- Joined
- Jan 15, 2001
- Location
- Denver, CO
Mods if you feel this belongs elsewhere please move it. I think it may apply to a wide range of motherboards and heatsinks, with only one thing in common - Athlon XP's which is why I'm posting it here.
This is going to be a long explanation, but I think it deserves further investigation and a general caution for those building new systems. I have only investigated this problem with Epox motherboards and a few brands of heatsinks, but it may very well be happening with all brands.
First a little background about the problem and myself. I have been building systems as a side job for several years. I am also A+ certified. In other words I'm not a new comer to hardware. A few months ago I bought my first XP chip. I decided to swap it with a tbird I had running with an Epox 8K7A. For a heatsink I was using a Thermalright SK6. I left the board in the case and replaced the tbird with the new XP, applied arctic silver, and reattached the heatsink. While booting up I briefly saw the post screen, then heard a pop and smelled the unmistakable odor of burned electronics. I pulled the plug and knew it was either my board or the new chip. When I took the heatsink off it was obvious the chip had burned up. It had popped at one corner and splattered arctic silver around the socket. A hasty investigation convinced me I had failed to notice that one edge of the heatsink was resting on the socket's cam box - the raised edge where the locking lever attaches. The Thermalright SK6 has a raised edge to clear the cam box, but it is possible to slide the heatsink over too far while attaching the clip and it will not clear the cam box. This prevents the heatsink from resting flat on the cpu. That's what I thought had happened, and I vowed never to make that mistake again.
Fast-forward a couple of months. I got several new XP chips for my Seti farm. I also picked up an 8KHA+. When I'm setting up a new system I like to mount the heatsink and add ram while the board is out of the case. This lets me work with the board on a solid surface without fear of bending it. I put my shiny new XP 1500 into the 8KHA+. A careful examination from the edge of the board assured me the SK6 was not touching any part of the socket. I finished putting the system together and fired it up. After only a few seconds I heard a pop and saw a puff of smoke. I instantly knew what had happened - but HOW? I was so careful this time! I immediately took the heatsink off and what I saw surprised me. The arctic silver had left a contact pattern on the heatsink that revealed the problem. The edge of the cpu closest to the came box had made good contact. The square outline of its impression was clear. The other end had not made contact. In fact only about 3/4 of the cpu had touched the heatsink. The silver grease tapered off in an upside down smiley face pattern, leaving that edge without any contact with the heatsink. It was that edge where the cpu had blown out too.
This incident frustrated me greatly so I decided to do a few tests. I took the board out and laid it on my workbench. I cleaned the chip and the heatsink, took the chip out of the socket, put it back in and held all four corners down while locking the lever. I put an even coat of grease on the chip and proceeded to attach the SK6 as carefully as possible. After the clip was in place I used a flashlight and checked the clearance carefully. Everything looked good. Next I slowly took the heatsink off making sure I didn't slide it around at all. Sure enough the grease pattern on the heatsink and cpu revealed only 3/4 of the chip had touched the heatsink. I tried this test 4 more times and the same problem occurred each time. Thinking I had a heatsink that was not flat, I grabbed a Tai Sol unit lying nearby. I repeated the same test with this heatsink and it showed the same problem, although not quite as bad. By then I was beginning to suspect the board/socket was warped. I was surprised the old tbird that had been in that board hadn't burned up too, so I grabbed it and put it back in. I repeated the same tests, but the old tbird passed with flying colors. So what's the difference? The tbird uses a stiff ceramic package, is slightly thicker, and has a rectangular shaped core. This prompted a couple of questions. Could the reduced height of the core above the package surface combined with the same old rubber pads be preventing the heatsink from coming close enough to the surface? Secondly, could the new organic package be bending under the stress of AMD's heavy clip retention specs? The first question seemed easy enough to examine. I simply pulled the rubber pads off and repeated my contact test. This resulted in no change - still no even contact with the cpu. Next I decided to check the bending theory. I took an old Alpha pal 6035 designed for socket 370 chips. The spring clip is relatively light on those old heatsinks. Finally - eureka! The pal made perfect contact with the XP core every time. It seemed safe to assume the XP was indeed bending. To check this out further I tried to look for evidence of bending with the SK6 and Tai Sol mounted. First I put the chip in the