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

yeah this is a good question!! an equivalent question could be why industry develop hydrogen car? the electrical car is more simple and efficient but industries have their reasons!!

 

Because this is not as convenient as plugging in a TB3/TB2 cable. And in most cases it will ground the laptop to the table.

 

Because Thunderbolt is external and rated to be used and reused. By comparison, MXM is rated for 16 insertions and removals if I recall correctly and requires removing the bottom of the laptop, meaning once you install this adapter you shouldn't remove it too often.

And, to be perfectly honest, PCIe x4 is enough bandwidth to run a videocard at full speed or nearly full speed. My goal with an MXM adapter is to convert an old laptop into effectively a power-efficient desktop, and because I think the whole idea is cool.

 

That's really good news, SourNinja, can you please a show device Manager pictures with a plug any nvidia/ATI card? And if possible, can you run 3d mark benchmark? And last question, when we see that adapter in mass production? Thanks!

 

Sompom, are you part of this project?

 

If you mean am I working with SourNinja, then no, I am not. Although our results look fairly similar, they were made independently, and theirs has the noticeable advantage of actually working
My last post was in answer to "Why are companies researching Thunderbolt when MXM already exists and could certainly be made to work"

(In answer to a potential follow up question: No, I don't work for Intel / Apple / etc. either. My day job is embedded software / studying to not fail every single class at the same time)

 

Actually, I didn't even think about asking about Apple and such. Not that I like them, but that issue is far more complicated

Anyway, your solution works too, no? I'm just looking for something that can be ordered from somewhere

 

Apple and Intel are relevant because they are the two biggest driving forces behind Thunderbolt, in case the connection wasn't clear

My solution "kinda works" -- Using the open-source Nouveau driver for Linux or the default Microsoft driver for Windows, neither of which really supports gaming (Nouveau gets closer every day, though...), I was able to boot and do basic web browser/y things. As soon as I used the official Nvidia driver on either Windows or Linux, I got only a black screen. I assume there is something wrong with my adapter since the problem was the same in both operating systems, but nothing significantly wrong. That means it could be that I wired all the pins correctly but the lengths need to be more closely matched, or the traces just need to have 45 degree turns rather than 90 degree.

I literally just slapped my prototype together without knowing anything about anything. I know more now than I did then about hardware design, but I still know very little. To be perfectly honest, I am astonished the thing works at all.

I am hoping @WhackingCheese completes his v2 "sometime" and that it works. If we don't hear from him, I will eventually come back to mine, but it isn't high on my list of projects right now, unfortunately

 

Hope you get it working. SourNinja's kickstater didnt quite pan out, and im not sure if they are also facing the same issues as you are. Maybe a collab will help ?

 

If @SourNinja wants to, I'm certainly willing to talk about possible collaboration. I don't know what form that would take, but it would probably be less work for me than fixing my currently-busted design

 

It will be the abomination that every one wants but does not deserve.

Jokes aside, i was sad to see how poor the community response was on the kick-starter, people are willing to drop $500 for a Razer Core, but not 1/4th the amount for something 10x better. Heck if you want a poster child and some one who is willing to experiment with you guys count me in.

 

People would pay, but I think the kickstarter campaign wasn't good
1. I bet people didn't know about it, like me. They could put some effort to press it up
2. The campaign page doesn't look that promising. They need some more refining
3. I guess that the description gives you the feel that currently they have problems they are not sure how to solve..

 

So... GTX 1080 SLI in P870DM anyone?

 

For several months I’ve been fighting NVidia’s optimus; at first I thought it was a software issue. However after looking at more information I found that the actual GPU and CPU are attached physically; this means I cannot record, stream or really do anything with the DGPU since the frames are dumped straight on the IGPU then to system RAM. So now I realized I was looking for a hard mod; this was the first result that came up.


So is the project still viable after the Kickstarter failed? There are hordes of people with these stupid optimus implementations that would love this!

Quite honestly I would be willing to pay 150 for it; this + an RX 480 would still be much cheaper than the desktop I’m planning.


(Yes I know I’m a new member but this is EXACTLY what I’m looking for)


Ps My laptop is a Sager NP8298 running a GTX 870M that supports the MXM 3.0 standard.

