eGPU experiences [version 2.0]

[angerthosenear]

-snip-

My max resolution on my internal monitor is 1280 x 800 ...... soo I don't have much of anything to test but at that resolution. Any particular game I should try? Rather, which did you use in your test?

Also, my CPU is set to 2.7 GHz (turbo is off due to consistent heat sourced throttling causing less performance. - read: it throttles more and longer with turbo on).

My CPU is a hyperthreaded dual.

Any my ram is 1333 not 1600.

Whoops, gotta update my HDD in my signature: 7200RPM 1TB HGST.

[MikjoA]

-snip-

Ok thanks for the details, and pointing out mistakes, I'm going to correct them and re-organize my table.

For the testing I used Skyrim main menu, where I use to get a lot of FPS with the external monitor.

When I have a lot of FPS on the external monitor (like 150+) then it becomes very easy to determine my FPS cap on the internal LCD.

So this can certainly be done with any game that provides a lot of FPS in the main menu.

I'm going to add different resolution to the table to include 16/10 1280x800.

it also seems that the FPS cap is completly dependent on the resolution on my system, for example I get 44% more fps when there are 44% less pixel to display (1080p to 900p).

[naturbo2000]

Could you do some bandwidth testing with SiSoft Sandra? (the lite version works)

SiSoftware Zone

(^ english dl page for the lite version)

[ATTACH=CONFIG]9154[/ATTACH]

I have the 60cm version. I get 357MB/s bandwidth through to my eGPU.

365MB/s

[Andrey2790]

Could someone who has successfully gotten past error 12 please help me get past mine? I have the card already installed with drivers and everything, but it will not run due to error 12. I have tried doing the DSDT override but that leads to a BSOD with the ACPI error. If I try and boot with the card plugged in then I just get a black screen from the very start. I can post screenshots of any of my steps to get some input on where I am going wrong.

HP 4530s

i7-2670QM

8GB RAM

1TB HD

Windows 7 64bit

EVGA GTX 650

[Tech Inferno Fan]

<h2>How to measure your bandwidth ?</h1>

You'll find examples of bandwidth measured using CUDA-Z (NVidia) or PCIeSpeedTest (AMD) for x1 2.0 and x2 2.0 links using a HD7870 and GTX660Ti at http://forum.techinferno.com/diy-e-gpu-projects/3062-%5Bguide%5D-2012-13-mbp-gtx660ti-hd7870%40x2-2-th05.html#post42483 (Port & Speed table column).

[angerthosenear]

365MB/s

[ATTACH=CONFIG]9160[/ATTACH]

Very nice! What is your CPU? How many monitors do you have hooked up? (I have internal+1 connected to laptop and then 4 connected to eGPU)

Ok thanks for the details, and pointing out mistakes, I'm going to correct them and re-organize my table.

For the testing I used Skyrim main menu, where I use to get a lot of FPS with the external monitor.

When I have a lot of FPS on the external monitor (like 150+) then it becomes very easy to determine my FPS cap on the internal LCD.

So this can certainly be done with any game that provides a lot of FPS in the main menu.

I'm going to add different resolution to the table to include 16/10 1280x800.

it also seems that the FPS cap is completly dependent on the resolution on my system, for example I get 44% more fps when there are 44% less pixel to display (1080p to 900p).

I notice some games have insane fps no matter. When I play osu! I can get 1,300 - 1,600 fps at anytime (even ingame) - eGPU disconnected even. Not GPU dependent at all it seems.

Skyrim might be a good test since probably most people on the forums has it.

Could someone who has successfully gotten past error 12 please help me get past mine? I have the card already installed with drivers and everything, but it will not run due to error 12. I have tried doing the DSDT override but that leads to a BSOD with the ACPI error. If I try and boot with the card plugged in then I just get a black screen from the very start. I can post screenshots of any of my steps to get some input on where I am going wrong.

HP 4530s

i7-2670QM

8GB RAM

1TB HD

Windows 7 64bit

EVGA GTX 650

As you said you performed the DSDT override. Did you allocate a Large Memory block (as view-able in Task Manager)? If so, are you running 36-bit compaction?

Did you follow my iGPU+dGPU+eGPU guide perchance? You can pretty much follow that minus anything dGPU related as reference (unless you have a dGPU and want that working side by side as well - for PhysX or something).

Go ahead and let me know your current procedure.

