[MOVED] Hacking NVidia Cards into their Professional Counterparts

[slis]

Just quick update for u guys... i have gtx680, i am using it for vga passthrough with kvm, and with some patches, you can SOFTMOD it to pro counter part and it works with latest geforce/quadro (both) drivers in virtual machine.

here is patch and how to use it:
https://bbs.archlinux.org/viewtopic.php?pid=1475170#p1475170

[gordan]

I seem to recall that there are now patches for QEMU that neuter the drivers' ability to detect it is running in a VM anyway. Between the no-snoop patch and passing through the CPU ID from the hardware (as opposed to reporting the CPU as something like QEMU) the driver can't detect it's running in a VM so it boots up the unmodified GeForce card just fine.

[mosmo]

I seem to recall that there are now patches for QEMU that neuter the drivers' ability to detect it is running in a VM anyway. Between the no-snoop patch and passing through the CPU ID from the hardware (as opposed to reporting the CPU as something like QEMU) the driver can't detect it's running in a VM so it boots up the unmodified GeForce card just fine.

Interesting. So, in terms of a 980 running in vmware and/or KVM we can still softmod to get them working, or that hasn't been tested yet?
I have just forked out on 2 x R9 280x on a C612 platform so that I can 4K game under ESXi 5.5u2.

[gordan]

No, you no longer need to do any modding, there are patches available that prevent the driver from detecting it is running in a VM which makes everything work without any modifications to the card, software or hardware.

[mosmo]

No, you no longer need to do any modding, there are patches available that prevent the driver from detecting it is running in a VM which makes everything work without any modifications to the card, software or hardware.

Excellent, thanks for the heads up!

[slis]

Geforce cards can work unmodified in kvm, but after 340 drivers you can't use hyperv enhancements, with softmod and mentioned patch to grid k2/quadro u can use hyperv (better performance) with latest drivers aka 344. Anyway i thought you guys want to know that u maybe can use your geforce as quadro in vm-s without soldering.

[gordan]

It's always good to have extra options.
Nvidia's crippleware practices annoy me sufficiently that I am only using their products until ATI get their drivers sorted out to the point where their products are usable for me. Then again, I've been waiting for this to happen for years, but hope springs eternal...

[pasha4ur1]

Hello, people!

I need your help. I want to convert MSI GTX 460 v2 Hawk into Quadro 4000m



my BIOS http://pasha4ur.org.ua/GF104.rom

Please, help me with commands for flashing. I broke my head while read about bytes here
http://www.altechnative.net/2013/11/25/virtualized-gaming-nvidia-cards-part-3-how-to-modify-a-fermi-based-geforce-into-a-quadro-geforce-gts450gtx470gtx480-to-quadro-200050006000/

And I found

Hi hishamkali,

you made a mistake…
you use the string from gtx480…
this work for me on a GTX460 (0E22)
nvflash –index=X –straps 0x7FFC3FC3 0x10006428 0x7FF1FFFF 0x00020000

bye

Will it work better in Adobe Apps? Will I can adjust clocks and fans? Games will work too?

Thx for help

[pasha4ur]

up

[mosmo]

Can the titan z be modded like the 780ti?
price of the titan z is down to 999 today.
I modded my 780ti by soldering the resistor on the eprom and curious if the same can  easily be done with titan z.

[mosmo]

Can the titan z be modded like the 780ti?
price of the titan z is down to 999 today.
I modded my 780ti by soldering the resistor on the eprom and curious if the same can  easily be done with titan z.

If not Titan Z, what about the Titan Black?

Was thinking of getting this one and mod to Quadro....
http://www.xbitlabs.com/articles/graphics/display/gigabyte-geforce-gtx-titan-black-ghz-edition_4.html#sect0

[high]

Hi,

I've read every post on this thread. Amazing!

I finally applied the mod to a Palit GTX680 to produce a K2 and passing this through XenServer (credence Beta) to Win 8.1 enterprise. Unfortunately, once I install the drivers, the card comes up with a yellow icon in Device Manager, saying that the card is not working. If I disable, then re-enable the card, Device Manager seems to like it.

Unfortunately, I don't believe the card is working correctly. For example, the GPUZ attachment detects the GRID K2, but doesn't detect any of the other parameters correctly. No BIOS info, none of the Senors are working, and the Memory is the wrong size. The card I have is only a 2GB card. None of the Clock information is shown. And this card supports OpenCL, CUDA, and DirectCompute, but none of those items are selected.

Please see the attached file.



