Toshiba romless MXM-II HD4650 DDR3 1 GB @ Acer 5920G soldering project for beginners

@triturbo: Thanks a lot for the initial information about such a plan!
@.NetRolller 3D: Thanks for your knowledge and great MXM support!
@Jacek of HQ Computers: Thanks for your patience and help!

This is my first thread and a need some help. I'd like to check, if it is possible to use a graphics card of a Toshiba notebook --> a romless HD4650 DDR3 1GB MXM-II <-- at my Acer 5920G. I am not an electrical engineering technician. That's why I need some beginners information. I also want to understand the process. Therefore, I have already read a lot things about the topic at the forum. - For the moment I have just a picture of the card and a seller, who would send me the hardware after the solution of the following tasks:

---

IMPORTANT UPDATE 2013: "valuxin" suggested a mod of the bios WITHOUT soldering a chip (thx to "triturbo" for the link):

My question: Has anybody successfully completed the procedure with a romless HD4650 (or other cards)? It would be interesting to know!

Be careful, modding your sBIOS could damage your notebook! No guaranties!
Needed files, system bios for the 5920G and video bios for the HD4650: Posting #60

---

1. Get pictures of the card. For a start I have only one picture and marked the spot for the EEPROM. DONE!


2. Send the pictures to the soldering pros namely .NetRolller 3D DONE!
3. Get the pictures back with marked spots for soldering DONE!
4. Get a soldering manual for the EEPROM and resistors DONE!
5. Get an adequate vBIOS for a HD4650, MXM-II, DDR3, 1GB card DONE! (Download: #60)
6. Ask an experienced friend to perform the soldering process DONE!
7. Buy an adequate EEPROM Got a flashed one by Jacek. DONE!

8. Buy the needed resistors DONE!
9. Get the reserved card DONE!
10. Do the soldering Plan B: Look at my post: It works! and take a look at the pictures below.

Unstable Construction V1 - Three classic 10 kOhm resistors next to the card

(Unfortunately, cracked V1, while removing the metal frame - RIP.)

Construction V2 - Backpacked Resistors @ M25P05A/10A @ Acer 5920G



11. Blind flash the vBIOS Not needed, cf. 7.!

12. Test it at my Acer 5920g, inform you about results Partially DONE!
13. Yes, you know the number, it is given for all unexpected problems! Wrecked V1!

14. COOLING THE SYSTEM: See triturbo's comprehensive tutorial: http://forum.notebookreview.com/not...tom-builds/569783-5920g-cooling-overhaul.html DONE!

I know I need a lot of help and there could be some hooks, let's go.
Thanks, ergoon

---

v v v v v v v v v v v v v v v v UPDATES SECTION v v v v v v v v v v v v v v v v

---

I am not responsible for any damages to your notebook's functionality by using the information provided here. Proceed at your own risk.

---

Overclocking HD4650 to the level of a HD4670
notebookcheck.net - HD4650, HD4670

Be careful, check the temperatures of the card, no guaranties! I've managed to change the clock frequencies of the GPU core and memory, using the RBE - Radeon BIOS editor (v1.28), loaded the Toshiba vBIOS (HD4650DDR3.ROM uploaded by .NetRolller 3D) and just used the "Clock tuning wizard" button with the settings: | GPU (MHz): 675 | RAM (MHz): 800 | Voltage (V): 1.100 |.

The Voltage should be 1.25 V as triturbo stated, but it isn't possible to change that setting to that value, so I used the highest possible value, saved the modified vBIOS and flashed it at DOS level with ATIFlash v3.99 ( ATIFLASH.EXE -f -p 0 MODBIOS.ROM ). I've checked the temperatures and stability using the tools GPUTool (CTP1), GPU-Z (0.6.6), Furmark (1.10.1) and MSI Afterburner (2.2.5). Additionally, I found a tool to check the video RAM of a card. There is a version for Windows and even a bootable image to test your card at a clean environment! I am always impressed to find such great tools at the net: Video Memory stress Test by Cherkes Mihail. - AND, found one more tool for checking your GPU: OCCT - Overclock Checking Tool by French developers with "Russian style".



At full load the temperatures are at ca. 85°C (normal use, e.g. writing this thread: 50°C) and the card works stable. Maybe, you don't even need to mod the vBIOS to overclock the card using this software tools. You just need to find a way to unlock the maximum settings for the card.

---

Construction V2 - Backpacked Resistors @ M25P05A/10A @ Acer 5920G
Due to the instability of my first design using wires and classic resistors, I've decided to rebuild the construction. This time I mounted three SMD resistors (10 kOhm, size 0805) at the top of a programmed M25P05A/10A EEPROM and connected the needed Pins (1-R-8, 2-R-8, 3-R-8) with conducting silver color. Click on the link for more information.

---

I have lost an SMD element while removing the metal frame of the card. Need to replace it, can't go on for now: Missing SMD element - Problem solved, but unfortunately damaged the version V1 permanently.

---

Thanks to Jacek for sending me better pictures of the card: One marker at the card tells SAMSUNG: I think, it is just the manufacturer of the RAM.



UPDATE: I've got a second card of the same type, but the RAM chips are by HYNIX now. Sorry, I'can't give you more info about the first card. It is wrecked for the moment. I am currently using the second card as "Construction V2".

