DIY eGPU experiences

Hallelujah! Thank you so much Nando! It works wonderfully! +rep

I have one more question for those who've had success with the DIY eGPU: is it possible to use the eGPU with a VGA-only monitor? I can hook up an HDTV using HDMI input and it works beautifully, but when I try using the GTX460's DVI port>VGA-DVI adapter>VGA input to my VGA-only monitor, it doesn't work. I looked at Device Manager, and my VGA-only monitor is apparently working properly and its location is "on NVIDIA GeForce GTX460" so the laptop is apparently reading it... anyone know the solution to this problem?

Or would I just need to use a VGA to HDMI Converter like this one?

 

I second this request as I just ordered a PE3A and the riser cable.
It looks like Nando just soldered in leads to the 12v here:
Imageshack - pe3amodded.jpg

I'm assuming that's power and ground (given on of the posts on the PCB is labeled GND). In the other modded PE3A I saw, the 12v pcb lane was drilled through to break the circuit there so the PSU could supply the power. But I don't see a clearly drilled hole in Nando's version. Can we get a better idea of how you did it Nando? This isn't rocket science, but I'd rather use a proven method than risk ruining $55.
I'm so excited to finally get the important bits to get this thing going. I've had my candidate laptop for months and I've since been acquiring things - gaming mouse, a nice laptop triple fan cooler, an awesome 1080p Asus LED monitor, and I even (successfully) made a flight sim head tracker out of old tv remote IR led's, BIC ballpoint pen tubes, and a 9v battery alongside a modified webcam (google Freetrack and DIY head tracking if you're interested) . Combine all that with Steam's insane holiday deals and my greatly expanded game library, and I am so ready to cease living in the real world. So hopefully BPlus still has those PE3A's.

 

So I would love to do this with my laptop (Dell Studio 1440 or 14z) but the initial post is full of technical jargon that I'm having trouble understanding. Hell, I don't even know which connector to purchase because the initial post doesn't say it in laymans terms.

I have a 26 pin ExpressCard port, so would that work me for PE4H-EC2C or PE4L-EC2C connector? Both say ExpressCard, but don't specify if there is a difference (though looking at the pictures I'm assuming its all dependent on the type of GPU you wish to use, but again there is no confirmation that I can work out of the post).

Also, I would prefer to use my laptop's internal display, but I can't find anything referencing what boards/chipsets are compatible with the program (nVidia Optimus) in layman's terms. The post is drowning in technical jargon that scares novice DIYs like me XD

Can anyone provide me some answers to my questions, so I'm less scared and more enthusiastic?

 

There is a 12V and GND post with a bit of plastic insulation over them. I simply soldered straight to those posts being sure the plastic insulator remained between them so there could be no possibility of the electricity sparking across the posts. After doing this my 2560P still wouldn't detect the eGPU so upon BPlus' instruction I dismantled the EC2C and remove the U3 and L2 ICs after which it detected OK. The dismantling is a delicate operation with the EC2C's bottom plate no longer fitting back on in it's original state.

 

So I'm guessing you pried the plates apart and then used a soldering iron to heat up the IC posts on the back of the PCB, and then pop them out? I'm making an assumption that the EC2C has a pcb inside those plates. Ah, now i see why you taped it back together. Nothin' like a little DIY handywork and a ghetto rigged gaming setup. Also, simply for assurance, you performed no drilling correct? So solder a power and ground lead to the posts from a Molex cable, which hooks up to a 12v line on the PSU. Remove said IC's. Patch EC2C together. Use riser flex cable to connect GPU to PE3A. Make power connections with 6 pin plugs (560ti). Bootup HP 6460B, standby with sleep button, power on PSU using paperclip (I have no ATX switch), plug EC2C into EC slot, resume system, configure card once connection has been made. Am I missing anything? Thanks Nando!

 

Just a heads up about trying to mod a PE3A. I tried it and it didn't work well. To get stable pci-e 1.1 operation I had to cut #PRST, and I would have to hotplug over and over again to get the graphics card detected since hotplug usually failed. PCI-E 2.0 was completely unstable.

You don't want to remove the posts, you need them to solder to. You can remove all the insulation if you want, but that makes it so that the posts can come out, giving you nothing to solder to unless you put the posts back in. It may not be necessary to remove the ICs. On my system there was no stability improvement after I removed the ICs and cut the 12V line.