socket and held it down firmly while locking the level. Looking at the chip from the side it appeared to be seated flat against the socket all the way around. Next I mounted the heatsinks and looked again. With the heatsink mounted I found just what I suspected. The edge of the cpu closest to the cam box was slightly raised above the socket - the pins clearly visible in between the cpu and socket. The opposite edge was still flat against the socket - hence the reason that edge never made complete contact. I tried this test with several AMD approved socket A heatsinks. Each one showed the same results to varying degrees, with the Tai Sol being the least affected. Why I do not know because it has a very stiff spring clip. Perhaps it has something to do with the thickness of the heatsink and material as well. Finally I put the Alpha pal 6035 back on several times and it never bent the socket/cpu. For one last test I tried the Alpha 8045 and Swiftech MC 462. Both use the four mounting holes around the socket. Both units showed no signs of bending whatsoever. So for now it appears that this is by far the safest mounting method, and most likely why AMD came up with this design. Intel adopted this mounting idea as well. Intel has also been using the organic packaging longer, but the socket 370 platform required little in the way of cooling, and those heatsinks generally put relatively little force on the socket itself. Perhaps both AMD and Intel saw a potential problem with the organic package bending under the stress of larger heatsinks which require increased clip pressure. Whatever the case may be this appears to be a serious problem. I have not had the time to test other boards yet, but it wouldn't surprise me if they have the same problem to varying degrees. Some may have the problem to a lesser extent, and perhaps with enough grease slathered on they get by without frying.
I presented this problem to Joe and Ed in the hopes they would do their own testing. In the mean time I would like others to be on the lookout for the same problem. Please note that I'm not saying this is a fault of Epox boards, but rather a result of the XP organic package in conjunction with heavy clip pressure. If you do run this test for yourself, make a note of the brand of socket used on your board, as well as any other observations you feel are relevant. Please feel free to email me your observations. I would like to compile a list of results.
This is going to be a long explanation, but I think it deserves further investigation and a general caution for those building new systems. I have only investigated this problem with Epox motherboards and a few brands of heatsinks, but it may very well be happening with all brands.
First a little background about the problem and myself. I have been building systems as a side job for several years. I am also A+ certified. In other words I'm not a new comer to hardware. A few months ago I bought my first XP chip. I decided to swap it with a tbird I had running with an Epox 8K7A. For a heatsink I was using a Thermalright SK6. I left the board in the case and replaced the tbird with the new XP, applied arctic silver, and reattached the heatsink. While booting up I briefly saw the post screen, then heard a pop and smelled the unmistakable odor of burned electronics. I pulled the plug and knew it was either my board or the new chip. When I took the heatsink off it was obvious the chip had burned up. It had popped at one corner and splattered arctic silver around the socket. A hasty investigation convinced me I had failed to notice that one edge of the heatsink was resting on the socket's cam box - the raised edge where the locking lever attaches. The Thermalright SK6 has a raised edge to clear the cam box, but it is possible to slide the heatsink over too far while attaching the clip and it will not clear the cam box. This prevents the heatsink from resting flat on the cpu. That's what I thought had happened, and I vowed never to make that mistake again.
Fast-forward a couple of months. I got several new XP chips for my Seti farm. I also picked up an 8KHA+. When I'm setting up a new system I like to mount the heatsink and add ram while the board is out of the case. This lets me work with the board on a solid surface without fear of bending it. I put my shiny new XP 1500 into the 8KHA+. A careful examination from the edge of the board assured me the SK6 was not touching any part of the socket. I finished putting the system together and fired it up. After only a few seconds I heard a pop and saw a puff of smoke. I instantly knew what had happened - but HOW? I was so careful this time! I immediately took the heatsink off and what I saw surprised me. The arctic silver had left a contact pattern on the heatsink that revealed the problem. The edge of the cpu closest to the came box had made good contact. The square outline of its impression was clear. The other end had not made contact. In fact only about 3/4 of the cpu had touched the heatsink. The silver grease tapered off in an upside down smiley face pattern, leaving that edge without any contact with the heatsink. It was that edge where the cpu had blown out too.