---

Position of the Resistors at the HD4650: Thanks to .NetRolller 3D for the information and the picture! It is taken for a HD4570 card sharing the same layout with the other model. - Remember: It is the backside of the board. - I've added a scheme for performing the soldering directly at the EEPROM PINs:

---

.NetRolller 3D reuploaded the vBIOS file for the HD4650: The VBIOSes were extracted from the original Toshiba L500 SBIOS. The only Toshiba HD4650 in existence is the 1GB DDR3 version. It should work as .NetRolller 3D and triturbo comfirmed. The info in front of the files is mixed up, but the filename should be clear. <s>

</s> Unfortunately the new links are down too. Here is the link for the videoBIOS ROM file and also for the optional MXM corrected systemBIOS: #60

I was able to open the file (HD4650DDR3.ROM) using Radeon Bios Editor v1.28. It tells me some information about the bios, but I cannot rate this info. I.e. the amount of RAM isn't given.
Download TechPowerUp Radeon Bios Editor v1.28 | techPowerUp

---

EEPROM: I will use a M25P10-AVMN6 SOIC8 for the ATI card: The PINs are numbered as shown at the shematic picture. A chip should have a marker point or bevelled edge or side to identify the direction of the component.

---

Information about the needed parts and a vBIOS: The manual was originally posted for a Toshiba ATI HD4570 card, but should be suitable for the HD4650. Unfortunately the links for the vBIOS are down and I don't know, if the vBIOS is/was for the "HD4650 DDR3 1 GB" version. NEWS: The links were updated - see previous part of the update section - and the version confirmed by .NetRolller 3D.

http://forum.notebookreview.com/acer/462852-acer-5920g-gpu-upgrade-problem-3.html

---

Info by .NetRolller 3D, comparing different procedures:
http://forum.notebookreview.com/com...-broken-8600m-gt-graphics-card-i-fl90-17.html

A HD4650 flashed to a HD4670, interesting detail; - found some BIOS versions (also a MSI) for a HD4670 GDDR3 there: Sorry, the linked BIOS versions DON'T work with my card. I cannot get a screen during bootup the notebook.
<s>Acer.HD4670Mobility.1024: VGA Bios Collection: Acer HD 4670 Mobility 1024 MB | techPowerUp
ATI.HD4670Mobility.1024: VGA Bios Collection: ATI HD 4670 Mobility 1024 MB | techPowerUp
MSI.HD4670.1024: VGA Bios Collection: MSI HD 4670 1024 MB | techPowerUp</s>

Of course, .NetRolller 3D - the most valuable NBR user for this topic - has the correct answer.

Look at the overclocking info above.

---

Soldering manual for a 230M GT card:
http://forum.notebookreview.com/acer/575866-acer-mxm-bios-mods-discussion-x920g-vs-9600m-gt-53.html

---

Some information about soldering and alternatives:
http://forum.notebookreview.com/acer/659923-acer-aspire-8920g-upgrade.html

---

I use this BIOS at my notebook, because that could solve some preliminary problems:
http://forum.notebookreview.com/acer/575866-acer-mxm-bios-mods-discussion-x920g-vs-9600m-gt.html

---

List and experiences about Acer models and mxm graphic cards:
http://forum.notebookreview.com/acer/396320-acer-mxm-models-cards.html

---

 

The resistor pads are on the backside of the board, near the board mounting hole. They are not next to the chip.

DoZe's pic shows a GT 230M, which has a completely different PCB.

 

@.NetRolller 3D: Thanks for joining this thread! Ok, so I need to correct my statements. I did not expect to see you so fast. I am still editing/updating my initial post. I've just uploaded pictures of both sides of the HD4650. There is a zoom of the backside of the card. Where are the resistor places situated? - I could guess, but would be wrong again:

 

Here is a pic I sent earlier to OS-Store, to help him sell pre-modded 4570s (it's the same for the 4650 as well):

 

@.NetRolller 3D: Congrats to the 1000th post! - Thanks for your picture, you are invaluable! - I've updated the information at the initial post.

Do I have a chance to get a suitable vBIOS file for the "MXM-II HD4650 DDR3 1 GB card" which is working for my Acer 5920G? You've posted one, but the link is down and a was not sure about the card version. What do I need to check?

UPDATE: Found the answer at thread: http://forum.notebookreview.com/acer/462852-acer-5920g-gpu-upgrade-problem-3.html#post8898804

 

Yes, the VBIOSes were extracted from the original Toshiba L500 SBIOS. The only Toshiba HD4650 in existence is the 1GB DDR3 version.

 

Don't worry about RBE not showing the amount of RAM. It usually acts like this, it doesn't shows mine as well. About compatibility - it should work, if it doesn't, I have some HP HD4650 1GB DDR3 vBIOS laying around. In the past (HD3650 times), a lot of the boards had HP vBIOSes and they worked on Acers, but anyway, as I said, it should work, considering everything else is done properly My vBIOS is a MSI one, just FYI.

 

@triturbo: Thanks for joining! It would be nice if you can post the HP bios.
@.NetRolller 3D: Thanks for your confirmation of the vBIOS file!

@ Jacek: Thanks for your patience! :thumbsup:

I've decided to order the graphics card. - I will inform you next week about the status!

Please keep your fingers crossed.

