam on linux so I dont break the rest of my system or have it all bleeding edge. Space isnt an issue now a days and there is no less performance. Annoying, but workable.

You could just install a newer glibc as required into a separate path and set LD_LIBRARY_PATH before invoking steam. LD_LIBRARY_PATH gets evaluated before ld.so.conf. No need for a full chroot. Just make sure your newer glibc isn't in one of the directories referenced in ld.so.conf or ld.so.conf.d.
 

Offline lightsol

  • Newbie
  • Posts: 6
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #358 on: May 28, 2013, 10:23:00 am »
anyone have luck with the actual VGX? It would be pretty damn nice if the gpu could be spread out to 10+ VM's at the same time
 

Offline gamezr2ez

  • Contributor
  • Posts: 30
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #359 on: May 28, 2013, 03:47:05 pm »
anyone have luck with the actual VGX? It would be pretty damn nice if the gpu could be spread out to 10+ VM's at the same time

I spent this past weekend working on that. Nothing.

I still have a few avenues to go down before I give up.
 

Offline winjet1

  • Newbie
  • Posts: 9
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #360 on: May 28, 2013, 04:03:36 pm »
anyone have luck with the actual VGX? It would be pretty damn nice if the gpu could be spread out to 10+ VM's at the same time

Here's the issue........

If you are just passing a 680 -> VGX card to a VM, there's really nothing special there.  You are just doing a hardware passthrough and since 5.1, pass-through has been pretty stable.

If you are using a 680 -> VGX to do sVGA, then you are truely trying to use a VGX as intended.  To use it here you have to enable a few things in VMware (see below).

http://communities.vmware.com/thread/415887?start=30&tstart=0

Sadly, VMware's 3D drivers are really not up to par to run things like games.  In addition you can only farm out 256Mb of RAM and 256Mb of vRAM per VM (can you remember the last game you played on a 256Mb card?  Bet it wasn't Crysis 2).

Microsoft's Hyper-V and Citrix may be a better solution.
 

Offline vsgan

  • Newbie
  • Posts: 4
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #361 on: June 02, 2013, 09:19:00 pm »
Does any vendor other than vmware have a stable shared graphic acceleration solution at this point? especially for *nix
 

Offline lightsol

  • Newbie
  • Posts: 6
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #362 on: June 04, 2013, 04:52:20 pm »
Well from my experience RemoteFX is great, but it needs a lot of configuration
 

Offline gamezr2ez

  • Contributor
  • Posts: 30
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #363 on: June 05, 2013, 01:09:36 am »
RemoteFX is great, seconded.

I have not tried this yet, I will setup a Server 2012 tonight to test that. I will try it 2 ways, natively and then through xen. Thats right, Server 2012 with GPU passthrough, running vms using the GPU. How cool would that be? I am not so familiar with Hyper-V to know if that will work inside of a Xen domU, I suspect not. Maybe it won't whine to much about the extensions it doesnt have access to.

According to an nVidia blog post about remoteFX, Quadro cards support remoteFX. This post was a few years prior to GRID cards, but I believe they should work too.

Will get back to the community on this.

PS: Through some testing and config verification, the offloading physx and poor GPU performance in passthrough was a problem with my mobo not truly supporting xen. Upgraded and everything is great. It is truly wonderful. (Plus I have an onboard iGPU to passthrough now, yay!)
« Last Edit: June 05, 2013, 01:11:25 am by gamezr2ez »
 

Offline lightsol

  • Newbie
  • Posts: 6
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #364 on: June 05, 2013, 08:06:17 am »
Try with the just released Server 2012 R2 preview , i heard that they upgraded the hyper-v engine even more, maybe there are some good RemoteFX upgrades as well :)
 

Offline function

  • Contributor
  • Posts: 11
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #365 on: June 06, 2013, 12:59:39 am »
I know that they, too, lack FLreset, but I am not all that convinced that FLreset is all that necessary. Sure, it makes it a little easier for the driver to do it's job, but think about this at a low level like an embedded engineer for a moment. On the lowest level it comes down to setting registers on the device. Unless the card is poorly engineered and buggy (e.g. it drops off the bus in a questionable, un-re-attachable and uncontactable state), the driver should always be able t o set the registers to whatever they need to be to get the card to a known, initialized state, without even any help from the card's BIOS. FLreset is a nicety that means your driver doesn't have to handle the initialization of the hardware itself, but it doesn't strike me at all as a necessity to get something like this working properly.

