Let's figure out how to convert internal MXM connector an external PCI-E x16 box

@god1729
Just out of curiosity, are you planning to work on this is ate you just spectators and for something to come out of this?

 

@WhackingCheese just a guy who's exited about tech for now.

Sent from my Micromax AQ4501 using Tapatalk

 

I'm most definitely still interested in this project. I have a machine with an MXM 2.x connector which differs from Sompom's MXM 3.0 design. I will hopefully be upgrading my notebook soon and this one (the one with the MXM 2.x) will be open for any modding since it will be decommissioned.

There were some really good points brought up as to why the project should continue and I agree with all of them. I was slightly discouraged when Nvidia released their statement about having desktop GPU's in notebooks (simply because I wanted to get a realization of this project working and in an engineering portfolio to present at a dreamed job interview), however; it shows that with the correct time, design, materials, etc, it is possible.

As far as the laptop CPU's go, yes and no. I have seen machines that utilize full desktop CPUs in a notebook application. The computer boutique company Xotic PC has a few machines that have both desktop CPU and GPU options available. Therefore, a desktop CPU is "possible", but I would imagine that the entire system is engineered from the ground up involving Xotic's engineers (I'm not sure if they have them) and/or notebook manufacturer's engineers.

I hope to contribute to this project even more because I am currently in a class detailing Computer Architecture and Organization (we study up to the electrical engineering point of hardware/software - the transistors and circuitry, I mean).

 

I am still very much interested in this, I've just been busy with other stuff recently I think there will still be demand for this, if your GPU dies and your laptop gas a gen 2 i processor or higher you can probably just take it out and use the integrated GPU for a working laptop, but if you still want to play games, it would be cheaper per $ to get a desktop card and be able to plug it in when your at home

Edit: I may be a little bias towards that sort of thinking because I have gone through 5 GPU's in my m18x XD

 

still waiting for any new developments.

 

OH MY GOSH!! So when I saw this last year I tried to make an account to log in to add my 2 cents, but found out that I already had an account and some how had been banned (mods have no idea why either XD) but it took like a month for my to log in because I'd lost the reply in my sea of emails, so I forgot why I wanted to log on other than to say I as interested! I woke up today remembering WHY!

If you still have this set up going, I know it's been a while, have your tried making sure the device ID is set right in the driver installation? I needed to do this when I installed my 675gtxm into my M18x
https://forums.geforce.com/default/...find-compatible-graphics-hardware-installer-/

 

Any updates? I'm willing to test this on my BIOS modded Clevo P150SM along a titan X XD

 

@Sompon hasn't been on this year :/ I'm hoping he jumps on soon, I don't have the money for a custom PCB atm so I've been trying to mod one of my old dead MXM cards to do the trick but there isn't much to solder onto with then PCI-e lanes :/ and if it doesn't work it's really not going to tell me if it does work or not because it could just be that the PCI'e lanes are too sensitive

 

I wonder what your mobile CPU thinks of possibly being hooked up to such a ravenous beast...


Subbed to the thread. I don't have the technical know-how to work on designs like this (more importantly I'm not sure how to acquire said know-how) but I'd love to help in any way I can.

 

You are underestimating stuff, im scoring higher than 6700k with this method too: http://hwbot.org/submission/2830783_0.0_cpu_frequency_core_i7_4700mq_4550_mhz

Even w/o OC, mobile haswell = desktop in die + and being close to a 4770k with turbo locked high, its not the 2000-2010 era people -_- , there is no reason for a mobile cpu not to equal desktop since haswell, all cpus have tdp of <100 watts

 

[DELETED]

 

https://www.techinferno.com/index.p...version-intel-haswell-cpu-microcode-bug-hack/

This makes it totally worth it now...

 

Rumours of my demise have been vastly overstated

I am still "watching" this thread, but apparently it's not sending me emails...
I guess I'll have to remember to check manually; I assumed it had just died!

@WhackingCheese -- I'm really sorry to hear your harddrive died That's the worst. The MSI I am using to test was decommissioned specifically because it had a SATA controller that literally eats harddives; I had three failures in two years. I feel your pain.
In particular, I'm sorry that you lost your super-length-matched solution. That is one of my thoughts for why my design does not work.

If I resume this project, which I hope to do, that is my plan. There is talk following this post discussing the worries of a laptop CPU being the rate-limit. While this could happen, my test-case is always a HP tm-2151nr with an i5-470um (dual-core 1.67GHz, 400MHz DDR3 controller) and a 560TI with PCI x1 link running Skyrim on ultra settings (with no mods). The CPU might eventually be the limiter, but my instinct is that point is some distance away.