 

@triturbo: MSI vs. Toshiba is not the only difference - your card has GDDR3 memory (basically DDR2 with extensions for high-speed graphics usage), while the Toshiba card uses JEDEC standard PC3-12800 DDR3 chips.

 

What I meant is, it doesn't matter that the vBIOS comes from different vendor. That's why I said that it should work, even though it's Toshiba. Anyway, looking forward to see how it's going to turn out

Here's the HP one.

View attachment HD4650.zip

 

I've downloaded the HP-BIOS and will pay attention to .NetRolller 3D's correction about the DDR type. Does this strictly mean, the HP bios should be incompatible? - Maybe, the experiment will answer the question. - I want to learn something more. I hope not to annoy you:

What is the difference between a ST Microelectronics M25P05-A and M25P10-A?
Just the memory size (512 Kbit respectively 1 Mbit)? Found something about it:
http://pdf1.alldatasheet.com/datasheet-pdf/view/22801/STMICROELECTRONICS/M25P05-A.html

What means the "A"? UPDATE: I've read at the description of the flash memory:
ENHANCED VERSION OF THE M25P05
This device is an enhanced version of the
M25P05. The enhanced features include: larger
page size, shorter programming time, higher clock
frequency, specific electronic signature.


Is that the answer? - Why it is important for my project to use an A model?

Will other models work, i.e. a Macronix MX25L512 or MX25L1005?
http://www.datasheetarchive.com/ind...27&keywords=MX25L512&database=user-highscore#

Is there one flash memory more compatible and should be prefered for the job?

Do the classic or SMD resistors need special specifications: design, 10 kOhm, tolerancy, power rating ... ?

 

That's more into NetRolller's field of knowledge, but I can't see why you don't want to use "M25P05-A or M25P10-A", since they are easy to find. I would understand if you couldn't find them anywhere, but it's not the case.

 

@ triturbo: You are right, it should be ok to use the chip, especially if NetRolller suggests that. - I've just asked to get some more information about the topic. - I also found some other pages who tell to use the other chip (with quite the same soldering manual), but I cannot decide, if there exist a reason to prefer one of the models:

http://peb.pl/sprzet/1205741-irytuj-ce-przegrzewanie-2.html

The other page:
http://bitboard.blogspot.de/2010/02/programming-mxm-graphics-module-bios-by.html

Finally, .NetRolller 3D also refers to the MX25L512's datasheet for soldering at a 230M GT board:

UPDATE: .NetRolller 3D has already answerd my question. I've just found the following post:

UPDATE2: I can only find a ST Microelectronics M25P05-AVMN6 at the internet. Is that the right chip?

UPDATE3: I was searching for an original ST chip, but some changes have been taken place since 1998: ...(1957) --> ST microelectronics (1998?) --> Numonyx (2007) --> Micron Technology (2010)

Read the specs, chapter "Ordering information":
Device type: M25P
Device function: 05-A = 512 Kbits (64 Kbit x8)
Operating voltage: V = VCC= 2.3 to 3.6 V
Package: MN = SO8 (150 mil width)
Temperature range: 6 = –40 to 85 °C

...and now?

 

The tutorials mentioning MX25Lxxx chips are for Nvidia cards. AMD cards need ST M25Pxx-A chips instead - or at least I'm not aware if AMD GPUs support MX25 ROMs.

And yes, M25P05-AVMN6 is the right chip AFAIK (anything that has an A after the dash, and has a SOIC8 package is good).

Why you need the "A" version? Because AMD GPUs don't support the original M25Pxx. If you pre-program an M25Pxx (non-A) in a SPI flash writer, the card will boot from it, but you won't be able to flash the VBIOS later.

 

Ok, fine, it is an compatibility issue and a "not A" verion cannot be flashed directly at the gpu. :thumbsup:

Would it be theoretical achievable to modify the 5920g Acer system BIOS (or any other system BIOS) in a way, so it would accept an unmodified Toshiba card by integrating the given Toshiba ROM information into the Acer system BIOS file or are the other hardware modifications needed? - Is it realistic to produce such a mod?

 

No, AFAIK the 5920G BIOS doesn't contain the facilities needed to support embedded VBIOS images. (However, this may be of interest to IFL90 owners!)

 

Hey ergoon, I've been looking at buying the card you're looking at on ebay as well When it comes through, let me know how it goes. I've read a few things about graphics cards bought from some parts of Asia being fake. I'm by no means an expert on computers, so I wouldn't be able to tell.

Sorry to hijack this thread, but I also have a question maybe for .NetRoller 3D to answer if he could be so kind; Do you think I could install this same board on my Aspire 6920G. I've flashed my BIOS with the mod you helped create in your other post. Would it be pretty much plug and play, or can you see a potential flaw? The HD4650 MXM-II 1GB DDR3 seems a lot better than the 9600M GT MXM-II 512MB DDR3 (Can't find any on sale with 1GB).
Edit: The HD4650 I'm looking at has a Toshiba laptop L500 Series EPROM. Will this matter with your updated bios?

Also, I have already undervolted (using flipfire's guide) to help cope with any potential overheating issues

 

@ .NetRolller 3D: Thanks! IFL90?? - Compal IFL90 User Review

@ Spearhead: Nice to see you. The Acer 6920G has also an MXM-II slot. I found some info at Acer 6920G MXM upgrade ,but at the main page they do not sell cards for this notebook anymore: "Following some issues we stopped offering the HD4670 for this machine."

