X600 Radeon PCIe gone bad

[mrmazda]

I have two near twin PCIe X600 Radeons. One started to give trouble a few weeks ago, but today it became useless. That one I've had in virtually constant use for over 6 years for OS/2 to run DOS apps in SVGA text mode. The other was bought to serve as backup, as cards any newer this and X800/X850 are not supported by the newest available OS/2 video driver. I'd very much like to make the bad good again, to minimize possible down time if the current one fails.

Neither card has a brand on it. Both have part number stickers that are vastly different. I know at least one was sold to me as MSI. The only visible difference between the two is the good one has three 470uF polymers where the bad has 4 Nichicon HC 470uF 10v 10mmX13mm H0524. The good also has 1 Nichicon HM 8mmX12mm, while the bad has 2 A0447.

Does this look or sound like a candidate for recapping success? If so, should I use polys for the 470s? For all? If polys, which should I try? If not, what should I use? Direct matches to HC and HM are absent from Mouser and Digikey, and I'm finding only dead links looking for specs on them.

[c_hegge]

Re: X600 Radeon PCIe gone bad

What is the symptom of the problem? Going by you other thread, it sounds like you mean it has graphic artifacts. If so, it's probably either the GPU or RAM that is faulty.

[stj]

Re: X600 Radeon PCIe gone bad

you mention drivers,
dont modern cards still comply with VESA ?
cant you fit a newer card and select the generic VESA driver? that is why they invented the standard!

[mrmazda]

Re: X600 Radeon PCIe gone bad

dont modern cards still comply with VESA ?
cant you fit a newer card and select the generic VESA driver? that is why they invented the standard!

Most do, but I'm not sure all do. Some gfxchips, including those written by people paid by Intel to write FOSS drivers for its own chips, don't even have Linux support. Over the years I've tested quite a number and found less than complete VESA support especially from NVidia, but others too. Very few PC users have any routine dependence on VESA, and most that do must depend on the most common modes, leaving a dearth to miss what isn't provided.

In OS/2 like in Windows, if you depend on VESA, you're limited to VESA modes provided by the video BIOS. Those modes are limited, and AFAIK, there are no VESA modes matching the preferred modes of 16:9 or 16:10 format screens. VESA text modes understood by DOS apps are even more limited than graphics modes. All are nominally 4:3. VESA also has no 1400x1050 mode, which is the native mode of the 20" Viewsonic LCD I use for OS/2. Not only are the modes limited, but also the speed. AFAIK, no hardware acceleration is ever available using a VESA mode. IOW, if you want the most a gfxcard is capable of, or even just respectable performance, you need a driver written to take advantage of non-VESA features. And if you have any type of wide screen, a non-VESA driver is required if you don't want everything to look fat.

The gfxchip makers make proprietary drivers for Windows and Mac, and for specific Linux kernels. The FOSS community provides its own drivers for a wider variety of chips for Linux, and without anywhere near the same kernel limitations.

For OS/2, the development community, paid and otherwise, is nearly non-existent. Until 2007 or so IBM was outsourcing a good video driver offered as a free download. It's best support (actually, the only decent support in this century) was/is for Radeon, but it ended before the r5## series was introduced, and barely includes any chips on PCIe cards. Those newer than what I have, and have VGA connectors instead of or in addition to DVI, were rare when new, and even more so now. Those of such configuration that don't have heat sink-specific fans that die within 2 years' use are even more rare. That driver is the one available and working splendidly without paying for an eComStation subscription to access the limited support available for some newer hardware.

For DOS apps in OS/2, same as booting DOS directly AFAIK, any newer gfxchip would work only for those who can be content with 80x25 text, or 640x480, 800x600 (and maybe 1024x768) graphics modes. Those are no good here. Without OS/2 to run DOS, access to HDs is very slow, and severely limited in disk and partition sizes.

[mrmazda]

Re: X600 Radeon PCIe gone bad

What is the symptom of the problem? Going by you other thread, it sounds like you mean it has graphic artifacts. If so, it's probably either the GPU or RAM that is faulty.

I don't know if people would classify what I see as artifacts. This starts as soon as the BIOS lights up the display in POST. It looks to me like blocks of the framebuffer with half the pixels missing are echoed in multiples of 4 or more, but also like sync is randomly missing. It makes figuring out where a mouse pointer points difficult, because it's a rectangular blob or blobs instead of a pointer.

The X600 currently installed looks like this one.

[mrmazda]

Re: X600 Radeon PCIe gone bad

The single pic and the middle of three is the fritzing card, which was reliable in virtually continuous use for over 6 years. The left two are MFI brand acquired many moons ago. The right I have no clues who made, and cost $8 on eBay last month. I noticed the heat transfer pad had become unglued on the frizter, so replaced it with fresh silver compound, but it didn't help. Cleaning its contacts with pencil eraser and spray contact cleaner and trying in another machine didn't help either. I bought an X800 from eBay last month too. It has the same problem on VGA output, no output on DVI, so I got a refund from its seller.

The OS/2 machine is multiboot on a 1400x1050 display. IIRC, Linux will use more than 128M of RAM when I connect it to a 1920x1200. I once made the mistake of buying a 128M RAM card, and hope I will remember not to do that again.

Caps on fritzer: Nich HC 470uF 10V H0524 10x13; Nich HM 470uF 16V A0447 8x12; SMD poly 100uF 16V E45 7x6. I removed the 2 HC closest to sink to check: near spot on uF, somewhere between 0 and .01 ESR. I put them back and the bad behavior that didn't always happen before warm up before began right away in cold POST.

[momaka]

Re: X600 Radeon PCIe gone bad

I don't know if people would classify what I see as artifacts. This starts as soon as the BIOS lights up the display in POST. It looks to me like blocks of the framebuffer with half the pixels missing are echoed in multiples of 4 or more, but also like sync is randomly missing. It makes figuring out where a mouse pointer points difficult, because it's a rectangular blob or blobs instead of a pointer.

That be artifacts are.
Time for reflow it. Or sell for parts or repair on eBay (or elsewhere) if you don't want to mess with it.

My guess would be those passive coolers are too small and eventually the heat cycles cracks the joints on the GPU BGA. RAM should be fine.

[mrmazda]

Re: X600 Radeon PCIe gone bad

That be artifacts are.

I fail to grok this.

Time for reflow it. Or sell for parts or repair on eBay (or elsewhere) if you don't want to mess with it.

This I may make into an experimental foray into reflow. These are cheap enough on eBay, and I already have spares.

Anyone remember where they've seen a reflow howto? Here somewhere maybe?

My guess would be those passive coolers are too small and eventually the heat cycles cracks the joints on the GPU BGA. RAM should be fine.

Makes sense to me.

Thanks for the reply.