As a point of interest, I read the other day that, officially, SLI GPUs require 8 PCI lanes each (While Crossfire requires 4 each). I'm not sure if this is the only requirement for SLI, but it seems to be an important one.

Regardless, I would like to get one GPU working, then we can start talking about features.

I still have all the bits if that's what you mean; they are currently 600 miles away because I study at an out-of-state university. I will be going home in the first or second week of May, so I could set it up again.

@ShadowsNight -- Does having the device ID set correctly just make the installer run, or does having it set incorrectly let the install go through but just malfunction? I was able to install the official NVidia drivers in both Windows and Ubuntu, but rebooting left me with a black display. I get the impression (from trying to remote-desktop in) that Windows had a resolution of 0x1080 or some such silliness. I didn't investigate Ubuntu as much as I should have.

 

Have you started using Circuit Maker? I was investigating KiCAD: http://kicad-pcb.org/ back in December, and it seems to be absolutely packed with features. You may want to have a look.

My complaint with Circuit Maker is it is windows-only, and my install of Windows barely runs these days... KiCAD is open-source and runs practically anywhere.

 

[QUOTE="Sompom, post: 10236482, member: 580099"
@ShadowsNight -- Does having the device ID set correctly just make the installer run, or does having it set incorrectly let the install go through but just malfunction? I was able to install the official NVidia drivers in both Windows and Ubuntu, but rebooting left me with a black display. I get the impression (from trying to remote-desktop in) that Windows had a resolution of 0x1080 or some such silliness. I didn't investigate Ubuntu as much as I should have.[/QUOTE]

It was about 4 years ago that I had the issue, and after I installed the drivers the first time I never needed to do it again before the laptop ate those cards as well, so I don't really remember. I have not been able to make my prototype, I de-soldered everything of one of my stuffed up 580's but there's virtually nothing to solder to and I've just decided it's not worth the solder breaking trying to test it and loose a laptop for testing, I have everything else for testing just not the MXM adapter. I live in AUS and custom circuit boards are more expensive here 0.o but I do have 2 laptops for testing if you think you've got it working, 1 with 2 mxm slots

 

Ah. Fair enough. I haven't read into it very deeply, but I think an ini hack just convinces the driver to install correctly, which was not a problem for me. Just that the driver (or something) appears to malfunction afterwards!

My prototype doesn't really work, unfortunately. I hope to have time "sometime" to make a new one, this time without using 90 degree corners in the traces and length-matching the PCI lanes. With how sloppy my first design is, it's a wonder it works at all

 

Hello all. Just to keep you all updated, I have been doing decently well in school this semester. As you may know, I am enrolled in a Computer Science course known as Computer Architecture and Organization. I have been looking through the chapters in our textbook and the chapters that remain after we finish (We finish up the course with chapter 3 and there are 8 chapters in total). Chapter 3 in particular goes into great detail of digital logic. Specifically section 4 examines CPU chips, and buses. A few examples including the Intel Core i7 are examined to explain the CPU part. Some examples of the buses include PCI and PCIe. These sections examine these examples in great detail and could prove very useful in our attempts to engineer this adapter. The title of the textbook is "Structured Computer Organization". It is authored by Andrew S. Tanenbaum. The specific edition that I have is the sixth edition. As for obtaining the textbook itself, that's up to you - I'm not getting banned for a link or explanations of questionable attainment methods. Just Google the book and you'll find some way of obtaining it.

As for updates for my hardware, I finally got a new machine to replace the older MSI laptop that I mentioned in page 28 or 29. I have a friend that sold me his older Asus Republic of Gamers G750GS. It has an Intel Core i7 Haswell architecture and an Nvidia GTX 870m GPU. It does in fact have an MXM 3.0 connector, so I may be able to experiment with Sompom and the others that have the PCBs. I have decided against rerouting the PCB traces/pinouts of Sompom's MXM 3.0 to MXM 2.0 to support my MSI machine. Instead, I will focus on troubleshooting the current design, and figuring out how to implement it with my idea for an all-in-one solution with my laptop.

