[MOVED] Hacking NVidia Cards into their Professional Counterparts

[eos]

I removed resistors R2 and R3 on the GTX 660 ti and now have a k10(deviceID 118F)!!

Since the 660Ti is still in the guessing game, you should remove resistors one by one.

Most likely by removing R2 you changed the last symbol from 3 to F, aka 1183 -> 118F. That is the Tesla K10.
R3 probably has nothing to do with the Device ID. Can you solder the R3 back on and check again?
R4 could be the resistor of interest... Replaced by a 20K resistor, aka "B"...

[verybigbadboy]

I stopped my investigation about modification Palit gts 450 Green (OEM) to quadro 2000. I think it is impossible to modify only OEM version with 144 cores. But I think my knowledge can help to someone who want's to modify another card.

gts 450 device id is 0x0DC4.
quadro 2000 id is 0x0DD8

Lets me explain how to set device id.
4th byte value is calculated by resistors 1, 2, 3.
Initially resistance and their values are:

index	meaning	resistance	value
1	shift	10k	1
2	value 0-7	20k	3
3	value 8-f	none	none

Result value equals sum of 1 and 2 or 3. If value more than range of resistor 2 or 3 then it equals to high value.
Example:
if 1 = 5k, 2 = none, 3 = 5k  then result 8.
if 1 = 5k, 2 = 25k, 3 = none  then result 4.

setting 3th byte:
After checking different cards I found that 3 byte always set by resistors connected to pin 6 of eeprom.

index	meaning	resistance
4	value C	35k
5	value D	none

Setting 3th byte was difficult for me because card have soft straps. (it is new for me)
I tried every resistor value for 5th resistor, and this is my result:
4 = none, 5 = 5...20k device id is 0x0DD8, and card don't want to start due to unable to read eeprom. (but clear works fine )
4 = none, 5 = 25k ... 40k device id is 0x0DC8, card works but have wrong id
After I cleared eeprom by nvflash, device id has become 0x0DD8.
Then I get soft straps from quadro bios and upload it to eeprom with original bios.
Now card is starting with 0x0DD8 device id and works fine until I installed drivers . When windows tries to enable 3d I always have blue screen . Tested with every 5th resistor value, also tried to remove other resistors.

Also tried to upload quadro bios. It not works at all. Random lines appears .

[eos]

When comparing Tesla K10 and GRID K2 (from the virtualization application perspective) there is very little info to go by.
In terms of specs, both are the pro version of the (underclocked) GTX 690 or two GTX 680 on one PCB.

One vendor announced to carry these cards was Dell. Here is the the PE R720 that has has such option
http://configure.us.dell.com/dellstore/config.aspx?oc=bectj4&model_id=poweredge-r720&c=us&l=en&s=bsd&cs=04
The GRID K2 is 15% more expensive than Tesla K10 for some reason...

[gnif]

The GRID K2 is 15% more expensive than Tesla K10 for some reason...

Niche market that NVidia want to squeeze some extra cash out of.

[gnif]

I have all the stuff for BGA reworking, if you manage to find a replacement chip I can swap it in for you

Thanks for the offer, I am not certain that it is the point of failure, but it the only thing there that would have failed like this, voltages on the surrounding circuitry all seem to checkout and the bridge chip does not even show up in lspci or via nvflash any more, so I think the GPUs are fine, but the bridge chip is dead. Apparently the same chip is used on some ASUS motherboards, if I can locate one I will be in contact .

[eos]

Niche market that NVidia want to squeeze some extra cash out of.

They definitely will do their best...

Maybe because Tesla system needs another NV video card when used in a Windows environment (no outputs).
http://nvidia.custhelp.com/app/answers/detail/a_id/2144/kw/tesla/related/1

GRID in its native form also doesn't have outputs
http://www.nvidia.ca/object/grid-boards.html
But modding a GTX670 to a GRID keeps them live, i.e. no additional card needed...

[gnif]

GRID in its native form also doesn't have outputs
http://www.nvidia.ca/object/grid-boards.html
But modding a GTX670 to a GRID keeps them live, i.e. no additional card needed...

Same for Tesla K10.

[amyk]

Sorry to hear your card died (I suppose a PCIe switch chip needing a heatsink is a little strange in and of itself) but maybe now you can remove the GPUs and see what balls the ID resistors are connected to?

[Neo_Moucha]

........Apparently the same chip is used on some ASUS motherboards, if I can locate one I will be in contact .

That sounds promising, even I have a "dead" ASUS motherboard (failed BIOS update, no boot, otherwise all OK), let me think what model it is... I think P5K PRO or something like that.

[gnif]

........Apparently the same chip is used on some ASUS motherboards, if I can locate one I will be in contact .

That sounds promising, even I have a "dead" ASUS motherboard (failed BIOS update, no boot, otherwise all OK), let me think what model it is... I think P5K PRO or something like that.

The chip the card uses is a PEX8747 (see: http://www.plxtech.com/products/expresslane/pex8747). Some boards I read somewhere are using it to expand the number of PCIe slots on it.

