Kepler TITAN returned after flash "defective"

Discussion in 'Mac Pro' started by macsforme, Dec 25, 2018.

  1. macsforme macrumors member

    Joined:
    Mar 16, 2007
    #1
    I picked up a secondhand kepler TITAN for my Mac Pro, thinking it would be the best case scenario for NVIDIA on Mojave. This was the item description: https://www.ebay.com/itm/nVidia-GeF...-DVI-D-HDMI-DP-Graphics-Card-GPU/253987370909 I tested it for a few weeks in my machine to make sure it worked, which it did, even in Mojave. Here are photos of the card:

    1.jpg

    2.jpg

    3.jpg

    I sent it in to MacVidCards to be flashed for a boot screen. The card was working fine before I shipped it. After a few weeks, I got it back from him with a handwritten note "card is defective can't be flashed." There was nothing more specific in the message about where along the process it failed, and I didn't see any email from him with more information either.

    I plugged it back in to my Mac Pro to try it, alongside my GT120 for the boot screen (my standard setup before sending it in). No boot screen on the monitor connected to the TITAN. The system also freezes after I type in my FileVault password and the progress bar goes about halfway (on the monitor connected to the GT120). I can't boot into macOS at all with the card in. Here's a video of the display output when booting Sierra from an external drive in verbose mode (the system freezes at this point and goes no further):



    I can get into Windows alright, albeit with no display output from the card. I checked GPU-Z and dumped the ROM to try to look at it (I REALLY REALLY wish I'd thought to dump the ROM before sending it in to him to be flashed):

    original.PNG

    Capture.PNG

    (ROM file attached to this post)

    Looking at the ROM, I don't get a grep match for anything "macvidcards" like I would expect if he flashed it. There is a "Version 80.10.2C.00.92," which doesn't match the label on the board, although I suppose it could have been updated at some point. There also appears to be a serial number in the ROM that doesn't match the label on the board. The vendor also shows EVGA in GPU-Z, although I thought the card was from DELL based on the model number.

    What should be my next steps? I'm trying to think of appropriate questions to ask him in an email, such as whether he tested the card before trying to flash it, whether that worked or not, and at what point the card was found "defective." Hopefully he responds. Is there any indication in the info I posted whether the card can be salvaged? Should I try to re-flash it with a stock BIOS available online?
     

    Attached Files:

  2. Macschrauber macrumors 6502

    Macschrauber

    Joined:
    Dec 27, 2015
    Location:
    Germany
    #3
    Do you have sent it in a antistatic foil or package ?
     
  3. macsforme, Dec 30, 2018
    Last edited: Dec 30, 2018

    macsforme thread starter macrumors member

    Joined:
    Mar 16, 2007
    #4
    Oh yes. I wrapped it in an antistatic bag and packaged everything very carefully.

    I am confident that he made modifications to the card, presumably after verifying it arrived working. I found thermal paste smeared on the card in a few places where I don't remember it being before, and looking at (I think) the EEPROM chip, it looks like he soldered on a new one.

    EEPROM chip.jpg

    I've never tried to desolder and re-solder a component like that on a board before, but it looks to me like that chip is soldered differently than the other ones, although all of the contacts seem to be in place and there don't seem to be any shorts or anything. The text is tiny, but it seems to say "adesto1413 25df041A," which seems to be an EEPROM chip from the googling I did. Also, the output of nvflash.exe reports it as "EEPROM ID (1F,4401) : Atmel AT25DF041 2.3-3.6V 4096Kx1S, page," which is a 512KB chip. I guess I don't know what size of EEPROM chip the card originally shipped with, but I assumed 256KB since all of the ROMs seem to be exactly 228.5KB.

    Anyway, I sent him an email asking for more information about where in the process it failed. Waiting on a response.

    Okay, so I started by dumping the ROM in the condition he sent it back to me with GPU-Z as I said before, and then I dumped it again with the nvflash-5.541.0 (seems to be the latest version), and I dumped it a third time with an older, modified version of nvflash-5.221 (more on that in a minute). Interestingly, all three tools produced a slightly different dump, with the difference starting at byte 211329 or 214723 (is that expected? maybe some kind of internal diagnostic information that changes between dumps?).

    When I attempted to use nvflash-5.541.0 to flash the ROM you linked for 80.10.2C.00.02, I got this error (I did not get this error when I tested the tool on a GTX 1080 I also have):

    latest nvflash errors.PNG

    I don't know what the issue was, or if the card was too old or something, but I was able to flash the card with the modified version of nvflash-5.221 that I mentioned above, and I also later verified that the DOS version of nvflash (which is also older than the Windows version, incidentally) can flash the card without this error.

    flash report.png

    So the flash seemed to work, with GPU-Z listing what I believe is the correct ROM for the device, although still a bunch of missing information, and the card doesn't work any better than it did when it arrived.