[MikjoA]

You'll find examples of bandwidth measured using CUDA-Z (NVidia) or PCIeSpeedTest (AMD) for x1 2.0 and x2 2.0 links using a HD7870 and GTX660Ti at http://forum.techinferno.com/diy-e-gpu-projects/3062-%5Bguide%5D-2012-13-mbp-gtx660ti-hd7870%40x2-2-th05.html#post42483 (Port & Speed table column).

Thanks Nando, I'm going to add these values, but what do yo think should I look at ?

Pinned or Pagable ? What do they mean ? Comparing to your captures from CUDA-Z it seems SiSoftware Sandra reports the Pinned.

Device to Host seems very interesting btw !

I notice some games have insane fps no matter. When I play osu! I can get 1,300 - 1,600 fps at anytime (even ingame) - eGPU disconnected even. Not GPU dependent at all it seems.

Skyrim might be a good test since probably most people on the forums has it.

But you can't get these insane FPS when your eGPU is connected and when you only use your internal monitor for it, can you ?

Also, you need to performs this test on fullscreen as windowed can affect the performance. (51 to 47 for me on 1080p)

If with Osu! it does just like I imagine, then it could be an even better test since it is free & light, don't you think ?

[angerthosenear]

But you can't get these insane FPS when your eGPU is connected and when you only use your internal monitor for it, can you ?

Also, you need to performs this test on fullscreen as windowed can affect the performance. (51 to 47 for me on 1080p)

If with Osu! it does just like I imagine, then it could be an even better test since it is free & light, don't you think ?

With osu! it doesn't matter what is hooked up, I pretty much always get the same insane fps.

eGPU not connected on internal = eGPU not connected on external = eGPU connected on internal = eGPU connected on external

So seems solely CPU dependent.

--

I can't think of any free games that are GPU intensive. Will try to think of something. *scratches head*

[naturbo2000]

Very nice! What is your CPU? How many monitors do you have hooked up? (I have internal+1 connected to laptop and then 4 connected to eGPU)

As my sig. i5-2520M

No external monitors hooked up. I connected my old main-rig monitor up to the eGPU once when I first set it up, but haven't since. My laptop is now my main rig (hence getting a Thinkpad) and use it either solo, or with the eGPU driving the (upgraded) internal LCD for gaming. Test above was via internal LCD if it matters.

[angerthosenear]

As my sig. i5-2520M

No external monitors hooked up. I connected my old main-rig monitor up to the eGPU once when I first set it up, but haven't since. My laptop is now my main rig (hence getting a Thinkpad) and use it either solo, or with the eGPU driving the (upgraded) internal LCD for gaming. Test above was via internal LCD if it matters.

Ah, very nice. Seeing mine was a little lower (not much though), I guess the monitor(s) used doesn't affect bandwidth that much.

[Andrey2790]

I just tried and followed your guide again, the programs seemed to compile (but with 140 errors and 5 warnings) then the asl successfully overrode. The issue for me came from the reboot, after a reboot I would get a very fast BSOD just as the WIndows dots are spinning around. I got a video snapshot of the issue and its 0x000000A5 "The BIOS in this system is not fully ACPI compliant. Please contact your system vendor for an updated BIOS"

I thought I had the most recent BIOS, but I will now update it and see if that helps in any way.

http://i.imgur.com/Izlio2y.jpg

[angerthosenear]

I just tried and followed your guide again, the programs seemed to compile (but with 140 errors and 5 warnings) then the asl successfully overrode. The issue for me came from the reboot, after a reboot I would get a very fast BSOD just as the WIndows dots are spinning around. I got a video snapshot of the issue and its 0x000000A5 "The BIOS in this system is not fully ACPI compliant. Please contact your system vendor for an updated BIOS"

I thought I had the most recent BIOS, but I will now update it and see if that helps in any way.

http://i.imgur.com/Izlio2y.jpg

Is your SATA mode in your BIOS set to ACPI ?

Good luck with the BIOS update @_@

[Andrey2790]

I just did the BIOS update and tried your method again. Here is a screenshot step by step process, but it lead to the same BSOD error as before.

Photo Album - Imgur

[MikjoA]

Now I'm picturing you using your eGPU while on the train (and scaring the fellow passengers).

Without any experience with it, I'd say stick with the 60cm one. 100cm is a lot of extra wire to have information pass through. Plus, if 60cm works, I'd don't see why you would go for extra cable length.

I'm guessing there isn't much hope for finding your adapter again..

It seems the length of the PE4L doesn't influence the performance since the bandwidth was not lower for naturbot2000

[angerthosenear]

I'm not at home atm, but decided to see the bandwidth on my dGPU.

5.37 GB/s lol.

I'll run it again when I get home (since there are new driver versions / sandra was updated).

