pinging FirstofOne (more info on X850xt cooling)

[pigdos]

First, I've discovered that the screwed-in, retaining clip on the back of
my X850xt is not the primary method by which force is applied to the GPU. If
you look closely at the X850xt there are two screws holding the copper
heatsink to the AL body of the assembly. These two screws are, in fact,
spring-loaded and determine (I beleive) how much force is applied to the
GPU. The retaining clip on the back merely holds the fansink assembly in
place.

On another note, I replaced the heat plate assembly on the back of my X850xt
with individual heatsinks. It probably makes some sort of difference because
the heat plate uses thermal pads that are attached to some sort of sticker
which is stuck on the heat plate.

I noticed the thermal pads you wrote about (the ones that interface the
front-side PCB memory to the heatsink). Those pads look so thick that they
must be insulating the memory on the front of the PCB rather than cooling
it, LOL. I'd imagine the reason my memory is incapable of running anywhere
near 600Mhz (which they are rated for) for sustained periods must be because
of these thick thermal pads. I wonder if the X850XT PE's feature
different/better thermal interfaces for their memory/GPU than the X850xt
does.

 

[First of One]

First, I've discovered that the screwed-in, retaining clip on the back of
my X850xt is not the primary method by which force is applied to the GPU.
If you look closely at the X850xt there are two screws holding the copper
heatsink to the AL body of the assembly. These two screws are, in fact,
spring-loaded and determine (I beleive) how much force is applied to the
GPU. The retaining clip on the back merely holds the fansink assembly in
place.

Note that I don't have an X850XT, so I have been working off of photos here:
http://www.hardocp.com/article.html?art=NzYwLDEsLGhlbnRodXNpYXN0
and in particular this photo:
http://www.hardocp.com/image.html?image=MTExMzg0NzA1Mk4wSWtjeTZaYVhfMV8xMl9sLmpwZw==

I see two gold-colored standoffs, to which the retaining clip on the back of
the GPU screws in. There also appears to be two screw tips visible on two
corners of the copper inlay. I didn't know those were spring-loaded. This
means you can independently adjust the pressure applied to the GPU and RAM
chips, bringing the fansink closer to the board without crushing the GPU.

On another note, I replaced the heat plate assembly on the back of my
X850xt with individual heatsinks. It probably makes some sort of
difference because the heat plate uses thermal pads that are attached to
some sort of sticker which is stuck on the heat plate.

Individual RAM sinks will allow you to use a thin layer of thermal adhesive
instead of a pad, improving heat transfer. You may also want to do a 50/50
mix of adhesive and paste to get "removable strength".

Be prepared to see no difference, though, even after modifying the fansink
interface in the front. As bad as a thick pad may be, the plastic RAM
chip package is probably doing more to insulate the silicon. Intel and AMD
went to the flip-chip design in the PIII days. ATi eventually followed suit
with the 9700 GPU. However, RAM is still packaged the old-fashioned way.

I noticed the thermal pads you wrote about (the ones that interface the
front-side PCB memory to the heatsink). Those pads look so thick that they
must be insulating the memory on the front of the PCB rather than cooling
it, LOL.

The pads were probably designed thick as a precaution, in anticipation for
large variation in RAM chip heights after mounting and soldering. If the RAM
chips are all at the same height (check with a steel ruler), you may be able
to replace the pads with paste, and bring the whole heat sink closer.

According to a Loctite engineer I worked with a few years ago, the thermal
resistance of thermal paste is actually 3-4x worse than aluminum, but
still orders of magnitude better than air. The purpose of the interface
material is to fill air gaps (both large and microscopic), nothing more.

I wonder if the X850XT PE's feature different/better thermal interfaces
for their memory/GPU than the X850xt does.

From what I remember, ATi doesn't market an AGP version of the X850XT, only
the XT-PE. The AGP X850XT is an effort from board partners like Sapphire or
Powercolor. If your fansink looks like the one pictured in the HardOCP
review, then the XT and XT-PE have identical cooling.

 

[pigdos]

Speaking of Loctite, First, what would you think of using one of the Loctite
thread fastener compounds for attaching heatsinks? Some of the Loctite
products are designed to be used pre-assembly and others post-assembly but
all of them seem to have high temperature tolerances (somewhere over 100
degrees C).

