Quote:
Originally posted by bigben2k
The core is only 80mm^2 in area, so why would I otherwise use a 2 in. by 2 in. plate, especially if the heat spreads from the center out?
|
Convective heat transfer = h * A * delta-T
h is mainly a function of fluid properties and for a given fluid, velocity is the dominant variable. If you fix A to 80 mm^2, then delta-T becomes inversely proportional to h. h has practical limits based on the maximum velocity we can develop. This means if you limit your block to 80 mm^2, you have a minimum achievable delta-T (block to fluid) that you'll approach asymptotically (sp?) as velocity heads toward infinity.
In the case where the block matches the die size, you can calculate the delta-T required for conduction across the block by assuming unidirectional heat transfer. This is simply q = k A / l * delta-T. Say q = 75 watts, k = 401 w / m - K, and A = 80 mm^2 (80 * 10-6 m^2). Then delta-T / l = constant. At l = 5 mm, delta-T = 11.7 K (same scale as °C). At this point, something should sound fishy. You know darn well that a copper heat sink with a Delta fan will keep a chip within 12°C of the air temperature, yet your "die sized" block can't possibly do as well because there's nothing left for block-to-fluid delta-T.
You absolutely require some spreading of the heat to take place.
If 2/3 of the heat goes to your die-sized block and 1/3 escapes to the 2" X 2" periphery, then your block's peak delta-T (which WILL occur over the die) drops from 11.7 to 7.8 °C. That's still not so good, but now in the ballpark of the best (read: Loudest) air cooled solutions.
Now you should understand where baseplate thickness comes into play. If the baseplate is really, really thin then not much spreading will occur. In this case, you best have some serious direct impingement going on. As baseplate thickness increases, the percentage of the thermal energy that gets dissipated "outside" the core region increases. You need to start balancing direct impingement against maintaining flow over a wider surface area.
IMHO, the best designs rely on direct impingement for a substantial portion of the heat transfer as this provides excellent flow instability (read: minimal boundary layer thickness) while carrying that flow through to the perimeter where it can absorb additional energy. You still need to maintain decent velocity outside the "die box" in order to keep the convection coefficient high, but you do get to trade some of the initial velocity of the impingement for the added surface area around the periphery.
I'll tell ya what, I really liked Paul's block that he posted a few days back (and wants $120 US to ship). Not only is it a work of art, but it also takes a solid (solid from the engineer's perspective) approach to maximizing heat transfer from direct impingement AND the outer perimeter. On a side note, it has a strong similarity to what I posted a while ago as my idea of a good low-flow block. Paul's is better than what I showed as he still maintains the direct impingement that I did not and he also has a better exit than I showed.
As a final footnote, I'll repeat what I said earlier about the 0.13 XP chips. Their smaller die size is a bigger pitfall than people realize. Before you had ~72 watts going through 130 mm^2 whereas now you have 62 watts through 80 mm^2 (numbers from AMD's tech docs on XP2100 chips, 0.18 and 0.13 construction). The heat flux (watts / mm^2) is over 37% higher on the new XPs. This means you have less total heat, but a lot higher concentration of what heat remains. Baseplate thickness should increase a little to compensate for the higher concentration of heat.
I haven't seen anything concrete from any heatsink or block manufacturers on this point. Probably because (A) I haven't looked and (B) they probably haven't thought too much about it. Anyway, this is the primary reason why people aren't reporting lower temperatures for the new XP chips and their lower power consumption.