Author Topic: [MOVED] Hacking NVidia Cards into their Professional Counterparts  (Read 1644915 times)

0 Members and 3 Guests are viewing this topic.

Offline victorngcm

  • Newbie
  • Posts: 9
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #300 on: April 18, 2013, 02:00:56 am »
The problem of mine is that I got a 4G gigabyte GTX 680....which is not the same as EVGA...I can't find the correct resistors on the board...
Could someone help please?

It should not make that much difference. What is the brand and model # of your card?

Gigabyte GTX 4G
I couldn't find the Y1 element on the front but on the back
 

Offline amigo

  • Regular Contributor
  • *
  • Posts: 108
  • Professional Wannabe
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #301 on: April 18, 2013, 02:46:53 am »
Gigabyte GTX 4G
I couldn't find the Y1 element on the front but on the back

Like this? http://www.overclockzone.com/articles/zolkorn/2012/12/gigabyte_gtx_680_windforce/IMG_9112.JPG

Right below the rightmost memory chip, the quartz oscillator is there Y1 and then below it to the right could be an EEPROM.

Technically around the EEPROM was where all the resistors were found on most cards, perhaps some of those there are what you would be looking for.

You need to do some tracing and probing with a DMM to see what leads to the EEPROM. Also, the resistors could be on the front side directly above the area where the EEPROM is on the back.
 

Offline solarbot

  • Regular Contributor
  • *
  • Posts: 60
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #302 on: April 22, 2013, 09:21:37 am »
Hello folks - I have just come across your very cool thread, thanks for sharing.

I'm very keen to give this a go and have been searching for specific info on how to do this to an ASUS GTX 660 DirectCU II but without luck - so wondered if I could run the question/image below past yo'all to see if anyone else has done this or can spot the resistors in question from afar?  I'm still waiting for my card so nothing in front of me at the moment but can hopefully post some decent marked up images soon.  Wondering if anyone has anyone produced a reference for the different cards - success and fails?

Cheers for now :-)



from:

http://rog.asus.com/161872012/graphics-cards-2/graphics-go-nifty-on-route-660-asus-geforce-gtx-660-directcu-ii-top-unboxed/
 

Offline lightsol

  • Newbie
  • Posts: 6
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #303 on: April 22, 2013, 08:01:37 pm »
You will probably need to take off the front cooler as well, because thats where they rest ;)
 

Offline solarbot

  • Regular Contributor
  • *
  • Posts: 60
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #304 on: April 24, 2013, 10:05:20 am »
Morning,

My ASUS GTX 660 DirectCU II arrived just now so I took it straight apart :-) - Thanks lightsol, you are spot on by the looks of it, the resistors are on the GPU side :-)

Below are the photos with resistor values - wondering if anyone can confirm that I am looking in the right place and perhaps which resistors I need to change/add/remove to make the most of the ASUS GTX 660 DirectCU II as a Quadro card- ie what is the most we can get out of this baby :-)

In case it helps the datasheet for the Serial Flash IC is here: http://www.mxic.com.tw/QuickPlace/hq/PageLibrary4825740B00298A3B.nsf/h_Index/3F21BAC2E121E17848257639003A3146/$File/MX25L2006E,%203V,%202Mb,%20v1.3.pdf

Might I also ask if anyone knows what size these resistors are.... 0603 or 0402 ?

Cheers.
« Last Edit: April 24, 2013, 10:07:47 am by solarbot »
 

Offline solarbot

  • Regular Contributor
  • *
  • Posts: 60
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #305 on: April 24, 2013, 02:30:58 pm »
Hi,

I was wondering if the info was available to allow us less able folk to work the resistors out ourselves - it might be in this thread already or written up somewhere else but I haven't found it yet?

I guess with the info below we would be able to track down the fix on a card by card basis?

- Which resistors are connected to with which pin on the Serial Flash chip - eg R1 connected to pin 1 on the Serial Flash chip.
- A list of resistor values that equate to device IDs - eg R1@10K, R2@25K & R3@1R = 11BA
- A list of Device IDs which reference card names - eg 11BA = Quadro 5000
- A list of GTX cards & what the metamorphose into - eg a GTX680 into a Quadro 5000

The pin out for the MX25L2006E IC is below.

