DIY eGPU experiences

[bjorm]

thanks for answer.

I've been considering 2560p but I'm afraid of its size. There is no a big differences between E6320 but direction arrows scares me. At this moment there is no bargains for 2560p in poland. I found a 8460p as bargain with WWAN, 8GB RAM an on-site warranty with price lower than for 2560p without WWAN and on-site. I do not need HD+ and QM for this moment because of use external monitors (U2312HM).QM upgrade maybe possible after IVB entry on market (to reduce QM sandy bridge prices). I wouldn't be keen on to pay huge prices for IVB QM.

I used to have Vostro 3450 and his size was good, according to descriptions 8460p is not bigger than V3450. another reason is troubles eGPU with E6320.

[SimoxTav]

I was looking to replace the flat mHDMI cable of the PE4L 2.1 with a standard one (round) or another type of cable (to increase its length and make the curves easier). Any suggestions on what kind of cable should i use?

[fr0z3n]

One of the cables will be connected to ExpressCard slot. What you need is one PM3N card to do x2 setup.

BTW, did you try boot your notebook with eGPU already connected?

Without Intel HD Graphics you can't do Optimus setup with Nvidia Fermi graphic card. So, the only way to get better/good performance is by doing x2 setup.



@nando, I sent you PM. Please reply.

Thanks for the information.

I have ordered PM3N card.

What would be the fastest single eGPU supported with this setup?

Will the ATI Radeon HD7970 work and perform well?

[V410]

[QUOTE=nando4;8402038]The GTS450 will require a 120W PSU to drive it, up to 75W for the slot and the remainder for the single pci-e power connector. Suggest use an ATX PSU to get it up and running. Would be great to have a Sony TT working example.



Many thanks for your reply, i will give a try tonight.
Yes idd would have been great to see vaio tt with egpu. i believe it's very capable for a lappy. i have already tried with psu 500w atx, makes led 2 a little greener but doesn't solve the issue for detecting . also tried with gtx550 powered with the psu i end up same results. At this point i believe the graphic card is not communicating with mpcie port

[User Retired 2]

I was looking to replace the flat mHDMI cable of the PE4L 2.1 with a standard one (round) or another type of cable (to increase its length and make the curves easier). Any suggestions on what kind of cable should i use?

BPlus' rounded cable certainly was easier to handle and might work at pci-e 2.0 speed if soldered onto the ends since it was the mHDMI connectors that caused the problems. I'd anticipate you'd need to shorten the cable to get reliable transmission. Consider that at pci-e 1.0 speed, their maximum rounded cable length for reliable transmission was 2m. With the flat cable they were able to extend that out to 5m. Running the individual wires in a parallel fashion minimizing crosstalk.

[QUOTE=V410;8402984]

Yes idd would have been great to see vaio tt with egpu. i believe it's very capable for a lappy. i have already tried with psu 500w atx, makes led 2 a little greener but doesn't solve the issue for detecting . also tried with gtx550 powered with the psu i end up same results. At this point i believe the graphic card is not communicating with mpcie port

Doesn't the TT have an expresscard slot? If so, then would a better choice to use an EC2C.

If you are using a PM3N then boot with the wifi card in the slot into Win7, sleep the system, *carefully* power on the eGPU first, then hotplug the PM3N into the mPCIe slot, resume the system and see if it's detected. If it is then it indicates an issue with either whitelisting and/or the CLKRUN pci-e signal.

[kizwan]

Successful eGPU allocation to 36-bit space using DSDT override

I changed the range a little bit but it doesn't matter (see below), I think. (Edit) It does matter. My memory range is suitable for up to 12GB of RAM while the suggested range (see below) is suitable for up to 48GB of RAM.

QWordMemory (ResourceProducer, PosDecode, MinFixed, MaxFixed, Cacheable, ReadWrite,
    0x000000000000000,  // Granularity
    0x000000320000000,  // Range Minimum,  set it to 12.5GB
    0x000000F0FFFFFFF,  // Range Maximum,  set it to 60.25GB
    0x000000000000000,  // Translation Offset
    0x000000BF0000000,  // Length calculated by Range Max - Range Min.
    ,, , AddressRangeMemory, TypeStatic)

Modified DSDT: dsdt_Dell 1557_C8000000_36bit.aml

Boot procedure:-

1. Boot with eGPU turn OFF
2. Enter Setup 1.x OR halt windows loading by pressing F8
3. Turn ON eGPU
4. If in Setup 1.x, press F5 & make sure eGPU is detected
5. If in Setup 1.x, Chainload
6. Boot windows

I also tried the suggested range from nando4's DSDT Override page.