    GPU-Z after flash of good ROM.PNG

    Additionally, when I tried to do a comparison between the ROM image I flashed and the one on the card after flashing it, the comparison failed.

    comparison after flash with good ROM.PNG

    The one last interesting thing I found is that I do get some screen output, albeit with some weird artifacts, when blind booting into FreeDOS in legacy mode, only when I remove my GT120 from the system. This is the only output I've been able to get from the card at all.

    good screen output in FREEDOS.jpg

    So, flashing the card with the same ROM version as shown on the label didn't seem to help. Any ideas where to go from here? I would be happy at this point to get my card working again like it was before (and hopefully get a refund or credit from him). Would it be worthwhile to try to find a replacement EEPROM chip and replace the one he installed, in case it's a dud? Other than some other kind of random hardware failure, I think I'm currently out of ideas.
     
  4. Macschrauber, Dec 30, 2018
    Last edited: Dec 30, 2018

    Macschrauber macrumors 6502

    Macschrauber

    Joined:
    Dec 27, 2015
    Location:
    Germany
    #5
    Well, the soldering could look a little better for a Professional's work. At least had I used isopropyl to clean it.

    I'd try to write a flash chip of the same type the gpu came with with an external programmer and solder it back on.

    Maybe the cause is a bad flash chip. Also it has twice the size, but that should be ok if it had to fit the addtional efi part. Or its another kind of copy protection, just make the prom bigger so it does not work without soldering...

    If you cant do that I bet there are people in the forum what can.
     
  5. macsforme thread starter macrumors member

    Joined:
    Mar 16, 2007
    #6
    In case anyone was curious about the end of the story...

    I think the card was damaged. I flashed multiple different ROMs to the card using a variety of different versions of nvflash in a variety of different operating systems. I even bought a CH341A USB flasher and a test clip, and tried flashing the original ROM that way. None of these methods was successful in making the card functional.

    I ultimately bought a second card (got a steal of a deal on an EVGA 06G-P4-2791-KR model, which is the superclocked version) and paid MVC for another flash, which was successful. MVC also refunded the cost of the original flash through the PayPal dispute process.

    I then bought some GD25Q40 EEPROM chips, flashed one of them with the known good MVC ROM, and soldered it to the board (ripping off one of the solder pads due to my impatience... grr... but I was able to run a jumper from that pin to the contact nearby after following the trace so I could continue testing). This still didn't resolve the issue, so I concluded that the problem wasn't due to a bad EEPROM chip.

    I guess if I had one suggestion after all of this, it would be to buy a card with a backplate if you're going to send it to MVC to be flashed. My best guess is that perhaps some component was knocked off the board, which has some risk of happening especially due to the layout of slot 1 on the cMP and how the back of the card may rub against the top of the CPU cage, especially if the card is put in and taken out frequently. The EVGA 06G-P4-2790-KR seems to be the model that the MVC ROM is based on. My superclocked model is now running at the frequencies of the base model, so buying the superclocked model seems to be a waste.

    All in all, you can get a GTX TITAN and have it flashed by MVC for around $350 to $400 these days, which will give you (what I believe is) the most powerful card available that works in Mojave and also has an EFI boot screen. I know the necessity of the boot screen is debated, but the peace of mind that comes with FileVault isn't something I'm willing to give up to move to Mojave (and yes, I know I will have to utilize workarounds in order to get it to work).
     
  6. startergo macrumors 6502a

    startergo

    Joined:
    Sep 20, 2018
    #7
    I am sorry Mojave? GTX Titan Mojave drivers? Did I miss something?
     
  7. thornslack macrumors 6502

    Joined:
    Nov 16, 2013
    #8
    Remember in Windows the Gt 120 drivers will interfere with modern nvidia cards. You must remove it’s drivers and use the generic windows display driver if you plan on having both cards in the system at once while using windows.
     
  8. tsialex macrumors 68040

    tsialex

    Joined:
    Jun 13, 2016
    Location:
    Brazil
    #9
    First TITAN GPU is a Kepler one, GK110 like GTX 780.
     
  9. startergo macrumors 6502a

    startergo

    Joined:
    Sep 20, 2018
    #10
  10. tsialex macrumors 68040

    tsialex

    Joined:
    Jun 13, 2016
    Location:
    Brazil
    #11
  11. startergo macrumors 6502a

    startergo

    Joined:
    Sep 20, 2018
    #12
    So the 780 and the GK110 Titan's cards are only flashable by MVC?
     
  12. tsialex macrumors 68040

    tsialex

    Joined:
    Jun 13, 2016
    Location:
    Brazil
    #13
    Yes, but the GTX 780 model A, the first one released, uses the same firmware as GTX 680, flash yourself GPU.
     
  13. zozomester macrumors regular

    zozomester

    Joined:
    Apr 26, 2017
    Location:
    Hungary
    #14
    So the GTX 780 can flassable with the GTX 680 Mac edition ROM? What does it write to the system information?
     