Any pointers appreciated :-)
 

Offline vacaloca

  • Contributor
  • Posts: 14
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #306 on: April 25, 2013, 10:08:03 pm »
This question is for those of you that have already modded to a Tesla K10, Quadro 4000 or Quadro 5000, or for that matter either GRID K1 or K2 variants:

Which (if any) settings are you able to change using nvidia-settings? See below:
http://microway.com/hpc-tech-tips/2011/12/nvidia-smi_control-your-gpus/
Are you able to enable ECC memory settings, for example? Please try to change as many settings as you can test. :)

In Windows installations, nvidia-smi is (usually) present in %ProgramFiles%\NVIDIA Corporation\NVSMI for those that are looking for it.

There still seems to be debate as to how performance differs after a modification is done, and I'm surprised that hasn't been looked into -- virtualization is a great feature, but others are more interested in CAD performance (Quadro mods) or seeing if other nvidia-smi settings can be changed... like these:

Code: [Select]
    -e,   --ecc-config=         Toggle ECC support: 0/DISABLED, 1/ENABLED
    -p,   --reset-ecc-errors=   Reset ECC error counts: 0/VOLATILE, 1/AGGREGATE
    -c,   --compute-mode=       Set MODE for compute applications:
                                0/DEFAULT, 1/EXCLUSIVE_THREAD,
                                2/PROHIBITED, 3/EXCLUSIVE_PROCESS
    -dm,  --driver-model=       Enable or disable TCC mode: 0/WDDM, 1/TCC
    -fdm, --force-driver-model= Enable or disable TCC mode: 0/WDDM, 1/TCC
                                Ignores the error that display is connected.
          --gom=                Set GPU Operation Mode:
                                    0/ALL_ON, 1/COMPUTE, 2/LOW_DP
    -ac   --application-clocks= Specifies <memory,graphics> clocks as a
                                    pair (e.g. 2000,800) that defines GPU's
                                    speed in MHz while running applications on a GPU.
    -rac  --reset-application-clocks
                                Resets the application clocks to the default value.
    -pl   --power-limit=        Specifies maximum power management limit in watts.
« Last Edit: April 25, 2013, 10:12:49 pm by vacaloca »
 

Offline Kniteman77

  • Newbie
  • Posts: 2
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #307 on: April 26, 2013, 07:13:59 am »
So upon reading through this thread I have a couple questions.

It Appears that you CAN unlock a GTX680 into a full K5000 /w the same pipelines, added features and performance boost.

However, it appears you CANNOT unlock a GTX670 into a K5000 /w the pipelines and performance boost.

Is there any information to the contrary on that I'm missing?

I'm trying to build a video editing MacPro and I'm looking for the best single GPU card I can throw in.
« Last Edit: April 26, 2013, 07:42:53 am by Kniteman77 »
 

Offline gordan

  • Frequent Contributor
  • **
  • Posts: 277
  • Country: 00
    • Tech Articles
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #308 on: April 30, 2013, 07:31:25 pm »
Has anybody tried to modify a Titan into a Grid K2?
I'd like the performance of the extra shaders over the GTX680, but I specifically need it for Xen VGA passthrough. And I'm not sure what the driver will do when it sees a card that claims to be a Grid K2 with an extra thousand shaders.

Alternatively, has anyone modified a Gainward Phantom 4GB GTX680? Is the PCB a straight reference design?
 

Offline allyman

  • Newbie
  • Posts: 1
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #309 on: April 30, 2013, 09:58:58 pm »
First of all, great work on discovering this.

Has anyone gotten Nvidia Mosaic to work on the modified cards? I modified my EVAG GTX670 to a Quadro K5000 successfully, however using the quadro driver rev 314.07 or 311.50 no mosaic support in nv control panel.  Do I have to enable it somehow?
 