There are some successful upgrades for this notebook described at http://forum.notebookreview.com/acer/396320-acer-mxm-models-cards.html - i.e. a HD4650, but DDR2: http://forum.notebookreview.com/acer/468070-6920g-9500m-gs-swapped-hd4650.html I think, the DDR3 model should be significantly faster...

I'll keep you informed about my progress, but I am an absolute beginner at this business, just look at my next question, to make that clear:

@ Everybody: Another day, another question: What specifications should the 10KOhm resistors have; - design/size, tolerancy, power rating? I visited my local electronics shop today and they ask me such important questions. Alternatively: What is the resistor type I should tell them? It feels like learning a new language. - All I know is that I know nothing.

 

Hey ergoon, mis-read your post at first. It is my understanding that while some illegitimate boards may still work, they may not do so at the same capacity of the manufacturers' build.

I too have read the 'Acer 6920G MXM upgrade' page and can see they have stopped offering the HD4670. They don't mention anything about the HD4650 though, so I don't know if I can safely assume it's still ok.
Also, I don't know how old that page is. I can see no date on it. I would have thought if it was more recent, they would have tried it with the DDR3 model.

Having read the '6920G 9500M GS swapped -> HD4650' link you sent, I think I'm going to give this a go. If you manage to get the one you ordered working and get some good benchmark results, I will be a lot happier taking the plunge.

Thanks for your help. At least you can use 'search' better than me.

 

The resistors should be SMD size 0402 US or 1005 metric, 10% or tighter tolerance, and the wattage doesn't matter (as it's only a strapping resistor).

Be prepared that it's VERY small. You will need a good soldering iron, and a steady hand.

 

The last sentence of .NetRolller 3D is quite a warning for me. I can't proceed with my equipment and my poor skills. I've found some (oversized) resistors. - The suggested 0402 resistor package type is even smaller.



Source: Surface Mount Resistors

Package type	Size in inches	Size in mm	Power rating
0201	0.024" × 0.012"	0.6 mm × 0.3 mm	1/20W
0402	0.04" × 0.02"	1.0 mm × 0.5 mm	1/32W 1/16W
0603	0.063" × 0.031"	1.6 mm × 0.8 mm	1/16W
0805	0.08" × 0.05"	2.0 mm × 1.25 mm	1/10W
1206	0.126" × 0.063"	3.2 mm × 1.6 mm	1/8W
1210	0.12" × 0.10"	3.2 mm × 2.6 mm	1/4W
2020	0.20" × 0.20"	5.08 mm × 5.08 mm	1/2W
2512	0.25" × 0.12"	6.35 mm × 3.0 mm	1W

That kills the project for the moment. Any suggestions about alternatives and new ideas would be welcome.

 

Well, you can just use my alternative suggestion of soldering regular (non-SMD) resistors to the pins of the ROM. It's a bit messier, but it should work.
If there is not enough space there for the resistors, you can try mounting the resistors and the ROM on a separate PCB (e.g. a small breadboard), and wiring that PCB to the place of the ROM on the card.

And yes, the 0402 is even smaller than the 0603. It is generally considered the smallest package that can still be hand-soldered (for 0201 and below, you need special tools).

 

The notebook boots and I can see an output at the internal Display!

I am totally confused, because it works: I shall tell you, what I did and you will see, it is crazy and not suggested to follow my way directly:

* I need to mention, that I use the graphics card and hard flashed EEPROM with the Toshiba BIOS-rom-file sent by Jacek. (More info: HqComputers Kontakt)

* I've mounted the EEPROM at the needed spot using two-component adhesive at its body.
* I adjusted the EEPROM in that way, that every PIN is exactly in position the connection at the board.
* I've put some of (silver) electrically conductive adhesive at every connection.
* I was measuring the resistance of every PIN after drying to check the connectivity.

* I've soldered two wires at the SMD resistors (0805), looks embarrassing, but the resistance is correct!
* I've twisted naked wire (just one strand) around the needed PINS of the EEPROM (1, 2, 3, and 8).
* I've connected the wired SMDs at the twisted wires, measured the result.
* I've secured it, using pieces of insulating tape, looks even more crazy.
* I've put the card into the Notebook, thinking, that can't work! ..........

Indeed, it works! I can see the boot screen and the system starts. As the cooling system doesn't fit for my Acer I stopped at this stage and powered off the notebook. I just wanted to see, if I am at the right way! I need to make a break there! I can't say anything about the situation at a booted OS, stability, workflow, drivers etc. The GPU is getting too hot very fast without adequate cooling. - My construction is pretty unstable, so I need to change that.

Ok, pros, if you like to mod your system. Let's go, but I am not responsibly for any damage at your notebook or graphics card...

A big !THANK YOU! again to the users .NetRolller 3D and triturbo and also to Jacek the hard and experienced worker in Poland.

 

WOW, CONGRATS! A picture would be nice Throw a copper shim and everything would be fine

 

:hi2: I have to tell, that there is a fifty fifty chance of sucessfully recognizing the card and boot. - I couldn't stop... I need to check for defective contacts, but I think it is normal for this beta construction. Here are three pictures for you:

 

I'd recommend replacing the silver adhesive with actual solder. Also, did you use 3 resistors as originally recommended, or just one?

