iM 4,1-14,4 2011 iMac Graphics Card Upgrade

[Ausdauersportler]

Hi everyone,
I've been following you for some time but it's my first post, I'm Italian

Hardware: Imac 27" 2011 last firmware + Nvidia K3000M flashed by Nikey22 rom

Problem: with the internal display connected, imac does not turn on (only the first LED turns on). With the internal display disconnected (internal cable) everything works ok but only in the external display, as shown in the pictures.
With the original Apple graphics card everything works perfectly

Attention: as you can see, macfan control sees 2 GPUs, maybe it can be useful to solve the problem?

Thanks for can answer me

Hi!

This is a known issue with some NVIDIA cards, it has been posted several times here and it is mentioned on the first post and there is no solution known.

We do not know what the root cause of this problem is. I had exactly the same problem with an iMac 2011 and had to move this particular card into a 2009/2010 system - where it performed completely normal.

Your iMac 2011 has an Sandy Bridge Intel CPU with an integrated GPU named HD3000.

 

[nolikeee]

Hello guys GTX 680M Vbios only for 4Gb version card? i have 2Gb Dell 680M and nvflash just says error for another version of ID mismatch. If i force flash GPU (-4 -5 -6) in windows gpu cant run 3D graphics - shows black screen and artifacts, when im back to dell original rom card works good. So i tested in Alienware 17 laptop.

I saw bios based on Clevo ROM. Clevo\MSI Cards have 4Gb but dell only 2Gb.

did somone try to work Dell 680M?

 

[rodolfo.araujo]

A useless post to thank everybody who helped develop the solutions that made legacy system fully usable with newer OS. So much hard work from these guys. I've been following this thread for a few years and can see how we have evolved from a completely experimental adventure to an almost out-of-the box solution. You guys are GREAT and I really hope we can push these systems a little further into the next decade!

THANKS!

 

[RioRio]

If the CPU and GPU performance is constantly below what is expected, it is often a thermal throttle in order to avoid over heating. I had this after swapping my CPU, when I forgot to plug in the CPU heatsink sensor cable.

Thank you for your reply. I just did the benchmark in another one of my iMacs with all fans on max speed. The GPU heatsink temp always stays below 64ºC in this one but the Valley score is still the same. So overheating seems not to be the problem, right?

Any other ideas?

 

[zzzeck]

Hi!

This is a known issue with some NVIDIA cards, it has been posted several times here and it is mentioned on the first post and there is no solution known.

We do not know what the root cause of this problem is. I had exactly the same problem with an iMac 2011 and had to move this particular card into a 2009/2010 system - where it performed completely normal.

Your iMac 2011 has an Sandy Bridge Intel CPU with an integrated GPU named HD3000.

Thank you !!

now I'am looking for a 27 imac 2009 or 2010 to complete the job.

 

[Ausdauersportler]

Thank you for your reply. I just did the benchmark in another one of my iMacs with all fans on max speed. The GPU heatsink temp always stays below 64ºC in this one but the Valley score is still the same. So overheating seems not to be the problem, right?

Any other ideas?

As always the first idea is to check back to the first post and read about issues or hardware problems or search this thread:

first post -> hardware problems -> missing sensor -> no boost of CPU => low Valley Score

 

[RioRio]

As always the first idea is to check back to the first post and read about issues or hardware problems or search this thread:

first post -> hardware problems -> missing sensor -> no boost of CPU => low Valley Score

To exclude this issue I already ran Geekbench 5. Scores seem normal. If the CPU wouldn't boost, I would expect to get much lower Geekbench scores, right? I also ran the Apple Hardware Test and no sensor issues were found. The problem appears in all my 3 iMacs, I think it's pretty unlikely that they have all the same sensor issue.

 

[Mischa85]

iMac 9,1/10,1 (2009 Core2Duo) GPU Change Project

In this post I try to post as much information on hacking the VBIOS to work on our precious iMacs. I'm starting literally from scratch. When I learn something new I will share it with the world. I might get some things wrong initially but I try to be as accurate as I can.

