X1900 Crossfire - Thermal Imaging

M

Mysterae

Gawd
Joined
Sep 11, 2005
Messages
648
I finally managed to borrow a thermal camera (Flir Systems Agema 570) and have been taking a few quick shots of the X1900XTX and X1900 Crossfire to see just how hot they get.

This is a little teaser for you before I can really sit down and get some meaningful images this weekend. Remember to look at the scale on the right when interpreting the images.

thermal_01.jpg



thermal_02.gif



thermal_03.gif



thermal_05.gif



thermal_07.gif


Some reference images so you understand what you are looking at:

stage_79.jpg



stage_82.jpg


P.S. They get hot.
 
That is the mock up of his custom case he will be building out of plexi, go over to case modding and check it out before you make fun of him.

Interesting images.
 
Im not going to mock anybody who has that kind of thermal imaging or anybody who can get those components in that case!
 
nobi125 said:
$1000+ on video cards and < $5 on a case.

Very interesting.
Click to expand...

:D Dude, the wood cost more than that! :rolleyes: Thanks goes to the guys that are with the program!

I am trying to answer a few questions like:

- how effective is my air cooling before water cooling,
- are the pads on the RAM there to absorb heat to the cooler, or to insulate it from heat from the core, or just there as as packing,
- the increase in temps when both cards are overclocked

It's a shame I don't have a zalman vga cooler, I could answer the age old question of ramsinks or naked ram. Academic really as I'm not going to fit one, but still would have been interesting.

Sorry, I had to post and run and have to do so again.
 
Thats pretty damn hot on the back side of the cards but you mean to tell me the VRMs are over 137ºC?

I couldn't find the camera model you list, but does it have active/passive cooling (Liquid N2) or not? The reason I ask is it greatly effects accuracy, if it doesn't use a reference cooling liquid, I wouldn't believe the temps, just the proportions.

Definitely glad I have some cooling all around mine.
 
Sorry I've been away so long!

First, here is the thermal camera I have been using:

thermal_08.jpg


I don't know much about it as it's quite old and is now discontinued, here's some details. When new, this would have set you back £40k, so I'm very lucky to be in the position to borrow it, and yes, I have to give it back! This will come in really useful testing the effectiveness of my water cooling loops in parallel HEX.

Speaking of which, I have spent some time tidying up my cables as it was really killing the airflow from the left crossflow fan. I also cut up the brackets holding the hard disks on as they were getting in the way of the air too.

stage_86.jpg


For this thread I'm concentrating on the heat the video cards produce, off load and on load, at varying crossflow fan speeds. First I'll run the cards as a normal user would, with the only overclock being done through Overdrive in CCC, which only affects the XTX and lifts it to 690/800. Then I'll move on to the overclocking using ATITool and overclock both cards to the limit. I'll use the X3:Reunion benchmark demo as that really stresses the cards out at high resolution and detail and is a lot prettier than 3dmark to look continuously!

Is there anything else you guys can think of?
 
You got a manual with the camera? you should try to calibrate it. I doubt your power cables are 50c hot :eek:
 
Right, where do I start...

After spending some time with the camera and wondering why many temperatures where just plain wrong, I researched a little. I apologise now for getting scientific on your ass in a video forum!

There is a factor I knew about but took for granted, and that is emissivity. Look it up, it's a huge subject and black magic[k] to many. Almost every material has an emissivity value. When I took the shots up above, it was with a fixed emissivity value of 0.44. Now, there are many different materials in the images above; copper, aluminim, plastic, tin, rubber etc. These all have different emissivity values. The camera only lets you set one value, so the values shown are potentially wrong.

Then there is the finish of the components, mat or gloss, dull or reflective. All play a part in reading the results.

I tried a little experiment with a known temperature - boiling water. We all know water boils at 100C and is a good constant. However, the steam produced fubar's the results slightlly, but even so when the water cooled down the temperature reading was still wrong. I entered in the correct emissivity value for water, 0.67, and it brought it in line. Emissivity of the target material is important.

To the amatuer, what the camera is good for is comparing the difference in temperature between multiple objects, or the rise in temperature of an object between states.

So I won't show anymore images, I wouldn't want to be misinformative. However I'm going to continue to use the camera to test the effectiveness of my air cooling and then water cooling. Sorry if I wasted your time, but at least we're all a little more educated!
 