Edit: The ASUS P8Z77-V Premium uses it and it seems it is not using a heatsync! Perhaps I have misdiagnosed the fault, I will have another go tomorrow and check things on it to see if I missed anything obvious.

[Neo_Moucha]

OK, so probably only on newer motherboards...
http://www.anandtech.com/show/6170/four-multigpu-z77-boards-from-280350-plx-pex-8747-featuring-gigabyte-asrock-ecs-and-evga/31

[amigo]

I'm going to order me some thin film resistors for this mod (0402 imperial/1005 metric) but I'm split between Vishay and Panasonic.

While I'm at it I'm ordering an assortment of values (5K-40K), but getting Vishay is generally twice the price of Panasonic.

Obviously I want to keep the cost low so would Panasonic be adequate in this case?

[Fsck]

Panasonic's a good brand. They'll be more than enough assuming you've selected adequate ratings.

[amigo]

Thanks for the heads-up.

The best I can get is 0.1% but that's about as low as they go and at 25ppm/C they should be stable even when the board gets hot.

Although in this application I doubt that'll serve any purpose after the system boots up and Device ID is acquired.

[poorchava]

I'm usually using Yageo, Vishay and Royal Ohm (whichever i can get cheapest). Never actually had any problem with them.

[gnif]

I am using Vishay, but through testing I found that the values can be +- 2K and it still works, I do not think you need to worry too much.

Also, the straps are read every time the video driver loads, if you are using Linux, this means every time you logout/login. Or windows, every time you get a 'driver is not responding' error and it recovers.

[reefjunkie]

Hi-
Well I ended up getting Two EVGA 04G-P4-3687-KR GeForce 4GB GTX 680. Core Clock 1084mhz and Boost Clock 1150mhz.
The boards are the same as the GV-N680OC-2GD except mine are 4GB. I modded both of them to Quadro K5000 (Thanks old Playstation 3 for the resistors  )
I ran the latest Nvidia 314.22 drivers and Quadro 311.35. It seems the 314.22 drivers are a little bite better so I'm using those.
I did some benchmarking to compare the cards before and after the mod's.
                                                   GTX 680 #1   GTX 680 #2   K5000  #1   K5000 #2
3DMARK 11                                      9022                   8987           9077           9016
Passmark 8 (3D Graphics Mark)     6044                   6091           6025          5996
PCMark Vantage (Gaming)             19336                 18956         18880        16177
PhysX                                         10158-166 fps   10003-165 fps    10176-167 fps   10123-166 fps
SPECviewperf 11
   Catia-03                                       6.05                       5.98               5.9             10.20
   Ensight-04                                  32.20                     32.23              32.20         32.27
   Lightwave-01                             13.23                      12.84            13.14           13.22
   Maya-03                                     12.77                     12.73             12.86           12.85
   Proe-05                                       0.96                      1.00                  1.00           0.99
   Sw-02                                         11.09                    11.37                11.36         12.78
   Tcvis-02                                        1.01                      1.17               1.02             1.02   
   Snx-01                                         3.42                      3.37                 3.40            3.42
As you can see all the scores between stock and modded cards are about the same. The problem is with the SPECviewperf 11 scores. This is  the benchmark for Graphic and CAD programs. This is what the Quadro cards were made  for. The scores for the modded K5000 should be MUCH higher. Take a look here.
http://www.xbitlabs.com/articles/graphics/display/nvidia-quadro-k5000_4.html

It looks to me that just because the computer thinks it’s a Quadro K5000 does not mean that it will act like a K5000.
I even tried this benchmark with the Quadro drivers and got the same results. Hopefully It's just a driver issue and not a hardware issue.

[Neo_Moucha]

did you try vga bios mod with nvflash?

[natiss88]

I ran the latest Nvidia 314.22 drivers and Quadro 311.35. It seems the 314.22 drivers are a little bite better so I'm using those.
I did some benchmarking to compare the cards before and after the mod

please, can you benchmark the cards with quadro drivers?

p.s. happy easter to everyone.

[victorngcm]

It looks to me that just because the computer thinks it’s a Quadro K5000 does not mean that it will act like a K5000.
I even tried this benchmark with the Quadro drivers and got the same results. Hopefully It's just a driver issue and not a hardware issue.
[/quote]

Thanks for your test!
I would like to make some suggestion as well...there is some another drivers which can be found in nvidia's homepage....like 191.66 Quadro win 7....would you mind try them as well?

That's a very useful information from your test
I just got my Gigabyte 4G 680 last night and will start modding in the coming days

[niopio]

1) Please can somebody check if this hack gives you 10bit colors output? Thanks!

2) Whats the progress in flashin quadro bios (and overclocking it) to the gtx cards?

[ss_march]

Hi-
Well I ended up getting Two EVGA 04G-P4-3687-KR GeForce 4GB GTX 680. Core Clock 1084mhz and Boost Clock 1150mhz.
The boards are the same as the GV-N680OC-2GD except mine are 4GB. I modded both of them to Quadro K5000 (Thanks old Playstation 3 for the resistors  )
I ran the latest Nvidia 314.22 drivers and Quadro 311.35. It seems the 314.22 drivers are a little bite better so I'm using those.
I did some benchmarking to compare the cards before and after the mod's.