As for the cooling, you will need a copper shim between the GPU and the cooler. Heatpads are useless for this purpose. And get rid of that metal-based (gold? copper?) thermal paste, as it may cause shorts on the card (metal-based TIMs are not recommended for notebooks). Use something like Shin-Etsu X-23 or Arctic MX-2 instead.

 

@ .NetRolller 3D: Yes, I've used all three resistors (at the picture of the card there was a 4th resistor, but, you have removed it with the info prevents updating), PIN connection and resistors, schematic:
__/-R-1
8---R-2
__\-R-3
I thought, they are all needed:



I think, I will replace the SMD resistors by classic 10KOhm resistors. Indeed, using real solder material should solve some problems, but I need a lot more soldering practice. I will make some tests at an old destroyed board, just to get a handling. I've seen some youtube videos about soldering SMD or other electronic components. No way at the moment...

Actually, it is not the best idea to start to learn soldering, using an MXM card, cause all parts are very small. For now, I've put back my old card - a 8600M GT - into the notebook. - I need it as working horse.

Ok, I need a copper shim and better paste to save the cards and notebooks life. Thanks, very useful hint and guess what, I was always thinking about the fact, why somebody is producing conducting heatsink paste. My answer was, metals have a very good thermal conduction; yes, but also conduction for electricity. - BTW: PCB were always polychlorinated biphenyls for me.

 

Weekend Impressions!
View attachment 86485

I needed to rebuild my 1st construction: I've replaced the wired-SMD-resistors by classic 10 kOhm resistors and rechecked the connection of all 8 PINs of the EEPROM and for the 4 wires connecting the resistors. You can see the crazy construction above! BTW: You can also see the boot screen with the card mounted. - I powered off the notebook at this stage. - Ok, that's the state at the moment.

Next step: Integrate the construction into the notebook and take care for the cooling system (use a non conduction paste for the card and copper plates for the GPU as suggested earlier). Maybe I need a second fan. I hope to do that next weekend... Be prepared!

It would be interesing to know: What would happen, if I had used a clean unflashed EEPROM? - I can't check that. My guess:

* Boot into a DOS-like system (by USB or partition or image) or maybe use a Linux-Live-CD/DVD
* Use a flashing tool and the Toshiba ROM file - BLINDLY and/or try to automatize that procedure.

 

With an empty EEPROM, yes, you would use a DOS boot disk (e.g. USB), with autoexec.bat configured to automatically force-flash the card. That's how I got my modded cards to work.

Metal-based TIMs (e.g. AS5 or that gold- or copper-based paste you are using) are great for desktop chips with an integrated heat spreader (though silicon/oxide-based pastes nowadays perform better even there). They are a potentially deadly choice for the "naked" chips found in notebooks; the service manual of the Toshiba machine from which these cards are taken explicitly warns against metal-based TIMs.

 

Did I lose an SMD resistor at my romless Toshiba HD4650 MXM-II card? It doesn't work anymore. Could someone please check if there should be a resistor at the marked position? Maybe I've lost it, while removing the metal frame at the backside of the card. It is near PIN 25 of the MXM board:

 

Do you have high res photos of your GPU? Sorry, but unless someone has photos of this exact GPU, there's no way to tell you if something's missing or not, based on different GPU. At least not without having it in person and testing it.

 

I have two new pictures of the card and the marked pic. Maybe someone has the answer:

View attachment 86543

I wrote an email to the seller of such fine cards. - That's my hope.

UPDATE: The resistor at the right side next to the missing(?) resistor has a resistance of 20.5 kOhm. Does this help? [...]

UPDATE2: Maybe, it is not a resistor missing, but a capacitor connected in parallel... Yes/No? BTW: If I measure the disconnected spots of the missing part, then I see also 20.5 kOhm resistance, same resistence also measuring at the 2 clear spots left to the missing part. Do I actually measure the resistor at the right side again and again...?

UPDATE3: I hope to get the answer soon and I hope the missing SMD element is a resistor and not a capacitor, but I need to be patient. Of course it is a problem to measure directly at the board:

http://www.ladyada.net/library/metertut/resistance.html
http://www.ehow.com/how_7537013_combine-parallel-resistors.html

ANSWER: The missing part is a resistor; - ca. 20.5 kOhm; - no guarantee! It seems to be connected in parallel. The total resistance at the local circuit is ca. 10 kOhm
(1/Rtot=1/R1+1/R2).

More info... next week!

 

Construction V2 - Backpacked Resistors @ M25P05A/10A @ Acer 5920G
Due to the instability of my first design using wires and classic resistors, I've decided to rebuild the construction. This time I mounted three SMD resistors (10 kOhm, size 0805) at the top of a programmed M25P05A/10A EEPROM and connected the needed Pins (1-R-8, 2-R-8, 3-R-8) with conducting silver color.

Currently I use the ATI drivers version 8.632.1.2000.- I think, Windows update automatically installed "ATI Technologies Inc. driver update for ATI Mobility Radeon HD 4650". Windows installed the drivers just while I was trying to install/check the "AMD Catalyst 12.11 Beta6 dotNET4" manually. - Now I am a bit confused, which routine installed the final driver version. Maybe, I will search for optimized drivers.