This incident frustrated me greatly so I decided to do a few tests. I took the board out and laid it on my workbench. I cleaned the chip and the heatsink, took the chip out of the socket, put it back in and held all four corners down while locking the lever. I put an even coat of grease on the chip and proceeded to attach the SK6 as carefully as possible. After the clip was in place I used a flashlight and checked the clearance carefully. Everything looked good. Next I slowly took the heatsink off making sure I didn't slide it around at all. Sure enough the grease pattern on the heatsink and cpu revealed only 3/4 of the chip had touched the heatsink. I tried this test 4 more times and the same problem occurred each time. Thinking I had a heatsink that was not flat, I grabbed a Tai Sol unit lying nearby. I repeated the same test with this heatsink and it showed the same problem, although not quite as bad. By then I was beginning to suspect the board/socket was warped. I was surprised the old tbird that had been in that board hadn't burned up too, so I grabbed it and put it back in. I repeated the same tests, but the old tbird passed with flying colors. So what's the difference? The tbird uses a stiff ceramic package, is slightly thicker, and has a rectangular shaped core. This prompted a couple of questions. Could the reduced height of the core above the package surface combined with the same old rubber pads be preventing the heatsink from coming close enough to the surface? Secondly, could the new organic package be bending under the stress of AMD's heavy clip retention specs? The first question seemed easy enough to examine. I simply pulled the rubber pads off and repeated my contact test. This resulted in no change - still no even contact with the cpu. Next I decided to check the bending theory. I took an old Alpha pal 6035 designed for socket 370 chips. The spring clip is relatively light on those old heatsinks. Finally - eureka! The pal made perfect contact with the XP core every time. It seemed safe to assume the XP was indeed bending. To check this out further I tried to look for evidence of bending with the SK6 and Tai Sol mounted. First I put the chip in the socket and held it down firmly while locking the level. Looking at the chip from the side it appeared to be seated flat against the socket all the way around. Next I mounted the heatsinks and looked again. With the heatsink mounted I found just what I suspected. The edge of the cpu closest to the cam box was slightly raised above the socket - the pins clearly visible in between the cpu and socket. The opposite edge was still flat against the socket - hence the reason that edge never made complete contact. I tried this test with several AMD approved socket A heatsinks. Each one showed the same results to varying degrees, with the Tai Sol being the least affected. Why I do not know because it has a very stiff spring clip. Perhaps it has something to do with the thickness of the heatsink and material as well. Finally I put the Alpha pal 6035 back on several times and it never bent the socket/cpu. For one last test I tried the Alpha 8045 and Swiftech MC 462. Both use the four mounting holes around the socket. Both units showed no signs of bending whatsoever. So for now it appears that this is by far the safest mounting method, and most likely why AMD came up with this design. Intel adopted this mounting idea as well. Intel has also been using the organic packaging longer, but the socket 370 platform required little in the way of cooling, and those heatsinks generally put relatively little force on the socket itself. Perhaps both AMD and Intel saw a potential problem with the organic package bending under the stress of larger heatsinks which require increased clip pressure. Whatever the case may be this appears to be a serious problem. I have not had the time to test other boards yet, but it wouldn't surprise me if they have the same problem to varying degrees. Some may have the problem to a lesser extent, and perhaps with enough grease slathered on they get by without frying.
I presented this problem to Joe and Ed in the hopes they would do their own testing. In the mean time I would like others to be on the lookout for the same problem. Please note that I'm not saying this is a fault of Epox boards, but rather a result of the XP organic package in conjunction with heavy clip pressure. If you do run this test for yourself, make a note of the brand of socket used on your board, as well as any other observations you feel are relevant. Please feel free to email me your observations. I would like to compile a list of results.