Your Video Card Cooling Mods

[ChaosLegionnaire]

Re: Your Video Card Cooling Mods - Sapphire Radeon HD4650 1G

well, i'd just like to affirm that your testing methodology is sound on both the 2400 xt and 4650.

with the ambient temp of 29°C and the max load temp of 37°C, u have a delta or rise above ambient of 37-29=8°C. since the 2400 xt has a tdp of 20w and the 4650 having a tdp of around 50w, the heatsink has to handle 2.5x the heat output of the 2400 xt. since the heatsink gave a +8°C delta above ambient with a 20w tdp heat load, so with 2.5x higher tdp to handle, we have 8x2.5=20°C delta above ambient. with an ambient temp of 29°C, so we should be expecting temps of around 29+20=49°C on the 4650. in practice, your result was a max load temp of 50°C.

so i just love it when theory matches up with practice! this means there are no mounting issues with the heatsink and it works perfectly exactly as it should! the above theoretical calculations and momaka's practical testing methodology above should be useful for anyone looking to measure the tdp of a heatsink and how much heat it can handle and what temps u should be expecting. therefore, this negates any newbie mistakes of using an inappropriate heatsink for a video card and ending up with an even worse overheating video card!

let me tell u all about it. recently, i bought a few video cards off junkbay that came with third party heatsinks. thing is, they all had inappropriate heatsinks for the tdp required. i had a 5870 with arctic twin turbo pro. arctic lists the max tdp handling capacity of the twin turbo pro as 125w and the 5870 has a tdp of around 200w! the video card turned out faulty when i tested it and i had to file for a refund.

then i got another 5870, this one has a zalman vf1000. i put a zalman vf1000 on my 6800 ultra before and it struggled to contain the 75w tdp of the 6800 ultra. i dont need to tell u what happens when u try a 200w heat load on a heatsink that can barely handle 75w! i also found that the plastic washers on the back of the card to insulate the pcb from shorting on the spring loaded screws had melted and fused onto the card's pcb. i wonder how hot the gpu ran to cause the plastic washers to melt like that. over 100°C?

then i got an asus 4870 dark knight. the dark knight cooler has 4 heatpipes and its design is almost exactly like the vf1000 except upside down, so its an inverted vf1000. the 4870 has a tdp of 150w. no way that dark knight cooler can handle that when the similarly performing vf1000 struggled with 75w tdp.

i find it mind boggling when ppl try to supposedly "improve" the cooling of their video cards but have no idea what they are doing and make the cooling worse instead than even with the stock cooler! so i hope some grey matter use with mathematics and practical physics above, teaches ppl to use the right cooler for your video card!

I was getting over 65°C on the core (under full load). notice the darkening of the PCB on the back side under the GPU chip? this card came out of a very dusty OEM PC

the 65°C u got isnt hot enough to cause the pcb to darken. rather its the most likely cramped oem casing and poorly ventilated oem pc case plus the combined effect of the dust that led to the gpu running even hotter than the 65°C u got. i estimate the internal case temps of the gpu running at max load hit over 80°C maybe even close to 90°C over a prolonged period. thats what caused the pcb to darken.

i highly doubt that gpu will last long. it might need a reflow soon. tho with the 50°C load temps now with the new heatsink, the gpu will be lovin' u for givin' it a chill from all the stuffiness it experienced with its previous owner!

I also had some Shin-Etsu thermal compound that I wasn't sure if it was good or not, as it looked rather hard to spread (thick, but not dry).

i have used shin etsu before and i have both the x23 7783d and g751 pastes and yes they are thick and i had to spread them with a spatula. the reason for that is that it has to be thick enough to iron out all the microscopic air bubbles to improve the heat conductivity or sumthing since air is an insulator, not a conductor of heat. i could be wrong tho on why it has to be thick to be good, so dont quote me on that.

i practice my tim application technique by practicing spreading cold butter on bread with a metal butter knife. kills two birds with one stone... or in this case, feeds two mouths with one toast.

I've played around with these small coolers quite a bit, I can tell you from experience that they are not capable of more than 30 Watts TDP max.

well dont toss em just yet. they should do well for cooling the entry level series of each generation from the 2000 to 5000 series all of which have 20-25w tdp. e.g. 2400 pro/xt, 3450/3470, 4350 and 5450. all of em should have 43mm square mounting hole pattern which fit the cooler. in fact, that 4650 stock cooler should work nice in replacing the stock passive heatsink of my 5450. u should sell that 4650 stock cooler to me! lol!

also, if u dont mind and have the time, u should try that 4650 stock cooler on the dell 2400 xt instead. i wonder how it would perform. i theorise it should perform great! not sure what the ambient temp was when u tested the 4650 stock cooler and got that 65°C load temp result but i am going to ass-u-me the ambient temp is 29°C. therefore, 65-29=36°C delta above ambient. since the 2400 xt has 2.5x lower tdp than the 4650, 36÷2.5=14.4°C delta above ambient. with an ambient temp of 29°C again, we get 14.4+29=43.4°C. so u should get around 43°C load temps give or take a couple of degrees with that 4650 stock cooler on your dell 2400 xt. well, if u have some spare time, that would be an interesting experiment to try out! give it a go and lemme see the results! thanks! hehehe...