 

Anyone here know how to get ahold of SourNinja? Website etc. have no contact information and I would like to discuss this project further with him.

 

@IOC-000 your external display ports for your laptop are generally powered by your DGPU directly and don't go back through your IGPU, I'm just finishing up the MXM side of my prototype as we speak hoping to get the order in before the Chinese Lunar New Year's holidays.

Haven't seen SourNinja since the kickstarter ended I'm afraid

 

Hey Guys,

I've been through the thread and have a couple of questions:

1.) MXM to PCIe could theoretically be done with a simple riser and PCB provided 8 pin or 6 pin power is adequately available via a PEG connector to the graphics card in use correct (ie. the MXM slot can easily provide 75W to the GPU)?

2.) A relatively simple PCB either end could route all of the relevant connections from MXM 3.0 into PCIe 16x

3.) An MXM to PCIe riser cable could work in a plug and play fashion or definitely requires custom drivers?

Sorry to seem like a bit of a noob here, but this isn't exactly my area of specialty (I'm a physiotherapist by trade). That being said I am extremely interested in see this product come to market in light of some very exciting news regarding the MXM form factor that will be announced at CES 2017.

Cheers,

Lampshade

 

i saw the failed kickstarter page and i can see why - only ship to the US.

 

I should give someone who actually monitors this my email address so they can bother me when I miss the notification...

One thing to be sure you understand is that, typically, these eGPU solutions rely on, or at least can leverage, Optimus. In the case of using an external ATI card, this obviously doesn't apply...

The answer is yes, this is still viable. The kickstarter was not affiliated with the original project on here. If you know anything about electronics, you could go look back and try to refine my original design. Alternatively, ShadowsNight appears to have something rolling.

Have you considered a direct PCI passthrough to a Virtual Machine of your dGPU? I have experimented with this, and I think it would work -- Pass the dGPU in, set up the VM with some kind of remote desktop, delete the "normal" VM display device and try to connect to the VM via the local network. I have played with this some, to the point of seeing the PCI device in the VM, but I haven't pursued it further.

This depends on the laptop. I know the dGPU + iGPU on my laptop is as IOC-000 descrived, but there are some laptops where the outputs go directly to the dGPU as you described
(Interestingly, the low-end nVidia dGPUs are not capable of being attached to a monitor directly. The MUST be used in an Optimus setup where dGPU dumps frames into the iGPU's memory)

@ShadowsNight Any news?

I have never been able to provide a convincing answer to this question -- the obvious thinking is yes, if the MXM slot can power a 100W 980, it could provide the 75W for a PCIe card. The subtler answer is I'm not sure if the 100W 980 wants the same voltages, etc. It was much easier to just connect everything to the power already needed for the GPU.

Yes. That is essentially what my prototype does.

I don't know any reason why not. The answer is most likely yes because my prototype works with the open-source Linux drivers and the default Microsoft drivers. The answer is complicated because the Nvidia-provided driver didn't work in either case. Other eGPU solutions don't need custom drivers.

 

Update, not that it's one you want to hear.
I jumped ahead a bit and started working on the finished product rather than the prototype, in my defence I did need to fly out to Tasmania for a few weeks and wasn't going to delv into it while I was there.
Right now I'm comparing a photo of a 675m with the MXM file. Interestingly the PRSNT_R# is tied strait to Ground and Wake# isn't used. But I guess that means # is positive from the MXM board side. I am also re reading all my notes because it's been almost a month since I was processing this stuff. I'm going to set up a DropBox just for this later so people can see what I've got.

Edit: RSVD 159/161/163/165/167/239/241/245 are also used, back when I was looking for replacement cards I vaguely remember that I could only use 2 brands of MXM cards, Alienware and Clevo From memory, I'm wondering if these are important? On a motherboard side of things.