This post seems to hint on the fact that these older iMacs are using LVDS instead of DP for the internal display. Indeed we find LVDS entries in the VBIOS of the iMac9,1. Adding these to the VBIOS of another card might just work.

iMac9,1 uses LVDS, iMac10,1 uses DP!

I have no idea why the vbiosses supplied here are not working on the iMac10,1. A GeForce 9400M dump of an iMac10,1 probably holds the answer, but I have not acquired it yet.

So the question to investigate is: What did Apple change between A1311 (iMac10,1) and A1312 (iMac11,1)?

iMac10,1:

iMac11,1:

Plan of attack:

• Put valid DCB Table Entry #0 X3 X1 00 01 34 00 01 00 / Copypaste whole DCB Table
• Put valid CCB Table Entry #X 03 00 00 05 / Copypaste whole CCB Table
• Put valid Connector Table Entry #X 40 02 00 00 / Copypaste whole Connector Table
• Fix Checksum

First off I gather the VBIOS from a iMac9,1 featuring a Geforce 9400M. This is done using rEFInd. Because the firmware of these iMacs is too old we also need the EFI shell of rEFIt. Replace the shell.efi in the 'tools' folder. Information is gathered from here.

- "devices -b"
E6 B - - 1 2 5 NVIDIA GPU

- "dh -b e6"
ROM Size......: EE00
ROM Location..: BE62D018

- "fs0:"
- "mem BE62D018 EE00 > 9400_output.txt"

The DCB pointer is found at 0x36 of the UEFI rom starting with 0x55 0xAA. All pointers are relative from this location, if an IFR header is present the offset needs to be added!

55 aa 77 eb 4b 37 34 30  30 e9 4c 19 77 cc 56 49  44 45 4f 20 0d 00 00 00  98 01 9a 1c 00 00 49 42
4d 20 56 47 41 20 43 6f  6d 70 61 74 69 62 6c 65  01 00 00 00 c0 00 14 be  30 31 2f 30 38 2f 30 39

So this is 0x14 0xBE -----> pointer to 0xBE14.

This is the header found at @0xBE14:

40 1b 10 08 af be cb bd  dc 4e 00 f4 be f1 bf fd  bf 06 c0 1b c0 01 00 00  60 c0

Header Size (this!) = 0x1b = 27 bytes long
Entry Count = 0x10 = 16 entries
Entry Size = 0x08 = 8 bytes long
Communications Control Block @ 0xBEAF.
GPIO Assignment Table @ 0xBEF4.
Input Devices Table @ 0xBFF1
Personal Cinema Table @ 0xBFFD
Spread Spectrum Table @ C006
Connector Table @ C01B
??? @ C060

Directly following are the DCB Device Entries:

DCB Device Entries

00 => 23 11 00 01 34 00 01 00 <-- TYPE = 0x3 = LVDS; EDID PORT = 0x2 = Reference to Communications Control Block?; CON INDEX = 0x1 = Connector Table Index
01 => 32 02 01 02 10 00 00 00 <-- TYPE = 0x2 = TMDS; EDID PORT = 0x3 = Reference to Communications Control Block?
02 => 86 02 01 02 10 00 22 0f <-- TYPE = 0x6 = DISPLAYPORT; EDID PORT = 0x8 = Reference to Communications Control Block?
03 => 0e 00 00 00 00 00 00 00 <-- TYPE = 0xE = EOL (End of Line)
04 => 0f 00 00 00 00 00 00 00 <-- TYPE = 0xF = SKIP
05 => 0f 00 00 00 00 00 00 00
06 => 0f 00 00 00 00 00 00 00
07 => 0f 00 00 00 00 00 00 00
08 => 0f 00 00 00 00 00 00 00
09 => 0f 00 00 00 00 00 00 00
10 => 0f 00 00 00 00 00 00 00
11 => 0f 00 00 00 00 00 00 00
12 => 0f 00 00 00 00 00 00 00
13 => 0f 00 00 00 00 00 00 00
14 => 0f 00 00 00 00 00 00 00
15 => 0f 00 00 00 00 00 00 00