  14. tsialex macrumors 68040

    tsialex

    Joined:
    Jun 13, 2016
    Location:
    Brazil
    #15
    Like I said, GTX 780 model A.

    I had this GPU around 2014 or 2015, don't remember exactly what System Information showed. Some people here still have the model A.
     
  15. joot macrumors member

    Joined:
    Nov 3, 2006
    #16
    So gtx 680, gtx 770 and gtx 780 are the ONLY Nvidia cards that will work on Macpro 3,1 on Mojave?
     
  16. tsialex, Feb 13, 2019
    Last edited: Feb 13, 2019

    tsialex macrumors 68040

    tsialex

    Joined:
    Jun 13, 2016
    Location:
    Brazil
    #17
    Any Kepler GPU supported by Apple NVIDIA native drivers, from 640 GT to TITAN. Some GTX 750Ti has problems with native drivers.

    Flash yourself are just GTX 680 and GTX 780 model A.

    No AMD GPUs for 2008 Mac Pros.
     
  17. macsforme thread starter macrumors member

    Joined:
    Mar 16, 2007
    #18
    Yes, and Mojave has problems when that card is in the system. The reason I wanted a flashed TITAN was so I can get rid of that GT120 card.
     
  18. zozomester macrumors regular

    zozomester

    Joined:
    Apr 26, 2017
    Location:
    Hungary
    #19
    Thank this info!
     
  19. zozomester, Feb 13, 2019
    Last edited: Feb 13, 2019

    zozomester macrumors regular

    zozomester

    Joined:
    Apr 26, 2017
    Location:
    Hungary
    #20
    Could you not specify the exact type of GTX 780 model A?
    Thank!
    I remember that the GTX 680 is the GTX 770.
     
  20. W1SS macrumors 6502

    W1SS

    Joined:
    Aug 20, 2013
    #21
    Unless you have a need for the extra 384 CUDA cores and double precision computes under High Sierra, the next best supported (with flashing) nvidia option is the EVGA GTX 780 SC 6GB with part no 06G-P4-3787-KR which gives you firmware update progress bar screens (on Model B versions) and of course filevault with the workaround I posted.

    Keep in mind though that MVC downclocks GPU base/boost speeds of the SC cards to stock speeds so these cards play nice with the internal PSU as they can draw 250W with their factory speeds which exceeds the cMP's 225W PCIe safe power draw limits.
     
  21. macsforme thread starter macrumors member

    Joined:
    Mar 16, 2007
    #22
    I’m speculating here, but I believe he just maintains one ROM for each model of card, and flashes that when you send it it. I don’t think he takes each card’s ROM and modifies it individually then flashes it back. I sent him two cards during this process, and they both came back with the same VBIOS version (for the EVGA 2790), neither of which matched the actual model of either card. In the first case, he flashed this EVGA VBIOS to a card which wasn’t even an EVGA card, from what I could tell.

    I do recall reading here that someone sent him a factory overclocked card, and it came back with the original frequencies, so that contradicts my own experience. Maybe he only does that for certain models of cards and not for all of them.

    According to the spec sheets and a comparison of the VBIOS of each card here, both cards are rated at 250 watts. The superclocked model is not rated at a higher wattage than the base model (I’m not familiar with how much each card tends to draw in practice). In my case, I use a dual mini 6 pin cable to one 8-pin cable, and I supplement that with a dual SATA power cable to a 6 pin cable, and both of those run through an EVGA PowerLink. Might be overkill, but I wanted to be extra safe.
     
  22. startergo macrumors 6502a

    startergo

    Joined:
    Sep 20, 2018
    #23
    For my R9 280X, which is 225W TDP I use 2 mini 6 pins to 6 pin connected to 6 pin to 6+2, 6+2. So I use on the second part 6+2 going to the 8 pin and 6 going to the 6 pin on the card. So currently I have 2x120W=240W spread out equally across 8 and 6 pin connector. I ordered an EVGA link, but I am not sure if the triple fan long card will allow EVGA link to be installed+the cable without interference with the fan.
     
  23. startergo macrumors 6502a

    startergo

    Joined:
    Sep 20, 2018
    #24
    Alex MVC is stating that the Titan black does not run in Mojave. Also there is no mention of the Titan Z (highest model). Do you know what is the story here?
     
  24. tsialex, Feb 16, 2019
    Last edited: Feb 16, 2019

    tsialex macrumors 68040

    tsialex

    Joined:
    Jun 13, 2016
    Location:
    Brazil
    #25
    Its news to me that the Titan Black don’t work, maybe Apple native driver don’t have the ID or it’s the weird binary blobs into the 1st and 2nd streams of the NVRAM. I know of two people with lots of Titans, but don’t know if they have Black models.

    Titan Z probably needs driver setup to work.
    --- Post Merged, Feb 16, 2019 ---
    Btw, I did two BootROM reconstructions this week with that weird binary blobs, one 4,1>5,1 and one 2010 made into 2012.
     

Share This Page