And I believe that is exactly the problem, GPU vendors do not want to reveal the initialization and setup of their hardware lest they end up revealing their IP. So the video bios does a lot of heavy lifting for bootstrapping the GPU. intel opregion is at least nice in that it allows for complete OS (in contrast to part firmware) based initialization.
 

Offline gordan

  • Frequent Contributor
  • **
  • Posts: 277
  • Country: 00
    • Tech Articles
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #366 on: June 06, 2013, 06:38:57 am »
I know that they, too, lack FLreset, but I am not all that convinced that FLreset is all that necessary. Sure, it makes it a little easier for the driver to do it's job, but think about this at a low level like an embedded engineer for a moment. On the lowest level it comes down to setting registers on the device. Unless the card is poorly engineered and buggy (e.g. it drops off the bus in a questionable, un-re-attachable and uncontactable state), the driver should always be able t o set the registers to whatever they need to be to get the card to a known, initialized state, without even any help from the card's BIOS. FLreset is a nicety that means your driver doesn't have to handle the initialization of the hardware itself, but it doesn't strike me at all as a necessity to get something like this working properly.

And I believe that is exactly the problem, GPU vendors do not want to reveal the initialization and setup of their hardware lest they end up revealing their IP. So the video bios does a lot of heavy lifting for bootstrapping the GPU. intel opregion is at least nice in that it allows for complete OS (in contrast to part firmware) based initialization.

I think you got that backwards. If they implemented FLR, they would need to reveal _less_ because the reset would be a single, standards defined call to reset the card without having to reveal _anything_ about the hardware. What using proprietary initialization does do, however, is make it more difficult for open source drivers to be written. This enables companies like Nvidia to charge you 5x the amount for the same hardware just for changing 2 resistors and half a byte of firmware to get access to the "pro" feature set of the driver.

It's not about protecting the IP - it's about protecting the high-margin revenue streams.
 

Offline function

  • Contributor
  • Posts: 11
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #367 on: June 10, 2013, 09:09:58 am »
I know that they, too, lack FLreset, but I am not all that convinced that FLreset is all that necessary. Sure, it makes it a little easier for the driver to do it's job, but think about this at a low level like an embedded engineer for a moment. On the lowest level it comes down to setting registers on the device. Unless the card is poorly engineered and buggy (e.g. it drops off the bus in a questionable, un-re-attachable and uncontactable state), the driver should always be able t o set the registers to whatever they need to be to get the card to a known, initialized state, without even any help from the card's BIOS. FLreset is a nicety that means your driver doesn't have to handle the initialization of the hardware itself, but it doesn't strike me at all as a necessity to get something like this working properly.

And I believe that is exactly the problem, GPU vendors do not want to reveal the initialization and setup of their hardware lest they end up revealing their IP. So the video bios does a lot of heavy lifting for bootstrapping the GPU. intel opregion is at least nice in that it allows for complete OS (in contrast to part firmware) based initialization.

I think you got that backwards. If they implemented FLR, they would need to reveal _less_ because the reset would be a single, standards defined call to reset the card without having to reveal _anything_ about the hardware. What using proprietary initialization does do, however, is make it more difficult for open source drivers to be written. This enables companies like Nvidia to charge you 5x the amount for the same hardware just for changing 2 resistors and half a byte of firmware to get access to the "pro" feature set of the driver.

It's not about protecting the IP - it's about protecting the high-margin revenue streams.