For cooling the system I added an old heatsink of an 3DFX card (BTW: a Guillemot MaxiGamer 3D Phoenix voodoo) spaced by a small pad to protect the SMDs to the back of the card and mounted a small cooler (5V) next to it. - The holes of the heatsink fit into the mounting holes of the MXM card. -The idle temperature of the card is now ca. 48°C and about 85°C at a FURMARK test. More checks needed.

 

Great work, nice to see it working again! The cooling is nice, I was thinking about the very same setup, and I even have the same heatsink. I didn't do it because - direct hit to the heatsink = hit to the GPU and it's quite possible to damage it. I have droped mine more than enough times

 

My construction V2 is working very well now for a week and the cooling is ok. :thumbsup:

---

@triturbo: It's a quite funny detail: You had the same ideas and even the same heatsink for your modding. I like such coincidences. It feels like "reinventing the wheel", but it tells me I have learned a lot things during the testing and maybe found the right way and access to the topic. I will be careful and try not to hit the system.

---

I just can't stop. Actually, I was thinking about a new experiment, but I want to be careful. .NetRolller 3D wrote, that it could be possible to flash a MSI BIOS for a HD4670 card at a different context:

I have such a Toshiba DDR3 card with 1024MB RAM. - Can I just flash the mentioned MSI vBIOS to the EEPROM? I have found some sources, I am not sure which of the BIOS files is correct:

UPDATE: The given BIOS versions don't work. I've just checked that. I can't see a screen while booting the system.

<s>Acer.HD4670Mobility.1024: VGA Bios Collection: Acer HD 4670 Mobility 1024 MB | techPowerUp
ATI.HD4670Mobility.1024: VGA Bios Collection: ATI HD 4670 Mobility 1024 MB | techPowerUp
MSI.HD4670.1024: VGA Bios Collection: MSI HD 4670 1024 MB | techPowerUp</s>

Should that work? Do I need to prepair something else? Too easy...?!?

---

Just for your information, I just wanted to test a flashing tool, I did not flash a HD4670 bios so far. Result: I've extracted the VBIOS from the chip with GPU-Z -- to check the identity -- and compared it with the original HD4650DDR3.ROM (uploaded by .NetRolller 3D ref. Post #5) The mounted M25P10A EEPROM was flashed by a programmer. That's why, I didn't need to blindflash the vBIOS to the chip.

The data are NOT identical. - There are some differences!
Example (ROM = original ROM file; CHP = extracted chip):
ROM - Line 2620: 04 00 80 FC 02 77 02 EB
CHP - Line 2620: 04 00 80 FC 02 02 02 02
...
ROM - Line 2960: 20 00 80 02 E0 01 48 00
CHP - Line 2960: 20 00 80 02 FF FF FF FF
...

The function of the card seems to be not not affected by the differences. - At least, I had no problems so far, but some parts seemed to be incorrectly. I checked it twice. - That's why, I've reflashed the ROM file using ATIWinflash 2.0.1.18 and now there are no differences left. I will recheck that regularly. - I had to use the -f (force) command line option for the flashing as the tool checks the BIOS version and tells, that the new and old BIOS are the same version.

---

 

Well, the MSI HD4670 has GDDR3 RAM (basically DDR2 enhanced for graphics operation), while the Toshiba cards come with JEDEC DDR3. So, simply flashing the MSI VBIOS may cause problems - it's better to mod the clocks in the Toshiba VBIOS to match those in the MSI one. (Radeon BIOS Editor can do that, I think.)

 

Those vBIOSes that you listed are probably for DDR2 GPUs, I know for sure that the MSI one is. The Acer one could be from MXM-A DDR3 card, but then it's different standard. I saw a post not long ago, in which was said that some MXM-HE nVIDIA has to be flashed with MXM-B vBIOS in order to work in different brand laptop. So it's not a definite no-go, but it's risky. You could edit your vBIOS, as .NetRolller 3D said, but I guess that you wont be able to increase the voltage to 1,2v, standard for MSI HD4670 GDDR3. If you can, then there is no point in flashing different vBIOS, just edit yours and that's it.

Edit: If you can't increase your voltage you can give a go to my vBIOS. As I said earlier RBE doesn't shows the amount vRAM (at least for our cards), which probably means that it is RAM independant, as far as they are the same type of course. DDR2 and the rest are very different - speed, timings and so on, but DDR3 and GDDR3 should be fairly close, packaging aside.

 

DDR3 and GDDR3 are not "fairly close" - they are completely unrelated kinds of memory. GDDR3 is closer to DDR2 than to DDR3.

 

The exact observer may have noticed already, that I own two Toshiba cards at the moment (guess why ), b ut they have different labels at the RAM chips!

@triturbo, @.NetRolller 3D and all other enthusiastic NBR users: Can you please check the information about the memory of my Toshiba HD4650 card(s)? GPU-Z tells, the RAM type is GDDR3, of course that information is NOT definite. - I made pictures of the RAM bodies to check the true types:

What is the correct answer to the RAM type now?
Is there a difference in the RAM type for the SAMSUNG and HYNIX model?
What does this mean for the plan to flash/mod the vBIOS?
The cards PCB layout is identical.