---

@MikjoA when you measured the fps for skyrim, did you have external monitor on or off? Or was your internal set as the main display (so it loaded on that screen), but your external display was still on.

It seems the length of the PE4L doesn't influence the performance since the bandwidth was not lower for naturbot2000

Seems so. Good to hear !

- - - Updated - - -

I just did the BIOS update and tried your method again. Here is a screenshot step by step process, but it lead to the same BSOD error as before.

Photo Album - Imgur

Are you able to add the QWord stuff then just save the DSDT file? Then load it?

I had the same issue with having a pile of errors. I may have circumvented the issue by doing that. I can't really remember tbh

[Andrey2790]

If I do not do anything right after adding QWord the compilation says that is it complete, but has a line above it that states "Maximum error count <200> exceeded." The AML file created is 0 KB so I do not think it would allow me any other way.

[angerthosenear]

If I do not do anything right after adding QWord the compilation says that is it complete, but has a line above it that states "Maximum error count <200> exceeded." The AML file created is 0 KB so I do not think it would allow me any other way.

Hmm, not sure. All I can say is fiddle with it for a while and hope something magical happens. That's how I got it working.

Sorry I can't really help more.

[Andrey2790]

It's alright, I think I will return the GTX 650 and get a 460, people seem to have fewer issues with it. If I understand correctly it is due to the 650s memory needs that I need the DSDT override, but with a 460 it should work normally.

[MikjoA]

I'm not at home atm, but decided to see the bandwidth on my dGPU.

[ATTACH=CONFIG]9161[/ATTACH]

5.37 GB/s lol.

I'll run it again when I get home (since there are new driver versions / sandra was updated).

Wow that's a huge difference but it makes sense !

It's nearly 15X more.

It's interesting to see how good the performance even with such a reduced bandwidth... I guess the driver are very well made too!

Btw, I've read a very interesting (french) article on GPU and their VRAM amount. (translated)

They've even tested their modded skyrim which consumed up to 2.5Gb of VRAM, and agaisnt my expectation they didn't suffer from FPS degradation at all (or almost) on their 2GB graphic card, but this seemed to be thanks to the very fast PCI-e gen3. This is not true at all with eGPU as far as I know. I had horrible lags with my 2GB GPU, and it became much much more smoother with a 4GB.

So this is an other issue which can be associated to eGPU I guess, due to limited bandwidth on the PCI-e port, exceeding in game the VRAM amount of the eGPU results in stutters and a significiant fps decrease.

Oh gosh, and I have advised more than 2Gb card to many people who were hesitating between 2 or 4gb video cards, although I always precised it could be usefull for skyrim only, while it doesn't make much difference.

But I wonder what it would be like with my insane skyrim which consumes over 4gb as soon as I enter a town (1080p) I guess it could affect the 2gb gpu.

my highest record btw VRAM + RAM was 8.5gb (more or less)

[angerthosenear]

Wow that's a huge difference but it makes sense !

It's nearly 15X more.

It's interesting to see how good the performance even with such a reduced bandwidth... I guess the driver are very well made too!

Btw, I've read a very interesting (french) article on GPU and their VRAM amount. (translated)

They've even tested their modded skyrim which consumed up to 2.5Gb of VRAM, and agaisnt my expectation they didn't suffer from FPS degradation at all (or almost) on their 2GB graphic card, but this seemed to be thanks to the very fast PCI-e gen3. This is not true at all with eGPU as far as I know. I had horrible lags with my 2GB GPU, and it became much much more smoother with a 4GB.

So this is an other issue which can be associated to eGPU I guess, due to limited bandwidth on the PCI-e port, exceeding in game the VRAM amount of the eGPU results in stutters and a significiant fps decrease.

Oh gosh, and I have advised more than 2Gb card to many people who were hesitating between 2 or 4gb video cards, although I always precised it could be usefull for skyrim only, while it doesn't make much difference.

But I wonder what it would be like with my insane skyrim which consumes over 4gb as soon as I enter a town

my highest record btw VRAM + RAM was 8.8Gb

5.55 GB of RAM and 3.25 GB of VRAM ._. ?

Yeah, I notice when I approach my cap of 2GB VRAM usage performance starts degrading rapidly.

Kindof makes sense though:

PCIe 3.0 x16 being about 15x better than PCIe 2.0 x1 (bandwidth wise). Glad to see we aren't only getting 1/15th of the performance though

[MikjoA]

5.55 GB of RAM and 3.25 GB of VRAM ._. ?

The other way round lol

Yeah, I notice when I approach my cap of 2GB VRAM usage performance starts degrading rapidly.