In the XenServer host, nvidia-smi give the following output:

Thu Nov 27 23:25:14 2014
+------------------------------------------------------+
| NVIDIA-SMI 340.34     Driver Version: 340.34         |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  GRID IceCube        Off  | 0000:02:00.0     N/A |                  N/A |
| 37%   47C    P0    N/A /  N/A |      9MiB /  2047MiB |     N/A      Default |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Compute processes:                                               GPU Memory |
|  GPU       PID  Process name                                     Usage      |
|=============================================================================|
|    0            Not Supported                                               |
+-----------------------------------------------------------------------------+

A Grid K2 card is supposed to show CUDA process names in nvidia-smi. "nvidia-smi stats" also says that my card is unsupported.

I also have an unsupported GeForce 6200 in the same XenServer host as the primary display. Not sure if that is causing some Nvidia conflict?

Any thoughts on what could be done to get this GTX680 (now K2) card to work correctly under XenServer with Win 8.1? I would love to get the vGPU stuff to work, and XenCenter gives me lots of vGPU options, but when  I start the VM with the vGPU, XenServer complains:

Internal error: xenopsd internal error: Device.Ioemmu_failed("vgpu exited unexpectedly")

Considering that the passthrough card is not detected correctly inside the VM, I suspect there is something fundamentally incorrect with either the mod or my XenServer setup? Any thought?

Thanks!

[gordan]

I've yet to see anyone get this working with a real Grid/Tesla/Quadro (they should all work, they are listed as supported for VSGA on ESXi), so it's rather difficult to begin guessing whether the problem is related to the card not being a real Grid. If you had a 4GB card you might be able to flash a real K5000 firmware onto it, which may give you better results, if the problem is the card being modified in the first place. I had a 4GB 680 modified into a K5000 and flashed with the K5000 firmware, but I sold it many months ago when I upgraded to 780Ti.

[high]

For those of you wanting to use VSGA type solutions, there are known bugs in the Kepler firmwares, and only the latest firmware for the K2 works properly. All the GeForce cards have firmwares where the bugs aren't fixed, so it probably won't work. You could cross-flash a K2 firmware (after you have edited the memory size initialization block) onto a GeForce, but you will lose all video outputs on the card, and I don't think anyone has tried this.

I'd be willing to flasha K2 bios onto a modified GTX680 and try this. Firmware is available from IBM and HP for GRID K2. Unfortunately, I only have a 2GB card (Palit). Whats the best place to start looking for how to alter the memory size initialization block in the BIOS? ...I think there was something about memory flags in this post somewhere....stay tuned.

[imanz]

Is anyone willing to take some pictures of their modified cards that are working as expected? Unfortunately because this post is 75+ pages its getting a bit hard to follow and find directions/information.

Specifically it would be nice to have a set of instructions including the recommended resistor type and location for an assortment of cards of each model. If anyone is interested I am willing to paypal someone 30 bucks for taking the time to write up something on the nvidia 680 with pictures of a finished mod, recommendation on which resistors to get and a list of models that have resistor locations determined (to make it a bit easier to find on ebay). If you want to share this with the community just PM the paypal.

[verybigbadboy]

Hello

I am interested in modding 750 maxwell card. I have one card on my job, and I did some research on it. it is Palit 750 with 1gb memory.

I tried to mod it similar to 650 but have some issues:
resistors around bios a bit changed -> my card have 0-ohm resistor connected to 6-pin of bios rom.

I tried to change resistors connected to this pin, but it is not affects on pci id.
initially it is 20k, I tried to solder 40k resistors to both places.

I checked bios, but bios is also changed. It is looks like unlocked, but I think it is not.


so anyone tried to mod maxwell cards ?

It will be really nice to mod 750 cards

[high]

Tried to flash my 2GB GTX680 with K2 firmware from:

ftp://ftp.hp.com/pub/softlib2/software1/pubsw-linux/p1141455199/v96072

There are three pieces of firmware in this package, 2 for GPUs and a tiny one for what I would assume is some other functionality on the K2, like the PCI bridge, or maybe even some bits of a hypervisor?

Anyway, I tried flashing both instances of the K2 BIOS to my GTX680. The card too the firmware, but wasn't functional. I didn't modify the firmware in any way, so I assume there is a mismatch with the memory allocation at least.

Any thoughts on getting a K2 firmware to work properly with the hard modified GTX680?

Thanks!

[gordan]

If the K2 is anything like the GTX690, then there will be 3 firmwares involved:
1) GPU 1
2) GPU 2 (no, they are not quite the same, the device hierarchy ID is different which you will need to edit)
3) PCIe bridge

Since the K5000 is supported for VSGA I would expect it to work just the same with other hypervisors. You will almost certainly find that easier to modify (I was running my 4GB GTX680 with the K5000 BIOS before I sold it). For a 2GB card you should be able to get away with just changing the memory layout initialization block, and you will retain the functioning of (at least some) video outputs), unlike with the K2 firmware where you will lose all video outputs since the K2 has no video outputs.