Communications Control Block Header:

40 05 10 04 02

Entry Count = 0x10 = 16 entries

Communications Control Block Entries:

00 => 00 00 00 05
01 => 00 00 00 ff
02 => 03 00 00 05 <------ Device Entry 1 refers to this??? 0x3 = I2C
03 => 02 00 00 05 <------ Device Entry 2 refers to this??? 0x2 = DDC2
04 => 00 00 00 ff
05 => 00 00 00 ff
06 => 00 00 00 ff
07 => 00 00 00 ff
08 => 00 00 00 06 <------ Device Entry 3 refers to this??? AUXCH 00
09 => 01 00 00 06 <------ AUXCH 01
10 => 02 00 00 06 <------ AUXCH 02
11 => 03 00 00 06 <------ AUXCH 03
12 => 00 00 00 ff
13 => 00 00 00 ff
14 => 00 00 00 ff
15 => 00 00 00 ff

GPIO Assignment Table Header:

40 04 08 01

GPIO Assignment Table Entries:

00 => 0f
01 => 0f
02 => 0f
03 => 0f
04 => 0f
05 => 0f
06 => 0f
07 => 0f

Input Devices Table:

40 0c 00 00 f0 00 00 00 0f 00 00 00

Header Size = 0x0c = 12 bytes long
Entry Count = 0
Entry Size = 0
So why the long header...?

Personal Cinema Table Header

41 05 04 01 00

Header Size = 0x05 = 5 bytes long

Personal Cinema Table

01 00 00 00

idk lol

Spread Spectrum Table Header

40 05 04 04 00

Header Size = 0x05 = 5 bytes long
Entry Count = 0x04
Entry Size = 0x04

Spread Spectrum Table Entries

01 => ff 00 00 00
02 => ff 00 00 00
03 => ff 00 00 00
04 => ff 00 00 00

Connector Table Header??

40 05 10 04 00

Connector Table Entries??

00 => 46 a1 00 00 <-- DisplayPort External Connector
01 => 40 02 00 00 <-- LVDS-SPWG-Attached <------ Device Entry 1 refers to this???
02 => 46 52 00 00 <-- DisplayPort External Connector
03 => ff 00 00 00
04 => ff 00 00 00
05 => ff 00 00 00
06 => ff 00 00 00
07 => ff 00 00 00
08 => ff 00 00 00
09 => ff 00 00 00
10 => ff 00 00 00
11 => ff 00 00 00
12 => ff 00 00 00
13 => ff 00 00 00
14 => ff 00 00 00
15 => ff 00 00 00

??? @ C060

10 04 12 05

Checksum:

8c

Last byte of UEFI firmware is checksum. It should be necessary to recalculate and change this when making changes to the vbios. Attached is a small C program that calculates this. It currently only understands the raw UEFI/GOP portion of the VBIOS, which is as far as I currently know the only thing we need to work in anyway.

• Do we need the IFR section before the GOP/UEFI section for iMac to operate?
• Do we need the oldschool BIOS after the GOP/UEFI section for iMac to operate?

Attached is the original VBIOS of an iMac9,1 equipped with a GeForce 9400M. This article is based on it and it may contain the information necessary to create a VBIOS that works on iMac9,1 and iMac10,1.

Under investigation (reverse engineering the modded nikey22 vbios):

GPIO:

41 06 20 05 F7 5B

00 => 00 6C 00 00 4F
01 => 01 6D 00 00 4F
02 => 02 21 88 80 4F <--- changes 0x88 (SEL_SOR2_TMDS_OUT_PWM) to 0x84 (SEL_SOR1_TMDS_OUT_PWM)
03 => 83 01 00 00 4F
04 => 04 00 8A 00 4F <--- changes 0x8a (SEL_SOR2_TMDS_OUT_PINB) to 0x86 (SEL_SOR1_TMDS_OUT_PINB)
05 => 05 FF 00 00 EF
06 => 06 6E 00 00 4F
07 => 07 FF 4B 00 43
08 => 08 FF 00 00 EF
09 => 09 49 59 00 1F
10 => 8A 2E 00 00 1F
11 => 0B 81 5D 80 4F
12 => 0C 2B 00 17 BF
13 => 0D FF 00 00 EF
14 => 0E FF 00 00 EF
15 => 0F 51 00 01 EF
16 => 10 FF 00 00 EF
17 => 11 52 00 02 EF
18 => 12 5E 00 03 EF
19 => 13 5F 00 04 EF
20 => 14 FF 00 00 EF
21 => 00 FF 00 00 0F
22 => 00 FF 00 00 0F
23 => 00 FF 00 00 0F
24 => 00 FF 00 00 0F
25 => 00 FF 00 00 0F
26 => 00 FF 00 00 0F
27 => 00 FF 00 00 0F
28 => 00 FF 00 00 0F
29 => 00 FF 00 00 0F
30 => 00 FF 00 00 0F
31 => 00 FF 00 00 0F

 

[Ausdauersportler]

To exclude this issue I already ran Geekbench 5. Scores seem normal. If the CPU wouldn't boost, I would expect to get much lower Geekbench scores, right? I also ran the Apple Hardware Test and no sensor issues were found. The problem appears in all my 3 iMacs, I think it's pretty unlikely that they have all the same sensor issue.

This is another classical. I really do not like to guess.

If you report an issue or error please describe as precise as possible what you already tested yourself.

You can use the FakeSMC package from the Catalina Loader to report the CPU core frequency as well as GPU memory and GPU core frequency. Alternatively you can check for other tools providing this data while you run your benchmarks.

 

[The_Croupier]

I 've bought three 2011 27" iMacs with broken video cards and I've ordered GTX 780M cards from China for them.
The cards are mounted and working. The problem is, that all three cards only reach a score of around 1700 in Unigine Valley. According to the post on page 1 it should be something around 2600.

The max heatsink temperature during benchmark is 72ºC, even with all fans on maximum. The machines all have 3 pipe (modded) heatsinks. I used Arctic MX-4 for the GPU and K5-Pro instead of thermal pads.

What could be the reason for the cards underperforming in all 3 machines?

Check you haven't damaged the 2 resistors at L5500 or L5501 AMB TEMP.
If you have and the pads are intact just bridge the pads and it will fix the throttling.

Below you can see L5500 resistor is gone and a pad missing, i ran a wire from the intact pad to the corresponding pin on the back of the temp connector and its working great.

 

[Ausdauersportler]

Check you haven't damaged the 2 resistors at L5500 or L5501 AMB TEMP.
If you have and the pads are intact just bridge the pads and it will fix the throttling.

Below you can see L5500 resistor is gone and a pad missing, i ran a wire from the intact pad to the corresponding pin on the back of the temp connector and its working great.
View attachment 1794603

Have the same problem, unfortunately the right solder pad of the upper L5500 (coil) is gone.
Which port of the ambient temp sensor do I have to connect to the wire from the remaining left solder pad?

 

[Radical78]

Seems more about hardware than software to my naked eyes.
You can swap the K3000m with a stock GPU to isolate the issues.

Yes, I believe it's hardware. It tries to load High Sierra (in the old SSD) but I only get a blurry image of the desktop.

Also has the same issue:

Replaced the 2N7002K mosfet with a new one and LVDS with a new one but when the solder they applied some green stuff on top...

I'll come back with news later...

 

[The_Croupier]

Have the same problem, unfortunately the right solder pad of the upper L5500 (coil) is gone.
Which port of the ambient temp sensor do I have to connect to the wire from the remaining left solder pad?

I cant remember but I can check later. If you have a multimeter you can check continuity from the small pad on the trace or if you solder from the intact left pad to the small pad on the trace closest to the 0 in L5500 it will work however when i did this first up I kept knocking the wire off when inserting the amb temp connector, so suggest routing the wire behind the socket. I think from memory its the pin under the V.