---

The first card (v1, broken resistor) has SAMSUNG memory chips:
| SAMSUNG 943 | K4W1G1646E-HC12 | YJI259GE | datasheetdir.com - K4W1G1646E-HC12

---

The second card (v2, working properly) has HYNIX chips:
| HYNIX | H5TQ1G63BFR | 12C 942A | NTJB4U22S1 | hynix.com - H5TQ1G63BFR

---

More interesting information: At the SAMSUNG product guide for Graphic Memory I found the labeling key for the different types:
alldatasheet.com - GRAPHIC MEMORY ORDERING INFORMATION - 2010

---

UPDATE: The "new" nomenclature for RAMS by SAMSUNG including type "K4B...":
samsung.com - PRODUCT SELECTION GUIDE - 2012 (ref. page 10)

It reads as follows for the K4W1G1646E-HC12 (added 2012 nomenclature in brackets):
01 - K: Samsung
02 - 4: DRAM
03 - W: gDDR3 SDRAM (sDDR3 sDRAM)
04 - 1: 1G, 8K/64ms
05 - G: 1G, 8K/64ms
06 - 1: x16
07 - 6: x16
08 - 4: 8Banks
09 - 6: SSTL_15 (1.5V,1.5V)
10 - E: 6th Gen.
11 - H: 84FBGA(Halogen Free & Lead Free)
12 - C: Commercial Normal
13 - 1: 1.25ns (1.6Gbps)
14 - 2: 1.25ns (1.6Gbps)

It would be interesting to find such a detailed information for the Hynix chips, but I didn't find an adequate source. UPDATE: Document with HYNIX nomenclature: skhynix.com - DDR3 SDRAM PART NUMBERING and an overview for the types DDR, DDR2, DDR3: skhynix.com - partnumberdecoder.

---

Interesting link, comparing SAMSUNG and HYNIX RAM types at Nvidia GeForce 210 and GeForce GT 220. Not the the same card, the label of the RAMS is a bit different, no ATI cards, not even MXM cards. - Just take a look: xbitlabs.com - GeForce 210 and GT220

---

I've just checked the different BIOS versions for a HD4670 card. They do NOT work with my HYNIX card! Sorry, can't tell you the answer for the SAMSUNG card.

The BIOS versions given by .NetRolller 3D (HD4650DDR3.ROM) and by triturbo (HP_HD4650.rom) work. The first BIOS is working with both cards (SAMSUNG and HYNIX memory), the second BIOS at least with the HYNIX version. I can't check that bios for the SAMSUNG card, because it is wrecked for the moment.

 

You are right. Both BIOS versions of the HD4650 can use 1.1 V as maximum voltage. - The BIOS editor says, the voltages of a card are limited as the connections are hard-wired. I've tried to use the maximum voltage of 1.1 V and 675 MHz GPU/800MHz RAM setting. The card boots, I can see a screen, windows loads, but then the card freezes. I use a dos boot image and created some batch files a.bat, b.bat, c.bat ... for the different bios versions. So, I can easily change the active bios. It's better to use the 8+3 filename convention for the rom files at dos level. The batch files load the flashing tool and I added some beep commands to get a notice of the state of the flashing, just for the case, the screen stays black after booting.

beep
REM ATIFLASH v3.99 EXAMPLE
ATIFLASH.EXE -f -p 0 HD4650.ROM
beep

I will search for a way to overclock the card after booting the system. I hope there is a tool around for ATI cards. I will also experiment a bit with the RBE tool, e.g. use the "clock tuning wizard" of the prg.

UPDATE: I've used the wizard to mod the bios, now it works. I have to test the stability and the temperatures a bit longer. Look at the stylish new GPU-Z skin, used just for fun. - Am I running a HD4650 at the level of a standard HD4670? - I just can't believe it.

 

Yes, Samsung's nomenclature is a bit confusing. "gDDR3" refers to DDR3 intended for graphics use - at the time this chip was released, JEDEC only approved the use of DDR3-800, DDR3-1066 and DDR3-1333 on DIMM modules, so Samsung had to mark higher clocks as "for graphics use only". In fact, AFA<!-- -->IK Samsung later re-released the exact same chip with the ID "K4B1G1646E" when JEDEC approved DDR3-1600 as standard.

EDIT: Googling Hynix's part number reveals that they suffered from the same restrictions - however, they kept the label H5TQ1G63BFR when JEDEC approved DDR3-1600, and simply changed their datasheets to indicate that this chip can now be used on DIMMs without violating the standard. (The old versions of Hynix's datasheets can still be found on Google.)

 

Nomenclature is all and nothing. I see! The "JEDEC" for electronics is as the "IUPAC" for chemistry: Lysergic Acid Diethylamide - PubChem - IUPAC name: (6aR,9R)-N,N-diethyl-7-methyl-6,6a,8,9-tetrahydro-4H-indolo[4,3-fg]quinoline-9-carboxamide

I think the DDR, GDDR, SDDR, 1, 2, 3, 'whatever' RAM labeling is very confusing. - Your correspondence to triturbo is just one example. If you search at the internet (e.g. search term "DDR3 vs. GDDR3"), there is just more confusion. Thanks for your explanation!

I was searching for the "K4B..." memory type at the internet, because the old "Graphic Memory Ordering Information" (2010) doesn't mention the type "B" only the "W". Though, the SAMSUNG page has a document called "Product Selection Guide" (2012, page 10 pp.). Interesting to compare:

2010 - W: gDDR3 SDRAM
2010 - B: not specified

2012 - W: sDDR3 sDRAM
2012 - B: DDR3 sDRAM

The HYNIX memory should have 'the same' specifications as the SAMSUNG then. Fine.