Offline moisyes

  • Newbie
  • Posts: 4
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #310 on: May 02, 2013, 03:41:31 pm »
Hi all,

I decided to have a go at finding the straps for GPU 1 on my card, with both success and failure as the result. I was able to locate them and modify the GTX690 to be a dual core Quadro K5000, but I made the stupid mistake of running it without a heatsync on the bridge chip in the middle of the two while testing. The chip quickly died from overheating when I got excited and let Linux boot into the graphical environment, and there goes my $1000 video card for the greater good, and as such donations are now more important then ever to replace this card now.

I am now running on a semi faulty GT220 (random lockups) and an AMD Radeon X300 to get my triple head working, but as you can imagine this is a very buggy configuration.

Thank you very much for you job, Gnif. I am working in a Desktop with a GTX 690 using Blender for Architectuiral Rendering. After finding these posts I adviced him And he purchased a GTX 680 GB Zotac, and I could (tks God) Mod it to Quadro K5000.I'm testing it and will take the GTX 690 for modding too. Will I need any extra heatsink to prevent what happenend to you hero card? In you opinion, will be better for my use to mod it to a dual quadro k5000 or to a K10? Thank very much in advance, Gnif.



Also that SOIC that sits near the straps I believe is the EEPROM.
 

Offline amigo

  • Regular Contributor
  • *
  • Posts: 108
  • Professional Wannabe
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #311 on: May 02, 2013, 10:03:36 pm »
You do not need an extra heat sink, just make sure you put your original one back in place.

The reason gnif's 690 went up in smoke is because in haste he did not put the heat sink back on, so learn from his painful example.
 

Offline gamezr2ez

  • Contributor
  • Posts: 30
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #312 on: May 05, 2013, 10:44:35 am »

Might I also ask if anyone knows what size these resistors are.... 0603 or 0402 ?

Cheers.

Judging from the pictures, these have got to be 0402. I am planning on modding my gtx680 when i get a chance.

I would like to know this ahead of time if someone knows the answer.
 

Offline justanothercanuck

  • Frequent Contributor
  • **
  • Posts: 391
  • Country: ca
  • Doing retro repairs...
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #313 on: May 06, 2013, 04:18:11 am »
Wow, I didn't know people were still doing this...  I guess I'm a little late here, I should really browse the forums more.  :P

I remember people used to do this so they could run 3d animators and video editors that wouldn't run on the desktop cards...  I'm surprised they are skimping on the linux drivers though...  Kindof sad really, as they're all I would recommend for linux systems, as the ATI drivers were an absolute hellhole for the past 10 years.  What can you do I suppose...  :-//
Maintain your old electronics!  If you don't preserve it, it could be lost forever!
 

Offline gamezr2ez

  • Contributor
  • Posts: 30
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #314 on: May 06, 2013, 09:24:00 am »
I remember people used to do this so they could run 3d animators and video editors that wouldn't run on the desktop cards...  I'm surprised they are skimping on the linux drivers though...  Kindof sad really, as they're all I would recommend for linux systems, as the ATI drivers were an absolute hellhole for the past 10 years. What can you do I suppose...  :-//

I would have to agree with you there. They were bad, but slowly getting better, though they still have some major problems.

One area where AMD/ATI shines is virtualization. I can pass a 7970 through to a Xen guest with relative ease and get native performance within the VM, useful for gaming no doubt. I believe AMD even worked to help build the code that Xen uses for the gpu passthrough.

In any case, I do not recommend nVidia on linux anymore. I did buy a gtx680 just to do this though.... mmmm FLReset.
 

Offline gordan

  • Frequent Contributor
  • **
  • Posts: 277
  • Country: 00
    • Tech Articles
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #315 on: May 07, 2013, 06:56:38 pm »

One area where AMD/ATI shines is virtualization. I can pass a 7970 through to a Xen guest with relative ease and get native performance within the VM, useful for gaming no doubt. I believe AMD even worked to help build the code that Xen uses for the gpu passthrough.

In any case, I do not recommend nVidia on linux anymore. I did buy a gtx680 just to do this though.... mmmm FLReset.

Both Nvidia and ATI have issues once you start straying from the basics. All you can do most of the time prioritize what you need to not be broken and pick the card which works for that specific environment. Just try getting things working reliable in both Linux and Windows with something like an IBM T221 and you'll find there's a number of pitfalls if you haven't done your research properly to begin with. Throw Xen virtualization into the mix and the number of complexities multiplies.