[brethin]

Re: Your Video Card Cooling Mods

LOL killing low profile cards must be the new thing.

[momaka]

Re: Your Video Card Cooling Mods

LOL killing low profile cards must be the new thing.

When manufactures are too lazy and/or too cheap to put a proper heatsink ... yeah, I'll put whatever I have that is better than their crap. I don't have any low-profile PCs, so I frankly couldn't give a crap about low-profile stuff. Besides, if you are stuffing a 50W+ video card in a small shoebox PC, you're doing it wrong. Most small SFF/USFF PCs already run way too hot inside. They don't need more heat. So a mid-range low-profile video card is a silly idea anyways... unless it really has a properly designed heatsink that exhausts all of the heat outside of the case - and those usually don't come in anything less than a dual-slot and full-profile.

well, i'd just like to affirm that your testing methodology is sound on both the 2400 xt and 4650.
...
the 65°C u got isnt hot enough to cause the pcb to darken. rather its the most likely cramped oem casing and poorly ventilated oem pc case plus the combined effect of the dust that led to the gpu running even hotter than the 65°C u got. i estimate the internal case temps of the gpu running at max load hit over 80°C maybe even close to 90°C over a prolonged period. thats what caused the pcb to darken.

Yes, 65°C isn't hot enough for that. But this was the temperature I was reaching only after a few _minutes_ of medium-high load... and at 18°C room temperature with _open_ case. Since GPU temperature rises proportionately with ambient case/room air temperature, @ 28°C (normal temperature in my computer room for the summer), I expect that stock cooler to easily hit 75°C and possibly leveling off around 80-85°C or higher with all of the dust it had - which is all hot enough to darken a PCB. (And for the record, I cleaned the stock cooler's fins from dust prior to testing it, so the overheating clearly wasn't from dust accumulation alone).

Meanwhile, this modded cooler keeps the card at around 50°C with the same 28°C room/case ambient.

I guess as classic car guys like to say: there's no replacement for displacement. In the case of cooling, there's no replacement for surface area (with some airflow, of course).

let me tell u all about it. recently, i bought a few video cards off junkbay that came with third party heatsinks. thing is, they all had inappropriate heatsinks for the tdp required. i had a 5870 with arctic twin turbo pro. arctic lists the max tdp handling capacity of the twin turbo pro as 125w and the 5870 has a tdp of around 200w! the video card turned out faulty when i tested it and i had to file for a refund.

Not surprised.

Whenever I'm bored, I browse for cheap video cards on eBay. I've been doing that for a few years now, and I've noted that there are video cards with certain coolers that are almost regularly listed for parts or as broken. I saw a few like my Sapphire above as well, so no doubt the stock HS is poorly matched for it. Other regulars are the single-slotted reference HD4850 cards, single-slotted GeForce 8800/9800 GT (which is over 90% what they come with), and GTX 560/Ti with single fan... that's just to name a few off top of my head. There are many more in this list. Most artifacted cards are due to crappy cooling.

i dont need to tell u what happens when u try a 200w heat load on a heatsink that can barely handle 75w! i also found that the plastic washers on the back of the card to insulate the pcb from shorting on the spring loaded screws had melted and fused onto the card's pcb. i wonder how hot the gpu ran to cause the plastic washers to melt like that. over 100°C?

Hey, it's a self-reflowing video card! How neat.
The feature doubles as self-drying/baking to remove moisture from air humidity. Awesome!

i have used shin etsu before and i have both the x23 7783d and g751 pastes and yes they are thick and i had to spread them with a spatula.

Thick pastes, I don't spread. I find that the heatsink pressure does a better job at that while pushing out all of the air bubbles indeed. But this only works on GPUs and CPUs with exposed dies. Applying thick paste on a CPU/GPU with a large heat spreader tends to yield not great results, IME. So for those, I use a less viscous (runny/watery) compound.

well dont toss em just yet. they should do well for cooling the entry level series of each generation from the 2000 to 5000 series all of which have 20-25w tdp. e.g. 2400 pro/xt, 3450/3470, 4350 and 5450. all of em should have 43mm square mounting hole pattern which fit the cooler. in fact, that 4650 stock cooler should work nice in replacing the stock passive heatsink of my 5450. u should sell that 4650 stock cooler to me! lol!

also, if u dont mind and have the time, u should try that 4650 stock cooler on the dell 2400 xt instead. i wonder how it would perform. i theorise it should perform great! not sure what the ambient temp was when u tested the 4650 stock cooler and got that 65°C load temp result but i am going to ass-u-me the ambient temp is 29°C. therefore, 65-29=36°C delta above ambient. since the 2400 xt has 2.5x lower tdp than the 4650, 36÷2.5=14.4°C delta above ambient. with an ambient temp of 29°C again, we get 14.4+29=43.4°C. so u should get around 43°C load temps give or take a couple of degrees with that 4650 stock cooler on your dell 2400 xt. well, if u have some spare time, that would be an interesting experiment to try out! give it a go and lemme see the results! thanks! hehehe...