QWordMemory (ResourceProducer, PosDecode, MinFixed, MaxFixed, Cacheable, ReadWrite,
    0x000000000000000,  // Granularity
    0x000000C20000000,  // Range Minimum,  set it to 48.5GB
    0x000000E0FFFFFFF,  // Range Maximum,  set it to 56.25GB
    0x000000000000000,  // Translation Offset
    0x0000001F0000000,  // Length calculated by Range Max - Range Min.
    ,, , AddressRangeMemory, TypeStatic)

36-bit allocation also successful.

Modified DSDT: dsdt_Dell 1557_C8000000_36bit_C2.aml

[Blackett]

After spending quite some time on following latest egpu progress and watching the recent deceleration with pcie 2.0 extender kit (ofc I know there're a couple of functioning units on sale currently ), I ultimately decided not to wait any longer and to come up with another, equivalent solution.
Well, at the time I spotted this post, I knew it is possible to make an egpu on my own with my (limited) ordinary engineering skills and instruments, since pinout and wiring has never been an issue with all the required documents being published by pci sig and bplus (schematics) for a long time.

To cut a long story short ... Since Ivq Bridge and therefore the development and commercialization of the (most) promising (in terms of speed and egpu purpose) interface Thunderbolt still lag behind its (former) schedule, I've built my own adapter for presumedly not expresscard 2.0 complaint dell 1340 (although spec sheets contradict in this case), just to bridge the 3-4 months till computex and products announcements:




I'm still working on suitable case and and system's performance e.g. due to a successful established 2.0 connection (wondering whether it is possible with my current device at all )

[thaatz]

im trying to get chainloading to work on my setup. i have a macbook pro with windows 7 installed via bootcamp. when i use disk utility in mac os x, it tells me that my windows partition is on disk0 partition 4, so in in my menu.win, i changed it to say

chainloader (hd0,4)/bootmgr

but it just sends me to a grub4dos prompt (grub>). so i typed "ls (hd[tab]" and i think my windows partition was being seen as disk1 partition 3, so i went back to change that but I get the same problem (grub prompt). how do i know for sure which partition my windows install is on, and why does it throw me to a prompt. i am missing something. can someone help me chainload my system?
thanks.

so i was able to get past the grub prompt by typing "quit", but then it gives "error, chainboot failed"

when i open the grub prompt to check my hard drives by typing "ls (hd" and [tab], it tells me my available drives are "hd0 hd1 hd2". i have 2 hard drives in my laptop. when i type "ls (hd1" and tab, it shows me the partitions in my first hard drive. My windows partition (the only one partitioned as ntfs) is labled as partition 3, so my menu.win looks like this atm (the only thing in the "user edittable section" now)

chainloader (hd1,3)/bootmgr

however i still get a chainload error

i have changed the chainload settings to chainload win7 in setup 1.x, but im not sure what the problem is. any help would be appreciated. perhaps i am overlooking something.
thanks

[User Retired 2]

so i was able to get past the grub prompt by typing "quit", but then it gives "error, chainboot failed"

when i open the grub prompt to check my hard drives by typing "ls (hd" and [tab], it tells me my available drives are "hd0 hd1 hd2". i have 2 hard drives in my laptop. when i type "ls (hd1" and tab, it shows me the partitions in my first hard drive. My windows partition (the only one partitioned as ntfs) is labled as partition 3, so my menu.win looks like this atm (the only thing in the "user edittable section" now)

chainloader (hd1,3)/bootmgr

however i still get a chainload error

i have changed the chainload settings to chainload win7 in setup 1.x, but im not sure what the problem is. any help would be appreciated. perhaps i am overlooking something.
thanks

The disk image is mapped in as (hd0) and the first harddisk is (hd1). So the top part of menu.win7 remaps hd1 as hd0. Based on your indication that (hd1,3) is your ntfs partition, your \config\menu.win7 should have the RED line modifications below. You can even just do a test at the grub> prompt with 'chainloader (hd1,3)/bootmgr; boot' to see if it boots Win7.