Regarding Xen virtualization, I haven't tried Nvidia yet (my Quadro 2000 for testing is in the post), but I sincerely hope the experience is less appalling than with the ATI. Granted, ATI cards almost work whereas desktop Nvidia cards don't work at all with VGA passthrough without a whole raft of extra Xen patches, but the experience is poor at best. The longest I've gotten out of my 6450 card I've been testing with is about 10 minutes before something odd happens and the driver decided to reset the card - at which point the VM crashes, and the only way you'll get it to boot again without BSOD-ing before the login screen is by rebooting the host. If you have more than one CPU you'll only get to the guest's login screen if you disable IRQ balancing. All in all, good enough for a demo, but absolutely not good enough for anything meaningful.

My plan is to test whether using a Quadro 2000 (drivers officially supports VGA passthrough) makes for a workable experience before I spend 4x as much on a GTX680 to modify into a Quadro K5000 or a Grid K2. Ideally I'd rather like to get a Titan and see what happens if I mod it's device ID to read as a K5000, but as far as I can tell nobody ever reported trying it, and I'd hate to end up with a Titan that I cannot use for it's intended purpose.
 

Offline gamezr2ez

  • Contributor
  • Posts: 30
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #316 on: May 07, 2013, 10:34:20 pm »
Regarding Xen virtualization, I haven't tried Nvidia yet (my Quadro 2000 for testing is in the post), but I sincerely hope the experience is less appalling than with the ATI. Granted, ATI cards almost work whereas desktop Nvidia cards don't work at all with VGA passthrough without a whole raft of extra Xen patches, but the experience is poor at best. All in all, good enough for a demo, but absolutely not good enough for anything meaningful.

The patches are only 5 files, about 100 lines of code in total. They are just to read the bios from an extracted bios rather than from the card at runtime as well as a few other things that Xen can't pull dynamically, unlike AMDs. It is fairly basic code, nothing fancy.

As far as only "good enough for a demo", I will have to disagree. You may have just had a poor experience and been unfortunate enough to have an uncooperative motherboard and graphics card. I can attest the fact that the passthrough is fairly stable once it is setup properly (that's the hard part). I had it running for 2 weeks as a gaming VM and it never had a hiccup with an older 5670 of mine. It was impressive! That said, I wouldn't put this into a production environment without alot more testing.

My plan is to test whether using a Quadro 2000 (drivers officially supports VGA passthrough) makes for a workable experience before I spend 4x as much on a GTX680 to modify into a Quadro K5000 or a Grid K2. Ideally I'd rather like to get a Titan and see what happens if I mod it's device ID to read as a K5000, but as far as I can tell nobody ever reported trying it, and I'd hate to end up with a Titan that I cannot use for it's intended purpose.

Keep in mind that the Quadro series does not support FLReset. It is probably not a good idea to use that for passthrough if you plan to start and stop the VM. It will work just fine, but Xen/linux won't be able to reset the card upon VM reboot. If you have to reboot the VM you'll still need to reboot the entire machine. You may have crashes or performance degradation otherwise.

The above also applies to the K5000 if you plan to modify a Titan, there will be no FLReset.

Good luck!
 

Offline gordan

  • Frequent Contributor
  • **
  • Posts: 277
  • Country: 00
    • Tech Articles
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #317 on: May 08, 2013, 06:46:03 am »
Regarding Xen virtualization, I haven't tried Nvidia yet (my Quadro 2000 for testing is in the post), but I sincerely hope the experience is less appalling than with the ATI. Granted, ATI cards almost work whereas desktop Nvidia cards don't work at all with VGA passthrough without a whole raft of extra Xen patches, but the experience is poor at best. All in all, good enough for a demo, but absolutely not good enough for anything meaningful.

The patches are only 5 files, about 100 lines of code in total. They are just to read the bios from an extracted bios rather than from the card at runtime as well as a few other things that Xen can't pull dynamically, unlike AMDs. It is fairly basic code, nothing fancy.