Already ahead of you there

Though I didn't use the Sapphire HD4650's stock cooler in particular, I did reuse a low-profile PowerColor Radeon HD6670 cooler on one of my HD2400 XT cards. I just didn't take pictures of it, as I didn't think it was that interesting of a card/mod, given how saturated the online market is with used HD2400, 3450, and 6450 cards (they are literally sold by the bucket). In any case, the PowerColor HD6670 low-profile stock cooler is about the same size as the HD2400's reference cooler. But it has much more fins and much thicker base, similar to the Sapphire HD4650 (but still quite inadequate for a 60W TDP). So the temperatures were much lower on the HD2400 - actually very very close to what you calculated/estimated. I don't remember exactly, but I think I was getting indeed around 40-43°C load with 28°C room ambient and closed case. Stock HD2400 HS in that same case with same room temperature was giving around 60-65°C under full load. I'll take some pictures when I take it out for replacement maybe this or next month and also make some graphs. But yeah, that one worked quite well.

And for brethin's sake, I didn't kill a low-profile card with that mod.

[ChaosLegionnaire]

Re: Your Video Card Cooling Mods

another candidate up for modding is a sapphire radeon 8500le 64mb. i bid on it on junkbay cuz those gf4 ti cards just have overinflated prices! before the seller shipped the item, he said that while testing the item before shipping out, there was some noise from the fan and that he included a second fan in the package.

when i got it and tried testing it on the abit ic7-g, i found that the fan had partially seized with a dry bearing most likely from the noise it made tho it was still spinning very slowly. i still had my dual 80mm fans blowing at it to keep it cool. i tested it in win98 by forcing the driver to see it as a radeon 9000 series card (because the omega drivers i was using no longer support the 8500 series) and it worked fine in many games i tested it with.

i tried removing the fan to lube and clean the bearing but the stator broke apart while i was trying to remove the rotors. grrr! the screw holes for the spare fan the seller provided dont line up with the original fan. there is no way to remove the heatsink because it seems to have been glued down onto the gpu with thermal epoxy. so the only way is to mod the heatsink to attach a more powerful fan onto it.

after cracking my head and having a headache from it for over a week! i finally had an eureka moment. i decided to use cable ties and to use the stator frame of the original fan as an anchor point to secure the cable ties onto. i had to run two cable ties in a "U" manner under the stator frame and then screwed the stator frame back onto the heatsink with the original screws to secure the cable ties onto the heatsink.

took me a few tries to get it right. then i just ran the cable ties sticking up from the heatsink through the mounting holes of the 40mm fan the seller gave me for free. used one of the mini 2-pin to 3-pin fan adaptors i got off aliexpress to connect the fan to the video card fan header.

i ran some game tests with dx8 games to see how hot it ran and if the mod cools well. i also had the usual 140mm fan blowing air over the pci and agp slot area. measuring the temperature of the rear pcb side of the gpu with the scythe ir thermometer, i got 53-54°C while gaming. so this means the gpu core probably runs at around 60°C at max gaming load. so i wouldnt say it cools well but this is the best it can do since i cant replace the heatsink as its glued down with thermal epoxy.

so anyone have any ideas for removing and weakening thermal epoxy? does acetone work?

[momaka]

Re: Your Video Card Cooling Mods

so anyone have any ideas for removing and weakening thermal epoxy? does acetone work?

No ideas, but I do know temperatures below freezing weaken the epoxy, though. In the past, I've removed the heat spreaders on PS3 GPU chips just with heating the heat spreader to around 60-80C and using brute force to pry it off. But that's always a bit risky.

Anyways, I do have a Radeon 7500 AGP video card with what appears to be the same heatsink as yours (in fact, the whole PCB appears to be very similar to your card with a few minor differences.) And on mine, the fan still works (though it was very noisy and almost stuck when I got it), but I do plan on changing the heatsink on it, just for the sake of having a more quiet card and maybe more reliable too when I use a different fan. I'm thinking to use the same HS as the ASUS V3800M I put in post #2 of this thread.

[momaka]

MSI GeForce GT 430 (N430GT-MD1GD3/OC/LP)

It’s been a somewhat long time since I posted a cooler mod. So let’s do this.

Here is an MSI GeForce GT430 (model N430GT-MD1GD3/OC/LP) video card I found in the recycle bin at a former workplace.


It came with a completely seized fan (50 mm PowerLogic PLD05010S12L). Some of you may have already spotted the glue/adhesive on the fan rotor hub, thinking that this might have something to do with the problem. But really, this is just me taking the picture after I attempted to repair the fan. I say “attempted”, because the fan was one of those sealed sleeve bearing types. Long story short, I did repair it. However, I had to bust its shaft to do that (at least that seemed like the best way to open it at the time - now I have a new and less destructive method… but that’s for another thread perhaps.)

