no-name Radeon 9600 XT recap and cooler

[momaka]

I've been seeing a lot of cheap-looking (but not always so cheap) no-name AGP video cards on eBay for a while. They started popping up quite a bit when the market for AGP cards became more alive all of a sudden due to interest in “retro” PCs. One thing that struck me about these cheapo cards is that they almost always seem to come with garbage caps - including the abominable Sacon FZ. That aside, though, I've been wondering about the rest of the quality (or lack of?) on these cards. So I decided to satisfy my curiosity yet again.

This one was actually suggested to me by ChaosLegionnaire. It's a Radeon 9600 XT with 128 MB of RAM, according to the sticker.




Right off the bat, I'm sure everyone saw the blown Sacon FZ caps, and you probably know by now I hate those with a passion. Perhaps, I should change my forum title again - Sacon Exterminator would seem fitting.

Garbage caps aside, the card appears to use genuine Samsung k4d261638f-tc40 (4.0 ns) RAM chips. No telling, of course, if they are recycled or brand new (and maybe reject?) chips. They do look clean, though - if that means anything.

The GPU chip, on the other hand, really does seem like a recycled part, as I found out after removing the GPU “cooler” for TC inspection. I forgot to snap pictures of that, but two SMD ceramic caps on the GPU chip substrate were nearly blown off their pads. The substrate PCB also had a bit of darkening around the die area. It could be that the GPU chip is NOS, and the darkening and offside SMD caps is from the manufacturer using their IR / hot air a bit more carelessly. But really, these kinds of things are something I've seen frequently on reworked / reflowed BGA chips, so I'm still leaning towards recycled GPU chip.

Whichever the case above, I was eager to test the video card. Before that, though, a recap was necessary. Knowing how bad Sacon FZ caps can read when they blow / go bad (typically in the picoFarad range), there was no way I'd test the card with them in there.

So before anything else, I made a quick “cap map” to mark what rail each electrolytic cap (or empty spot) is connected to.

(Hopefully the above picture is not too small to read my marks… and also note there was a cap in spot C29/CA29 that I hastily removed before realizing I should take a picture with the marked caps/rails.)

For those who prefer to go by the cap designators on the PCB rather than the cap map above (if you can't view attachments, for example), below are listed all of the rails present on this card and which caps they contain.

* 3.3V rail from PSU / AGP slot: 3 cap spots, 2 caps total:
C29 / CA29, C8 / CA8, CA3 / CP3 [unpopulated]

** GPU V_core: 2 cap spots, 2 caps total:
C38 / CA38, C45 / CA45

*** RAM V_dd: 2 cap spots, 2 caps total:
C19 / CA19, C20 / CA20

The below cap spots are all unpopulated free spots, but listed for completeness:

**** GPU V_tt: 1 cap spot, 0 caps total:
CA4 / CP4 [unpopulated]

5V rail from PSU / AGP slot: 2 cap spots, 0 caps total:
CA2 / CP2 [unpopulated], C9 / CA9 [unpopulated]

12V rail from PSU / AGP slot: 1 cap spot, 0 caps total:
CA1 / CP1 [unpopulated]

Also note that this PCB can take both regular through-hole or SMD caps. In addition to that, the cap spots associated with the GPU V_core high and low sides (C29 / CA29, C38 / CA38, and C45 / CA45) have vias for both 8 mm an 10 mm diameter caps. The rest of the cap spots are for 8 mm diameter only.

Here is also a list of the cap values used on this card:
C29 / CA29, C8 / CA8, C19 / CA19, C20 / CA20: 4x Sacon FZ, 6.3V, 1000 uF, 8 x 13 mm (dia. x height)
C38 / CA38, C45 / CA45: 2x Sacon FZ, 6.3V, 1500 uF, 10 x 13 mm (dia. x height)

So for a full recap, one needs 4x 1000 uF caps and 2x 1500 uF caps, though I reckon 820-1200 uF capacity can be used without problems in place of the 1000 uF caps and 1200-2200 uF capacity for the 1500 uF caps.

In fact, here is how my recap turned out:




Basically, I (again) used whatever I had on hand in stock:
- 1x Nichicon HC, 10V, 470 uF cap for the filter in spot C29 / CA29, which is the 3.3V rail input / high-side supply of the GPU V_core. I went with much lower capacity here, because there are two more cap spots on the 3.3V rail, and for this one, the ESR is probably just as important as the capacitance, so I thought this was the perfect time to use some of my aging old stock of pulled Nichicon HC caps.
- 2x Rubycon MFZ, 6.3V, 2700 uF caps for spots C38 / CA38 and C45 / CA45 (GPU V_core rail / low-side), because like my Nichicon HC cap stock above, I have too many of these and they are starting to get rather old. Worth mentioning is that before I completed my above shown recap, I used only ONE Ruby MFZ 2700 uF cap at first to test the card, and it ran fine with that under full load. So this card is definitely not too picky on the capacitance.
- 2x OST RLP, 6.3V, 1000 uF caps in spots C8 / CA8 (3.3V rail PSU) and C19 / CA19 (RAM V_dd). I initially put these just to get the card ready for a function test. But after checking their ESR (showed OK) and due to being short on Rubycon ZLH 6.3V 820 uF caps, I opted to keep the OST caps. The only place I put a single Ruby ZLH 820 uF cap was in spot C20 / CA20 (RAM V_dd rail.) And that completed the recap for now. I might revise this recap someday when I get more Ruby ZLH caps. But for now, it's more than fine, considering the low power of the Radeon 9600 GPU (even the XT version.)

More to be continued in part 2 below…

[momaka]

no-name Radeon 9600 XT recap and cooler - part 2

You may also have noticed I changed the original cooler with one that looks like it came from a Radeon 9800. This new cooler, in fact, did NOT come from a Radeon 9800. Rather, it is a cheap eBay clone of the 9800 cooler. I bought it over a decade ago when there were all kinds of cheap clone coolers available. I think this one ran me $7 shipped at the time, if not less. It's actually pretty close to being as good as the original Radeon 9800 cooler. I bought it a while back with the intention to use / test some Radeon 9700 cards that I acquired for parts/repair… and I did use it on a few. While it was able to keep them (barely) functional, the heat from a 9700 GPU is simply too much for such small cooler. So this 9800 clone cooler ended up in my spare parts bin. I've been using it on various cards with 55 mm screw distance, and so far the cheap fan has help up fine.

On that note, this is precisely why I wanted to change the stock cooler on this 9600 XT card. Although the cooler is probably just adequate enough for the low TDP of the 9600 die, I would have had to keep everything the same – meaning to leave the small fan on the stock 9600 cooler to run at full speed. But it's very loud at that speed and will likely wear out quickly. The one on the 9800 clone cooler is slower-turning and thus quieter. Not only that, but the extra fins and plate thickness of the 9800 clone cooler meant that I could possibly run its fan at even lower voltage for further reduction in noise. And sure enough, that's what I ended up doing above, as visible by those extra wires you see going between the fan connector and the PCB. Essentially, I wired the fan to 7V (i.e. fan wired between PSU 12V and 5V rails, for a difference of 7V). Even with that voltage and running the card in a hot 82-83F room temperature, the heatsink on this Radeon 9600 XT card was completely cool to the touch.

Again, the original cooler was probably adequate, but just loud. It also didn't have much thickness / thermal capacity:

This would have left the possibility of the GPU frying pretty quickly if the fan failed. Meanwhile, the 9800 clone cooler is almost as large as the passive coolers on some regular 9600 cards.

That aside, the hottest components were, surprisingly (to me), the RAM chips. With the card idle, they ran just warm. But under full load or when gaming, they got almost too hot to touch – enough for me to consider adding heatsinks on them (when I get to it. )

Strangely enough, the V_Core MOSFET, SQ1, (APM7313) didn't run as hot as I thought it might, given that it's only a single SO-8 device doing the whole 3.3V --> GPU V_core conversion. On the other hand, the RAM V_dd MOSFET, Q3, (ST D1703L) ran fairly warm… but that's expected, since it is used as a linear regulator to drop 3.3V down to the 2.5-2.6V the RAM needs. Then finally, there's SREG5, an 1117 adjustable regulator for the GPU V_tt voltage. And to complete the list of major components used, the GPU V_core PWM controller for SQ1 is a SO-8 part labeled “1575”.

One last mod I had to do on this card was to sand/cut away slightly from the top part of the PCB that's used for the AGP lock. It was simply too tall, and motherboards with lock levers on the AGP slot would get stuck, not letting the card go all the way down in the slot. Cutting down the PCB leg for the lock fixed this issue.

Doing all of this left me with a well-working Radeon 9600 XT card. Or did it?
While the card does work fine, I don't think this 9600 has quite the specs for an XT version. Checking it out, this is what GPU-Z spit out.


A proper 9600 XT should run stock at 500 MHz on the GPU core and 300 MHz (600 MHz DDR) on the memory. This no-name 9600 “XT” is running only at 400 MHz core and 225 MHz on the memory, which is closer to the 9600 Pro… and even that is not quite right, because a stock 9600 Pro still runs with 300 MHz on the RAM. Thus, this 9600 “XT” is really somewhat more in between a 9600 pro and non-pro.

What's interesting is that the Samsung RAM on this no-name 9600 “XT” has 4 ns chips, meaning the RAM should be capable of 250 MHz. I tried “overclocking” the RAM to 250 MHz, and the card ran with that RAM speed without any issues. However, this made the RAM chips run even hotter than before. And as I mentioned, they were already running nearly too hot to touch @ 225 MHz, so the extra 25 MHz increase didn't seem worth it, at least until I add some RAM heatsinks (not to mention I didn't really notice any significant performance gains.)