 

AnandTech - MSI's GUS II: External GPU via Thunderbolt

Possibly not useful to us unless we use another PSU to power the GPU, but a great proof of concept!

 

@Slagatron Prime, yes, that's the exact process I did. Worth noting to steps required to get pci-e 2.0 working on a PE3A varied across different systems:

Lenovo X220: add 12V/GND (serialkllr)
HP 2560P: add 12V/GND and removing U3 + L2 ICs (nando4)
Dell E6520: no success when adding 12V/GND and removing U3 + L2 (Khenglish)

Since you have a 6460B it's likely going to need the same process used on the 2560P to work.

 

My thoughts exactly sir. I will let you guys know how it goes for the sake of adding knowledge to this impressive database . It won't be for a while though, as I'm guessing HK and Taiwan to Midwest US shipping transits are a bit lengthy, and I still have to purchase my GPU. I'm leaning towards an MSI Twinfrozr II or an EVGA 560ti of some sort, or possibly the Asus one as that looks good as well. My friend is cloning my system and already has the EVGA, so I will have results for that as well. The PSU I will be using is a CoolerMaster 460w with 18A each across two rails and I believe 320w on the 12v, so that should be good. Best of all it was free with rebate from Fry's on black friday. Anywho, good luck to all in their EGPU pursuits and thanks for the help!

 

An ultrabook with Thunderbolt is announced. It will be available in Q2-2012 for purchase.

Acer announces 'world's thinnest' 13.3-inch Aspire S5 Ultrabook at CES 2012 (update: pictures) -- Engadget

 

​

That PSU is going to be cutting it very close. Only one of the 12V rails will be going to the graphics card, leaving you with only 18A.

 

Somebody gotta help me! I bought all materials and I need but when I plug everything on it turns out that display adapters is standard VGA graphic adapter instead of HD 6850. I don't know if I did anything wrong. Please help me!

 

Hey Guys,

Anyone here have any luck getting a PE4H running on a Macbook Unibody (late 2008, MB470) on windows 7? Is the large PCIe memory window a hardware limitation or a OS X thing?

TIA

 

Please, anyone want to help me? read my latest post.

 

@Lord_Epithis.

Your eGPU GTX460 doesn't perform better than dGPU GT330M because eGPU is running only at PCIe x1 bandwidth. A lot of data is struggling to move between eGPU & notebook through the small PCIe bandwidth. Your notebook also doesn't have active Intel HD Graphics iGPU to utilized Optimus performance. That's why you can't change primary GPU in Nvidia control panel (because this feature for Optimus setup only). The only way to get better performance is by doing x2 setup. This will require two PCIe link/ports. These PCIe ports can be either two mPCIe ports OR ExpressCard+mPCIe, depending on the PCIe ports arrangement on your notebook. x2 only possible with either of these combinations:-
- PCIe port#1 + port#2
OR
- PCIe port#3 + port#4
OR
- PCIe port#5 + port#6

For example, if ExpressCard is PCIe port#4 & one mPCIe port is PCIe port#3, you can do x2 link setup using PE4H-EC2C-PM3N kit. Use AIDA64 to check the PCIe port number & also visual inspection to see whether the mPCIe port is physically available or not. To determine PCIe port number for mPCIe port(s), move WLAN card to that mPCIe port & check it in AIDA64.

What is the PSU specification? The max current on the +12V? If you don't know, post the exact brand & model of the PSU.

If you googled "ExpressCard", you'll find there are two type, 1) PCMCIA CardBus & 2) ExpressCard 34mm/54mm. The latter is what we need for eGPU. If you have the latter, you can use either PE4L-EC2C ¹ or PE4H-EC2C ¹. If not, then you can use PE4L-PM3N ² or PE4H-PM3N ².

Both PE4L & PE4H deliver the same job. There are no difference of performance between them. The only difference are:-

• PE4L is cheaper than PE4H.
• PE4L only have one PCIe x1 port, while PE4H have four PCIe x1 ports.
• PE4L is recommended for Optimus³ setup with Nvidia Fermi desktop graphic card. PE4H is recommended for x1/x2/x4⁴ setup.
• PE4H have sturdier construction than PE4L.

To be able to use internal LCD (acceleration on internal LCD), you need to do Optimus ³ setup using Nvidia Fermi desktop graphic card. If Optimus ³ not possible then you can do x2 ⁴ link setup to get equivalent performance.

