albertfu

For driver mod (DX10 cards), the idea is to rename all driver files and modify the INI file, so system could use a different set of file and registry for older card. Then you can install normally the new Catalyst for your DX11 card. The reason to modify Catalyst for legacy cards and not normal Catalyst for DX11 and newer cards, is that the former is less frequently updated. (13.9 was the last version if i am not wrong) Some functionnalities aren't working with this driver mod, like openCL, but for gaming it works fine. Another annoying thing is that in x64 Windows you have to force disable driver signature during boot, i didn't figure out how to make it permanent, so I was forced to press F8.

I have the same XPS 1645 laptop and did an eGPU with 7950 in 2013 It was a PCIE 1.1 x4 configuration, I used all mini PCIE+EC ports available + PE4H adapter. The tricky part is from the Catalyst driver, for the eGPU to work, the internal Radeon 4670 has to work first, so you can't just install the latest Catalyst since 4670 is no longer supported by it. So basically without driver mod you are stuck with Catalyst 12.4 or 12.6, the last version which supports both 4670 and 7950 If you have dx11 internal GPU (5730 for XPS 1645) ? You don't have to go through this trouble

The most time consuming part was to modify the driver and to make two versions of Catalyst coexist, I tried a dozen INF files and finally it worked out. I did it this way : Extract Catalyst 13.4 Legacy (latest version for legacy cards HD2000 - HD4000) Find the INF file appropriate to your system (Vista, 7, 8 / x86 or x64), in my case win7 x86 = CW156445.inf Extract and rename all file listed in the INF (in my case all file except difxapi.dll) to some thing like xxx_134L.xxx (134L for 13.4 Legacy). Don't know if it is necessary to extract them or not, but extracting can't do wrong so I did it. Delete all parts non necessary in the INF file (find your device ID and delete parts who are not related/used by this Device ID), doing so also makes sure that it will not be installed on other Radeon cards like eGPU Delete digital signature and catalog related lines in INF, delete MSI and CAT file Modify all file names mentioned in INF to xxx_134L.xxx In some REG lines (notably DX related parts), when it comes to a file name without extension, don't hesitate to add the extension, for example atiumdva >>> atiumdva_134L.dll Then for installation, make sure to clean all old Catalyst files in the system (\Windows, \system32, \system32\drivers, \system32\DriverStore\FileRepository, \WinSxS, etc.), you can manually do it and/or use the AMD clean up utility Manually install the driver, ignore the digital sign warning, restart <<< this may be skipped, but I didn't try Install 13.6 Beta2 or any version later than 12.4, restart After installation of new Catalyst you will find old card not working, so reinstall modded driver for legacy card, restart Then you will find your legacy and new AMD card using 2 sets of Catalyst files, in dxdiag, it shows 2 set of files for 2 cards, in CCC or GPU-Z, also 2 different versions of Catalyst On benchmark side : 13.6 Beta2 shows 33% improvement on DX11 benchmarks compared to 12.4, so it is clear that this solution is working. Dual screen (extend the desktop), hot plug eGPU (plug and unplug) are working, and everything was fine for me till now.

Finally I overcomed the Catalyst issue and have installed 13.4 Legacy (inf file rewritten) on HD4670 and 13.6 Beta2 (untouched) on HD7950 And I also got my PM3N yesterday, now I got a PCIE 1.1 x4 configuration, and here is a first benchmark GPU/VRAM OCed to 1220/1630 3DMark 11 X3302 is a better score than HD7970 GHz edition or GTX680 (of course not OCed) on a standard desktop PC, it is a big deal for a laptop bought in 2009, so I am just happy about it

Yet another weird issue : I just got my DVI-VGA adapter, so ditched my old LCD TV who has only 1366*768 panel resolution, but who displays at 1920*1080 on HDMI port of HD7950 (everything looks blur and the TV itself is hot) My D-Sub only monitor is an old yet Full HD one, now everything looks crytal clear on it. So both TV and PC monitor display at 1920*1080 resolution. I runned Cloud Gate and Fire Strike with PC monitor, it constantly shows a graphic score drop of 4.2-4.5% compared to the results obtained on TV (this time I am sure it's not an error), and switch back to TV solves the issue. So I am wondering this performance drop is related to interface (HDMI vs D-Sub), or to panel resolution (1366*768 vs Full HD) ? I tried to set a lower resolution on my PC monitor, it doesn't help, so I guess it is related to interface, which I couldn't understand...

