stealthjoe wrote on 2023-02-15, 04:49:

After switching on the board, a few minutes later noted that the area at the backside of the main IC was quite hot (not sure of the temp but could be around 60-65 deg C) and the heatsink was cold.

It's pretty normal for the backside of a Voodoo 3 to feel very hot to the touch (55 - 60 degrees C would be typical, even with more advanced cooling). At this temperature, you can't keep your finger on the PCB for more than just a few seconds. If the temperature hits 75+ degrees, it could end up being a problem in the long run, but not necessarily.
The problem is the chip itself and its subpar plasticky packaging, so most of the heat remains trapped inside. Had it been a FC-PGA (or FC-BGA, more accurately), the Avenger would've been much easier to cool.

It's worth noting, though, that silicon can withstand and run just fine at temperatures that are 'too hot' for us humans. Just think about it: many of these cards worked for years in ancient cases without any airflow whatsoever and, even worse, the cards themselves being passively cooled. In such conditions the back of the PCB could easily hit 90+ C. 😀

Of course, lower temps never hurt, which is why I also strive to achieve them whenever possible (even with modern parts). All in all, just keep in mind that "quite hot" is not indicative of a problem, you'd need an IR gun to properly see what the actual temperature is.

I haven't actually done this myself, but one option is to put thermal paste in the center of the chip with dabs of super glue in the corners to hold it on. I think this is a common method to reattach a heat spreader on to a CPU, but I'm not sure if it's resilient enough to hold a heatsink on. One of the downsides to this old hardware is that it will happily just sit there and cook itself to death if the heatsink falls off. No throttling, no temperature warnings etc.

Come to think of it, I wonder what would happen if you used good quality copper or aluminum foil tape and then glued the opposite side to the bottom of the heatsink? Then you can just stick the tape side down to the chip. Surely the metallic tape would conduct better than whatever plastic\rubber stuff is used as the structural layer of most double-sided tapes. The adhesive itself would obviously do nothing, but it might still work well enough. Remember, the chips on these things actually have plastic tops so heat transfer is always going to be limited.

I see there is some double sided copper tape on Amazon, but it's probably trash and I wouldn't trust the adhesive.

If you just search for thermally conductive glue there are tons of options out there.

I used to use Arctic Silver thermal epoxy (I think it was Alumina) and it was an amazing product, but you will basically ruin whatever you attach it to, so you need to be absolutely 100% sure that you never ever want the cooler to come off. My first DOS\9x retro PC I built to play old games back in 2001-2003 had an Evergreen Spectra 333 K6-2 (I think) in a board that could otherwise only use 133Mhz Pentium chips. One day the stupid tab on the CPU socket sheared off when I wasn't even using the system (heard a clang and found the cooler laying in the bottom of the tower) so I couldn't install the heatsink for the CPU anymore. Well, genius that I am, I just purchased some fancy arctic thermal epoxy and slathered it in there... voila. No more mounting clip needed. I still have that Spectra to this day, and the heatsink is still firmly attached. I would imagine that the heat spreader would come off of the CPU before the epoxy would come off of the heatspreader. The stuff really turned into metal... it was cold to the touch, shiny and hard. Great stuff... but way too scary for me to use now that these parts are actually valuable.

I have a mac voodoo3 that has no cooler. would be nice to get a passive cooling-solution.
Having the lines on the cooler parallel to the agp-slot is not the best solution.
I really would love such a cooling solution. as always professional stuff, not consumer stuff from MATROX!
https://www.techpowerup.com/gpu-specs/millenium-p650.c3728

You definitely need to add a heat sink and fan because these cards run “Very HOT”.

Copper is the best conductor of heat so I would look for a cooper heat sink.
Maybe From Zalman. Search on eBay for Cooper heat sink.
Then you can just add a fan on top of that.

Here is what I did.
See link:
Re: Ultimate AMD K6-3+@500mhz.

Ozzuneoj wrote on 2023-02-15, 07:15:

I haven't actually done this myself, but one option is to put thermal paste in the center of the chip with dabs of super glue in […]
Show full quote

I haven't actually done this myself, but one option is to put thermal paste in the center of the chip with dabs of super glue in the corners to hold it on. I think this is a common method to reattach a heat spreader on to a CPU, but I'm not sure if it's resilient enough to hold a heatsink on. One of the downsides to this old hardware is that it will happily just sit there and cook itself to death if the heatsink falls off. No throttling, no temperature warnings etc.

Come to think of it, I wonder what would happen if you used good quality copper or aluminum foil tape and then glued the opposite side to the bottom of the heatsink? Then you can just stick the tape side down to the chip. Surely the metallic tape would conduct better than whatever plastic\rubber stuff is used as the structural layer of most double-sided tapes. The adhesive itself would obviously do nothing, but it might still work well enough. Remember, the chips on these things actually have plastic tops so heat transfer is always going to be limited.

I see there is some double sided copper tape on Amazon, but it's probably trash and I wouldn't trust the adhesive.

If you just search for thermally conductive glue there are tons of options out there.