Kindof makes sense though:

PCIe 3.0 x16 being about 15x better than PCIe 2.0 x1 (bandwidth wise). Glad to see we aren't only getting 1/15th of the performance though

Yeah I was thinking the same lol, fortunately we got way much more performance than 1/15th, it's hard to understand that X16 2.0, in some games, allows some more fps than X16 1.1 which is still much faster than our poor X1.2opt

I bet 1.3opt wouldn't increase a lot the performance on the external LCD (not as much as X1.2Opt did compared to X1.1opt) but on internal it would be really enjoyable! especially for me with a 2million pixel internal monitor x__x.

[angerthosenear]

Quote

Spoiler

Hey guys, I was wondering something about nVidia GPU + opt & internal LCD performance.

I'm not much interested in your GPU this time, except that it has to be an nvidia. But rather about the FPS cap I believe we all encounter when playing on internal LCD. I believe a better CPU could potentially gives slightly better performance, as it seems a better CPU in general seemed to compress the bandwidth, done by optimus, more efficiently.

So what is your maximum FPS cap on the internal monitor depending on the resolution you set ?

User & his/her eGPU setup	Measured bandwidth	32bit	16bit
MikjoA Sony vaio SVZ1311 i7 3612QM@ 4x2.1Ghz4x2.8Ghz HT ON 2*4Gb DDR3 1600Mhz PE4L-PM060A 2.1b GTX 670 x1.2opt	not tested	1920x1080	1600x900	1366x768	1280x720	1920x1080	1600x900	1366x768	1280x720
51 FPS (measured)	73 FPS (measured)	100 FPS (calculated)	114 FPS (calculated)	/	/	/	/
1680x1050	1280x800	1680x1050	1280x800
/	/	/	/
angerthosenear Toshiba T901 i7 2620M 2x2.7Ghz HT ON dGPU NVS 4200M 2*8GB RAM DDR3 1333Mhz 60 GB Intel 520 series SSD 1TB 7200 RPM HGST PE4L-EC060A GTX 660 Ti x1.2.opt	364Mb/s	1680x1050	1280x800	1680x1050	1280x800
/	/	/	/
naturbo2000 Lenovo T420 i5-2520M 8GB DDR3 512GB OCZ Vertex 4 PE4L-EC200A 2.1b GTX 560 Ti (448) @ 849Mhz X1.2opt	365Mb/s	/	/	/	/

The only thing I can conclude so far, is that the length of the PE4L doesn't seem to affect the performance at all.

Redid test, this time with all updated stuffs , and as a bonus, nothing else running in background. [ATTACH=CONFIG]9162[/ATTACH]

364 MB/s

Perfect! I guess we can confirm cable length makes no difference on bandwidth speed.

[MikjoA]

Well then I might go for the PM100A, not sure yet

[angerthosenear]

@MikjoA

For your FPS cap testing, I cannot seem to find a game that has decently stable menu activity.

For Skyrim, depending on the number of little clouds in the bottom, that determined what was the stable fps. It was either ~65, ~75, or upwards of ~110 fps.

Can you think of anything that would be rather uniform?

---

Saints Row:The Third :: Has a stable main menu, however I cannot determine if it is a video in the background or something actually rendered. I get ~110 fps.

---

Borderlands 2 :: Pretty intensive game. I get a measly 12 fps at the main menu (since it is all rendered). Changing the video options doesn't change this. Essentially - low video settings fps = high video settings fps. I just think this game has a lot going on GPU wise no matter. Not sure. Maybe it is trying to do everything with the iGPU instead? At least with Skyrim I was able to select the GPU (launcher video options).

[jacobsson]

@MikjoA

For your FPS cap testing, I cannot seem to find a game that has decently stable menu activity.

For Skyrim, depending on the number of little clouds in the bottom, that determined what was the stable fps. It was either ~65, ~75, or upwards of ~110 fps.

Can you think of anything that would be rather uniform?

---

Saints Row:The Third :: Has a stable main menu, however I cannot determine if it is a video in the background or something actually rendered. I get ~110 fps.

---

Borderlands 2 :: Pretty intensive game. I get a measly 12 fps at the main menu (since it is all rendered). Changing the video options doesn't change this. Essentially - low video settings fps = high video settings fps. I just think this game has a lot going on GPU wise no matter. Not sure. Maybe it is trying to do everything with the iGPU instead? At least with Skyrim I was able to select the GPU (launcher video options).