Thanks for your clarification, an nVidia card will suffer more from lack of PCIE bandwidth, so I will try to stick to AMD cards. For now i did a driver mod, and GPU-Z and Catalyst Control Center are both reporting that HD4670 is using Catalyst 12.4, HD7950 using 13.6, I guess it is good for now... And in 3DMark it shows some minor improvement compared to 12.4 (around 3%, not sure it is error or really an improvement)

@Tech Inferno Fan : Is it possible to disable dGPU (the notebook has no iGPU) with Setup 1.x ？ I read in this article that is not possible for Win7 Video cards.Hybrid gfx Hoping that I was wrong.

Yes, both cards run simultaneously without issues, as long as the Catalyst driver is compatible with both of them. But the problem is that AMD no longer supports HD 4000 series in newer Catalyst, and I saw bug fixes and performance enhancements with these new versions. I tried to do a DSDT override before I figured out the real issue comes from driver. I installed Catalyst 13.6 and since the primary dGPU is not supported, I can't use eGPU neither. And after installing the right version of Catalyst, there is no need to do a DSDT override, plus, it is causing BSOD when eGPU is not attached, so I just undoned it. For number of PCIE lanes, with PCIE 1.1x1 (PE4H), 7950 already largely out performed 4670 (43M vs 5M transistors, 3DMark Cloud Gate GPU score : 34260 vs 5767). And I also ordered 3 PM3N to build a PCIE 1.1 x4 configuration, since my laptop can do that (XPS 1645). Can't wait to receive them to see performance improvements with higher bandwidth - - - Updated - - - I searched for modded Catalyst drivers, but they will replace core files with legacy ones (AMD updates in parallel Catalyst for legacy cards, latest version : 13.4 Legacy) Legacy core files works with 7950, but the performance drop compared to 12.4 is between 30%~40%, so I just abandoned this way of modding drivers. Maybe I can modify in an opposite way to force legacy card to use new card's core files. Worth a try

Hello, I finally got my eGPU up and running, but I have an issue with calalyst driver The notebook has a Radeon 4670 inside, and for the eGPU I got a Radeon 7950, the latest Catalyst who works for both is version 12.4. How could I disable dGPU (not possible according to here Video cards.Hybrid gfx since I don't have an iGPU) or install 2 versions of catalyst in 1 PC and make them coexist ? Thanks

Unfortunately it is a BGA CPU in E6330 so you can't change it easily

One week ago on BPlus's site there was no PCIEMM-060B cable and now it is! Finally a detachable solution for PCIE Gen2 eGPU!

Hello Nando4, I plan to build a @x4.1 eGPU configuration on Dell Studio XPS 1645 (PM55 + first gen i7), will it be impacted by half-duplex issue if I choose an AMD card like 7870 or 7950 ? If yes, is there a way to fix this half-duplex issue ? Or should I really choose nVIDIA cards (660 or 660Ti with same price tags as 7870 and 7950) to get better performance ? I saw AMD cards are more suitable for eGPU purposes as they are less sensitive to lower PCIE bandwidth here and in another post comparing 7970 and GTX680 (Ivy Bridge PCI-Express Scaling with HD 7970 and GTX 680 Review | techPowerUp). Thanks in advance