You can calibrate the camera to the surface or you can make the surface a uniform emmissivity. When I have to deal with this problem, I use either some black electrical tape, or a can of aarid extra dry aerosol deodorant. Believe it or not, it will allow for a fairly uniform emmissivity of .92-.95 when sprayed even on reflective surfaces.
Neither of these seem to be desirable for use on a video card, but if you try the aarid extra dry, I'd recommend you sacrifice, say, an MX400 or radeon 9000 to the [H] gods rather than your crossfire setup.

Another thought is to let the card cool to room temperature, then adjust the emmissivity until the component you're interested in (power mosfet, ram, etc) displays the correct temperature. The temperature for that component only will then be correct when the unit is under power.

You bring up a good point about emmissivity. Folks with these $50 temperature "guns" don't generally understand that their gun is #1 giving them a correct value only if the surface is a .95 (nearly black body) emitter, and #2, that at any distance over about 4' from the surface your shooting, you're getting reported the average temperature of a circle roughly 2' in diameter.
 
The chap that wrote the article on emissivity had a good idea for measuring it, and that was paint the object matt black, simillar to your idea of black electrical tape. That gives the entire object the same emissivity, the paint absorbs the energy and emits the temperature differences. Although all over black X1900 would look damn sweet, I don't think I will do it at this time!

Your suggestion on bringing the object to a known temp, in this case room temp, and having the emissivity value of the object entered is a great one. If I knew the values for pcb, anodised aluminium and the material the memory IC's are made from, I could produce some meaningfull results.

Brent, do we see a good idea for your reviews ;)? As I have proved, it can be very complicated and time consuming, but none the less can be done!
 
nobi125 said:
$1000+ on video cards and < $5 on a case.

Very [H]ard.
Click to expand...

You should have see the mod he is doing, it a realy kick ass project in the worklogs, its called parallel HEX, check it out under Case Mods> mod worklogs.
 
lockheed2266 said:
You should have see the mod he is doing, it a realy kick ass project in the worklogs, its called parallel HEX, check it out under Case Mods> mod worklogs.
Click to expand...

Thanks for that, I really should have put a link in the first post, but hey, you can only pimp your own work so much! ;)
 
I would like to see some shots of 7900GTXs for comparison's sake...and I'd laugh if the wooden case caught on fire from the video cards :)
 
Wow I totally forgot about this thread!

Heh.

Thanks for bringing it back. The discussion about emissivity is an interesting one.
 
Nice thermals! :)

I have noticed from my 1900s in CF on air (default sinks) that if I OC (675/1600) the cards for long duration gaming I get aritfacts after an hour or so... not OC'd everything runs great.. I have extra fans blowing right on the cards also.. I think those thermal pads are there to help get heat from the chips to the sink.. not to protect the memory from the core.. otherwise I think they'd have gone with a core only sink.. when there's any OCing going on the cores overheat the memory... those sinks just can't shed heat fast enough.. with the fans on auto. Fans on full blast is not an option. wow! wayyy to loud.

So, I have a set of EK waterblock due in end of this week. While I have never been a fan of full face blocks for fear of the core overheating the memory, but I hear great things about these ones.. there's a lot of copper on these blocks.. they have the extra voltage reg sink add-ons also.. it will be interesting to see what kind of numbers I can hold for long durations on water with these suckers... seems like putting them on water is the only option if you want to do any serious OCing.. not like the 7900GT which was happy with vmods and air with insane OCs.. anyways..
 
revenant said:
for fear of the core overheating the memory
Click to expand...

Just a baseless paranoia...there is no chance the core is going to have any appreciable adverse effect on the cooling of the memory. In fact you should be more worried about the effect that adding in the extra restriction of a full cover block lowers the performance of the waterblock for cooling core (But neither are enough to keep you from using that awesome shiny EK block).
 
Paranioa, perhaps yes.. maybe, more like ignorance for not having done my own testing.. I would not say completely baseless either. I've chatted with a lot of people (who do a lot of extreme OCing) who said for best cooling performance keep the core and memory seperate.. anyways.. these peeps now have these EK blocks (well, one of them, the other has the alphacool) and say they're great. lol. whateva.. the flow path seems pretty good.. for a full face block. so I am hoping for good things.. like with everything else.. if it doesn't work well, I'll try something else.. I reckon.
 
Good luck with the EK blocks, they do look cool, pun intended!

The only downsides I see to the aquacomputer full-face blocks I've used is they are made from alu and there isn't enough metal on the block for the power regs on the card - should be extra fins for a little extra heat dissapation by any passing air.

Although my water cooling is far from the norm, I was getting a max core temp of 72C when at full tilt and oc'd (XTX, the CF was at 67C). I'd expect the EK blocks to better that because of the copper and better water channels, let us know!
 
You must log in or register to reply here.
Back
Top