While on a roll, I also tried OC'ing the GPU core. I pushed the slider right up to 500 MHz – i.e. what a true XT runs at stock. This no-name 9600 XT took it just fine. No crashes or any other funny business. However, after running the card for a while like that, the GPU V_core MOSFET, SQ1, started running a bit warmer than I like to see. Being that the card uses only a single SO-8 dual-channel MOSFET for the V_core buck conversion, I decided it's probably not a good idea to push it that much either. And again, whatever performance gains there were from going from 400 MHz to 500 MHz core, I didn't notice them with any of the old games I use for testing. So I returned everything back to the stock values in BIOS. At the end of the day, I'm not really disappointed that this Radeon 9600 XT is not really a true XT card. As with anything cheap / knock off / no-name on eBay, I rarely expect items to be as good as their labels claim. With that said, I'll just treat this one as more or less a regular 9600 non-pro card with a slight OC… or a slightly slower 9600 Pro. Whichever the case, it's still not bad for an older mid-tier AGP card.

So overall, I'm happy with the results and I can say the card itself is probably OK in terms of quality… once it is recapped, of course. Would I recommend anyone to buy one? - Meh… maybe. If it's for a good price (under $10-15 shipped), I'd say OK. Any more, and it's probably not worthwhile, considering the recap it needed. Also, many major brand Radeon 9600 cards can still be found for under $20 somewhat frequently, and these almost always come with good caps, since ATI doesn't slack when building their cards.

[ChaosLegionnaire]

Re: no-name Radeon 9600 XT recap and cooler - part 2

With that said, I'll just treat this one as more or less a regular 9600 non-pro card with a slight OC… or a slightly slower 9600 Pro. Whichever the case, it's still not bad for an older mid-tier AGP card.

yea im trying to be a positive, bubbly, glass half full type of guy, so im gonna call this an overclocked 9600 non-pro.

however, u didnt provide any die shots of the gpu chip, so im not sure if its a rv350 chip there or rv360. i'd like to know what u saw printed on the die. if its an rv360, then it must be a dodgy rv360 that cant do 9600 xt speeds or slightly more than that, so it got binned down. in that case, then its more like a 9600 xl card. same with dodgy r360 cores being binned down as 9800 xl.

[momaka]

Re: no-name Radeon 9600 XT recap and cooler - part 2

however, u didnt provide any die shots of the gpu chip, so im not sure if its a rv350 chip there or rv360.

Yeah, sorry, I kind of recapped this one in a rush. By the time I remembered that I should take some pictures of the GPU die, I already had put new thermal compound and the modded cooler, so I didn't feel like taking it apart again.

But it's an RV360 core, just as GPU-Z reports.

And the card indeed "OC's" without problems to 500 MHz core, just what a normal XT would run at, so I don't think it's necessarily a binned down chip. Again, the only reason I didn't try pushing further (and I suspect this is also why the manufacturer lowered the core clock) is because the GPU V_core rail is generated from a small SO-8 2-in-1 MOSFET, which starts to run a bit warm when the core is pushed to normal XT speeds, especially under load. It's not getting very hot by any means, but that's not to say it couldn't fail all of a sudden without warning. After all, this being a cheap China fab PCB, I don't have 100% confidence that everything is designed properly or that they used adequate parts. And I think the manufacturer probably knows that too and maybe that's why they didn't clock it all the way up. As for the RAM... it also runs fine at rated spec, but gets very hot. So it may be dodgy and/or reject chips. Or recycled chips near EOL (hence the reason for running hot too.)

That said, the RAM being capped at 250 MHz max (in terms of running according to printed spec, that is... I haven't tried OCing further than 250 MHz to see how far I can go), this still makes the card more like a 9600 Pro -at best-... or 9600 XL, like you said. Whichever the case, it still performs like a 9600 series card and there really isn't that much of a difference whether I OC it or leave it running all stock speeds, in terms of performance. On games that get FPS dips down in the 20-25 range, even a 20% increase in performance from OC doesn't give more than a few FPS extra. So as you know me, I rarely see a reason to push the GPU and RAM to its thermal limits just to get a few tiny FPS more. To me, if it's not running at 50 FPS or more, it's running low... and once it's running low, I don't care that much, so long as the FPS is at least above 20-22 FPS. Below 20, though... oh I remember those days with my ASUS Radeon 9200 SE. If I had even this cheap no-name Radeon 9600 "XT" back then, I would have been sooo much happier. Playing through Half-Life 2 on 640x480 @ 18-22 FPS average was kind of rough... but I still enjoyed it all the same.