Thanks a lot for your very detailed reply ! A very sad story I found is that PE4H 2.4 can not support reliably PCIE 2.0, which has something to do with HDMI physical connection. So on newer platforms like Series 6-7 based laptops, PCIE slots must be restricted to PCIE 1.0 in order to use PE4H 2.4. So there is not really much reusability of PE4H 2.4 on newer laptops like M4700. I suppose that changing better HDMI cable could not solve the issue/negotiate a stable PCIE 2.0 connection neither. And for old platform like my Studio XPS 1645, which supports PCIE 1.0 x4 (3 Mini PCIE + 1 ExpressCard, numbered 1+2+3+4), the best and yet expensive solution (around 180-200 USD) for eGPU remains PE4H 2.4, who can deliver 8Gbps = 1GBps bandwidth. According to a study of scaling done on Radeon HD 5850 on x1 1.0, x1 2.0, x2 2.0, a x2 2.0 (x4 1.0) eGPU could deliver 90% of desktop GPU performance in 3DMark Vantage GPU score, which is great. Since the laptop itself only supports PCIE 1.0, PE4H 3.2 or PE4L who can only form x1 1.0 on it aren't comparable to PE4H 2.4 who can form x4 1.0. Questions here : My Studio XPS 1645 is equipped with Radeon HD 4670 GPU, is it an issue to put a nVIDIA card like GTX 660 Ti along with it ? Any possible driver conflicts ? The rest of main parts of system are : i7 920XM OCed to 2.4GHz~3.33GHz 4 cores, 4GHz single core, 8GB RAM, Crucial M4 256G SSD, will them be a strong bottle neck in gaming and 3D benchmarking ? A GTX 660 Ti is an overkill for it or not ? Especially due to weak heat dissipation, most of the time CPU operates at only 2.4GHz (passmark around 5200-5300) and it gets passmark around 6000 only under very strong ventilation. Should I trade it for newer laptops ? Comparing the price of PE4H 2.4 + PM3N *3, which will cost me 196$ excluding shipping and custom tax, to 230 EUR for a GTX 660 Ti card, PE4L seems to be much more acceptable (70$), and delivers x1 2.0 optimus connection on newer laptops. And just to mention : I don't move my laptop often so anything lighter than 3kg and with fullHD screen is welcomed. You talked a lot about nVIDIA's PCIE bandwidth compression technology (Optimus ?), I didn't get the prerequisites very clear. Does the laptop must have an Intel integrated GPU to make it work ? If I only have a discrete GPU, Optimus mode can not be activated so I can't benefit from PCIE compression ? I don't plan to use the internal screen of laptop for eGPU since it seems to consume more PCIE bandwidth. ------------------------------------------ I digged around a little too, and it seems that PE4H 3.2 and PE4L both only supports 1 PCIE 2.0 connection (soldered connection or PCIE 1x slot), and in 1 post of Tech Inferno Fan about TH05 thunderbolt adapter, he mentionned that x4 2.0 config will be limited to 10Gbps = 1.25GBps, only 25% more bandwidth than x2 2.0 or x4 1.0, instead of 16Gbps. So I wonder before the launch of Thunderbolt 20Gbps version in 2014, the highest bandwidth of an eGPU solution on the market is how much Gbps and what is this "most powerful" solution ?

Hello, I am very interested to add an eGPU to my old Studio XPS 1645, I already have an ATX PSU with 27A for 12V output and an external monitor, and I am considering a PE4H PCIE 1.0x4 configuration with GTX 660Ti. So I have some questions here : Is my PSU output sufficient for 150W TDP GTX 660Ti or even 200W TDP Radeon 7950 Cards ? Will I encounter issues with memory resource allocation ? I have 8GB RAM installed on it. Can PM3N adapter be easily cut to fit in half size Mini PCIE slot ? Can I replace cables of BPlus by longer Mini HDMI to Mini HDMI cables bought otherwhere ? (1m or 1.5m) Any recommandation on eGPU choice or any cautions to take ? And other questions about reusability of PE4H on other notebooks : Can Precision M4700 profit from PE4H to form a PCIE 2.0 *4 configuration ? It has 1 EC slot and 3 Mini PCIE slots. In general, can Series 6-7 chipsets based notebooks form a multi-lane eGPU connection ? (like PCIE 2.0 *2, *4) I saw in first post that for Series 6-7 chipsets that is not possible but later I saw some posts saying that it is possible but with some constraints, so I am confused here. Question about MXM slots : Is there a solution to externalize MXM port to create a PCIE 2.0 *16 connection for eGPU ? Thanks in advance