I did some snooping the 3.3v rail in the MXM is only 1A and the Pci-e is 3A
(Side note, for people looking into making something like this, I'll be using SFX or SFX-L PSU's in my end product, if I get it done....but they are cool)

 

MXM is not really a standard in that a lot of companies just do whatever they feel like and swap pins around as is convenient for them. This means that, for certain laptops, someone would have to design a custom MXM card in order to be able to attach to the PCI-e... There are a lot of problems here that just aren't explored. Step 1 is make it work with one laptop, step 2 is make it work with all laptops

 

Yeah further exploration and it appears like Alienware are using those reserve pins for Audio out while Clevo are using the OEM pins for it (Clevo have the right idea imho).

I'm actually almost done making the PCB's here. Last night I thought of a couple changes that I'm implementing but things are looking good

 

Ok, I've sent in my order, here are my files if anyone wants to try them out, I haven't done anything regarding length management, figured the traces are small enough it shouldn't matter too much. I also haven't really marked anything so you'll need to look at the tracing to pins (It's only really the clock pins and SMD pins that you'll need to look for really). I've done it all using FreePCB, if this set up so much as almost works I'll look into getting a more sophisticated program to measure out lengths/resistances/ect.
Also, if this works I might be happy to mock up some left overs for a few people, this is just a prototype, I have plans for a product, a pretty epic product imho.
https://www.dropbox.com/sh/db9dne9cdtdrt0p/AADqdVgI2Tx72nokbtnJZ5B4a?dl=0

 

I have a somewhat ridiculous request if you would be so kind to indulge me; would you be willing to create an MXM riser? By that I mean instead of ending in a female PCIE 16 slot it would end in a female MXM slot in other words MXM-A male to MXM B female.

Why would anyone want this? Testing. I for one want to testing how my laptop would work with an MXM-B card in it's MXM-A slot; if it can handle the extra power requirements and works well I'll shave of the optical drive connector that is in the way of fitting in an MXM-B card

 

hmm would be easy enough to make, but I don't know where you'd get a MXM female slot. The place where I got my PCB's made make a minimum of 10 per board and it'd cost $40USD for the MXM male side, the female side would probably be $10 plus postage then about $25 in cables and connectors (not including the female MXM connector), depends how much you want to spend really. I just had a quick look through the MXM specs and it doesn't specify a difference between the A and B power rail limits. It will just be your cooling that will need to make sure is rated for X watts because more power=more heat.

 

Oddly, unlike the rest of anything to do with MXM, the female connector is reasonably easy to buy:
http://www.mouser.com/ProductDetail...=sGAEpiMZZMtWxQ3D3EXBN9OrIGFtQGWfbEoJTEdwazg=

@ShadowsNight makes a good point, though. Figuring out how to attach a cooler to the card without it being in the proper place is likely to be a big challenge.
(Figuring out a cooler for an upgraded GPU is going to be an issue in any case because it always is... This just adds an additional hurdle)

Unfortunately, I don't have time to make such a card, but given the card blanks both ShadowsNight and I have already posted, it would be quite straightforward to just make a board with 1:1 pin mapping from inputs to outputs. I don't think he or I had any experience with the tools before we started doing this project. They're easy to learn, so don't let that stand in your way!

 

Yeah it only took a couple day's to figure out the basics, @Sompom was a huge help with the MXM side, I used their pins and boarder. If you look around you MIGHT get lucky and find a different model of laptop that has a cooler that is close enough to your old heat-sink and supports a B size card, (Take into account that there are different layouts for the cards as well, While a heat-sink for a 580 will work for a 675, it won't work with a 780. Here are 2 different MXM 3.0 B cards https://cdn.discordapp.com/attachments/161759056766173184/278053715347243008/unknown.png the memory on the right side shouldn't be an issue (most coolers just have a long strip going along them anyway), it's just the parts in the red that really needs the heat-sink to mach the GPU

 

Let me know what you would want for something like this. Also I'm not too concerned about the cooling during the testing phase as I have heatinks out the yinyang and if nothing is inherently compatible I'll drill new holes

I had noticed that the other day while looking into MXM stuff on direct from china sites

Definitely a good point I'll proably look into wheter or not i can get the top tdp support 17" Zbook cpu/gpu heat pipe setup to fit into the case of the 15" if I can't figure out a reasonable way to piggy back off the existing heatpipe combo

 

I'm not sure if $150USD plus shipping from Australia for an adaptor just for testing is worth it 0.o, plus the time to draw up the board and put it together :/ I mean... if your still interested you can PM me, but there are probably better people out there

 

Good Point lol

 

Damn. You guys got me all excited now. Any images or pictures you can post ??