I'm not saying it's a big deal, but it's unstable and not merged into the mainline last time I checked, which means that if you care about such niceties as package management, especially on a stable (e.g. EL6) rather than unstable bleeding edge (e.g. Fedora) distribution, you have to hunt around for the specific versions that the patches are against or versions that the patches apply against cleanly (and work). And considering that some of these things don't work too well even when extensively tested, I don't fancy my chances much with something that has been only tested by a handful of individuals.

As far as only "good enough for a demo", I will have to disagree. You may have just had a poor experience and been unfortunate enough to have an uncooperative motherboard and graphics card. I can attest the fact that the passthrough is fairly stable once it is setup properly (that's the hard part). I had it running for 2 weeks as a gaming VM and it never had a hiccup with an older 5670 of mine. It was impressive! That said, I wouldn't put this into a production environment without alot more testing.

It's not just about the motherboard or the card being uncooperative. Sure, having to disable irq balancing took a while to think of to try (google didn't find much obvious mention of it, and neither did anyone mention it on the threads I posted on xen-users). But if things work for 10 minutes, they should stay working and not randomly crash after that. And since the BSODs always refer to a device reset failure/timeout, it looks like the problem is mainly related to the driver not knowing how to reset the card properly. (Also see below re: FLreset.)

My plan is to test whether using a Quadro 2000 (drivers officially supports VGA passthrough) makes for a workable experience before I spend 4x as much on a GTX680 to modify into a Quadro K5000 or a Grid K2. Ideally I'd rather like to get a Titan and see what happens if I mod it's device ID to read as a K5000, but as far as I can tell nobody ever reported trying it, and I'd hate to end up with a Titan that I cannot use for it's intended purpose.

Keep in mind that the Quadro series does not support FLReset. It is probably not a good idea to use that for passthrough if you plan to start and stop the VM. It will work just fine, but Xen/linux won't be able to reset the card upon VM reboot. If you have to reboot the VM you'll still need to reboot the entire machine. You may have crashes or performance degradation otherwise.

The above also applies to the K5000 if you plan to modify a Titan, there will be no FLReset.

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.

I'll know one way or the other soon enough.

Edit: Having tried a Quadro 2000, the experience thus far is that it is even more unstable than using an ATI card. Most disappointing. I guess I won't be wasting my time modifying a GTX into a Quadro.
« Last Edit: May 09, 2013, 05:21:56 pm by gordan »
 

Offline gamezr2ez

  • Contributor
  • Posts: 30
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #318 on: May 10, 2013, 04:25:46 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.

I'll know one way or the other soon enough.

Edit: Having tried a Quadro 2000, the experience thus far is that it is even more unstable than using an ATI card. Most disappointing. I guess I won't be wasting my time modifying a GTX into a Quadro.


Today I have found the time to test and successfully modify my GTX 680.

I tested a K5000 and Grid K2. To my surprise neither of them supported FLReset! At least not according to `lspci`. This means I cannot issue a reset to the card through the kernel. Despite that, it worked fine. It would appear you are correct about FLReset not being necessary.

The Grid K2 was my first to test, but both cards behaved identical in my testing with a small difference after rebooting the VM.

I complied xen-4.2.1 (stable) without any patches. I did a standard vga passthrough. The real video card is actually a secondary video card in the system with this method. I installed the appropriate drivers and reboot the VM. Upon reboot everything worked as desired! It was quite easy compared to the previous mess I have been through trying to get this to work.

The "K5000" card had performance issues upon VM reboot, but it did not blue screen. I ejected the K5000 using Windows "Safely Remove Device" feature. After it came back up the performance degradation was gone and it performed perfect as far as I could test it.

Thanks go to verybigbadboy and gnif!

The Xen portion of this is a bit offtopic, so if anyone has any questions on that feel free to send me a PM.
 

Offline Kniteman77

  • Newbie
  • Posts: 2
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #319 on: May 12, 2013, 09:26:53 am »
I'm looking at getting this card.

It seems like most of the 680's people are modding are the 2gb version, how can I tell I'll be able to mod this 4gb version?

