This engineer has decided to return to the world of ground-up PC assembly, starting with a CPU upgrade experiment.
20 years ago, I was a pretty prolific PC-builder, if I do say so myself. Various EDN hands-on projects were the primary motivation; my late-1999 article, for example, showcased Intel’s SE440BX and SE440BX-2 motherboards, respectively mated to Intel Pentium II-400 and Pentium III-550 CPUs. Take a look at Table 1 and you’ll likely have a good laugh at what a “high-end PC” of the era looked like (frankly, the whole piece is chuckle-provoking, in the spirit of “640K ought to be enough for anyone”).
Less than one year later, I’d graduated to Intel’s Rambus-based VC820 and OR840 motherboards, respectively coupled to Intel 600 MHz “Katmai” and 800 Mhz “Coppermine” Pentium III CPUs. But beyond that point, to the best of my recollection, I down-throttled to two subsets of the full build-from-scratch experience:
After two decades away, I’ve decided to get “back in the saddle” with another ground-up PC build (several, actually). I thought I’d ease in with a processor upgrade, specifically to one of the Hackintosh systems to which I’ve devoted a multi-part series. Last May, I mentioned that the SFF (small form factor) system came standard with an Intel Core i5-3570 CPU (HD Graphics 2500, 3.4 GHz base clock, 3.8 GHz Turbo clock, 4 cores/4 threads, 6 MB L3 cache).
I also mentioned that I’d separately purchased a used Core i7-3770 (HD 4000, 3.4 GHz base clock, 3.9 GHz Turbo clock, 4 cores/8 threads, 8 MB L3 cache) in case I ever wanted to update the SFF to support an additional 4 virtual cores via HyperThreading (along with gaining 100 MHz higher Turbo clock performance). That CPU update is precisely what I decided to tackle over the recent holidays.
If you look at the two processors’ specs, you’ll note that they document the exact same 77 W TDP (thermal design power). Frankly, I suspect that the Core i5-3570 is probably made from the exact same die as the Core i7-3770, just with HyperThreading disabled in the former case. So I should be able to use the exact same cooling solution already in the system for the new CPU. Let’s find out. Here’s what the system looked like prior to the processor upgrade, in Windows 10’s System Information panel:
Time to dive in. Let’s begin with an overview of the system’s insides, complete with the as-usual obligatory 0.75″ (19.1 mm) diameter U.S. penny for size comparison. The system looks brand new, even though I bought it used, doesn’t it?
In the center are the four 4 GByte DDR3 1600 (PC3 12800) 240-pin DIMMs, in aggregate comprising 16 GBytes of system DRAM. In the bottom left corner is the uncommon CFX form factor power supply I’ve mentioned before. Underneath it are two 500 GByte WD Blue 3D NAND 2.5” SATA SSDs in combination with a 3.5” to dual 2.5” mounting bracket, one for Windows 10 and the other (someday) for MacOS; the red SATA cable is a tipoff to which one I added (thankfully there was a spare SATA power supply feed already available). In the bottom right corner is the optical drive. In the top left corner you’ll (barely) see the ESATA adapter I added; below it is the (much more obvious) MSI Radeon RX 560 4GT LP OC graphics card. And in the upper right corner, underneath a heat sink, is the object of our attention: the CPU. Let’s take a closer look:
On one end is the heat sink; on the other is a fan. In-between them is a passive plastic cowling piece, to route thermal emissions from the heat sink to the fan. It lifts right off after you de-clip the various cables hooked to it for neatness purposes:
Only a thin coating of dust on the fan (along with some stray scratches on the system case) betrays the fact that the system’s not brand new (and I’m happy to say that the others I purchased were in equally near-pristine condition):
The heat sink is held in place by four screws, one in each corner. I could have used a flathead screwdriver to remove them, and in fact I started down this route, but the grooves in the screw heads were pretty shallow. So I switched to the T15 Torx bit from my iFixit driver kit, which easily did the trick:
Next, let’s take a look at the heat sink underside, along with the topside of the socketed CPU normally underneath it:
The fact that the previous total thermal paste payload is now subdivided between the two is no surprise; this is why you need to remove the old paste from both (70% isopropyl alcohol and a lint-free paper towel do the trick nicely) and re-apply fresh every time you separate a CPU from the heat sink, regardless of whether you’re replacing either (or both, for that matter). But if the paste looks dry to you, you’re right. I’ve warned about this already with respect to older systems, and by the way, now’s as good a time as any to correct an inadvertent error I previously made.