 

[Mischa85]

I'm working on this post.

Most of the edits are being done in the DCB area.

But there's still some unknown undocumented vbios wizardry outside of there. Could anybody tell us what it is?

 

[The_Croupier]

AMB TEMP pin bridging diagram if anyone needs.

 

[RioRio]

This is another classical. I really do not like to guess.

If you report an issue or error please describe as precise as possible what you already tested yourself.

You can use the FakeSMC package from the Catalina Loader to report the CPU core frequency as well as GPU memory and GPU core frequency. Alternatively you can check for other tools providing this data while you run your benchmarks.

Thank you so much for your help, it is very highly appreciated! I will do my best to provide all relevant information from now.

I ran Unigine Valley again monitoring frequencies with FakeSMC / HWMonitor, as you suggested. During benchmark the GPU core is constantly at 796MHz, and the GPU memory at 4.99GHz. The frequencies of CPU Cores 1-4 jump between 1.58GHz and 3.46GHz.

I don't know if these values are normal. According to Techpowerup the GPU boost clock should be 797 MHz, but according to Notebookcheck the clock speed should be 823 MHz.

On this iMac I have the ODD sensor unplugged, so the ODD fan always runs at high speed. What I've noticed is, that with the ODD fan plugged in, the maximum ODD fan speed is shown as 2500 rpm, while with the ODD sensor unplugged it is shown as 3800 rpm, and the fan spins constantly with 3200 rpm in one iMac and with 3500 rpm in the other. It truly spins faster as it also gets louder. I've measured max 51dB with ODD sensor plugged in and fan at maximum rpm vs. 57dB without ODD sensor and fan also on maximum rpm. But the low GPU score also occurs on my Big Sur system, where I have the ODD sensor plugged in and moved it to the GPU heatsink, so it seems not to make any difference for the low Valley score.

EDIT: I've plugged in the ODD sensor, just to be sure. The frequencies and the final score stay the same.

The cards are all GTX 780M boughed from the same Chinese seller on Aliexpress. They came with Nick [D]vB ROM and I flashed them with the nikey22 780_BR3 ROM. The performance with Nicks Rom was slightly higher (around 1800).

I've added screenshots of HWMonitor before and during benchmark.

 

[RioRio]

Check you haven't damaged the 2 resistors at L5500 or L5501 AMB TEMP.
If you have and the pads are intact just bridge the pads and it will fix the throttling.

Below you can see L5500 resistor is gone and a pad missing, i ran a wire from the intact pad to the corresponding pin on the back of the temp connector and its working great.
View attachment 1794603

Thank you so much for this post! I have e a fourth iMac that has exactly this issue! Now I know how to fix it, thank you!!

Unfortunately in my other 3 iMacs with GTX 780M and low GPU performance the resistors look fine. I will get the schematic and measure those resistors, but I don't think that this is the problem in those machines as the Geekbench 5 CPU benchmark gives normal results in all three of them.

Any other ideas?

 

[The_Croupier]

Thank you so much for this post! I have e a fourth iMac that has exactly this issue! Now I know how to fix it, thank you!!

Unfortunately in my other 3 iMacs with GTX 780M and low GPU performance the resistors look fine. I will get the schematic and measure those resistors, but I don't think that this is the problem in those machines as the Geekbench 5 CPU benchmark gives normal results in all three of them.

Any other ideas?

I think the GPUs are fine, when running unique valley make sure your settings are as recommended in the guide post 1

Spoiler: About graphics card performance
1. OpenGL

OpenGL performance can be measured using the free Unigine Valley. Use these standard settings as resolution of 1920x1080, quality Medium, stereo 3D disabled, Anti-aliasing Off to compare results

Your benchmarks should now reflect results in line with others on this forum.

 

[RioRio]

I think the GPUs are fine, when running unique valley make sure your settings are as recommended in the guide post 1

View attachment 1795421
Your benchmarks should now reflect results in line with others on this forum.

