For the Intel Colorado LANFest 2013 event, my co-workers and I decided to do some sort of neat set of demo systems just because Intel always has a table with neat stuff. Last year, we put together the 5th fastest 3Dmark computer in the world, and this year my co-worker de-lidded a Core i7 4770K and had a lab at Intel "lap" or mill it to a thickness of 50um which is less than half the thickness of a standard sheet of printer paper (which are ~0.1mm) and then he stuck an Intel Xeon Phi 7120p coprocessor (a 61 core, 16GB, PCI-E supercomputer-in-a-card) and 8 Intel 320-series with a high-end RAID controller.
So I couldn't touch that... but I wanted in on this demo thing too, so I decided to do something that I've wanted to do ever since I got my Core i7 4770K earlier this summer - de-lid it and directly attach a heatsink to the back of the die. Not as cool as a 50um thick CPU and a Xeon Phi, but still pretty neat.
The core of the problem with directly attaching a heatsink to a Haswell - to me, anyway - is the part where you take the lid off the part, so I took the opportunity of the event - for an Intel sponsored demo system - to have the pros take the lid off of my part (which I bought myself). So I sent it off to the Intel team in Oregon to have the lid removed and they sent it back happily bare-die. So that was the hard part out of the way... the rest was fun.
Having read (and re-read) IDontCare's excellent Bare-die testing: A delidded 3770k, an H100, and 9 different TIMs post, I had the confidence to hack something today, so I bought an H100i and I headed to Lowe's for parts. I bought a 4 pack of nylon spacers that were about 12.7mm (0.5") long and then tested the ones in the store until I figured out that I wanted 3.5# x 25mm screws. Then I bought some #6 nylon washers. Total cost was about $4. See photo if you want the Lowe's part numbers.
Similar to Idontcare, I carefully measured the thickness of the IHS and matched his number of 2.5mm and the height of the H100i standoffs and measured 12.5mm - which is 1mm shorter than his H100 standoffs... which was odd. And then I cut the nylon spacers to 10mm with an Xacto knife, eyeballed them and measured them and they were within ~0.5mm so I called it good enough. Total time to do this was about 15 minutes - most of which were spent puzzling over why a H100i standoff is shorter than an H100 and trying to figure out if I was just measuring it wrong.|
So, in summary, take the 4 nylon standoffs and trim them to 10mm as close to exactly as you can. It doesn't need to be perfect, but the closer they are, the better. You can stand them up next to each other to make sure they all match. Then remove the CPU cover but keep the backing (see Idontcare's thread), don't use the backplate for the H100i (but expect a bit of motherboard flex as a result of this). Just thread the screws through, with the nylon washers on the backside of the board, and then put the screws through the nylon standoffs. I used a bit of painters masking tape to hold everything in place. Put the CPU in the socket, use a good TIM - I like Liquid Ultra the best but use only a really small amount - and then attach the H100i and screw down with the thumbscrews but not too tight, and start in one corner, then the opposite corner, then the other two opposites so that you don't crank down too much one one particular corner.
When I put the thing together, and put on some Liquid Ultra TIM and ran some Linpack and soon I found the results to be well beyond my expectations.
Originally, my Core i7-4770K could overclock to 4.3GHz at 1.05V (stock voltage), but running Linpack it would hit 100C and thermal throttle.
After, with the direct die attach, at stock voltage I can run 4.5GHz at 1.05V, but running Linpack, it hits a max temp of about 81C with the case cover on - and 76C with the cover off. It boots and runs Linpack at 4.6GHz too, but there were signs of instability in the Windows log files while Linpack was running.
Overall, I'm very happy with my direct-die attached 4770K. The H100i is pretty quiet (although I had to install some rubber grommets on the fans to stop a resonance problem), the temperatures are low enough, and 4.5GHz on stock seems pretty good to me. And a Linpack result of 185GFlops seems amazing.
One might ask why not boost the voltage and push the frequency higher - and I will probably do that at some point. But I just got everything set up and I'm going to just leave it as it is for now.