I used to use Arctic Silver thermal epoxy (I think it was Alumina) and it was an amazing product, but you will basically ruin whatever you attach it to, so you need to be absolutely 100% sure that you never ever want the cooler to come off. My first DOS\9x retro PC I built to play old games back in 2001-2003 had an Evergreen Spectra 333 K6-2 (I think) in a board that could otherwise only use 133Mhz Pentium chips. One day the stupid tab on the CPU socket sheared off when I wasn't even using the system (heard a clang and found the cooler laying in the bottom of the tower) so I couldn't install the heatsink for the CPU anymore. Well, genius that I am, I just purchased some fancy arctic thermal epoxy and slathered it in there... voila. No more mounting clip needed. I still have that Spectra to this day, and the heatsink is still firmly attached. I would imagine that the heat spreader would come off of the CPU before the epoxy would come off of the heatspreader. The stuff really turned into metal... it was cold to the touch, shiny and hard. Great stuff... but way too scary for me to use now that these parts are actually valuable.

The thermal paste / super glue in the corners will work for a while but it doesn't hold up to the thermal cycles and the heatsink will eventually pop off.

Sometimes it feels like I'm the only person to use hand tools any more... I've taken one of these, filed the mounting holes to move them inwards a bit so they fit the PCB and mounted it on the backside for extra cooling. No glue necessary, just some reasonably long screws and nuts in place of the plastic standoffs. Buy two and sandwich the board inbetween, maybe? Iirc I needed to use pads on the backside because of components getting in the way, but on the chip a paste should work perfectly.

Or, since the heatsink I linked may be a bit on the small side for a main, you can always use a bigger one, drill holes where needed and cut threads using a tap. Both aluminium and copper are pretty soft and really easy to work with. If you can't find what you need, a little DIY goes a long way

Yep, modification goes a long way. That's what I would do except with a larger heatsink.

The laptop cooler idea mentioned earlier would also work provided there is somewhere to put the mounting holes.

Thank you all for your replies. Now going back to the original question 😀. Which fan orientation do you think is better for this card? Air blowing towards the heatsink or otherwise?

stealthjoe wrote on 2023-02-16, 08:45:

Thank you all for your replies. Now going back to the original question 😀. Which fan orientation do you think is better for this card? Air blowing towards the heatsink or otherwise?

Always blowing towards the heatsink.

stealthjoe wrote on 2023-02-16, 08:45:

Thank you all for your replies. Now going back to the original question 😀. Which fan orientation do you think is better for this card? Air blowing towards the heatsink or otherwise?

Has there ever been a video card that had a different type of cooler? Ever? 😀
So, if you're thinking of reversing the fan to "extract" heat from the heatsink... no, just... don't! 😁

EDIT: appiah4 was faster. 😁

bloodem wrote on 2023-02-16, 08:53:

Has there ever been a video card that had a different type of cooler? Ever? 😀
So, if you're thinking of reversing the fan to "extract" heat from the heatsink... no, just... don't! 😁

EDIT: appiah4 was faster. 😁

Thanks. Agree with your points 😁. Side effects of reading too much literature 😜 😜

bloodem wrote on 2023-02-16, 08:53:

So, if you're thinking of reversing the fan to "extract" heat from the heatsink... no, just... don't! 😁

How much of a difference would it make? Are there any thermal tests of this?

Shponglefan wrote on 2023-02-16, 12:14:

How much of a difference would it make? Are there any thermal tests of this?

Well, the fact that such an approach was never used in the past 2+ decades of GPU releases... pretty much tells the whole story, IMO. 😀

bloodem wrote on 2023-02-16, 08:53:

stealthjoe wrote on 2023-02-16, 08:45:

Thank you all for your replies. Now going back to the original question 😀. Which fan orientation do you think is better for this card? Air blowing towards the heatsink or otherwise?

Has there ever been a video card that had a different type of cooler? Ever? 😀

Yes. And it's even weirder than you'd think.😄
https://www.techspot.com/news/49385-gigabytes … on-hd-7970.html

But really with a tall heatsink it may be beneficial to pull air through instead of pushing. Do you remember CPU coolers like the Kanie Hedgehog that would come with the fan pulling air through in default? Those often tested better in a pull configuration.
Of course on single slot coolers pushing air is usually much more effective though.

bloodem wrote on 2023-02-16, 12:35:

Shponglefan wrote on 2023-02-16, 12:14:

How much of a difference would it make? Are there any thermal tests of this?

Well, the fact that such an approach was never used in the past 2+ decades of GPU releases... pretty much tells the whole story, IMO. 😀

I'd still like to know how much of an impact there would be if the fan direction was reversed. Is it a small impact? Big impact? Situational dependent?

I'm assuming someone has tested this so I'm curious what data is available.

Shponglefan wrote on 2023-02-16, 13:13:

bloodem wrote on 2023-02-16, 12:35:

Shponglefan wrote on 2023-02-16, 12:14:

How much of a difference would it make? Are there any thermal tests of this?

Well, the fact that such an approach was never used in the past 2+ decades of GPU releases... pretty much tells the whole story, IMO. 😀

I'd still like to know how much of an impact there would be if the fan direction was reversed. Is it a small impact? Big impact? Situational dependent?