Anyways… regarding the stock heatsink/fan on the MSI GT430 - it was barely enough to keep the card cool. With 70°F / 21°C room temperature, the GPU core load temps were about 53-54°C… and that’s with an under-clocked core of 670 MHz (which I did to keep the GPU stable – more on that below.)

As my repaired fan started getting too wobbly and noisy again (the repair lasted only about a month of daily use ), I knew it was time to replace it for something “better” – even if temporarily. That’s when this “mod” (read: ghetto handiwork) was born:


Yes, folks, that’s half of an Xbox 360 fan! ghetto: Its “other half” has been married to another video card for many years now (a GeForce 7600 GS AGP), and still working!

The Xbox 360 fans are a little noisy, but they push quite a bit of air and have decent static pressure. With that said, it was clear I wasn’t going to run it full tilt @ 12V. Instead, I ran it on 7V (fan wired between 12V and 5V rails for + and GND, respectively.) This actually kept the fan quieter than the original 50 mm PowerLogic fan (which, according to the GPU BIOS, should always be running at 50%, except in cases of elevated temperature, possibly.)

So how much of a temperature improvement did I get with this 70 mm Xbox fan?
- About 5-6°C, meaning my max load temps were now around 48°C.
Overall, I don’t consider this a very significant improvement. However, given that I was still using the stock Aluminum heatsink, this isn’t bad either. The GT430 is roughly rated for 40-45 Watts TDP. So in all honesty, I expected worse (especially with the stock fan) and was pleasantly surprised to see the temperatures above.

Now, I can’t tell you if this mod will be permanent or if I will replace the heatsink as well. 48°C seems all fine with 21°C room temperature. But in the summer, of course, I expect my room to hit 28-29°C as usual (and peak up to 30°C). This means a 7 to 8°C rise in ambient temperature. As such, I expect the load temperatures to rise by a proportional amount – i.e. possibly hitting ~55°C or more. If more, I will likely mod it. I will also need to see how the card does in terms of stability at those elevated temperatures. Again, this card seems to be stable enough when under-clocked to 670 MHz core (sometimes still crashes after several hours of gaming sessions… which isn’t too bad IMO.), but certainly not at the stock 785 MHz core (30 minutes, typically).

Speaking of which here is a GPU-Z screen of the stock clocks.

It’s interesting to note that this is probably the highest-clocked GT430 video card I have seen anywhere. Most other GT430 cards are set to run at 700 MHz stock, with very few over-clocked to 730 MHz. (Also, is the number of ROPs and TMU's correct above? Techpowerup suggests this card has only 4 ROPs. But the Quadro 600 -equivalent has 8 ROPs.)

Perhaps MSI pushed the clocks too far, and now after a few years, the GPU chip has degraded in such a way that it can’t overclock anymore. Or perhaps not, seeing how I still sometimes get crashes even at 670 MHz core (it just takes a lot more time and loading for that to happen – as stated, at least 3-4 hours of gaming typically.) I guess it's also possible that when the original fan failed, the GPU chip overheated, and that somehow is now affecting the stock core clocks – i.e. a failing GPU chip? Ah well, it is a FERMI GPU, after all – got to expect that to happen sooner or later. This is why it’s important to keep your graphics cards cool.

[momaka]

BFG GeForce 7300 GT AGP [BFGR73512GT]

I’m back!
Been busy lately with my new job and also spending most of my free time building up computers with some of the junk I’ve accumulated through the years. But none of that has stopped me from still doing the occasional video card cooler mod.

For today’s post, I have a BGF GeForce 7300 GT, model BFGR73512GT. There’s nothing special about such an old card… except note that this is an AGP video card. I don’t expect most folks to be following up on “trends” with old “retro” PC hardware (I don’t really either), but the value of AGP video cards has steadily gone up in the last few years – especially the more powerful cards (and especially VooDoo cards, if you haven’t noticed. )

So, when I saw this card on eBay for $10 (including shipping), I couldn’t help but snatch it up. This is what I ended up getting:



https://www.badcaps.net/forum/attach...1&d=1565533885

Now, does anyone remember that GeForce 8400 GS PCI card I showed on the last page? If yes, do you see resemblance anywhere? I’ll give you a hint: look in the center of the card.
.
.
.
- The above 7300 GT uses the same crappy cooler as that 8400 GS card with a tiny 40 mm screamer fan. At least the fan wasn’t seized this time, though. In any case, the cooler on this BFG 7300 GT wasn’t adequate to keep the card cool at normal room temperatures, just like the 8400 GS. And here is proof of that:

At 20°C (68°F) room, the GPU core temperature under full load hit 57°C and would have kept going up if I didn’t stop the testing. The 8400 GS did much worse, of course, because it has a higher TDP, IIRC (maybe 5-10 Watts more.)