I'm still working my way through the thread, it's quite long. :-/
 

Offline vacaloca

  • Contributor
  • Posts: 14
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #320 on: May 13, 2013, 12:41:16 am »
I posted this a while back but no one addressed it:
https://www.eevblog.com/forum/projects/hacking-nvidia-cards-into-their-professional-counterparts/msg223546/#msg223546

The TL,DR version is, does modifying any particular card into Tesla K10, Quadro 4000 or Quadro 5000, or for that matter either GRID K1 or K2 variants enable nvidia-smi support for changing the settings listed in the link above? e.g. ECC/TCC support, application clocks, power limit?

Note that not all the settings may work with a particular (transformed) card. If anyone could try to modify each of the settings for their modified card I would very much appreciate it!
« Last Edit: May 13, 2013, 12:49:54 am by vacaloca »
 

Offline gamezr2ez

  • Contributor
  • Posts: 30
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #321 on: May 13, 2013, 03:47:33 am »
I posted this a while back but no one addressed it:
https://www.eevblog.com/forum/projects/hacking-nvidia-cards-into-their-professional-counterparts/msg223546/#msg223546

The TL,DR version is, does modifying any particular card into Tesla K10, Quadro 4000 or Quadro 5000, or for that matter either GRID K1 or K2 variants enable nvidia-smi support for changing the settings listed in the link above? e.g. ECC/TCC support, application clocks, power limit?

Note that not all the settings may work with a particular (transformed) card. If anyone could try to modify each of the settings for their modified card I would very much appreciate it!

I have a GTX 680 modified to a Grid K2 passed through to a Windows 7 x64 xen VM. I am running the nVidia quadro/tesla/grid drivers version 320.00

Here is a pastebin of my nvidia-smi out

ECC would have to be supported by the RAM, which they wouldn't install in a consumer grade card. The power features and other things I would _assume_ are also added hardware bits that physically don't exist on the card.

I am no expert on this subject or with nvidia-smi, though. If you would like me to try something else, I will. I may have just not used the correct commands.

After reading this entire thread can i conclude the following?
a GTX680 can be fairly easily modded into K5000 / K10 / Grid K2 by fixing some ID-resistors
this results in additional features (like gpu passthrough for VM's and Mosaic support)
but no performance gain for Pro apps (specviewperf 11)

Or has anyone (Gnif, VeryBigBadBoy, ReefJunkie, etc.) discovered
how to actually boost the OpenGL performance of a GTX680 ?
For many self-employed pro-users like me that would be truly awesome!

The performance boost you are asking about would have to come from a disabled feature or something to that effect. As far as I am aware, there are no disabled features related to performance. This just removed the arbitrary limitations placed on nVidia cards that have no purpose other than to get the user to upgrade to a professional card to gain access to those features.

Also, in the case of the Grid K2, the real card supports VGX. I have no gotten that to work with this card sadly. I haven't spent to much time with it yet. Just want to see it in action, play with it a little.
« Last Edit: May 13, 2013, 03:56:57 am by gamezr2ez »
 

Offline vacaloca

  • Contributor
  • Posts: 14
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #322 on: May 13, 2013, 02:30:24 pm »
I have a GTX 680 modified to a Grid K2 passed through to a Windows 7 x64 xen VM. I am running the nVidia quadro/tesla/grid drivers version 320.00

Here is a pastebin of my nvidia-smi out

ECC would have to be supported by the RAM, which they wouldn't install in a consumer grade card. The power features and other things I would _assume_ are also added hardware bits that physically don't exist on the card.

I am no expert on this subject or with nvidia-smi, though. If you would like me to try something else, I will. I may have just not used the correct commands.
Thanks for the output. I'll give a bit of background. Under Linux, there are some card settings that cannot be read, and to my knowledge there isn't any sort of application that provides the equivalent control/monitoring of the Windows NVIDIA Inspector and EVGA Precision X/MSI Afterburner applications. At least with nvidia-smi support, monitoring under Linux would be possible if the settings are reported, however at least for your converted Grid K2, it doesn't seem like anything else is reported, boo. (i.e. no data for GPU Utilization)

The other major thing for me would be TCC support -- it turns off the card outputs and doesn't have the overhead of driving a display when you're running computational CUDA codes under Windows, for example.