 

I can post the parts I have, but the PCB's are still on the way I'll provably just wait and post when I make one up

 

Sweet!

 

soo cooked my test GPU with a faulty PSU :/ should have figured it was faulty, I pulled it out of a new computer that the mobo had died, then in a test computer a year back it fried the GPU in that.... but I've made 2 proto types so far, I haven't been able to test the second one due to not having a GPU compatible with it, I've tried it with a HD5670 and a HD5770 neither of them were recognised by bios or Windows, the 750ti sorta worked, bios saw it as a NVIDA GTX and I installed drivers and have a screenshot but that was with my first prototype, didn't get to test the second one which I should have sorted the issue I had with it

 

Is there any way to back feed the DP signal coming out of a eGPU back into eDP laptops to be displayed on the laptop screen ala most modern dGPUs?

 

Not unless the laptop has a mux that allows switching between the dGPU and the iGPU. Or is wired up for Optimus/MSHybrid.

 

Assuming that was the case would an eDP motherboard and screen be able to use the DP signal coming out of an eGPU?

 

The signal is converted from PCIe through the PCH to the iGPU that routes it to eDP, afaik. That is why external GPU have even more of an overhead (even more than when they output directly to an external monitor) when using the internal display as the output.

 

Other than certain manufacturers like Razer IIRC getting eGPU's to display to the internal monitor over Thunderbolt is it possible to accomplish this without their proprietary implementation?

 

There is no proprietary implementation about this. The overheads are because of signal conversions and need a mux/Optimus/MShybrid capabilities.
Any sort of PCIe signal for the internal display will need to go through the PCH. Only thing that depends is what is the source, since a dedicated PCIe signal will be faster than a Thunderbolt one.

 

Good to know, thx bloodhawk.

Just so I understand, an optimus supporting laptop can take the output of an nVidia eGPU and display it on the internal screen?

 

Yes, if it has optimus it should work, providing that the eGPU is new enough, Nvidia was 500+ from what I gather? I was using the laptops internal HDMI port for the test because the laptop won't boot without a monitor, I didn't get to test it with the GPU's HDMI before the faulty PSU fried the GPU :/ have a new one now, with a new PSU, just double checking all my wires before I test again, I noticed something odd with the graphics card that didn't match up with the PSU issue, not sure if it was broken before I got it though, the GPU was from a friend who had it lying around for ages and was going to throw it out.

EDIT: I've fixed it, it works windows recognizes it and HWinfo shows the clock speeds, haven't run benchmarks yet, will keep you posted

 

Hello, I am very interested in your progress and myself have plans to build an converter, but still a long way ahead of me...
@ ShadowsNight if you could take some pictures of your converter and share them?

 

I get that the following is kind of a divergence from this threads primary subject matter but it can only work with knowledge built up by this threads contributers and everything comes in phases and right now I'm in a very deep diveish phase lol

I have started thinking about the feasibility of getting a ribbon that goes from m.2(up to 4 lanes of pcie) and soldering the other end to an MXM female connector to allow for the use of an MXM-A type card in an Lenovo T540p as the next project to get into after I'm done with my HP Z-Book 15 which will leave me with an unused MXM-A k1100m that trounces the the 730m BGA of the lenovo

I'm thinking I could take the leftover Mxm-A Heatsink plate, solder/epoxy it to 2 small heatpipe heatsink fan combos I got from broken netbooks or 1 that I got from a t44p fit that over the k1100m and place it all in the 9.5mm DVD drive slot.

My only concern here is power delivery I doubt m.2 is designed to satisfy the power requirement of 55w TDP device so I was thinking I could augment that with the 5v of the ODD connector but I'm not sure how much wattage those are designed to do at max I know they can support laptop hard drives if in a conversion caddy but I haven't seen a laptop hard drive use more than 1A at 5v.