One challenge of mine for the AIO solution will be the electrical engineering. I want to power the desktop GPU with an appropriate ATX supply (more than likely using a Pico PSU with appropriate wattage), but I also want to incorporate the Pico into the same case as the laptop charger if not inside the laptop case itself. If I do have it in the charger (of course I would re-engineer a custom enclosure and ensure that the charger electronics are physically separate (meaning a separate, shielded, enclosure) from the PSU electronics. Separation ensures no RF/EMI will occur). I will need to engineer some kind of way to have one cord from the AC outlet that will then split to the charger and the PSU. I will probably have the split occur inside the case to simplify cabling and only carry one AC cable. There will be another cable coming out of the charger "brick" that will be the PCIe power connector (if I do decide to do it this way). From there, I would bundle the charger cable and the PCIe power cables into a single cable (Remember, simplicity for storage and mobility), that has two connectors sticking out. To imagine what I envision, imagine a cable reel with two cables coming out. Another method I could use would be to somehow use the charger port to power the PSU (this would also require extensive engineering and understanding of the specific PSU components and circuitry). Which would you pursue if you were in my situation?

 

My first thought would be to not count any chickens before they hatch; we still don't have a working eGPU solution, so don't go crazy trying to make an enclosure yet!

If the laptop has an easily removable/replaceable AC jack board (It was a fad in laptops a few years ago, but I haven't seen it in awhile) you could remove that and engineer a little "vampire" that goes between the AC jack and the motherboard and has additional power outputs.

If the jack is soldered directly to the motherboard, note that it is (usually) fabulously difficult to de-solder. Don't let me stand in your way, but I would probably pursue a different path if it were soldered on.

Externally, you could do a similar "vampire" connection between the male AC plug of the power supply and the female of the laptop; just insert something that has additional wires for your use. Or, since power supplies are reasonably replaceable, just carefully slice the outer plastic open and attack the wires directly...

In any case, unless your laptop's power supply is vastly over-spec'ed or you can buy a higher-power one (which you often can, to be honest), you're not going to be able to power much of a GPU.

If you're lucky, your laptop runs on 12V and you can just clamp on a PCIe power connector and be done! It probably doesn't, though (even my netbook is 19V), so you're going to have to changed the voltage anyway, which is not easy with DC...

For all the difficulties suggested above, I would suggest you go with the Pico PSU (or any other clean source of 12V. Literally all you need is 12V and ground, just make sure you don't have a massive floating voltage between the laptop and the PSU!) and modify the laptop's power brick to hold the Pico PSU

 

I'm electrical newbie here, you mentioned that 90 degrees trace angle is a mess, guess length affects high speed data transmission, why wouldn't you go with zig-zagged overlapping traces (2 layers), you will achieve the same length (with varying zigzag radius depending on length), is it costy, for sure im missing something here XD

thanks

 

I don't understand the electrical/physics reason for why 90 degree corners are bad, but the hand-wavy reason is that the electrons "crash into" the side... Something like that, anyway. See idq000's post on page 33. Key point: they are bad. Avoid them

The idea with making the traces zig-zaggy and therefore the same length is basically the plan. However, a slight modification to your plan is that you should avoid vias (places where a trace changes layers). Again, I don't understand why, but I know that one's bad too.
Of course, if you use nice PCB layout software, it can do the length-matching for you. That's my plan

 

Huh, this is weird isn’t it? I stopped receiving emails about new posts and assumed noone was posting anything. Alright, heres a quick little update: The hard drive is propper dead, no data can be recovered. I have started working on a new design in CircuitMaker. It might be a free piece of software with some restrictions but it's made by the same people that made Altium Designer and has pretty much the same feature set with the only two real restrictions being that it has to be always online and only works on windows, neither of which are a problem to me. I will take a look at KiCad though. My current design is far along but not finihed yet. This years finals are coming up in 3 weeks and I will probably not finish anything until then but after that I can work on it in the afternoons after work and hopefully finish it within the next month. So if everything goes according to plan i will release gerber files before the end of may for you to take a look at and give feedback, but for now ima be off.

 

EE with specialization in HF design here. Main reason why this is bad is pretty mundane and has nothing to do with electrons crashing into the side or current crowding. It adds capacitance (on the order of a few pF per corner) which can cause significant impedance mismatch and reflections on the transmission line, especially problematic at HF. The waves on the line will sometimes cancel each other out and cause signal corruption.

 

Seems like what i was taught a couple of years back in school isn´t accurate. Not surprised to be honest. Thanks for giving an accurate depiction of what actually goes on.

 

This series ha been really helpful to me while learning PCB design for high speed design:

EDIT: Found this also for USB3 which also mentions differential pairing width design and angular traces: http://www.mobilewill.us/2016/04/routing-high-speed-signals-and-usb.html

 

Hello, Sompom! For MXM and PCI boards files in zip identity.I don't know, but is correct file for mxm board? Can you fix it?
Thanks.