In that earlier writeup, I mentioned the “CPUs’ reliance on thermal paste versus the fluxless solder in prior-generation designs.” Turns out the controversy I alluded to wasn’t actually with respect to the material between the spreader plate and the heat sink above it, but instead in reference to the material between the spreader plate and the silicon die underneath it. Nonetheless, I was spot-on in my recommendation to just go ahead and “proactively remove the CPU’s heat spreader plate, scrape off the old thermal paste and replace it with a brand-new high quality alternative compound.”
Removing the old processor (and installing the new one in its place, for that matter) is pretty straightforward, especially because in this particular case it’s in a LGA (land grid array) package with no pins to bend or break. First, unclip and lift away the retention lever:
Then lift away the latch plate:
Before removing the old CPU, here’s a look at its successor:
See those two alignment notches, one on either side near the “top” (direction reference based on the product mark orientation)? They thankfully make it basically impossible to place the CPU in the socket the wrong way. And turning the package around you can see the 1155-contact (!!) LGA array.
The original CPU lifts right out of the socket, once its retention mechanism has been released. Here it is after the aforementioned isopropyl alcohol cleanup (does anyone out there want to buy a used Intel Core i5-3570?):
Drop the new CPU in the socket in its stead, replace the retention latch plate and lever and … wait, don’t put the heat sink back on yet! First, we need to apply new thermal paste to maximize heat transfer off the CPU, by filling in the microscopic gaps between the CPU’s spreader plate and the heat sink’s base plate.
What thermal paste to use, how much of it to use, and what pattern (or not) to employ when applying it are topics of great controversy in the overlocker community. iFixit’s tutorial, for example, references Arctic Silver’s documentation, which has different application-pattern recommendations for AMD versus Intel CPUs, and even for different generations of those CPUs! And this guy, who’s done lots of videos on fixes and upgrades for the HP 6×00 Pro and 8×00 Elite series, suggests putting the new thermal paste on the heat sink’s base plate instead of the CPU’s spreader plate.
Thankfully, my situation’s simpler, since I’m running the processor at stock speed (which also explains why I’m using the standard heat sink instead of something more exotic). I ended up just going with Intel’s consistent, conservative recommendation to put a tiny dollop in the center and let the CPU-plus-heat sink sandwich take care of subsequently spreading it out:
Speaking of Arctic Silver, the company’s popular “5” product variant is what I ended up using:
The 3.5 gram package contains enough compound for numerous CPU-plus-heat sink installations; if you firmly replace the cap after each application, the paste won’t dry out. And although it’s thermally conductive (duh), it’s not electrically conductive; good news in case you apply too much of it and it spills over the “sandwich” edge and onto the motherboard below!
Speaking of not drying out, let’s speedily get that heat sink back on. Sequentially and conservatively tighten each screw, repeatedly round and round the four corners of the heat sink until all are firmly locked down, then reverse the remaining steps to get the cowling back in place and the internal cabling neatly tucked back down. Reattach the system side panel and all the external cables, punch the power button, and bingo!
One last testing step: how well did my thermal paste application process pan out? Here’s what CoreTemp says with the system at idle:
Now let’s fire up HeavyLoad and see how the paste performs under more stressful conditions:
69°C (or less) is right in line with what the community says it should be, so I’m pleased!
With that, I’ll wrap up for today and save discussion of my next system-construction steps for future posts. Until then, I look forward to your thoughts in the comments!
This article was originally published on EDN.
—Brian Dipert is Editor-in-Chief of the Embedded Vision Alliance, and a Senior Analyst at BDTI and Editor-in-Chief of InsideDSP, the company’s online newsletter.