Thank you for tests.
Here i made one photo how to edit stock rom (people comments help here) with HxD and checksum fix with Kepler Bios Tweaker.
And one photo with my guess for GTX660Ti.

[NiHaoMike]

Maybe someone could put on a pin header in place of those resistors and have a play with some resistance substitution boxes? Combine that with a liveUSB designed to boot quickly and automatically do a lspci and it shouldn't take too long to figure things out.

[Marius3D]

Hi-
Well I ended up getting Two EVGA 04G-P4-3687-KR GeForce 4GB GTX 680. Core Clock 1084mhz and Boost Clock 1150mhz.
The boards are the same as the GV-N680OC-2GD except mine are 4GB. I modded both of them to Quadro K5000 (Thanks old Playstation 3 for the resistors  )
I ran the latest Nvidia 314.22 drivers and Quadro 311.35. It seems the 314.22 drivers are a little bite better so I'm using those.
I did some benchmarking to compare the cards before and after the mod's.
                                                   GTX 680 #1   GTX 680 #2   K5000  #1   K5000 #2
3DMARK 11                                      9022                   8987           9077           9016
Passmark 8 (3D Graphics Mark)     6044                   6091           6025          5996
PCMark Vantage (Gaming)             19336                 18956         18880        16177
PhysX                                         10158-166 fps   10003-165 fps    10176-167 fps   10123-166 fps
SPECviewperf 11
   Catia-03                                       6.05                       5.98               5.9             10.20
   Ensight-04                                  32.20                     32.23              32.20         32.27
   Lightwave-01                             13.23                      12.84            13.14           13.22
   Maya-03                                     12.77                     12.73             12.86           12.85
   Proe-05                                       0.96                      1.00                  1.00           0.99
   Sw-02                                         11.09                    11.37                11.36         12.78
   Tcvis-02                                        1.01                      1.17               1.02             1.02   
   Snx-01                                         3.42                      3.37                 3.40            3.42
As you can see all the scores between stock and modded cards are about the same. The problem is with the SPECviewperf 11 scores. This is  the benchmark for Graphic and CAD programs. This is what the Quadro cards were made  for. The scores for the modded K5000 should be MUCH higher. Take a look here.
http://www.xbitlabs.com/articles/graphics/display/nvidia-quadro-k5000_4.html

It looks to me that just because the computer thinks it’s a Quadro K5000 does not mean that it will act like a K5000.
I even tried this benchmark with the Quadro drivers and got the same results. Hopefully It's just a driver issue and not a hardware issue.

Check in nvidia control panel in profile menus. If you have GeForce like profile menus with games on it, then the mod is useless, if you have pro 3D apps we are getting somewhere. I remember when I softmoded a 8800 GTS and a 8800 GTX. Some times the pro apps in NVCP didn't show and I have no performance gain. Also the same when I did bios mod on a 8800GT/9800GT to FX 3700. The performance gain only when I had the right profile in NVCP.

I would also like to donate but since Easter, I can't (bank holidays). Also I would like to buy a card to mod it my self. Currently I have a GTX 580 in my system and as far as I know it cam only be modded into Tesla M2090 since this is the only pro card with GF110. I also have on my desk an reference PNY GTX 465 which can be modded to a Quadro 5000 since both have GF100.


I was really interested in modding nvidia card to Quadro a few ears ago. I work in Maya, Mudbox, Zbrush, 3D Coat etc. Now I am very interested again .

This is what I did in the past

http://translate.google.com/translate?hl=en&sl=ro&tl=en&u=http%3A%2F%2Fwww.mygarage.ro%2Fghiduri-si-tutoriale%2F29854-geforce-g80-g92-mod-quadro-fx-tutorial-si-rezultate.html

http://translate.google.com/translate?hl=en&sl=ro&tl=en&u=http%3A%2F%2Fwww.mygarage.ro%2Fghiduri-si-tutoriale%2F29854-geforce-g80-g92-mod-quadro-fx-tutorial-si-rezultate.html

The thread language is in Romanian (my native one also) so I tried to give a translated version. Anyway one picture is like 1000 words right  ?  Hope you can see them.

Hope I didn't bored you

[reefjunkie]

Here is a list of drivers I have tried and I get the same benchmarks from all of them:
NVIDIA DESKTOP DRIVER
314.22-desktop-win8-win7-winvista-64bit-english-whql

QUADRO DRIVERS
314.07-quadro-tesla-win8-win7-winvista-64bit-international-whql
311.35-quadro-tesla-win8-win7-winvista-64bit-international-whql  (PERFORMANCE DRIVER)
311.35-quadro-tesla-win8-win7-winvista-64bit-international-whql  (ODE DRIVER)
307.45-Performance quadro-tesla-win8-win7-winvista-64bit-international-whql   (CATIA PERFORMANCE DRIVER)

I'll keep trying.......