On that note, I’d guesstimate the G73 core is probably rated for about 25-40W, depending on clock speed (maybe closer to 25-30W for the 7300 GT). The G98 of the 8400 GS is probably at least 30W. Whatever the case, that small finless U-shaped “aluminum bracket” cooler above is certainly not adequate for either card. And you know what this means here: it’s cooler modding time!

Step 1: remove the old cooler and clean the GPU core.
https://www.badcaps.net/forum/attach...1&d=1565533885

Closer look of the G73 core:
https://www.badcaps.net/forum/attach...1&d=1565533885
It’s a pretty small die, so be careful if doing your own home-brew coolers. It’s not hard to crack these at all.

Step 2: make/put on a new heatsink/cooler.
You might be thinking this is going to be yet another cut-up, Frankenstein-ed Xbox 360 heatsink… but, nope! Instead, I decided to try something different and re-used the stock heatsink that came with this XFX GeForce 8600 GT video card that I recapped last year (complete post here.) I figured the TDP of the 7300 GT GPU should be low enough for the stock 8600 GT cooler to handle. After all, regular GeForce 8600 GT cards have a TDP of 44W, and the XFX version above is ~60W TDP with the factory OC. Thus, it *seemed* logical that the 8600 GT cooler should be able to handle at least 30W of TDP. Moreover, the cooler from the XFX 8600 GT also matched the stencil outline drawn on the 7300 PCB, as did its screw pattern. So this cooler seemed like a perfect fit every which way… well, almost:
https://www.badcaps.net/forum/attach...1&d=1565533885
I had to do a bit of cutting/drilling for the GPU’s crystal reference clock to fit. But other than that, it did fit well.

Up next, I put on a 50 mm fan that I had from another XFX 8600 GT (the first one that I recapped.) Of course, I knew this 50 mm fan would be very loud at 12V (and likely not last very long), so I attached a resistor in series with it. I don’t remember what resistance exactly, but I think 100 Ohms. All I remember is that the fan was running on ~6V. Here is the “new” cooler after these mods.
https://www.badcaps.net/forum/attach...1&d=1565533885
https://www.badcaps.net/forum/attach...1&d=1565533885

Step 3: Test your handiwork!
In terms of noise, this modded cooler was absolutely silent. Meanwhile, the cooling did improve a bit, but not as much as I would have liked to see. Temperature graphs:

The maximum GPU core temperature peaked around 50-52°C after a long session of testing (30 minutes or so) with the same room temperature. Compare that to 57°C with the original cooler, and this was with only a few minutes of loading. More than likely, that stock cooler would have gone a bit over 60°C if kept at 100% load long enough. Thus, this mod not only lowered GPU temperature by 5°C minimum, but the 50 mm fan running at 6V made the card pretty much silent.

Of course, it should be noted here that 20°C / 68°F is a fairly cool room temperature (at least for my twig structure .) “Normal” room temperature (at least according to most science and physics books even today) is considered to be 25°C (77°F) and it’s what most hardware is rated for anyways.

Since temperature of the cooler rises proportionately with the room temperature – that is, for a set fan speed… at 25°C, even my cooler mod above would make the GPU get a bit hot. That means, an increase in 5°C of Ta will result in 5°C rise for the GPU (i.e. I’d get around ~56°C.) As a matter of fact, I let room temperature go as hot as 28-29°C in the summer. So with that kind of temperature, I expected even my cooler mod would have a hard time keeping the GPU under the 60°C mark. Thus, it seems like more work is needed on my moded cooler. The first thing I thought of was to put on a fan shroud, so that all of the air from the fan actually goes towards the heatsink and not elsewhere. If you remember from that 8600 GT thread above, this third card did not come with a fan shroud on the cooler. As such, I made my own. Using cardboard, I quickly put this together:


And here it is on the cooler:
https://www.badcaps.net/forum/attach...1&d=1565534244
Not very pretty, but it does the job.

This helped lower the temperature by 1-2°C under load. However, I still wasn’t satisfied. So I removed the series resistor for the fan and instead hooked up the fan directly to 7V (by grounding the negative lead of the fan to the 5V rail.) I couldn’t find a convenient test point for the 5V rail, so instead I removed the entire fan connector and had it connect as follows:


Final look on the fan mod…


This yielded the following SpeedFan GPU temperatures:

… which were still not great, considering the GPU again reached close to 57-58°C after a few minutes @ 100% load and would have likely gone over 60°C if left running over 10-20 minutes. However, note here that the above tests were completed at 82°F / ~27.7°C room temperature, which is considerably hotter (about 7°C) than before. With less demanding games that didn’t stress the GPU to its absolute maximum, I as getting closer to 52-53°C on the core.

So overall, this cooler is a decent improvement over the stock one. Despite being used on the 8600 GT (a 47-60W TDP card), I still find it wasn't adequate for this ~30W TDP card. On the plus side, it still only takes a single slot (which is rather rare with my mods. ) And, I was finally able to reuse some of these stock cooler parts that otherwise keep accumulating in my junk boxes.

I just wish this card had implemented some form of fan control that would allow me to crank the fan under heavy load and then go back down to 6-7V in idle mode. Unfortunately, there is no missing or empty circuit on this card that I can fill in to enable this – original fan connector is simply hard-wired to 12V. I’m sure that 50 mm fan would improve the cooling quite a bit at 12V and likely keep the GPU under 60°C with the abovementioned room temperature. But it would be loud (I actually tested it before) and it certainly won’t last as long.

Anyways, I’m not all that worried, though, as I still haven’t decided in what PC this card will go in. So with that said, this cooler mod should suffice for now. If I do any revisions on it, I’ll update this thread.

[momaka]

BFG GeForce 7300 GT AGP [BFGR73512GT] - GPU-Z

Oh, and I forgot the GPU-Z screenshot for the 7300 GT above:


The GeForce 7300 GT isn't exactly a powerful card, but it's not weak either. In practice, it offers about the same performance more or less as a GeForce 6600 GT, GeForce 6800 XT, and Radeon 9700 non-Pro (depending on the games and 3D features used in the rendering, of course.) Even the 7600 GS is not really that much faster, and that's considered a “mid-range” card (at least was for its time.) This shouldn't be surprising, considering this card uses the same G73 GPU core as the 7600 GS and GT cards. Only difference is the 7600 cards have 12 pixel and 5 vertex shaders, whereas the 7300 GT has only 8 and 4 (same as a GeForce 6800 XT and Radeon 9700/9800). Otherwise, ROP and TMU count is the same as the 7600 cards with 8 and 12 respectively. As for the RAM: it's 128-bit bus. Funny thing is, my 7600 GS AGP actually uses slower RAM than the “Apogee” –branded chips on this 7300 GT (270 MHz vs. 333 MHz). Go figure. However, the higher pixel and shader count of the 7600 GS is what makes it perform slightly better. As for the core clock: the 7300 GT is only 50 MHz slower than the 7600 GS. On that note, since this BFG GeForce 7300 GT uses the same exact PCB as my PNY GeForce 7600 GS card, I had no problems overclocking the core to 7600 GS speeds (400 MHz.) With a better cooler, it might even do 7600 GT core speeds (540 MHz.) From my short experiments with OC on this card, however, performance hardly improved at all when I bumped the GPU core to 7600 GS levels. And I'm not sure the RAM will take much OC either, as it already runs rather warm. Thus, the gains from OCing this card didn't seem to matter much. Nevertheless, the 7300 GT is still a very decent AGP card to have around. Should play early and most mid-2000's games at medium-high settings with decent FPS. Probably best matched with a Windows XP machine running on a mid-high-end Athlon XP or Pentium 4. I tested mine on a socket 939 PC with an Athlon 64 3200+ OCed to 2.5 GHz, and the CPU was almost never the bottleneck.

[momaka]

Gigabyte Radeon HD6850 1GD5 [GV-R685OC-1GD] – Fan Issue

Another day, another video card from eBay – this time a Gigabyte Radeon HD6850 1GD5 with a Windforce dual fan cooler. I got it for real cheap (less than $13 shipped to my door), but that's because it was originally not working, as mentioned in this thread.

Although the repair in the abovementioned thread was a success, there was another issue with the card: it was hard to keep it from overheating. Even before repairing the video card, I noticed that one of the fans never turned and the other turned just barely. I thought that was an intentional feature programmed into the card's BIOS fan control. But even after flashing the BIOS with my own fan control curves, the fans still didn't seem like they were turning faster. Thus, I figured it's time to clean the heatsink and change the thermal compound. After taking the HS off, I noticed there was quite a bit of thermal compound applied on the GPU core, suggesting the previous owner did likely change the thermal compound. The TC felt very much like MX2 or MX4 and was quite fresh, so I scooped up all the excess for re-use. (Yes, TC can be re-used if it hasn't dried or degraded from extended exposure to high temperature! Don't fall for those myths suggested by gaming “experts” that thermal compound needs to be changed once every few months to a year, because it “goes bad”. In most normal circumstances, it doesn't.) I then gave the HS a wash with hot water and soap to get rid of all the dust in the fins. Pictures after the cleaning:
https://www.badcaps.net/forum/attach...1&d=1576468110
https://www.badcaps.net/forum/attach...1&d=1576468110
https://www.badcaps.net/forum/attach...1&d=1576468110

Also, a few shots of the PCB with the cooler removed:
https://www.badcaps.net/forum/attach...1&d=1576468110
https://www.badcaps.net/forum/attach...1&d=1576468110
https://www.badcaps.net/forum/attach...1&d=1576468110

Despite this, I still kept getting high temperatures – even with my custom fan curve in MSI Afterburner. With 50-70% GPU load (gaming), the GPU core was already running at 52-54°C. As such, the memory IO area of the GPU was touching or lingering around 60°C. A few times, I forgot to turn on my MSI Afterburner software, which let the card temperature jump to a blistering 72°C on the GPU core and 85°C on the mem IO area. Ouch! My flashed BIOS didn't seem to want to work any better than the stock either.