 

I don't know the specific purpose of any of the individual files in the .zip :| . As I keep saying, I am as surprised as anybody that this design works at all. I see that the .GTO files in both zips is the same, but it just looks like the top silkscreen to me, but the top_silk in both zips is different... All I can say is when I sent in the .zip to be manufactured, I got back what I expected!

If you trying to modify it, the "source" files are here:
https://www.dropbox.com/s/i1wupbj8zu18bvc/FreePCB_Files.zip

I put these layouts together with FreePCB: http://freepcb.com/
FreePCB has the advantage of being free, but the disadvantage of not having very many features.

In case you are interested, KiCAD: http://kicad-pcb.org/
Is also free, appears to have lots more features, but is, therefore, more complicated...

 

Ok! I all understood. Thanks you very much for your job! It amazing!

 

@I Hunt Demons your probably better off putting the Pico in your laptop and have it replace the battery, given the effort given trying to modify your laptop power supply, all laptops I've seen use 17.5/19/19.5 volts which is not what you need, the reason for that voltage is to properly charge the battery, it's easier to step the voltage down than it is to step it up. You can get little DC/DC converters cheap but they are generally only rated at no more than 5 amps and the ones that are higher need extra cooling, you could solder onto the bottom of the plug itself if it's not a separate board, (I was going to make a N64 laptop but I lost interest in the project) but Cheese is right, we still don't have something working enough too start butchering working laptops.

@WhackingCheese I've had a lot of luck recovering data for people, if your interested in having me have a whack at it PM me

I'm currently trying to get my friends 3D printer to print a circuit but the only conductive "ink" they have isn't conductive enough, we're seeing if we can get something better

 

Guess this doesn't happen with round 90 degrees corners?
I have seen thermaltake p5 case builds with 25 cm flexible pcie riser 3.0@x16 running without a problem (shielded epoxy thin cables not cheapo IDE cables), and flexing the cable does not affect anything. So my understanding is: square corners = thick segment (if u convert it back straight unless its originally rounded like i mentioned, then it should have the same thickness) = capacitance = ~signal loss , my guess is that length is more likely to be it though in sompom's design, lol just my guess, corrections are appreciated.

Also, shouldn't the pcie trace max length from controller be considered (desktop pcie 3.0 is 20cm from header, donno about mxm connector to controller, as well as from connector) when specifying the length of the adapter cable.
I was also recommended to avoid thro-hole VIAs copper layer as it affects impedance when soldering temp is not controlled, ill try to make a very short 2 sided pcb without VIA (wires coming from bottom and top just like how regular pcie riser's pcb are printed). BGA connection start points to lane use blind type VIA for signal integrity maybe(each layer etched/drilled separately). Thro-hole is electroplated from what iv learned. Lol maybe i'm overkilling here

I'm using Clevo P150sm, and I have the ability to change pcie gen speed(prema bios). Most likely gen2 to be used with the mess lol

Edit: mxm converter is out of stock (for playing with first, since 1.2mm pcb is kinda expensive as sompom mentioned, especially for a beginner like me, yet ill give it a shot)

 

Hey guys, thought you might like to see this.
Someone has figured it out.

 

Hello,
hween is on the ball, we just put that video up!

We were contacted by 'Sompom' in the comments of the video.
"As you can see, I am also keenly interested in MXM eGPU technology. I wonder if you would be interested in having a chat with me? You are very cagey about the roadblock you've hit. Maybe we know something you don't! Please feel free to make an account and join the conversation on the forum, or message me directly. Best of luck! I would love to see this for sale in the near future!"

We could always use help, to be blunt our main problem right now is funding. We have setup a kickstarter page, it is still under review as we just launched today.

The main troubles we are having is data sheets for older MXM version and version such as ASUS MXM card.

Right now if you are interested in helping be sure to share our youtube video and check out our site where there is an option to donate or pre-order the product. Thank you for your interest and future help!
www.sourninja.com/dualwield/

 

News travels fast, apparently!

When I first joined this thread, I read every post in it... I vaguely remember someone having the MXM 2 datasheets. You may try searching for it. My memory occasionally plays tricks on me, so I don't suggest reading everything unless you have the desire to.

I don't think we have any data on non-standard MXM connections. Happily there are a lot of people on here who have laptops with standard connections.

