Windforce R9 290 3X OC + 10.11.2

[Fl0r!an]

Can you verify if the same setup as in the other thread (I think HDMI & DVI at the same time) works on one screen?

I'm not sure if that ioreg stuff matters, but it's good to know.

 

[justr]

I will try, but I'm not sure I have a DVI cable/monitor around. I do have a DVI to HDMI cable, so I can try that.

I'm going to give this another day or so of tinkering, and then I'll probably remove this R9 290... and go back to my 5770, or see if another 290 works better with my setup. Or, a 7970/280x.

But I'll report my results over the next day, in the hope that I find a possible solution-- and that my struggle helps someone else!

 

[justr]

Can you verify if the same setup as in the other thread (I think HDMI & DVI at the same time) works on one screen?

I'm not sure if that ioreg stuff matters, but it's good to know.

Didn't work.

Monitor 1 (Samsung 4K TV) hooked up to HDMI. Monitor 2 (LG 768p TV) hooked up via (GPU) DVI->HDMI (tv end) cable. Result: Both monitors on until Clover, at which point Monitor 2 goes black. After Apple logo w/ loading bar, Monitor 1 blinks white, followed by reboot.

Tried the same setup except with Monitor 1 to ACD 27" via (gpu) DP->mDP. Same behavior, except blinks white screen, then Monitor 1 goes to blackscreen w/ backlight on. Does not reboot-- I can log in with remote management. Monitor 2 stays black.

Tried same setup as previous (ACD via DP), but now tried Monitor 2 (Samsung 4K TV) via HDMI. Same result as previous.

Tried DVI->HDMI on LG 768p TV with no other monitors connected-- blinky white screen + reboot.

Tried a number of these combinations and monitors alone with various patches, w/ Radeon as FB, with Exmoor. Always seems to give the same results: ACD alone, with Florian's suggest Baladi patch on ACD via DP gives me blinky white to blackscreen that I can remote log in. Same setup to any other single monitor alone is white screen to reboot. Adding a second display doesn't change this behavior; second display simply turns black (no backlight; signal disconnects), and primary follows the pattern above.

It doesn't matter whether I use either BIOS switch position on the card (which should be, I think: Quiet and Uber modes). The card supports UEFI, but I don't know if this matters (I'm on a Legacy install of Clover).

As mentioned previously, I have strange behavior when trying to use this card in Windows 10 (already installed) or booting from Win7 Installer CD:

This is with the SSDs with Clover removed from the machine, with Easy BCD set up to chainload Chameleon (first screen lets me choose to boot Windows, or go to Chameleon, I go directly to Windows).

With R9 290 alone, and set as primary, booting to Windows/CD takes 30 minutes (I timed it), super-super slow once in Windows, and after a while (like trying to download something ~300GB) I end up with a blue screen.

With Samsung 4K TV hooked up to R9 290 in PCI-E 16x2 set as primary in BIOS, and nothing connected to 5770 in PCI-E 16x1 (AMD Crimson reports it as "disabled" in this configuration), with latest AMD Crimson and drivers installed (16.1 Hotfix)-- the machine boots very quickly AND runs my Samsung at 4K30p (which makes sense since the GPU only has HDMI 1.4a) in 10bit, 4:4:4 mode!

If I next shut down, change nothing but remove the 5770-- back to taking 30 minutes to boot Win10! This is even after removing 5770 drivers, and after doing registry edits to turn off EnableULPS (ultra low power bug that sometimes causes slow boot w/ AMD card-- but when I say it causes slow boot, it's usually like a minute extra, not 30 minutes!).

From here (no SSDs with Clover/El Capitan) I removed my Windows 10 HDD, and installed a fresh HDD attempting to use my old Windows 7 install CD to do a totally fresh install to this fresh HDD. Only the R9 290 installed. BIOS set to boot from CD as primary boot device. Everything started normally, with the Install CD loading (takes a minute or two, which is normal from CD), then gets to the Welcome to Windows splash screen from the CD (Win7 not installed yet).... and after 20 minutes of that I had had enough, and rebooted the machine. Same super slow boot, apparently.

I've checked with the person I purchased the card from and he says he had zero trouble with it, and didn't load a custom vbios. Gpuz, on my end, shows that both bios positions on the card report F12 bios, the highest currently available for this card/run. So I assume the card is functioning properly. It does indeed run Valley benchmark in Win10, as well as Cinebench R15.

In El Capitan 10.11.2 and 10.11.3 (I have each installed on an SSD), running via remote management (which likely has significant overhead considering the virtual display it represents is my ACD 27 at 2560x1440), both bios switch positions give me Cinebench R15 results of about 70fps with my Xeon X5650 overclocked to 4.0 Ghz on all cores. Not a great result (5770 does 64-66 fps at the same settings, though not via remote management), but not bad. Also, Chess, FCPX, Lightroom w/ GPU acceleration turned on... all work on the R9 290 via remote management.

I'd love to keep tinkering, but I have a ton of work to attend to, and I've spent the last week trying all sorts of combinations of things. On top of that, I'm wary about too much SATA, PCI-E, Molex, PCI-E 6/8-plug inserting and removing. So, short of any last minute magical advice, I think I'm going to give up and uninstall the GPU later today.

Perhaps there's an issue with legacy Clover installs on 5-series boards and R9 290s, or just this model card, or some setting I just can't get right. After all, there is evidence that this R9 290 (albeit with black screens) is recognized, with full VRAM, and full acceleration, with zero crashing using accelerated apps, even recognizing two (albeit with black screens) monitors (ACD via DP, 4K Samsung via HDMI) at the same time with full acceleration!

I hope this thread provides some insight into what might be required to get this card working on this or another system. The closest I got was using Baladi w/ Florian's suggested patch from the other thread (linked to earlier here). I just never got any combination of solo or dual monitors to boot beyond a black screen no matter what I tried.

Many, many thanks to those who tried, particularly Florian who is an incredible resource and a testament to this place. I want to pay it back, so I'll be keeping and eye on R9 290 threads and attempting to help with whatever knowledge I have. I may also try another model R9 290 myself.

Cheers!

 

[justr]

New Update (haven't quite given up just yet).

Installed 10.11.3, and removed one DIMM of DDR3 RAM. I had had 4 sticks in a special 3-channel mode, but I'd read that this caused issues with the GPU. And in fact, once I removed that DIMM-- the R9 290 booted right into Windows 10 at normal speed with no other GPUs installed. (Strangely, my start menu disappeared, which is a common issue with Win10).

I've also been messing with a lot of connector patches via clover. So far I have been unable to get my DP/mDP monitor to come back from black screen + backlight on, even when booting with only DP plugged in. But with two patches I can get HDMI with full acceleration on my 4K TV (though the resolution is WAY too high: About This Mac reports the screen as a 120 inch 8K or so screen; changing the resolution crashes the system to black screen that does not come back).

Ironically, one of these patches that works on HDMI was supposed to work only on DP! The other was for HDMI. The only difference, that I can see, is a switch between the hot plug IDs.

These are the patches:

00080000040200000001020022050203 <---- HDMI connected here works, but resolution is too high
00040000040300000001010112040101 <---- matches DP info from ROM (12 04 01 01), but black screens
00040000040300000001020021030202
04000000140200000001030011020304
00040000040300000001050010000505
00020000040200000001040110000406

and

00080000040200000001000022050103 <----- HDMI connected here works, but resolution is too high
00040000040300000001000712040201 <----- almost matches DP info from ROM, but wrong hot plug ID, black screens
00040000040300000001000021030302
04000000140200000001000011020404
00040000040300000001000010000505
00020000140200000001000010000606

Another interesting tidbit is that if I boot plugged into HDMI, I can then plug the DP cable in to my other monitor and that screen turns on as a black screen but with the backlight on.

I may still give up, but I'm getting closer.

 

[justr]

Getting closer!

I used this as the connector patch (minus the comments, of course):

00080000040200000001000022050103 <-- HDMI
00040000040300000001000012040201 <-- All the rest here down are DP
00040000040300000001000012040302
00040000040300000001000012040404
00040000040300000001000012040505
00040000040300000001000012040606

These are the steps I took:
- Boot with DVI plugged in to 4K TV (I get "no cable" and obviously no display)
- Log in with Remote Desktop to check that system has booted
- Once booted, plug in HDMI to 4K TV (TV display comes on at 1080p, mirroring a phantom display on DVI)
- Quite Remote Desktop, state stays the same, I can use my TV monitor.
- Plug in DP (ACD 27" comes on at 1440p as a second display! I have 4K TV via HDMI at a bizarre 7K4K resolution, and ACD at native)
- Set displays to Mirror, and choose optimize for ACD: Both display 1440p, full acceleration!

I will continue working, because it would be outrageous to need to go through these steps on each boot, but this tells me the card is working, and somehow the patches are starting to line up.

Additional peculiarities:

Using IOReg to look at some parameters:

Booting with only DVI plugged in gets me, following this format "ATY,Baladi@X, port-number Y" where X and Y are numbers:
The numbers are out of order, but I believe: ATY,Baladi@0, 0x3

When I plug in the HDMI and TK comes on:

@0 gets LCD @ 0x0 <—HDMI, 4K TV, picture
@1 gets LCD @ 0x3 <— DVI, no picture (control 304, active 04, DP config)
@2 gets 0x1
@3 gets 0x2
@4 gets 0x4
@5 gets 0x5

When I unplug the DVI and now plug in the DP (so now I have DP and HDMI monitors both working at the same time!)

@0 gets LCD @ 0x0 <—HDMI, 4K TV, picture works
@1 gets LCD @ 0x1 <— DP, ACD 27, picture works
@2 gets 0x2
@3 gets 0x3
@4 gets 0x4
@5 gets 0x5

I will keep tinkering.

 

[justr]

Latest experiments with the previous patch:

- Boot with DVI plugged in (no monitor signal)
- Remote Desktop in (for some reason this is necessary to prevent a freeze!)
- Plug in HDMI, picture shows up on 4K TV w/ full acceleration at 1080p
- Unplug DVI, picture stays on 4K TV
- Plug in DP, picture shows up on ACD 27" AND 4K TV at 4K (strangely, at full 4K-- 4096 x 2160--not just UHD).
- Turn off 4K TV, picture remains on ACD.
- Quit/log out of Remote Desktop, picture stays on ACD, with no lag, and full acceleration.

That's a hell of a process to use just to get acceleration without glitches on a single DP display. But it's a proof of concept that it works. Note: I have to leave the HDMI cable plugged into the computer or else the ACD screen is glitchy (artifacts)-- though it runs Cinebench a few frames per second faster.

 

[Fl0r!an]

Interesting stuff you keep finding there!
I think this can be explained with some mismatched sense IDs (maybe hotplug IDs also). Maybe your GPU is gathering monitor information (EDID stuff) from the wrong port, resulting in strange behavior (e.g. wrong resolutions).
Possibly transmit and encoder ID of your HDMI port is correct (basic requirement for any output), but the hotplug ID is falsly assigned to a DVI port and the sense ID is assigned to DisplayPort (just examples, could be different). This way you'd need something connected to those 3 ports to initalize display output, and it would still be a bad solution since all your displays usually have different EDID.
Do you have a monitor with all those different inputs? Would be interesting to see what happens when hooking your GPU up to one display via DVI, HDMI and DP at the same time.

That's no ideal solution of course. I can't really explain why those IDs should be wrong, since we constructed the patch from your VBIOS data, but maybe we're missing something.

 

[justr]

What's strange is that the patch we constructed from my VBios data never worked at all, although I'm not sure I tried this convoluted plug and switch process with that patch.

The patch I'm using, which should be very similar, started with a 290X 8GB someone patched on InsanelyMac-- which happened to have identical Bios Decoder output as mine to result in the same four connector strings I got when I did the process myself. But than that patch was experimented with, and there are some changed things that don't make sense to me.

In order to attempt to focus on DP (since that's my only must have monitor), I made a few modifications so that all ports besides the working HDMI are set up for DP, with the second port the closest match to what my output should have been according to the manual patch I came up with based on my VBIOS.

My next step is to try to integrate the DVI and DP patch info you pulled and came up with, Florian.

It's a bit of a puzzle, because you move one thing forward and the next goes back, in an iterative cycle.

For now I'll keep tinkering since I know I can at least take breaks and, with a bit of hoop jumping, get back to DP for a day of work. So far it's totally stable and benches well (BruceX 5K is between 16-17 seconds, vs 44-45 seconds with my 5770).

 

[justr]

So, a quick update (I'm only getting to tinker when I have free time + energy in the evenings/weekend).

So far, I've gotten a few more strange/interesting results.

This is the latest patch I've been trying:

00080000040200000001010022050103
00040000040300000001020012040201
00040000040300000001030012040302
00040000040300000001040012040404
00040000040300000001050012040505
04000000140200000001060011020606

Notes:

- This will boot and the screen will come on with only the DP/ACD 27" plugged in, with full acceleration, at 1440p. However, the screen is VERY glitchy (every few seconds I get horizontal tearing), to the point that it's not really comfortably usable. If I try to make any changes to the resolution, rotate-unrotate, or plug in another display-- the computer instantly freezes. Putting the display to sleep or system to sleep and waking does not remove the tearing/artifacting. My highest benches, though, come from this config/boot protocol (Heaven 1080, medium - 86 fps/21xx score; Cinebench R15 ~77fps, BruceX 5K 16-17s).

- With this configuration I can get normal DP use on this screen with ZERO artifacting/tearing, but it requires a very specific, strange way of booting:

1) Boot with DVI plugged in (cable must be connected to a monitor, even if it's off).
2) Log in to machine with a Macbook using Remote Management/Screen sharing.
3) Plug in DP.
4) Turn off Remote Management/Screen sharing, turn off (if it was ever on) monitor connected to DVI.

Notes:

This works flawlessly, but takes a small hit (10-15%) on Heaven, and 5% on Cinebench. The reason, I believe, is that while I can turn off the monitor connected to DVI (I don't turn it on, because it gets no screen anyway), I cannot disconnect that cable or my main monitor freezes. So I'm left operating with a phantom 1080p DVI monitor throughout the whole process.

Of note: HWsensors, like everyone with an R9 290(x)/390(x) on El Capitan reports constant 0-degree GPU temps. Fans spin up and down under load, though. Interestingly, in iStat, the GPU processor load appears to by dynamic, moving up and down-- until it reaches full load, at which point it reports full or near full (say 90%) load until I restart the machine.

If I can figure out how to boot with only DP plugged in, without the horizontal lines/glitches/artifacts... I'll be satisfied. Performance is outstanding when this thing works.

I do have an old 5770, so another alternative would be figuring out how to use that merely as a dummy card, boot with no monitors plugged in, and once at login screen see my ACD plugged in to my R9 290. But whenever I boot this way I get kernel panics.

 

[justr]

And down the rabbit hole I go...

With my previous patch, as I mentioned, I can boot directly connected to my ACD via DisplayPort. Interestingly, I have almost non-stop artifacts: horizontal lines/tearing.

When I load the Heaven benchmark, the tearing stops. I can leave it running in a window and do anything I want with nary an artifact/tear. Of course, this is horribly inefficient, but it's a clue. The issue seems to be something related to the idle state of the card, or perhaps the moving between idle and non-idle states.

Further evidence to support this idea comes from the fact that when I have two monitors working, or one monitor and one phantom monitor-- I don't get the artifacts. Guess what? These GPUs are sometimes coded so that when more than one monitor is plugged in the GPU doesn't never drops to its idle state. No idle state, no artifacts!

A quick google search reveals this is a common problem for R9 290 cards-- on Windows.

Enter VBios editing. The next round of experiments will focus on trying to eliminate this problem via a few possible directions:

1) Increasing voltage to the GPU either only at idle clocks (DPM 0 state) or across all 8 power states.
2) Trying to edit the vbios so that the idle state core/mem clocks are something other than 300 mhz/150mhz.
3) Trying to experiment with how the core/mem clocks change via the intermediate power states (though mem clocks seem to only ever go between idle and full clocks, which some think could be the issue).
4) Increasing the VDDCI voltage (memory) by just at touch.

It could be as simple as the idle clocks get too little voltage to be stable.

It could be that voltage is fine, but the idle clocks (300/150) are simply too low to drive a 1440P monitor via DP without issues.

It could be that the big jumps between states and voltages cause the issue.

I've read most of this thread already... http://www.overclock.net/t/1561372/hawaii-bios-editing-290-290x-295x2-390-390x