And if you can't get the K5000 to work with GPU sharing, there is a good chance K2 won't work either.

[high]

just changing the memory layout initialization block

Any pointers on how to go about that? "Changing the memory layout initialization block".

Thanks!

[gordan]

I posted on this thread a quite a while back with some details of how to do it, and somebody else successfully used the approach on a different card. Couldn't tell you what page off the top of my head, though.

[high]

I posted on this thread a quite a while back with some details of how to do it, and somebody else successfully used the approach on a different card. Couldn't tell you what page off the top of my head, though.

I'm guessing that this was the post:

https://www.eevblog.com/forum/chat/hacking-nvidia-cards-into-their-professional-counterparts/msg292021/#msg292021

The hurdles in this approach is that K2 vs GTX680 bios comparisons have many, many differences. And the K2 doesn't come in two separate memory configurations....

https://www.eevblog.com/forum/chat/hacking-nvidia-cards-into-their-professional-counterparts/msg326647/#msg326647
Did oguz286 ever reply to you on the bios diffs?

[gordan]

Yes, that is the correct post, and I just spotted a typo in it - what I meant to say is that I'd be surprised if the Q6000 BIOS didn't work on the GTX480.

What are you referring to by "G2"?

You should be able to disassemble the BIOS using the envy tools and use similarities in surrounding code to identify the block where the memory layout is being initialized, and replace that block in the K2 BIOS with the equivalent from your 680 BIOS.

The strap info is still applicable, although at least the device ID straps are now also done in hardware, and the hard strap can be checked in a register in Kepler GPUs, although the soft strap still does affect the PCI device ID.

[high]

Apologies. "G2" is a typo. Meant "K2". I also found this post where you clarified the previous post:

https://www.eevblog.com/forum/chat/hacking-nvidia-cards-into-their-professional-counterparts/705/

[high]

I disassembled the K2 and GTX BIOSes using the "nvbios" tool from envytools. First, I had to strip off the UFFI headers from both BIOS before nvbios would parse the ROM with any sensible output.

I was able to find MEM TYPE and MEM TRAIN blocks in both the K2 and GTX bios. Interestingly, the GTX 2GB and GTX 4GB bios had the same exact binary code at the specified offsets

GTX
MEM TYPE table at 0x4d6f, version 10, 16 entries
Detected ram type: DDR2
0x4d6f: 10 07 07 10 00 e6 4d
....


The K2 Bios has only the header slightly different in the MEM TYPE block:

K2MEM TYPE table at 0x4e88, version 10, 16 entries
Detected ram type: DDR2
0x4e88: 10 07 07 10 00 ff 4e

The difference in the header could be a clue, but the GTX 2GB and 4GB ROMs had the same layout, so this block probably isn't it....

Desperately, I copied the pieces out of the GTX BIOS into the K2 Bios and flashed the modified GTX680. As predicted, it did jack shit and the though the system started, trying to access the card cause the host to crash.

>>surrounding code to identify the block where the memory layout is being initialized,
With the reference to "code" are you talking about the init scripts that nvbios decompiles? Or is it some other memory configuration table/flag that I should look at?

Thanks!

[high]

Based on the conversation I read here:

http://people.freedesktop.org/~cbrill/dri-log/?channel=nouveau&highlight_names=&date=2013-11-13

Yes, I am stalking you, Gordan! 

I modified the corresponding parts of the K2 bios with the bits you found to take the GTX680 from 4GB to 2GB. Flashed the modified BIOS to the card and lost control of the GTX's fan (because the K2 doesn't have any!!). But the modded card didn't play nice with vGPU under XenServer and I had the same errors are before. Additionally, passthrough to windows with the modified firmware causes GPU-Z to not detect anything.

So, I flashed the card back to my original BIOS.

In summary:
Modified GTX680 with Xen vGPU. Stock BIOS. -> Doesn't work.
Modified GTX680 with Xen Passthrough. Stock BIOS. -> works for Windows 8.1.
Modified GTX680 with Xen vGPU. K2 BIOS. The g1 and g2 BIOS -> Doesn't work.
Modified GTX680 with Xen Passthrough. K2 BIOS. The g1 and g2 BIOS -> Doesn't work.
Modified GTX680 with Xen vGPU. Mod K2 BIOS with the memory changed from 4GB to 2GB. Only the g1 BIOS-> Doesn't work
Modified GTX680 with Xen Passthrough. Mod K2 BIOS with the memory changed from 4GB to 2GB. Only the g1 BIOS-> Doesn't work

It looks like this ugly duckling GTX will never turn into a swan K2.

Any other thoughts or ideas to try?

What's the cheapest card available to mod that can be passed through properly?

PS. How many times can I flash the EEPROM before it dies?