damir001

This confirms a x1E link (port is set to x2 but link is running in x1 mode). Your method of attaching the eGPU by sleeping, powering on eGPU then resuming is causing the bios to reset the port status and revert back to x1 mode. Rather than use that method, do this to get the eGPU on the system in x1E mode:

1. set the PERST# delay switch to 15s, SW1=3 on your PE4H 2.4

2. power on the eGPU

3. power on your notebook. boot to Setup 1.1x and hit F5 until the eGPU is detected.

4. set your port to x1E mode, run your PCI compaction/startup.bat/etc so it chainloads to Win7.

5. Perform your RE5-var-dx9-1280x768 benchmark and PCISpeedTestv0.2.

6. Poweroff the systems, repeat steps 2-5 but in x1 mode.

7. Upload results for comparison. Does x1E give better results as it did with older HD5xxx cards?

I chuckled when i saw "explain it to me like i'm five" approach

Thank you both (khenglish and nando) so much for your patience and effort in this matter!!!

Short story:

I can report success with running 1xE mode but it was not so straight forward and nando explained and it took some guesswork as i'll explain bit later in post.

There is noticeable difference in speed between 1x and 1xE mode!

Here are requested benchmarks.

Long story:

x1E

1. Switched delay to 15 sec on PE4H 2.4 and powered on eGPU and notebook.

2. Setup 1.1x did not recognize eGPU even after extended period of time, i noticed port3 missing in system overview section of setup 1.1x

3. Manually enabled port3 in setup 1.1x and it appeared in system overview together with recognized eGPU, didn't have to do F5

4. Seting up 1xE mode on port3

5. Did Compact and "reinitialize/hot-reset ports" to get 256 MB in system overview...Compacting did not give me 256MB (at one point it even said 64MB)....reinit/reset did 256MB

6. Executed "startup.bat", i did not run this step before..ever..didn't know it was necessary, i'll look up in docs and thread what does it do.

7. Executed chainload to resume booting to windows

--- windows boots displaying on iGPU and eGPU is on but has black screen ---

8. Checking windows graphic properties gives just one single video output, "Fn+F4" does not switch to eGPU

9. Checking Device manager, seeing eGPU recognized OK but iGPU with error

10. Running setup of latest official drivers for iGPU from HP site

--- After setup is finished, eGPU powers on and on question to restart i click "No" so i end up with working iGPU and eGPU and device manager looks ok now

11. Run tests as requested for 1xE

12. Restart to do x1 tests

x1

Steps 1 to 7 same but without step 4 of seting up 1xE

--- windows boots ok displaying with eGPU as main display ---

8. Run tests as requested for 1x

9. Shutdown

Next step for me is to try to automate booting to windowsto 1xE mode. I will do my homework on this on threads and Setup 1.1x help but can you guys give me some pointers?

EDIT1: Found post from nando on automating startup.

EDIT2: Here are results for x1E mode in 3DMark

Cheers,

D

Fullscreen mode, disable the AMD audio, run RE5 var 1280x800 (or 768 if no 800). Your fixed results were a bit lower than expected given a T6600-2.2 with HD5750@x1E had a better result. Maybe that will fix it? If you've changed the port to be x2 mode in Setup 1.x prior to chainloading, that is all that is required. You could use AIDA-64, looking at the southbridge to confirm if it's still in x2 mode if really want to be doubley sure.

is this what i should be lookign at in AIDA-64? because it says x1?!

EDIT1:

This is how it looks like before going to sleep, powering on eGPU and powering on laptop back on.

EDIT2:

If i just power on eGPU while in windows this is how AIDA-64 looks like and this is how device manager looks like.

Put a hold on the additional testing. Something is very wrong with your setup. Your performance is less than 1/4th of what is expected of your hardware on x1.1 with the old half duplex problem. Are you dragging the RE5 window over to the internal LCD? If so please don't, since there are many more results to compare against when using an external monitor.

Also the full variable test is usually run.

All tests where run fullscreen (didn't see "window" option in RE5 benchmark option), not on docking bay and laptop lid closed. Getting screenshoot on dx10 results gave me blank screen in fullscreen and [PrintScrn] so i had to play with [Alt+TAB] and [PrintScrn] to get results.

I will re-check whole setup including disablement of AMD sound on eGPU and redo tests but i do have questions:

1. Are variable RE5 tests suppose to be run in full screen or window mode?

2. Do i need to verify in windows is x1E mode really engaged, if so how do i do it?

3. Regarding 1/4 expected performance, is there anything in device manager you suggest i re-check except disablement of AMD sound?

Cheers,

D

This is great news. Other test I'd ask you to do is run the Resident Evil 5 benchmark in x1 and x1E mode. It showed +20% improvement with x1E mode when using a HD5750. REF: x1E tweak. If the result is the same then yes, AMD have fixed the issue and HD7xxx is elevated in stature for older systems.

Here are tests for RE5, as before are same for x1 and x1E.

Rest of tests will follow...

Post your results like shown in the DX9, DX10 and DX11 tables on the first page, along with your system and eGPU details for comparison with and inclusion on that table.

Run pcispeedtest up to "4194304" is sufficient, so can CTRL+C out then. As explained previously, x1E is when your port3 is in x2 mode but only 1 lane connected. The screenshot in Setup 1.1x shows x1E was engaged. That pciespeedtest matches what I saw on my [email protected] on a Ivy Bridge here. Was that with x1E engaged or was it just x1? We require both to see if your HD 7770 performed differently for each as explained in the x1E tweak info. If they perform the same then AMD has fixed that handshaking bug on older chipsets.

Sorry, I was not clear with "ran PCISpeedTest few times but it did not show much of difference in longer tests.". I ran tests few times on both setup x1 and x1E that is why i didn't post second results, they are almost identical and there is no indication of mode in output. Only difference i could see was when i was clicking around in windows but they where quite unresponsive during tests, values dropped from 211227771.709277 to like 210xxxx.....

Cheers,

D

Thank you for posting your results.

Now your system probably has port3 as the expresscard slot making a great candidate to test if the x1E tweak still applies to the HD7xxx series. The way to test is do a run of PCISpeedTestV0.2 using x1 mode. Then reboot, apply Setup 1.1x's PCIe ports->Link Width.x2 or x1E to your port3 and redo the test.

Is there any performance difference between x1 and x1E mode? Previously we identified the x1E mode (running port1 or 3 in x2 mode but with only 1 lane) gave 15-30% better performance than x1. If they perform the same now then AMD have corrected a x1 link handshaking issue with that series which will mean the

First steps -> 2. Which video card do you recommend for best performance on my system? will be updated favoring AMD cards.

Attaching screen shoot for verification did I manage to get x1E more correct.

I ran PCISpeedTest few times but it did not show much of difference in longer tests. Im not sure was i suppose to let it run whole night but last one took a while so i stopped it every time when i say numbers got stable.

Here is output from one of files but as I mentioned numbers ran almost identical every time.

0 16 89968.259142 185073.791189

0 32 264436.646707 267712.923881

0 64 508931.263730 512307.656613

0 128 1173078.852779 1100926.643814

0 256 2045374.786510 2187707.159997

0 512 4138520.159608 4754876.985195

0 1024 10220309.650590 9662560.273504

0 2048 23288073.993202 26787975.878206

0 4096 42159968.375909 47693700.448241

0 8192 72036291.644763 57129406.961026

0 16384 103872245.503618 118089197.532234

0 32768 131568111.083229 159280685.367089

0 65536 154148072.070306 188248344.810719

0 131072 165666646.810459 199187185.772590

0 262144 172923254.749798 205094538.601962

0 524288 170026410.666416 208189452.025242

0 1048576 178261843.616446 209743127.171530

0 2097152 179206636.237480 210536159.496889

0 4194304 179684994.392547 210927962.755469

0 8388608 179934116.429364 211133051.296862

0 16777216 180115474.058833 211231469.309891

0 33554432 180225310.337301 211227771.709277

0 67108864 180255012.653471 211311752.398513

0 134217728 180269109.770626 211325601.466004

0 268435456 180159762.243375 211334413.467605

0 536870912

Is there a way to check which mode is currently engaged? I did see explanation for nvidia but did not see one for amd.

Also one novice question, guess i could try this one by myself but since i'm writing here goes , if i want to save all adaptations with F3 in Setup 1.1x do i have to press F3 after last command just before chainloading to windows or after each executed command?

Cheers,

D

Hi,

I got this configuration working last nigh

1. OS:

- Win7 64bit, DNA-AMD GFX 12.7B.0 - x64

2. System

- HP6930p, CPU P8400 2,26 GHz, PE4H ver2.4 with EC2C

3. RAM

- 3GB

4. eGPU used

- HD 7770 Sapphire

5. PSU specification (especially the +12V max current)

- 550W, LC POWER Silent Series, ATX V2.2

DSDT method faield but Setup 1.x worked, needed 256MB free as mentioned.

3dmark results

Cheers.

D

Tech Inferno Fan >> damir001 successfully engaged x1E mode and provided RE5 and 3dmark06 benchmarks here.