Over time, the issue got even worse. Even at 100% PWM for the fans, they seemed to turn slower than before. This didn't make any sense – especially how one of the fans was not turning at startup and turning much slower than the other at 100% PWM. After all, both fans are identical (save for the fact that one has an RPM monitoring wire) with the same part number. And both have their ground and Vcc pins connected together, suggesting that both should turn with the same speed. But I dismissed this sign at first.

Instead, I started investigating the fan control circuit. As expected, the ground pin is connected straight to ground and the positive Vcc pin for the fans is connected to a transistor to vary the speed. What wasn't expected: the RPM monitoring and Vcc pins on the connector of the fans were swapped (in reverse) to what they should be on the board! A-ha! So that's the issue. This means both fans were being fed power through their RPM monitoring pins. The fact that the fans turned at all with this connection is startling. I'd say these Everflow fans are quite good to still work with abuse like that.

Upon testing the fans individually with a 12V external source, I heard them roar like I never did before when they were connected to the GPU. Even when I fed them 5V from an external power adapter, they both spun with the same speed and were still blowing quite strong. So what gives? – I suspect the previous owner pulled the fan connector with the plastic connector guide on the board when changing the thermal compound and then installed it backwards by mistake.

Thus, I swapped the fan wires on the fan connector and plugged the card back into the PC. Results: no more overheating! The card ran nice and cool. With the same room temperatures as before, the GPU core hardly went over 45-46°C, and memIO area didn't go above 52°C. Even without my custom MSI Afterburner fan curves, the GPU temperatures didn't go more than 1-2 degrees up. The only downside is that the fans are much louder now – I can actually hear them. Before? – No.

So I suppose this Windforce cooler is capable of cooling the card, after all. Originally, I thought it wasn't. But after fixing the fans, now it's working quite well. Maybe at 100% GPU load, things might get a bit hot as this is a ~120W TDP GPU and the HS looks like it may handle only about 100W with those two thin heatpipes. Nevertheless, I don't think I'll be doing any cooling mods to the heatsink of this card. At most, I might re-visit my custom BIOS again and change the fan curves to reflect the fixed fan wiring issue with the now properly working fan control. Other than that, this Windforce dual fan cooler seems to be good enough for now. I'm going to leave a stock cooler alone, for once, lol.

[momaka]

ATI Radeon HD2400 Pro/XT OEM

Promised I was going to post this one a way while back: Radeon HD2400 GPU with a PowerColor Radeon HD6670 low-profile cooler. Better late than never, right?
(And judging by the lack of interest in this thread, I don't think it really matters. )

In particular, the ATI Radeon HD2400 Pro/XT video card I have is an OEM low-profile version with DMS-59 output.





These cards are quite irrelevant today. Although they do support DirectX 10, they are nothing more than a very basic 2D display adapter for a modern PC (no proper H.264 video acceleration, no HDMI audio pass-through, and laughable gaming performance, just to say the least.)

However, this was one of the first video cards I used for my PC builds when I started moving onto motherboards with PCI-E. At the time, I just didn’t want to spend much money. Also, good PCI-E cards not only cost more but ran hotter too. So what really sold me this card was the 19 Watt TDP rating. I didn’t really have that many high-power PSUs at the time either, so this seemed like a universal card I could use with any build (and I did.)

The other thing that sold me was that I found an auction on eBay at the time for 5 of these cards for $10 shipped - just a few $ more than if I was to buy only one of these cards. Of course, there was a reason why they were so cheap: DMS-59 output connector that requires a breakout cable for either D-sub or DVI. Most of the time, the DMS-59 breakout cable is not included and often costs almost as much as the card itself. But I found another sub-$12-total auction that included 3x DMS-59 to dual D-sub cables and… more of these low-end OEM low-profile cards in the package (2x HD2400 Pro and one X1300/1550 - all 256 MB versions.)

Anyways, back story aside, I’ve been using these HD2400 GPUs quite a bit. However, just because their TDP is 19 Watts doesn’t mean that they actually run cool under load. I only found this after a few years of using the cards.


With room temperatures around 27°C / 81°F, these OEM HD2400 GPUs easily get their cores to reach 65°C at high load. The worst I got was 67-68°C with 28°C / 82-83°F room temperatures in a PC case.

Not that it would be a big loss if one of these died on me, given how many I have and their value (or lack thereof). Still, it didn’t hurt to try to improve their cooling - at least for fun, if nothing else.

So, move forward a few years after I had started my GPU collection and restoration hobby. I had accumulated quite a few different stock coolers from dead and upgraded cards.

In case of this HD2400 Pro/XT, I had a similar-looking (same profile, fan size, and PCB footprint, actually), but much better designed cooler that came from a low-profile PowerColor Radeon HD6670 card I got for farts/repair. After managing to get the HD6670 working with a reflow, it was clear its stock cooler was not doing its job well enough (more material for this thread someday ). Because the HD6670 uses the same heatsink hole pattern as the HD2400 (as do most low-end and mid-end Radeon HD and X series video cards), the HD6670 cooler was a perfect fit on the HD2400.