Why do any of this?
Because I want to I guess and the idea of rendering games with the k1100m and using the 730m to do Physx amuses me for some reason and the laptop and all mentioned nvidia GPUs suppport Optimus so why not?

 

@ChanceJackson you might need to look more into the total power for your laptop, if your power pack is only putting out 60-90 watts for eg. and then the laptop it self will have unknown rail limits, the MXM uses 12v,5v,3.3v. I'm not at home at the moment so I haven't looked at the laptop your talking about but it definitely sounds like fun I know nothing about the m.2 standard other than it's a monster paired with my gf's Samsung pro in her desktop XD

@remek002 I'm currently working on a product at the moment, but the files in my drop box work with a lil tweaking, all the info you need is in this forum I'll have pictures up later today, when I go home I'll be running more tests and even testing it on a different laptop if I get time. The one I've been using so far doesn't have a screen so I can't boot it without a monitor hooked up to the laptop, I haven't been able to see if I can boot off the card yet i just get the same 8 beeps from the laptop saying that there is monitor issue.

 

Thanks, Didn't realize mxm used 12v

 

running tests now 46 min no issues

you might be able to pull the 12v from somewhere else, or if your really keen you could hook something like this to your battery http://www.ebay.com.au/itm/DC-DC-Vo...106138?hash=item465b28b8da:g:IbAAAOSwImRYYLLb you might need 2 of those though, that one is only rated at 3A

 

I've done some more tweaking have it working better, here is my DX11 Heaven Benchmark 4.0 results.

I'm only getting PCI-e 1.1 speeds at the moment, I've looked into it and found this:
"The Gen3 specification requires a pre-validation of the link before any data transmission, and if the automatic equalization training cannot establish a reliable link, then it won’t allow the transmission of data at 8Gb/s speeds, resulting in slower than expected card-to-card as well as card-to-SBC data transfers within a typical system design." Quote taken from " https://www.trentonsystems.com/industry-applications/pci-express-interface"

My guess is that because of the way I've run the traces, and the the way I've patched up the cables it's not getting a good first impression and sticks to the PCI-e 1.x speeds. But, it is working I just need to save up for another prototype. I'm thinking of trying out a kickstarter after I get it all going smooth

Edit: Here are some 3Dmark scores:
http://www.3dmark.com/3dmv/5594298
http://www.3dmark.com/3dm06/17960254

I don't know why it's saying it's a Generic VGA card, the software reads it as a 750ti
But it also say's "Time measurement data not available. The validity of the result cannot be determined." Sort of unsure how to go about fixing that, I've read the FAQ and can't see how anything I've done could effect it.

Update: Finally figured out the new 3DMark software, needed to download 3.7gig at 500Kb/s though.... here is result here. http://www.3dmark.com/3dm/18496711

 

If work only PCI-E 1.1 speeds, I do not see much sense to create Converter, because this is the same as eGPU.
If it through the bad connection or interference, it would be good to check the oscilloscope.

 

I don't have an oscilloscope, nor do I know how to use one would love to be able to use one though, I've seen a few videos where people use them and think it's cool how much more information you can get from them. I'm like 95% sure it's just my prototype that is the issue, I don't even think the 2x20 connectors I'm using are gold plated, none of my PCI-e lanes are length matched, I'm currently using unearthed aluminum foil as shielding for the cables and have no shielding for the lanes on the PCB's XD

I'm not sure what link speeds the other eGPU's run at, I say that because apparently it's hard to get a 8gh/z link to a separate board. But even they are 8gh/z, 16 lanes at 2.5gh/z is still slightly better than 4 lanes at 8gh/z and double that of 4 lanes at 5gh/z.

I am sure I can get what I have to run at at least 5gh/z, I'm going to make some better cables today and see if that helps, might even try shielding every pair individually.

But right now, it's looking like I'll need to draw up a new PCB and maybe some better connectors to get the better speeds, I'll keep playing with what I have but I've got to move house soon so I'm not sure how much money I'll have for this project.

Also another note, I've been meaning to post pictures but my phones camera is to strong XD file sizes are too big so I'm just going to make a drop box folder with them in it.