Another question, I remember you mentioned that instead of increasing RAM
clocks I might be able to adjust the timings. I have some knowledge of this
(I studied memory technologies in college) so I'm assuming I'd have to use
some sort of BIOS editor. The thing is, if I guess wrong and tighten up the
timings too much, would this just make my X850xt into a doorstop?

Hope you're having a good weekend First.

 

[First of One]

Speaking of Loctite, First, what would you think of using one of the
Loctite thread fastener compounds for attaching heatsinks?

Loctite is primarily known for threadlocking, retaining and gasketing
compounds. The company also supplies the electronics industry with thermal
adhesives, TIM pads, die-attach encapsulants, etc., often with the
corresponding automated plant equipment. See:

Some of the Loctite products are designed to be used pre-assembly and
others post-assembly but all of them seem to have high temperature
tolerances (somewhere over 100 degrees C).

Yep, and this is what you want for a thermal adhesive. It wouldn't be nice
if the heat sink fell off at high temps, where it's most needed.

Another question, I remember you mentioned that instead of increasing RAM
clocks I might be able to adjust the timings. I have some knowledge of
this (I studied memory technologies in college) so I'm assuming I'd have
to use some sort of BIOS editor. The thing is, if I guess wrong and
tighten up the timings too much, would this just make my X850xt into a
doorstop?

I'm not sure if you need to touch the BIOS. ATiTool can be used to adjust
the RAM timings. Timing and overclocking settings can be included in the
same profile, loaded at Windows startup. Consequently, if your system
crashes, ATiTool just applies the default profile on the next startup.

 

[Barry Watzman]

Do not use loctite. Sometimes it works too well. I've had laptops with
hard drives that were installed using it on the screws, and they had to
be drilled out to remove the hard drive. You could create such a
situation with a motherboard. In this instance (and I'd argue in the
case of the hard drives as well) loctite is a [bad] solution to a
problem that didn't exist.

 

[First of One]

Read my other post. I was referring to Loctite thermal adhesives, not
threadlocking compounds.

BTW, Loctite threadlocker comes in both permanent strength (red) and
removable strength (blue) flavors. If the red type was used by mistake, it
can be softened with a heat gun. Be sure to remove the screw while it's hot;
the Loctite will regain its effect once cooled.

 

[pigdos]

They also have Loctite threadlocker green now.

BTW, First, I was asking your opinion on using the threadlocking compounds
to attach heatsinks, not their thermal adhesives.

--
Doug

Read my other post. I was referring to Loctite thermal adhesives, not
threadlocking compounds.

BTW, Loctite threadlocker comes in both permanent strength (red) and
removable strength (blue) flavors. If the red type was used by mistake, it
can be softened with a heat gun. Be sure to remove the screw while it's
hot; the Loctite will regain its effect once cooled.

--
"War is the continuation of politics by other means.
It can therefore be said that politics is war without
bloodshed while war is politics with bloodshed."

Do not use loctite. Sometimes it works too well. I've had laptops with
hard drives that were installed using it on the screws, and they had to
be drilled out to remove the hard drive. You could create such a
situation with a motherboard. In this instance (and I'd argue in the
case of the hard drives as well) loctite is a [bad] solution to a problem
that didn't exist.

 

[First of One]

My opinion on using threadlocker to attach heatsinks? I have no experience
with it, though I foresee two possible issues:

1. Loctite threadlocker cures anaerobically, so you will need to apply great
pressure to squeeze out any trapped air. This is easy to do with a screw
because the thread form is naturally a continuous wedge. On a 0.5-sq. in
flat surface (i.e. RAM chip) it'll be very difficult to apply uniform
pressure.

2. Threadlocker may be much more insulative than a purpose-made thermal
adhesive, so it shouldn't be used unless you do instrumented testing.
Probably not worthwhile when there are proper adhesives for the job:
http://www.arcticsilver.com/arctic_alumina_thermal_adhesive.htm

 

[pigdos]

First how do Loctite's thermal adhesives compare to arctic silvers?