I'm assuming someone has tested this so I'm curious what data is available.

The difference in temperatures on the heatsink base should be less than 5°C in this situation assuming there's nothing constricting the airflow in either direction.

Shponglefan wrote on 2023-02-16, 13:13:

bloodem wrote on 2023-02-16, 12:35:

Shponglefan wrote on 2023-02-16, 12:14:

How much of a difference would it make? Are there any thermal tests of this?

Well, the fact that such an approach was never used in the past 2+ decades of GPU releases... pretty much tells the whole story, IMO. 😀

I'd still like to know how much of an impact there would be if the fan direction was reversed. Is it a small impact? Big impact? Situational dependent?

I'm assuming someone has tested this so I'm curious what data is available.

I haven't really seen any data on pushing vs pulling air from a very small flat heatsink, but I'm reminded of the PCs from the late 90s that used a plastic duct attached to the power supply fan to pull heat from the CPU... usually a Pentium II. In those situations you're cooling relatively low heat loads that are not only spread over a larger area due to good contact with the metal casing of the SECC cartridge but usually attached to fairly large coolers with long fins and a decent amount of mass (for the time). In other words, lots of metal to pull heat from the chip and lots of surface area to release that heat. Very little CFM or static pressure are needed to move "enough" air to keep a CPU cool in that setup, so it doesn't really matter in that situation how you move the air over the heatsink, even if one way is better.

With a video card like a Voodoo 3 though, we're talking about a very small circuit board with very little thermal mass that is being pushed to it's capacity limit (as shown by all the brown heat damaged Voodoo3 PCBs out there), a plastic chip package that transfers heat poorly, and usually small heatsinks without much surface area. In that situation you really just need more CFM and pressure to get as much air moving over as much of the heatsink as possible. I don't think it's really possible to design a fan with a very strong and focused channel of air coming IN to it, but most are designed to have that air focused going OUT. So, with a fan pulling air from a heatsink, there is no way to ensure that that air is moving over all of the hot areas of the heatsink. It could very well be pulling a good portion of the air from right around the edge of the fan (across only the tips of the heatsink), which will not help as much. So, I can't really think of a situation where it would ever be better to have a fan pulling heat away from a small video card heatsink like this. Of course, pulling heat away with a fan is nearly always going to be better than going totally passive, so if you have to do it for some reason, it's probably fine... just not necessarily better than having the fan go the other way.

The situation is totally different when talking about a tower heatsink or a radiator where the air is being pulled through the heatsink from open air on the other side. Sadly, it seems that those are the only situations anyone has done push\pull comparisons for.

Here's my take on the Voodoo3 3000 AGP ghetto cooling.

Works well, there's a few millimeters difference with the holes so the plastic pins are a bit skewed, but that's absolutely no problem, it's strong and snug.

I would think pushing the air against the card would more likely have added effects of cooling other parts of the card such as the memory as well. It seems less likely pulling the air through that the incoming air would be drawn across the card as much as pushing it might spread it out across the card.

Hard to tell the air could just cycle right around the fan area only, I prefer to put a fan card in the slot below my V3. Like a Vantec SP-FC70-BL or better, that way my whole card gets air.

Here's a push/pull comparison for the Alpha PAL6035 S370/462 CPU cooler: https://www.overclockers.com/alpha-pal-6035-heatsink-review/
In their test configuration with the original (26CFM) fan the cooler would perform 5°C or ~6% (85°C vs 89°C) worse under load when the fan was pushing air through the heatsink. In idle there was no difference.
With only a 20CFM fan it would perform worse under load when pulling air through the heatsink by 1°C or ~1.1% (89°C vs 90°C). Idle was again identical.

There are many factors at play here and cooler design as well as fan design are among the most important but the orientation of the test setup, case and case airflow are also relevant.

Keep in mind that for heat transfer it's irrelevant if air is pushed or pulled in parallel over a flat surface because in the end the same amount of air makes the same amount of contact at the same rate.
But with the air being moved at an 90° angle relative to the HS base its direction has to change by 90° creating turbulence. The fan also generates turbulence. Multiple variables in answering what is better.
Also an axial fan has no air flow directly in its center. Depending on blade design the 'blind spot' can be moved further out or in. That adds another variable to HSF performance: Do I accidentally create an air flow 'blind spot' in the center of the cooler right above the core?
My theory is that the Alpha PAL6035 with the stock fan was able to create more turbulence close to the center of the HS base resulting in lower CPU temps. It no longer managed to do this with the lower CFM fan.

My personal recommendation for OP's V3 DIY heatsink? Mount the fan sideways (or at an angle) so it blows air in parallel to the PCB. This should ensure optimal air flow, especially close to the center of the HS base. It also minimises the amount of hot air being recycled through the heatsink, directly cools the LDO and reduces possible air flow restrictions by cards underneath the fan.
Also cool the backside of the PCB. A significant amount of heat generated on V3 cards is transferred into the PCB copper. There's even a whopping 100 thermal vias ending in a 2.25cm² copper area on the PCB backside. They aren't there by accident and I bet they also connect to the ground plane internally. Hardly anybody ever thinks about those when trying to improve V3 thermals. 😉