So how did that affect the temperatures? - Have a look at this temperature graph.


Basically, with the same room temperature (27-28°C / 81-83°F ), the GPU core temps maxed out at 42-43°C with 100% GPU load, even after hours of loading. That’s over a 20°C drop in temperatures! And from what I observed, the stock fan on the HD6670 heatsink did not run faster than the stock fan on the HD2400 heatsink (both being 50 mm fans), nor did the BIOS make it go faster under load (it ran at 13% PWM under full load and mostly off when the GPU was idle.) Kind of amazing what few mm of extra aluminum on the base thickness can do for the temperatures. Of course, I’m sure the extra cooling fins (approx. 2x more) helped too. Just as a comparison, here you can see the thickness of the stock HD2400 cooler (along with stock pink “bubble gum” thermal compound.)
https://www.badcaps.net/forum/attach...1&d=1602288323

So with the HD6670 cooler, the 50 mm fan has to do a lot less work - which is good, because this is the main reason the HD6670 died – the stock HD6670 heatsink was too small, which made its BIOS crank up the fan speed to keep itself cool, and that led to the fan failing/seizing. What’s even more annoying is that the 50 mm fans that typically come with these cards are sealed sleeve bearing type. Fortunately, I have figured ways now to open these (the stuck fan on the HD6670 was one of the first I experimented on.) The pictures of the HD6670 cooler you saw is with the repaired 50 mm fan. It’s not 100% perfect (had to cut through the back to service it), but it does work OK now. The HD2400 card with this modded cooler has been installed in one of my GP PCs for 2 years now and I usually use it in the hottest months of the summer. So far, all is well.

Anyways, folks, that’s all I have for this old, slow GPU. It still serves me well, along with the others when I’m setting up PCs. So as Garage54 would say, this was a massive success, lol.

[ChaosLegionnaire]

Re: ATI Radeon HD2400 Pro/XT OEM

they are nothing more than a very basic 2D display adapter for a modern PC (no proper H.264 video acceleration, no HDMI audio pass-through, and laughable gaming performance, just to say the least.)

no. there are some 2400 driver hacks in the registry u can do to enable h264 acceleration. ATI HD Registry Tweaks for HTPCs

however and unfortunately, i only managed to get FHD mpeg-2 acceleration to work. FHD h264 acceleration didnt work for me at least not in mpc home cinema with the mpc dxva h264 decoder.

what really sold me this card was the 19 Watt TDP rating.

based on the clocks in the images u have posted, this is a 2400 xt card which has a slightly higher tdp of 25w. u can just downclock it to 400-500 mhz which is at 2400 pro speeds to keep the temperatures down as well.

What's even more annoying is that the 50 mm fans that typically come with these cards are sealed sleeve bearing type. Fortunately, I have figured ways now to open these (the stuck fan on the HD6670 was one of the first I experimented on.) The pictures of the HD6670 cooler you saw is with the repaired 50 mm fan. It's not 100% perfect (had to cut through the back to service it), but it does work OK now.

or as an easier alternative, just unscrew the fans and swap the fan from one heatsink to the other.

[momaka]

Re: ATI Radeon HD2400 Pro/XT OEM

no. there are some 2400 driver hacks in the registry u can do to enable h264 acceleration. ATI HD Registry Tweaks for HTPCs

however and unfortunately, i only managed to get FHD mpeg-2 acceleration to work. FHD h264 acceleration didnt work for me at least not in mpc home cinema with the mpc dxva h264 decoder.

Yeah, I downloaded those registry tweaks a long time ago.

They didn't seem to make any difference for H.264 acceleration in Youtube or in VLC... so as far as I'm concerned, they haven't been very useful for any of my intended purposes.

based on the clocks in the images u have posted, this is a 2400 xt card which has a slightly higher tdp of 25w. u can just downclock it to 400-500 mhz which is at 2400 pro speeds to keep the temperatures down as well.

Nah, these are slow cards already, I don't see why I should downclock them.

Also, there's no difference in specs for these low-profile cards between the ATI "XT" models and Dell OEM "Pro" models - both run with 650 MHz core and 500 MHz mem. On the other hand, the Pro AGP cards do run with slower clocks... or at least my VisionTek one does (525 MHz core and 390 MHz mem.)

Not that any of that matters, though. With only 64-bit memory, these cards are pretty slow for games, regardless of which variant you get.
They do well under Colin McRae Rally 3 / 04 / 2005, though, if anyone cares about some older rally games to try. In particular, CMR 3 and 04 can be maxed out, pretty much (minus AA, of course.)

or as an easier alternative, just unscrew the fans and swap the fan from one heatsink to the other.

Did that at first, but wanted to reduce / reuse / recycle more ... so the seized fan got fixed and reused.