<sup>1</sup>	EC2C: ExpressCard adapter/card
<sup>2</sup>	PM3N: mPCIe device
<sup>3</sup>	Optimus only possible on notebook with either Intel GMA 4500MHD or HD Graphics or HD 3000 iGPU
<sup>4</sup>	For notebook without the compatible iGPU^^, you can do x1 or x2 or x4 setup. Recommended to do x2 at least to get better performance. x2 only possible if you can access two PCIe ports, either two mPCIe ports OR ExpressCard+mPCIe. x2 only possible with either of these combinations:- - PCIe port#1 + port#2 OR - PCIe port#3 + port#4 OR - PCIe port#5 + port#6

I hope my explanation is layman enough for you.

 

Mr. Kizwan

See my problem from the previous 2 pages.

 

Try knocked it with a hammer.

Please post detail notebook specification & details on the kit you use (e.g. PE4H-EC2C, PE4L-EC2C, etc). Also post details specification of the PSU you use OR just post the brand & model of the PSU. Please refer to Main FAQ#1 on how to ask help properly.

It's BIOS/firmware thing. If you can get it to work in Windows 7, you should be able get it to work on Mac OS X. Of course, You'll need to get graphic card that natively support by Mac OS X to be able to use it on Mac OS X though.

If it suddenly stop working, possibly the mHDMI cable, used to connect eGPU & notebook, is faulty.

 

Even when the mHDMI cable is not connected, the gpu fan spins fast, this is abnormal.

 

The GPU fan should spins fast & loud when not connected to notebook. If it still spins fast when connected to notebook (in ON state), something wrong there.

 

I plug everything. But the EGPU did not show in device manager
________________
Laptop:sony VAIO VPCEB1SFX
OS: window 7 ultimate 64 bit
RAM: 4GB DDR3
iGPU:Intel HD Graphics
PSU: cooling master silencer 610W-12V/49A
eGPU:ATI radeon hd6850

 

Worlds first TB eGPU solution, previewed by MSI. Article seems to claim that the enclosed gpu can only be powered by the PCIe bus of the laptop, which seems to be insanity. Can anyone please dig up more details?

AnandTech - MSI's GUS II: External GPU via Thunderbolt

 

The two 12V rails can deliver 22A each, or 35A combined. It's a "be quiet! Pure Power 530W ATX 2.3 (L7-530W/BN106)". I think it should be more than powerful enough to drive the GPU.

I'm also trying to use the GPU on Linux for GPGPU like bitcoin mining and hash calculation, but so far not succesfully. While aticonfig sees my GPU,

Code:

aticonfig --lsa
* 0. 01:00.0 ATI Mobility Radeon HD 4650
  1. 04:00.0 ATI Radeon HD 5800 Series

* - Default adapter

pyopenCL does not:

Code:

>>> pyopencl.get_platforms()[0].get_devices()
[pyopencl.Device 'ATI RV730' at 0x11a8230, pyopencl.Device 'Intel(R) Core(TM) i7 CPU       Q 720  @ 1.60GHz' at 0x11bcde0]

(that's just my CPU and my internal GPU)

Seems like I'm missing the right driver or configuration.

 

I was thinking about the Thunderbolt Technology.
Did i read wrong or EACH TB port is able to provide only 20Gbps (10 + 10) instead of the erroneously 40Gbps sponsored (that are available only if you have both a lightridge/eagleridge controller + 2 ports available)?
Considering that notebooks will probably have the cheaper cactus ridge controller (able to handle one channel only) and for sure the 90% of those NB for some times will have one TB port only, we could easily compare each TB port to a "PCIe 2.0 2x", right?
In this way, considering that with the upcoming solution we can achieve about 80% of the performance with 1x (and even the newer HD7970 demonstrated that can't saturate the 16x 2.0 bandwidth), the remnant 20% will not be a great boost if those commercial solution will cost too much (and i think they have to due the competition with the gaming NB segment) So in the end, I think that even the Express Card/mPCIe solution will last for longer than expected) and will live happily together with the TB solution. Probably only who's interested in internal screen gaming will have the major benefit of a bigger bandwidth due the need of the "reservation bandwidth" for screen signal.

 

Actually, it happens randomly, but what's importantly, is that the gpu is not detected, no matter what method I try.

 

Well I already knew that, I was just asking how do I tell if my computer is using the egpu correctly.

 

You do certainly help in explaining what I need +1 to you sir.
Now I just need to find out where my mPCIe ports are and if they are available to use... more research and opening my laptop.