Can you try these? If there was no change, nvidia-smi should say 'not supported'. If it nvidia-smi acknowledges that it's able to change any of the settings it would be great. :)

Code: [Select]
nvidia-smi -e 1
nvidia-smi -dm 1
nvidia-smi -fdm 1
nvidia-smi --gom=0
nvidia-smi -ac 2000,800
nvidia-smi -pl 250

Also, if anyone else can try these with a converted Quadro or Tesla to confirm those cards behave the same way, that'd be awesome too. (nvidia-smi doesn't explicitly state full support for GRID cards, just Tesla/Quadro)
« Last Edit: May 13, 2013, 02:37:50 pm by vacaloca »
 

Offline gamezr2ez

  • Contributor
  • Posts: 30
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #323 on: May 13, 2013, 03:59:07 pm »
Also, if anyone else can try these with a converted Quadro or Tesla to confirm those cards behave the same way, that'd be awesome too. (nvidia-smi doesn't explicitly state full support for GRID cards, just Tesla/Quadro)

Quote
Supported products:
- Full Support
    - NVIDIA Tesla Line:
            S2050, C2050, C2070, C2075,
            M2050, M2070, M2075, M2090,
            X2070, X2090,
            K10, K20, K20X, K20Xm, K20c, K20m, K20s
    - NVIDIA Quadro Line:
            410, 600, 2000, 4000, 5000, 6000, 7000, M2070-Q
            K2000, K2000D, K4000, K5000, K6000
    - NVIDIA GRID Line:
            K1, K2, K340, K520

It does say it fully supports GRID K2. I can mod it over to a Tesla and check the difference (if any). I am not at home at the moment, but I will run your other commands to test it out.

My extent for using features beyond gaming is just some oclhashcat and some experimental x264 gpu stuff. Also some very rare 3D modeling stuff, not enough to care about performance. But if I am reading your post correctly, then this could give added performance in those aspects, yes?

EDIT:
Code: [Select]
nvidia-smi -e 1
nvidia-smi -dm 1
nvidia-smi -fdm 1
nvidia-smi --gom=0
nvidia-smi -ac 2000,800
nvidia-smi -pl 250

All features came back as "not supported for GPU"
« Last Edit: May 14, 2013, 07:10:44 pm by gamezr2ez »
 

Offline vacaloca

  • Contributor
  • Posts: 14
Re: Hacking NVidia Cards into their Professional Counterparts
« Reply #324 on: May 13, 2013, 05:46:40 pm »
It does say it fully supports GRID K2. I can mod it over to a Tesla and check the difference (if any). I am not at home at the moment, but I will run your other commands to test it out.

My extent for using features beyond gaming is just some oclhashcat and some experimental x264 gpu stuff. Also some very rare 3D modeling stuff, not enough to care about performance. But if I am reading your post correctly, then this could give added performance in those aspects, yes?

Well, TCC mode is a Windows exclusive setting. It disables the display outputs, so it would only be useful for running applications that don't depend on a display -- running CUDA code that just crunches numbers is one of those situations. Here's a bit more details about it: http.developer.nvidia.com/ParallelNsight/2.1/Documentation/UserGuide/HTML/Content/Tesla_Compute_Cluster.htm

I guess nvidia-smi executables have different verbiage between versions or perhaps even architectures. Under Win 7 x64, 314.14 driver, nvidia-smi 4.314.14 version reports:

Code: [Select]
Supported products:
- Full Support
    - NVIDIA Tesla Line:
            S2050, C2050, C2070, C2075,
            M2050, M2070, M2075, M2090,
            X2070, X2090,
            K10, K20, K20X
    - NVIDIA Quadro Line:
            4000, 5000, 6000, 7000, M2070-Q, 600, 2000, 3000M and 410
    - NVIDIA GeForce Line:            None

which is why I thought GRID K2 wasn't supported fully. Thanks for testing, look forward to seeing if by any chance any of the mode changes work. :)
 


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf