Testing the flashed FireGL X3-256MB

[jeremywork]

I picked this card up here from @herd a week ago, and I've been quite happy with its capabilities.



I've had it installed in an MDD/FW800 with a Sonnet MDX/Duet 1.8 and its performance complements the processor speed very nicely.

I was pleasantly surprised to find the 'top' port works as a dual link DVI port with the X850 Mac ROM, and happily drives the Apple 30" Cinema at native resolution. Some research suggests the FireGL's other port is also dual-link, but the X850 ROM would be expecting an ADC port, which were (probably) never dual-link capable. The single-link port still works great for any mode up to 164.990MHz. One of my first tests for the output drivers was an IBM T221 I've had on the bench (recent obsession) since it benefits from multiple DVI links.

For what it's worth, the 24Hz twin SL mode works just fine, and the card responds well to custom resolutions from DisplayConfigX, but the 'Detect Displays' command ends up in an infinite loop if two of the same DCX display are plugged in at once (which is how T221s work, unfortunately.) I can run either stripe at 33.1Hz, but not both at once. It's easiest just to uninstall the DCX profile for the T221 and let the EDIDs specify the output (24 or 25Hz max, depending on EDID mode.) The two stripes are less in-sync than the non-genlocked solutions I've tried from nVidia, but it's still a somewhat nicer experience than 17.2Hz. @gordan touches on a similar finding with later ATI/nVidia cards in point 5) of this post: https://www.eevblog.com/forum/gener...HPSESSID=5quoakqme3ndkc3pfhemije030#msg265450

Running only one link brought to light a curiosity I've had with ATI's software support for extra large desktops:
- The FireGL card enables the full 3840x2400 and most screen elements work normally across the whole desktop (might be some exceptions, such as screensavers.)
- The Radeon 8500, 9000, Mobility 9000 (TiBook) are essentially in the same boat (I've noticed quicktime videos and iTunes visualizer draw much more smoothly near the Apple menu than toward the far edges of the screen.)
- The Radeon 9800 Pro and Mobility 9700 (PowerBook DLSD) I've tried will also drive the display properly in SL (9800) and DL (9700) and draw the desktop in its entirety when enabling the display. However, the screen only redraws normally for the leftmost 2656 pixel columns, and rarely updates elements to the right side of this limit. Since both older and newer ATI generations don't have this limitation, I'm curious if this limit/bug could be overcome.
This video might help if my description is confusing:

I'd love to know if anyone's been poking at this already, and what you've found so far.

All this aside, I would happily recommend one of these flashed cards for whatever performance-oriented G4 you might be building. It's competent enough run the 30" without becoming choppy during Quartz Extreme/Core Image effects.

 

[herd]

I don't have much to add. I usually have two screens connected at 1920x1200. I have one bigger screen, an ultrasharp 2719, that has display port and HDMI inputs. I got an adapter for DVI to HDMI and the screen works with this card at 2048x1080. The native resolution is 2560x1440, and this option shows up in the display preferences, but when I pick it the display shows an out of spec message. Could DisplayConfigX get the higher resolution working?

Anyway, this is a data point, but well below what you're doing with it.

 

[jeremywork]

My (amateur) guess is the HDMI input is accepting a single TDMS link from DVI port which common passive adapters provide, but while still transmitting the EDID with full native resolution back to the host even though sufficient bandwidth isn't available. DisplayPort would do the same thing using its version of passive adapter too, though there are a couple options (Atlona, Gefen) for active adapters which allow dual-link DVI inputs to be converted to "high bandwidth" displayport outputs. Gefen's solution even includes a build-in EDID override switch with presets for 2560x1440 and 2560x1600 in case the default 'passthrough from monitor' behavior isn't behaving. https://gefen.com/product/dual-link-dvi-to-mini-dp-converter/

I've only begun poking at things with DisplayConfigX. Theoretically on displays without some form of mitigation, invalid inputs may cause harm to the display driving electronics as unix x modelines once could and did to many CRTs.

My testing so far has been only with Apple's 30" Cinema display, IBM's T221-DGP, and a few times with high end CRTs where I stuck strictly to GTF timings to see which monitors might happily achieve 2048x1536 even when not available by default (Sony's GDM-FW900 will happily accept 2560x1600@60Hz, for instance.) Each of the monitors I've tested has had an OSD message/LED pulse pattern for timings that exceed the hardware's capabilities, which should mean damaging these models with invalid modes isn't something to worry about. I still try not to leave the invalid mode running for too long since it may fatigue the sensing components over time.

That aside, even with the existing setup you could customize the input to achieve any input your display will accept that falls within the bandwidth limitation of single-link DVI (165MHz, often limited to 164.99MHz in the software.) The first thing to understand is that Mac OS will check all modes at boot time and will eliminate any modes which exceed the maximum parameters available on the video card (this is why some cards accept 165.00 and others only 164.99.) All modes which the video card accepts will be available (the modes provided by the display's EDID can be ignored or added to the pool of additional custom resolutions depending on your preference) and then it is up to the display's electronics to decide whether it wants to accept and attempt to drive that mode.

Some displays don't provide auto setup information (30" Cinema) so determining which blanking intervals to use may require a trial and error approach (so far it seems to basically inflexible to different resolutions, but will run happily between about 45 and 62Hz, which can significantly improve on its namesake, watching cinema content. It does flicker slightly at precise combinations of brightness and refresh rate, which is seemingly why Apple locked it to 60 and were so happy to boast about the PWM's flexible frequency on the 6K display.

The T221-DGP isn't a good sample because it isn't like much else that exists. Its panel driver is primed by a full 96MB scaler which is extremely flexible at interpreting and retiming signal (since it needed so many signals to run natively in the world of single-link DVI) and it basically accepts any mode within the bandwidth and physical pixel limits. Feeding it pretty miniscule intervals also basically works, although at its absolute limit the panel will sometimes flash an 'invalid mode' LED pulse instead of entering standby after the host has gone to sleep/shut down. The signal is always picked up again, and this is the closest thing I've seen to a modeline causing a malfunction on the T221. The T221 scaler's ability to split resolution into refresh (3840x1712@24/48Hz, etc) is not to be expected from most consumer displays in my experience.

Anyways, I'll be curious what other modes you can get the 2719 to display...

 

[pc297]

Is this with the stock EEPROM or with a larger one soldered on? What ROM file was used, the one form the Wikidot Mac Elite?

 

[jeremywork]

Is this with the stock EEPROM or with a larger one soldered on? What ROM file was used, the one form the Wikidot Mac Elite?

On this example an EEPROM swap is mentioned in the original sale thread to accomodate the Apple X850 ROM; I believe the Wikidot Mac Elite was the source. @herd also had some thoughts about reducing the ROM to fit in an unmodified card, but I haven't attempted such a thing yet.

Also, to paraphrase another question: Some lower tier ATI cards aren't compatible with flashing Mac ROMs (even with proper EEPROM) because Mac ROMs expect a fully featured chip, and ATI often sold lower tier PC cards with pieces of the chip disabled (maybe related to binning defects.) If a Mac ROM could be produced with some added limitations, perhaps many more cards would become viable flash candidates.

I should've linked the thread in the top post but I forgot:

https://68kmla.org/bb/index.php?threads/fs-ati-firegl-x3-256mb-agp-video-card-for-quicksilver-da-mdd-usa-shipping.43340/