I'll keep an eye out for your Kickstarter to go live. I have subscribed to your mailing list, so if you send a note out on there I'll be sure to see it!

 

WOW I totally forgot about this topic! Kudos for the great work everyone did!

@SourNinja - Search for the schematics of Acer 5920G (Quanta ZD1), it's MXM 2.1 Type-II. As you probably know, the connector and wiring is the same for every other MXM 2.1 (well minus Type-IV and HE, but the extra part is just power), so the 5920G should do. Same goes to the proprietary ASUS and Toshiba, but they actually are not proprietary at all, they just have odd shapes, the connectors and wiring are the same. OK they might be flipped, so it's best to check the schematics.

A couple of suggestions - the cable looks to be a PATA one, correct? If so, use coaxial wires between the laptop and the board the shielding would do miracles for the signal, or if the coaxial is too much of a hassle at the very least use twisted pairs. The second (a lot more work) - outputting to the internal display - route one of the display ports from the GPU to the DP-A, DP-B, DP-C or DP-D. I see it like a cable which can be attached to the DP of the GPU (obviously) and the other end to either of the four which are connectors on the board, so the user would plug it into A, B, C or D depending on where his/her internal panel is attached. Of course it would only work if the internal display is DP/eDP. Well, there's always the option to convert DP to LVDS, but this gets way too cumbersome, but also A LOT more appealing for users with older machines (actually not that old, since LVDS was used till 2014, maybe even 2015).

Keep up the great work!

 

hi, that is a great idea!! is it just on sale? is it compatible with a mxm 3.0b port of a clevo p751dm?

 

Yes the Dual Wield is compatible with a 3.0 port. Please visit our compatibility page for more information regarding your laptop: http://www.sourninja.com/dualwield/compatible.html

 

what about selling? kickstarter? delivery for italy is possible?

 

Sorry guys it double posted, (I know its a sin) Cant find where to delete the post

 

We do have a store for pre-orders, http://store.sourninja.com/sourninja-dual-wield.html
We have submitted for a kickstarter, requires 3 days to be approved we are on day 2. https://www.kickstarter.com/projects/1672091393/desktop-graphics-card-in-laptop-x16-speeds

Just about every modern country in the world has some sort of required electronics certification. Italy is part of the EU and does require a CE certification. The only modern country, to our knowledge, that runs on a 'buyer is responsible' basis is the United States.

To sum it up, we are only able to ship the USA right now. If anyone knows how to get a CE certification (or UL certification) please let us know. We will be looking into one eventually.

Sorry we didn't have better news for you!

 

Do keep us posted on the kick-starter, would love to back you guys up.

 

there are some infos in a page of us governent
http://www.export.gov/europeanunion/eustandardsandcertification/index.asp

here some informations by european government EU
http://ec.europa.eu/growth/single-market/ce-marking/index_en.htm
in particular you have to see the directive of european parliament 2014/35/UE for electronic systems with a voltage betwen 50/100V DC and 75/1500V AC.
http://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=CELEX:32014L0035&from=IT

 

Kickstarter just launched: https://www.kickstarter.com/projects/1672091393/desktop-graphics-card-in-laptop-x16-speeds

 

Donezo.

I think i popped another cherry.

 

Haha, Thanks for your support! Looks like your the first one.

Be sure to share it around, can't wait to get started.

 

Thank you for the information. We will be looking into this further!

 

ok. i obviously wait for a EU certification of your device!!

 

i'm talking about your device in an italian IT forum, and they say it is a good idea! but they ask if it is it plug and play? bios modding is created by you? if egpu, for example a nvidia m4000 with a tdp of 120w, is connected with your device to the laptop with mxm 3.0b that support 250w, a power supply is essential?

 

$78 USD seems super cheap. How are you able to do this so inexpensively?


Sent from a 128th Legion Stormtrooper 6P

 

Since its pretty much a pass through. With added power. Even if they sell it for a good price, they can make up for their amazing hard work.

 

Wow this is incredible! I'm glad to see a fully fledged solution that solves this problem truly come to life! I can do a bit of research to see what I can dig up about my ASUS machine. I read that @SourNinja is having trouble with those cards specifically. I have an ASUS ROG G750JS if that helps at all.

I would love a module to toy with if that's at all possible. I am totally willing to back the kickstarter, but I want one now to mess with.

I also have an MSI machine. I believe that it's a G1722 (uses MXM 2.x) or something like that. I should have it listed in my signature.

 

If this works why is the industry focused on pushing the much slower thunderbolt 3's pcie x4?