---

UPDATE: Finally found a nice tool to check the video RAM of a card. There is a version for Windows and even a bootable image to test your card at a clean environment! I am always impressed to find such great tools at the www. Here we go:

Video Memory stress Test by Cherkes Mihail

...AND, found one more tool for checking your GPU: OCCT - Overclock Checking Tool by French developers with "Russian style".

---

A funny detail about the linked pdf documents is the following: If one just reads the 2012 document section with the info about the RAM types, then you will read "W: SDDR3 SDRAM" (all capitals at the pdf view), but if you copy and paste the text into an editor, then you get "W: sDDR3 sDRAM" as answer. Ok, that is not the difference between "gDDR" and "GDDR", but also quite inconsequent:

 

Good to know - "SDDR3" is much less confusing. So, SDDR3 (aka "gDDR3", not to be confused with "GDDR3") is standard DDR3 running at non-standard speeds.

Another mystery solved with this document is that of "GDDR2" - there is no such memory standard, yet many cards are claimed to have been manufactured with it. In reality, this is the Samsung K4N series (e.g. K4N51163QZ) "SDDR2" (earlier "gDDR2"), which is just standard DDR2 with non-standard clocks. (If anyone has a graphics card - desktop or laptop - with "GDDR2" memory, please confirm that this is the chip you have!)

 

Updated my post about the SAMSUNG nomenclature. Guess what! - I've finally found a document with the nomenclature for the HYNIX memory: skhynix.com - DDR3 SDRAM PART NUMBERING and an overview for the different types DDR, DDR2, DDR3: skhynix.com - partnumberdecoder.

H5TQ1G63BFR-12C reads as follows:

01 - H: SK Hynix MEMORY
02 - 5: DRAM
03 - T: DDR3 SDRAM
04 - Q: VDD=1.5V & VDDQ=1.5V
05 - 1: 1Gb, 8K/64ms Refresh
06 - G: 1Gb, 8K/64ms Refresh
07 - 6: x16
08 - 3: 8 Banks
09 - B: 3rd Gen.
10 - F: FBGA SDP (Single Die Package)
11 - R: Lead Free & Halogen Free (ROHS compliant)
12 - 1: ??? not specified ???
13 - 2: ??? not specified ???
14 - C: Commercial Temp & Normal Power

 

I think the 12 in "12C" means 1.2ns between clock cycles, AKA 800MHz clock.

 

@.NetRolller 3D: You are completely right. It is the frequency of 800 MHz and the 1.2 ns are correct too compared to the "12" label for the SAMSUNG chips and their spec. of 1.25ns. Hynix is providing a special document for graphic card memory at their skhynix.com - partnumberdecoder page. If somebody is interested in the topic, just scroll a bit down to the "Graphics Memory" section. There are three documents: GDDR5 SDRAM, DDR3 SDRAM and GDDR3 SDRAM. Actually, they all refer to the same GDDR.pdf document). There is the "12" translated into: 800 Mhz.

I've - just for fun - tried to translate my older 8600GT RAM chips of HYNIX: HY5PS561621A FP25. Therefore, I managed to get the old specs. (e.g. "HY5" instead of "H5") skhynix.com - DDR_Old.pdf 2002

01 - H: HYNIX MEMORY
02 - Y: HYNIX MEMORY
03 - 5: DDR-II
04 - P: DDR-II
05 - S: VDD=1.8V & VDDQ=1.8V
06 - 5: 256M, 8K Ref.
07 - 6: 256M, 8K Ref.
08 - 1: x16
09 - 6: x16
10 - 2: 4Banks
11 - 1: SSTL_3
12 - A: 2nd Gen.
[...]

 

I'm pretty sure that clock and timing wise DDR3 and GDDR3, if not the same, are pretty close. Here's an old post of mine, where you can get the HD4670 vBIOS, give it a go if you want. The worst thing to happen is that you'll need to solder new EEPROM

 

@triturbo: Yes, of course, my friend, I've just flashed the linked vBIOS of your "M96XT.rom". The result is the same as for the other GDDR3 bios versions. I can flash, but after reboot I can't see a screen output. - So, I need to blindflash back to a working version! You just can't overrrule the physical requirements for the different RAM types. That's my guess. Maybe there are some additional problems towards the vBIOS.

BTW: Is there a way to tell the RBEditor the amount of RAM, if it is not included naturally at an unmodded vBIOS? Would there be an advantage to do that?

It seems, not a lot people are interested in this thread. I was reading the "Acer MXM BIOS Mods and discussion" thread. It is very interesting and informative to see, what companies do and how the experienced community is solving problems. I was also reading threads about the so called e. g."ERROR: MXM Structure not found or invalid AKA 30 seconds MXM bug" and the first workarounds. The final fix solves the problems and that is very important for many users.

 

The "MXM structure not found" error is specific to Nvidia cards. Note that the suggestions in that thread are not really the proper fix - the correct way of fixing it is crafting a new MXM-SIS structure for the laptop, and injecting it into the SBIOS. (This is what the 5520G and 9920G SBIOS mods do, for example.)

 

So no luck there, I guess you are limited to 1,1v, unless you find someone to increase the voltage on your vBIOS.