 

Did you connect the 6-pin PCIe power connector to 6850? Did you connect the floppy power connector to PE4H/PE4L? Did you connect the mini-HDMI cable to the first port on the PE4H board?

eGPU draw power only from first rail (+12V1) but the 22A (264W) already more than enough for 5870 (188W). Make sure you use D.N.A. drivers for both of your ATI cards.

On linux, only one GPU can be use. You need to set in xorg.conf file. You'll see something like this: BusID "PCI:7:0:0". Just need to change it to your eGPU.

Faulty mHDMI cable might explain the problem. I have no idea other than this.

eGPU already running correctly.

 

So to make this work I would need to do a fair bit of soldering and end up with a solution that might not stand up to frequent rough travels?

 

Thanks mate. While I understand that my egpu is only running at half of what the on board is, I'll wait till the pe4h ver3.0 comes out. So far the egpu is the only way I can use my external monitor for my laptop for some dumb reason.

 

Hm, that's a shame - if I install the D.N.A. drivers, I'll have to do the F8 switch everytime I boot Windows, right? (can't use the bootloader since it conflicts with grub)

Are you sure? When I use my 5870 in my other linux PC (which also has onboard graphics), I can plug my monitor into the mainboard and still use the 5870.

I don't want to use the 5870 to drive a monitor on linux, just for processing.

 

If I'm not mistaken your notebook have first gen Intel Core processor. This means your eGPU already running at full. The chipset on first gen Intel Core processor only support PCIe 2.0 2.5Gb/s, not 5Gb/s. Only Sandy Bridge notebook have chipset that can support PCIe 2.0 5Gb/s. Both first gen & SB chipsets are PCIe rev 2.0 but only SB support 5Gb/s. Even newer PE4H available later, you will not be able to utilized it.

If you have SB notebook, please ignore this post.

 

Try uninstall all ATI drivers then reboot. Press F8 & choose disable the driver signature. Install D.N.A. drivers & reboot (boot without F8). Let me know whether it produced any error or not.

I use D.N.A. drivers before & now using official driver from ATiAMD with modified INF file (to add support my dGPU ATI Mobility HD 4570). I only need to boot with F8 when I first installing the D.N.A. drivers. I don't remember what I did, if not I can share it with you.

Other workaround that I have tested it myself is:-

• For dGPU, install the official mobility driver using Device Manager. In Device Manager, right click at your dGPU, select Properties, go to Driver tab & click Update Driver.
• For eGPU, install the official desktop driver using setup file.

This worked for me.

I did test briefly a couple weeks ago. What kind of processing/software are you running? Are you running boinc?

 

I'm afraid I myself don't know the accurate numbers or the details of the upcoming implementation of TB/Light Peak ports on PC, but even 20% means a lot for gamers when it comes to just...a...few...frames...more.

On the other hand, I would really like this because this also greatly increases the amount of laptops compatible with *simple* and functional eGPU implementations.

Secondly, a lot of people, myself included, would also love to have more choice when it comes to choosing laptops with this capability. I often dabble in video production as well as 3d work and prefer to have just one machine that is 'modular' in nature. i.e. can hook up to eGPU for intensive work/gaming and unplug for portability. you would be surprised how hard it is to find compact, quad core rigs without niggling flaws that support expresscard. Or one ends up in mobile workstation territory pretty soon, with exorbitant sizes, prices and weight.

No doubt, initially these will be quite expensive, and people ofcourse will not want to upgrade that much for only 20% more frames at best, so yeah, I think current expresscard/mPCIE setups will continue for quite a while.

Still, a welcome addition overall.

 

Yes. I did all of them. External monitor is working. And my laptop detected hd6850 but it didn't show in device manager. And then I tried to solve it with PCI Reallocation. I free 256MB window by pci compaction but I don't know what to do to chainload. I mean I don't know how to use grub to finish it.

http://imageshack.us/photo/my-images/580/87840404.jpg/

 

So your saying that even with the new PE4H i'm still going to under preform? If so, then I have to go double rather than my single connection?