 

[First of One]

Compare to Arctic Silver thermal epoxy? Performance should be within a deg
of each other. Note that on a high-quality lapped surface, an adhesive will
always perform slightly worse than thermal paste, because the amount needed
to get a good bond (.004"-.008") is more than optimum for purely heat
conduction.

From a usability standpoint, most Loctite thermal adhesives are
one-component, whereas the Arctic epoxies are two-part and require mixing.

 

[pigdos]

I did the 50/50 mix you suggested for the RAMsinks and saw almost "no
difference." Just like you said. It did seem like I could overclock the RAM
an extra 2Mhz.

I wonder if ATI/Sapphire just manufacture X850xt's and XT/PE's on the same
line and then cherrypick the PE's.

ATI definitely makes AGP X850XT's, unless my box and video card are clever
counterfeits (down to the BIOS level).

It's interesting that while ATITraytools can test out to 605Mhz on the RAM,
when I run Far Cry it only takes a couple of minutes before large-scale
artifacting occurs. Maybe the RAM on the front-side of the PCB is what's
overheating <shrug>. I'd have to dump the existing, stock, cooling solution
to find out.

I've been seriously thinking of getting the Zalman Vf900CU, since I'm going
to be stuck w/this video card a relatively long time. Any opinions on the
vf900CU?

 

[First of One]

I did the 50/50 mix you suggested for the RAMsinks and saw almost "no
difference." Just like you said. It did seem like I could overclock the RAM
an extra 2Mhz.

It is as you say, the only way to find out is to put individual sinks on the
RAM chips in the front of the card, too. Even then you may need to raise the
voltage (can't do it on-the-fly like X1K cards, unfortunately) to get a
stable overclock.

Have you tried adjusting timings yet?

I wonder if ATI/Sapphire just manufacture X850xt's and XT/PE's on the same
line and then cherrypick the PE's.

I think the GPUs may be binned prior to board assembly, and go through the
same assembly line with the same PCB and components. I can even see it as
ATi's way of legitimizing a standardized overclocked part, instead of
leaving the board partners to set their own speeds like nVidia does.

ATI definitely makes AGP X850XT's, unless my box and video card are clever
counterfeits (down to the BIOS level).

Yep. In fact, re-reading the HardOCP review, it even says "X800 XL, RADEON
X850 XT, and X850 XT-PE into the AGP realm". I simply didn't remember
correctly...

Does your XT have the same grade of RAM as the XT-PE shown in the HardOCP
review?

It's interesting that while ATITraytools can test out to 605Mhz on the
RAM, when I run Far Cry it only takes a couple of minutes before
large-scale artifacting occurs.

I think the artifact tester in ATi Tray Tools (and probably ATiTool as well)
is more effective for stressing the GPU than the RAM.

I've been seriously thinking of getting the Zalman Vf900CU, since I'm
going to be stuck w/this video card a relatively long time. Any opinions
on the vf900CU?

Never used a Zalman part before. Cooling performance seems to be very good.
On the X1900XT, people report a 9C drop with the Vf900 compared to the stock
cooler at 100%. Keep in mind the X1900XT dissipates 110W while the X850XT
does around 63W. The only gripe is the mounting screws, which stick out from
the back of the card by about 1 cm. This makes the Zalman unusable in my
Crossfire setup.

 

[pigdos]

First, the memory type is Samsung 504 k4j55323qf gc16. It should be good for
600Mhz but it isn't. I installed the Zalman Vf900CU just yesterday along
w/individual RAMsinks. The GPU temp hasn't gone above 50 degrees celsius (it
used to peak at 61), but my max overclock only increased to 599 (from 592
w/the stock heatsink). The RAMsinks did allow me to overclock my RAM up to
586 (from 572 w/near stock cooling solution (individual RAMsinks on the
backside RAM). So the RAMsinks actually made more difference than the GPU
cooler. ATITray tools tests out the memory to 605 or so, but Far Cry gets
noticeable artifacting at that setting and locks up if I push the GPU to
anything much over 600Mhz.

For most modern games it seems like my X850xt is limited by my CPU (a mobile
Barton at 2.63Ghz).

I'm not up for making any voltage adjustments, it's just too much hassle.