The end result at the Intel LANFest was that the demo was a success as well. The milled part was pretty neat and so my co-worker got the big crowds, but my system seemed to have almost as many people since it seemed like something that anyone could do. So people seemed pleased with the demo, and LANFest raised a bunch of money for the United Way charity (over $5000), and the gamers all had a great weekend.
So I couldn't touch that... but I wanted in on this demo thing too, so I decided to do something that I've wanted to do ever since I got my Core i7 4770K earlier this summer - de-lid it and directly attach a heatsink to the back of the die. Not as cool as a 50um thick CPU and a Xeon Phi, but still pretty neat.
The core of the problem with directly attaching a heatsink to a Haswell - to me, anyway - is the part where you take the lid off the part, so I took the opportunity of the event - for an Intel sponsored demo system - to have the pros take the lid off of my part (which I bought myself). So I sent it off to the Intel team in Oregon to have the lid removed and they sent it back happily bare-die. So that was the hard part out of the way... the rest was fun.
Having read (and re-read) IDontCare's excellent Bare-die testing: A delidded 3770k, an H100, and 9 different TIMs post, I had the confidence to hack something today, so I bought an H100i and I headed to Lowe's for parts. I bought a 4 pack of nylon spacers that were about 12.7mm (0.5") long and then tested the ones in the store until I figured out that I wanted 3.5# x 25mm screws. Then I bought some #6 nylon washers. Total cost was about $4. See photo if you want the Lowe's part numbers.
Similar to Idontcare, I carefully measured the thickness of the IHS and matched his number of 2.5mm and the height of the H100i standoffs and measured 12.5mm - which is 1mm shorter than his H100 standoffs... which was odd. And then I cut the nylon spacers to 10mm with an Xacto knife, eyeballed them and measured them and they were within ~0.5mm so I called it good enough. Total time to do this was about 15 minutes - most of which were spent puzzling over why a H100i standoff is shorter than an H100 and trying to figure out if I was just measuring it wrong.|
So, in summary, take the 4 nylon standoffs and trim them to 10mm as close to exactly as you can. It doesn't need to be perfect, but the closer they are, the better. You can stand them up next to each other to make sure they all match. Then remove the CPU cover but keep the backing (see Idontcare's thread), don't use the backplate for the H100i (but expect a bit of motherboard flex as a result of this). Just thread the screws through, with the nylon washers on the backside of the board, and then put the screws through the nylon standoffs. I used a bit of painters masking tape to hold everything in place. Put the CPU in the socket, use a good TIM - I like Liquid Ultra the best but use only a really small amount - and then attach the H100i and screw down with the thumbscrews but not too tight, and start in one corner, then the opposite corner, then the other two opposites so that you don't crank down too much one one particular corner.
When I put the thing together, and put on some Liquid Ultra TIM and ran some Linpack and soon I found the results to be well beyond my expectations.
Originally, my Core i7-4770K could overclock to 4.3GHz at 1.05V (stock voltage), but running Linpack it would hit 100C and thermal throttle.
After, with the direct die attach, at stock voltage I can run 4.5GHz at 1.05V, but running Linpack, it hits a max temp of about 81C with the case cover on - and 76C with the cover off. It boots and runs Linpack at 4.6GHz too, but there were signs of instability in the Windows log files while Linpack was running.
Overall, I'm very happy with my direct-die attached 4770K. The H100i is pretty quiet (although I had to install some rubber grommets on the fans to stop a resonance problem), the temperatures are low enough, and 4.5GHz on stock seems pretty good to me. And a Linpack result of 185GFlops seems amazing.
One might ask why not boost the voltage and push the frequency higher - and I will probably do that at some point. But I just got everything set up and I'm going to just leave it as it is for now.
The end result at the Intel LANFest was that the demo was a success as well. The milled part was pretty neat and so my co-worker got the big crowds, but my system seemed to have almost as many people since it seemed like something that anyone could do. So people seemed pleased with the demo, and LANFest raised a bunch of money for the United Way charity (over $5000), and the gamers all had a great weekend.
Last edited:
Facebook
Twitter