OMG you are right! When I've read those directions I must have read erroneously "use the standard settings" instead of "use these standard settings". I never compared the resolution. I was convinced doing everything right by not changing anything and running the benchmark on default settings. 🤦‍♂️

Thank you for finding my error!! Now the results are fine! Thank you!!!!

 

[RioRio]

Check you haven't damaged the 2 resistors at L5500 or L5501 AMB TEMP.
If you have and the pads are intact just bridge the pads and it will fix the throttling.

Below you can see L5500 resistor is gone and a pad missing, i ran a wire from the intact pad to the corresponding pin on the back of the temp connector and its working great.
View attachment 1794603

Mission accomplished! I have finally another fully working machine with normal Geekbench scores!!

L5500 was missing on my board. Pads were still there.

What I've learned:

- Those are not resistors, they are very tiny coils. Type: FERR-220-OHM ; Size: 0402 (.04 x .02 inch)
- Most older MacBook Pros use these exact same coils!

I didn't want to order them somewhere online, cos I would probably have had to pay at least 20 times their value for shipping. So I desoldered one from a dead Macbook Pro board and soldered it to the iMac board. Works perfectly, as can be seen in the benchmark screenshot!

 

[RioRio]

Have the same problem, unfortunately the right solder pad of the upper L5500 (coil) is gone.
Which port of the ambient temp sensor do I have to connect to the wire from the remaining left solder pad?

You can scratch away the coating of that trace leading to the missing pad and solder a jumper wire to it, to rebuild the missing pad. Then you can replace L5500 with a FERR-220-OHM 0402 coil from a dead iMac or MacBook logicboard. Then your board will be as good as new. You only have to be careful not to melt the two connectors for CPU fan and ambient temp sensor if you're using hot air. You can protect those with kapton tape. Alternatively you can use SMD hot tweezers for the coil.
If you don't have the right equipment for this, you can send me the board and I will do it for you. I own a microscope and microsoldering equipment.

 

[RioRio]

Yes, I believe it's hardware. It tries to load High Sierra (in the old SSD) but I only get a blurry image of the desktop.

Also has the same issue:

View attachment 1794644

Replaced the 2N7002K mosfet with a new one and LVDS with a new one but when the solder they applied some green stuff on top...

View attachment 1794678

I'll come back with news later...

That green stuff is UV curable solder mask.

 

[Radical78]

Seems more about hardware than software to my naked eyes.
You can swap the K3000m with a stock GPU to isolate the issues.

By mistake I said that I had the same problem with my GPU's.

Made two tests again with my 4850 and K3000M:


My Nvidia K3000M:

Three Chimes, keep in the eternal loop.
Did the PRAM reset with the GPU, check every point in Hardware Issues in page 1.

It's possible that the card is DOA?

I installed back my Radeon HD 4850:

I flashed with the nikey22 K3000M ROM and didn't made any backup from my original card.

Can someone post the 4850 ROM so I can flash it back?

 

[Ausdauersportler]

By mistake I said that I had the same problem with my GPU's.

Made two tests again with my 4850 and K3000M:


My Nvidia K3000M:

View attachment 1795474

Three Chimes, keep in the eternal loop.
Did the PRAM reset with the GPU, check every point in Hardware Issues in page 1.

It's possible that the card is DOA?

I installed back my Radeon HD 4850:

View attachment 1795475

I flashed with the nikey22 K3000M ROM and didn't made any backup from my original card.

Can someone post the 4850 ROM so I can flash it back?

This looks a little bit like you have a K3000M with a different and currently not supported memory type, there is another K3000M version for download. Give it a try. Would be interesting to get a high res picture of the card to identify the memory type (always recommended to take those pictures before installing the card).

 

[Nguyen Duc Hieu]

By mistake I said that I had the same problem with my GPU's.

I flashed with the nikey22 K3000M ROM and didn't made any backup from my original card.

Can someone post the 4850 ROM so I can flash it back?

I think you can use the ROM from techpowerup

Apple HD 4850 VBIOS

512 MB GDDR3, 503 MHz GPU, 850 MHz Memory

www.techpowerup.com