Spoiler :

#############################################################################
# menu.win - boot using VISTA/W7 /bootmgr                                   #
# Use 'grub4dos.Prompt' menu, type 'ls (hd[tab] to see drive mapping        #
# DRV var passed by grub4dos: A=USB-FLOPPY(fd0), C=USB-HDD/DISKIMAGE(hd0)   #
#                                                                           #
# NOTE: Acer 3830TG will fail with a "vector probe" error. For those systems#
# comment out the MAP lines below and chainload off a specific partition.   #
#############################################################################
# Quiet operation
debug 0\ntimeout 0

# Appears need the following line for highest compatibility across machines
# If still see a A20-related chainloading halt then edit %DRV%:\fdconfig.sys
# and change the himemx.exe /METHOD parameter as described in config.sys
map --a20-keep-on=0

#::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
# DRV=boot drv. If it's C then swap real hdd(hd1) as hd0 and set harddrive 
# count=1 so the find doesn't use USB drive (clobbered USB sup)
# NOTE: Acer 3830TG users getting "vector probe" hang during chainloading 
#       will need to comment this out and chainload off a specific partition.
#::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
# Cater for USB-HDD boot/clobbered USB controller
echo .
echo ÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄ
echo If hangs with "Vector Probe" error then edit your config: Chainloader->!Edit
echo ÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄ
if "%DRV%"=="C" map (hd1) (hd0)
if "%DRV%"=="C" map --harddrives=1
if "%DRV%"=="C" map --rehook

#::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
# User edittable section below
#::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
#Use find to set root to location of Win7/Vista's /bootmgr then chainload it.
#find --set-root --ignore-floppies --ignore-cd /bootmgr
#chainloader /bootmgr

#OR chainload bootmgr off a specific partition, eg:hd0,1
chainloader (hd0,3)/bootmgr

[dav_jw]

I've been reading this thread since a long time, and thought I should contribute.

I recently sold my G73 and, thanks to this thread and Nando4's compilation of egpu-approved notebooks, I chose to purchase an Elitebook 8460p.

Today, I received my PE4L V2.1a kit, which I ordered on ebay (redtrontech) on 01/03/12. I read about issues regarding this particular model, but my sample did not have any problem (shipped on March 14th).

I used an antique LiteOn 300W PSU (19A on the 12V) and an almost-as-antique Geforce GT 430. ExpressCard Port set at x2 in BIOS, switches set accordingly with no delay. Notebook specs in sig, OS is Win7 Ultimate x64.

The installation was surprisingly simple and straight-forward; you wouldn't believe at first that those kind of things are plug-and-play, but it was (almost).

Basically, all I had to do was to connect everything, plug the card in the notebook with the computer booted-up, open the Device Manager, scan for hardware changes. The card was detected as a generic VGA adapter. Then, I installed the latest Nvidia (desktop) drivers (296.10), unplugged the card and rebooted. According to a post by danenick1212 describing his experience with the same notebook, it is necessary to plug the card just after the POST to have the card working with the integrated display, which is what I wanted. I did no further testing on the matter, so I cannot confirm if this is really necessary. With the drivers installed, the only thing left to do was to set the e-gpu as default in the Nvidia control panel.

I did not do any benchmarking yet, I will probably update this post in the next few days, I, however, tried to play Mass Effect 3. The results surpassed my expectations: around 40 FPS, with all graphic options enabled, 8xAF and 1600x900. On the internal LCD.

Edit: Ran some benchmarks, will update when I have time for more.

External screen, pcie 2.0
3dm06: 7488
RE5 : 57.4 FPS
DMC4 : 67.8 FPS
3dmV : 3611
Heaven : 318
3dm11 : 964



Internal screen, pcie 2.0





Edit: Upgraded to a GTX 460 1GB and a Corsair CX430. That card is impressive, from a price/performance standpoint. I did a quick benchmark on "The Witcher 2", with a light overclock. 43 fps average on a 25 minutes gameplay run. 1600x900 (internal screen), Ultra preset with ubsersampling and cinematic DOF disabled.

Some benchmarks:

All scores on the internal LCD, @ 849/1698/1920.

3dm06*: 16883
3dmV: 14033
3dm11: 3595

*1280x800 resolution used