 

I did exactly that (plus pci compaction so it would recognize the card to begin with), and that was the error I got after rebooting (after installing the DNA driver): http://i.imgur.com/NbKkF.png
Tried installing the driver again, still same error message. ;(

 

I mostly agree with you, for sure doubling the bandwidth could give the performance of the today solution on the external screen but on the internal one plus some more (and that is a lot). My thoughts were only about the market positioning of these products:
having a modular machine that can match gaming rigs providing the same benefits and less penalties (you can't have heavy gaming on battery anyway, but if you won't you can have a lightweight notebok with good battery life) but with a minor expense, sounds like a suicide for a company like MSI that has gaming rigs in its portfolio. I think they will place them in the same range of high end configs (because it's uncontestable that a midrange desktop GPU with enough bandwidth can easily destroy 95% of the notebook cards today (GTX 560 Ti (180€) = GTX 580m (500€) ) so if they want to keep both solutions in the market, they will have to match their price that actually doesn't sound like a deal for a only 20% of more performance (always if my math is correct about thunderbolt)
Then there will always be solution like the Bplus one that should be more than affordable

 

Device manager show driver is not installed. You need to install the driver for HD6850.

BTW, TOLUD is 3.5GB & you have 4GB of RAM installed, how did you got to free 256MB PCI space? You should get Nvidia Fermi desktop graphic card instead. With Fermi you will be able to utilized Optimus performance.

Yes.

Use the workaround solution in my previous post.

 

Can't install driver.When it comes to install, the driver stop working "catalyst install manager has stopped working ". free e00000000 256Mb by 36-bit pci compaction but I don't know how to keep going to chainload.

 

Excuse my pestering, but my friend is buy a laptop and he is also interested in making an eGPU. So I am posting his question about this:

What features should my laptop have to run the eGPU at full efficiently? (such as optimus)

 

If you want to be able to use Optimus, then notebook must have active Intel HD Graphics/3000 iGPU. Optionally, must have ExpressCard slot. I personally recommend Nvidia Optimus powered notebook (switchable graphic IntelHD + Nvidia) because for sure you will not have problem with TOLUD.

 

Just need to choose Chainload in Setup 1.x. It will bring up Windows boot menu. Then just choose Windows to continue booting into Windows.

Please reduced the memory to 2GB & try install the driver again.

 

I did what you said. But when I choose chainload in setup 1.x. The windows just turns into black screen. And reduced the memory to 2GB is not working neither.

 

When with 2GB of RAM only, do not use PCI compaction. Just try boot into windows & install the driver.

 

Hello guys. I just receieved my HID 6870 and i am desperately trying to hook it up on my Sony Vaio VGN-AW31 M/H laptop. I have connected everything, the card powers up on the PE4H but that's all it does. My laptop won't detect anything! Any ideas??

 

Okay, so optimus is a must. Would a sandy bridge intelcore bump it up more or would that be overkill?

 

Hello, I intend to build an eGPU to game with. I've read through a couple of posts and think I understand what I need to do, but before I actually purchase any additional thing, I would like to know if I'm missing out anything or doing anything wrong.

I have the following:

1. Lenovo X220 - BIOS 1.24, QM67 chipset, i7-2620M, no dGPU, iGPU is Intel HD3000, Express Card 2.0 slot.

2. 26" External LCD with 1366x768 resolution.

3. Salvaged 305W ATX power supply (will upgrade if necessary)

I will be getting:

1. PE4H Ver 2.4 with EC2C and SWEX from hwtools

2. One of these GPUs:
Products :: Gainward GeForce® GTX 570 1280MB "Golden Sample" Goes Like Hell
MSI Global ? Graphics Card - N560GTX-Ti Twin Frozr II/OC
GIGABYTE - Graphics Card - NVIDIA - PCI Express Solution - GeForce 500 Series - GV-N560SO-1GI

I will be using a x1.Opt link.
No setup 1.x is reqiured.
To start up,

1. Connect eGPU to noteboook, power on eGPU
2. Power on notebook
3. Switch ExpressCard to Generation 1 in BIOS
4. Install relevant drivers for GPU
5. Restart notebook.
6. After which, eGPU will be plugnplay.

 

Still can not install the driver

 

We really need more information on what you've done. When do you power on the card relative to the laptop? Are you hotplugging, booting with the card, or doing sleep/resume? Do the fans slow down on hotplug/turning on the laptop, or do they stay at full speed? Where are you seeing that the card is not being detected?

 

Thank you for your answer. I have a 700w psu unit supplying the card. i tried connecting the card while it's powered on via the expresscard slot and it wont auto detect it. i also checked it in device manager and there's nothing new there. fans stay at full speed when i turn on the laptop or connect the expresscard while the laptop is on. but when i connect the card on the laptop the red light on the pe4h adapter turns into green.
If you need more info please ask