Originally Posted by Puck
The issue you are encountering is that without a decent amount of radiator surface area to dissipate the added heat of the TEC, you are completely negating the benefit of running one. You are already "under rad'ed", as seen by your very high water load temp. Ideally that needs to be as close to the room temp as possible. On the bright side, with a chiller you do not want a lot of radiator area since all it does is warm your coolant back up to ambient. How do you plan on cooling the hot side? They basically need to be watercooled since they work on a delta, so now you are talking ~200 bucks in custom blocks for the cold side, plus another pump, plus some more blocks for the hot side, plus the extra tubing and fittings you will need...a chiller setup will add up fast. You may be better off with a QDD TEC and PSU, then just adding another rad or two to your current WC loop. If you don't want to insulate or want to keep the power draw down on a daily use system you will need a controller too, but it may be worth calculating the cost of both methods. Mindchill closed the shop for the holidays to catch up on orders, but from what I hear will be opening up again eventually. I haven't seen him on in a while, but maybe he will chime in with his suggestions. Him and Krow both know a LOT about TEC systems, so I would seriously consider both their advice.
Also, a bit OT, but are you sure you need all that voltage for 4.7? That's a ton of heat output, and I don't think two of those "waterblock chillers" will do much more then a good WC setup...
I have the same chip now, and one of the Arqtik SDD TEC setups, so once I get some free time from work I can play with it and let you know it handles the 3770K at what voltages and clocks. I can tell you it doesn't even bat an eye at 4.3ghz @ 1.25v, and holds the block at 8c with lots of headroom to spare. Plus another member says 4.7ghz full load is under 70c but pretty sure it was less voltage then you are using, which makes a huge difference.
To be clear, I'm not running a TEC yet. I'm just running the numbers to see how much a few potential setups would cost me, and what I could expect to gain in performance from implementing one of them. To me, if a SDD setup isn't going to work on my rig, a rad+chiller design makes the most sense for my goals. While having a high overclock and low temps is fun and all, I'm not willing to pay for it in my electric bill, in addition to having a huge cancerous growth of radiators and a veritable vortex of fans to achieve it. I like my system to be cool, quiet, and efficient when I'm doing things like web browsing (most of the time), and roar to life with power when I'm gaming or benchmarking. I'm big on efficiency, I guess you could say.
If the only way to get temps meaningfully lower than what I have now is to add an 4 TECs, 4 rads, and 8 fans then count me out. haha. My IBT temps are high, but linpak is not representative of a normal amount of CPU load anyway. As far as I can tell, my CPU never sees temps higher than 75c ona daily basis, and those are just peaks.
Regarding the setup I have now, I'm currently just running an H100i. I plan on upgrading to Swiftech's H220 when it comes out, though. When that happens, I plan on running a 240mm rad and two 120mm rads to cool the CPU. I don't believe my case (Corsair 600T) will allow for dual 240mm rads. Once I upgrade, I'll end up having an H100i sitting around doing nothing. In trying to find a use for it, I stumbled onto TEC cooling and here we are. (What I plan on doing with the H100i's rad given my previous statement I haven't quite figured out yet...haha) This is why I want to use a lower powered TEC + Rad system - it's more appropriate to the spare parts I'll have laying around at the moment.
However, if a single-TEC direct die setup similar to Arqtik's device will adequately cool my rig, I'm open to purchasing one of those and coupling it with the H100i (probably not) or the Swiftech loop when I get it. It's just that according to Aqrtik's own calculator it really won't even come close to performing adequately enough.
Regarding my OC voltage, yeah, it's pretty high. This OC is just what Asus's Auto Tuning software put in place with a few minor BCLK adjustments from me to increase stability. Before anyone berates me for not OCing my own system, it should be known that this is my second 3770k. My first one was one of the worst OCing 3770k's chip I've heard of. I spent 2 weeks trying to get the chip stable at 4.5ghz without success - even at 1.4 vcore. I tried everything I could think of to increase performance, but nothing helped. It ran hot, too. Upon getting this second chip and seeing it auto-OC to over 4.7, I was happy to keep it there, at least for the time being. Also, I want to dynamic OC with offsets because I hate the idea of the chip sitting there at idle drawing max voltage and producing so much waste heat. OCing with voltage offsets can be a pain in the ass, and Asus's Auto-Tuning algorithm figured out where it needs to be automatically. An unfortunate byproduct of this means reducing the vcore manually is not as simple pushing the down-arrow.
Originally Posted by Puck
Quick update on the SDD setup. Simple fast-n-dirty OC to 4.5ghz, 1.3v (1.35v under load), 62c hottest core. Block still @ 11c because of 48% humidity. Early mornings I can set block to 7-8c so in a better environment then humid FL you will get even lower. Looks like the Single is *plenty* for an IB.
Note that while I did delid to swap TIM, I'm using old nasty AS Ceramic both on die and IHS, which is trash by todays standards. A few C is still in there for sure with a proper TIM like Liquid Pro.
Those seem like pretty standard voltages for 4.5ghz. I think you'll find that higher clocks will need significantly more voltage. That's what I found, at least. Your vcore may not look so different from mine in the end. I'd really suggest using a dynamic overclock, though. I've heard it said many times that any vcore above 1.35 is into the danger zone. Even with my stupid-high voltage, the dynamic OC keeps my CPU at around 1.32 most of the time.
Also, de-lidding at all has probably helped your temps more than you think.Edited by JazzMac251 - 1/18/13 at 6:46pm