FLR however is only for resetting the device at the PCI bus level, not actually bootstrapping and loading of firmware to the ICs, DSPs etc on the board. E.g. PCI devices also support PCI reset via D3-D0 transition but that doesn't reinitialize the hardware after it has been reset either. I would also think that moving this complicated device initialization to the HDL would make board specific customizations unnecessarily complex.
 

Offline gpvecchi

  • Newbie
  • Posts: 3
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #368 on: June 10, 2013, 08:38:41 pm »
Hallo! I'd need to convert my MSI GTX 680 Lightning into a GTX 770 Lightning as I need a card to SLI (680 are unavailable!)...
Flashing the bios doesn't change device ID, so I suppose I need this hack... Any help, please?
 

Offline verybigbadboy

  • Contributor
  • Posts: 38
  • Country: ru
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #369 on: June 10, 2013, 09:10:18 pm »
Hallo! I'd need to convert my MSI GTX 680 Lightning into a GTX 770 Lightning as I need a card to SLI (680 are unavailable!)...
Flashing the bios doesn't change device ID, so I suppose I need this hack... Any help, please?

Hello, You may try to change device ID using gtx 680 guide
I updated resistors values for gtx 770.
Thank you ;)
6'7''
 

Offline verybigbadboy

  • Contributor
  • Posts: 38
  • Country: ru
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #370 on: June 10, 2013, 09:20:24 pm »
May be it is offtopic but I have question:
Did anyone try to flash bios from 670 to 680? will it lock cores?
If we unable unlock cores, may be we need to find how to lock it? ;)

It is really interesting to know function of near located resisors. ;)
6'7''
 

Offline gpvecchi

  • Newbie
  • Posts: 3
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #371 on: June 11, 2013, 07:57:26 pm »
Hallo! I'd need to convert my MSI GTX 680 Lightning into a GTX 770 Lightning as I need a card to SLI (680 are unavailable!)...
Flashing the bios doesn't change device ID, so I suppose I need this hack... Any help, please?

Hello, You may try to change device ID using gtx 680 guide
I updated resistors values for gtx 770.
Thank you ;)
Great! Thank you very much! I just hope that Lightning PCB doesn't differ in that point...
EDIT: PCB is very different... Could you please help me find resistor n.3 in this hires image?
http://img823.imageshack.us/img823/9379/backfullb.jpg
Thanks!
« Last Edit: June 11, 2013, 08:12:57 pm by gpvecchi »
 

Offline gamezr2ez

  • Contributor
  • Posts: 30
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #372 on: June 12, 2013, 05:45:25 am »
@gpvecchi

An image of the other side may be useful. I am assuming this is the "MSI GTX 680 Lightning" you spoke about previously.

Check to see if that is a 5k resistor. It should be if this is a GTX680. All the pieces seem to be there, just shifted around a bit.



@Anyone following GPU passthrough or GPU sharing

I moved this past weekend. It was a bit overly ambition of me to try and test a new server setup before. I will attempt Hyper-V on Server 2012 this weekend.
 

Offline gpvecchi

  • Newbie
  • Posts: 3
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #373 on: June 12, 2013, 09:23:57 am »
Yes, that's the PCB of the MSI GTX 680 Lightning (more power phases and military class components on custom PCB)...
This is the front side:
http://img405.imageshack.us/img405/8878/frontfulle.jpg
Thank you!
P.S.: Do someone know the difference between GK104-400 of the 680 and the GK104-425 of the 770?
« Last Edit: June 12, 2013, 10:22:29 am by gpvecchi »
 

Offline ret

  • Newbie
  • Posts: 1
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #374 on: June 12, 2013, 11:07:25 am »
Hi all!

First of all thanks for all your hard work, i really appreciate the "jailbreaking" way :)

Which card would you recommend for modding to a K1 card? I'd like to setup a lab to show off all the new 3D features which were introduced by Citrix - XenServer and vSphere supports GPU virtualization and this workaround would be awesome. Otherwise i'd have to spend ~ 2.000$ for a K1 card.

Thanks!
 


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf