Question about running Haswell-E at stock but with 2800Mhz Memory.

[BonzaiDuck]

I don't see that a lot, unless we are talking about sub-zero temps. The temparature drop is very considerable but the increase in OC potential is just not there.
http://www.tomshardware.com/reviews/visiontek-cryovenom-r9-295x2,3951.html

It's a GPU but it happens to CPUs as well, I can make a winter-cooling to see if HW really needs less V for a given frequency. The biggest advantage is pushing lots of voltage and have it still not overheat.

I wouldn't disagree appreciably. Even so, fellow member IDontCare, with his threads on de-lidding, was able to show perhaps a 20 mV difference on an IB processor with a temperature that was some 20C lower. Again -- I don't remember his exact numerical results and my memory may have exaggerated what I cite here; it was something he'd pointed out in some test outcomes.

One might never see it at all. But temperature exacerbates electrical leakage and power consumption -- something else he showed with test results. Generally, there is a consensus that higher temperature increases electrical noise, which either requires more voltage for stability or reduces stability at the same voltage.

If you don't see any appreciable change in required voltage, the lower temperatures will likely improve processor longevity -- IF -- one were planning to run the processor at full load for longer periods of time.

I could almost see myself spending money on water-cooling just to avoid throttling during stress-tests!

On the general situation you report -- what seems like the ability to significantly lower load voltage below what "Auto" features provide -- I spoke to my friend and retired physicist today. I suppose the question would be "how -- or why -- do the chipset's 'auto' features now seem to exceed processor requirements for stability?"

If one were to be extra cautious and looking for a voltage "safe upper limit," you would think that a CPU at stock settings would give a voltage "baseline" under load. You'd find the lowest stable voltage first, and then work upwards as you approached that baseline.

I only say this, because you've reported lower voltage settings for higher clock speeds than the default for either factor. Did I read that correctly?

 

[Lepton87]

I only say this, because you've reported lower voltage settings for higher clock speeds than the default for either factor. Did I read that correctly?

Auto settings seem to change the voltage with the CPU clock so it's not stock, they are just conservative so that the majority of CPUs should work under auto settings at least 4.2GHz.

 

[BonzaiDuck]

Auto settings seem to change the voltage with the CPU clock so it's not stock, they are just conservative so that the majority of CPUs should work under auto settings at least 4.2GHz.

I'd be willing to bet that 4.7 is possible with the right cooling. If it isn't, you'd have great cooling and probably an OC up to 4.5. I just think it would depend on how you want to deal with the temperatures, what you think a good target load temperature should be.

Given the base clock for the processor and the turbo-speed spec, even 4.2 seems pretty good -- for six cores.

It's a good bet that 4.2 for the 5820K might score as well in benchmarks as a 4930K or 39xx SB-E at higher clocks. Maybe someone posted some comparisons somewhere. Maybe even in the HardOCP review.

 

[Lepton87]

I'd be willing to bet that 4.7 is possible with the right cooling. If it isn't, you'd have great cooling and probably an OC up to 4.5. I just think it would depend on how you want to deal with the temperatures, what you think a good target load temperature should be.

Given the base clock for the processor and the turbo-speed spec, even 4.2 seems pretty good -- for six cores.

It's a good bet that 4.2 for the 5820K might score as well in benchmarks as a 4930K or 39xx SB-E at higher clocks. Maybe someone posted some comparisons somewhere. Maybe even in the HardOCP review.

4.7GHz maybe with water I think with AIR cooling 4.4GHz should be possible or even higher provided I didn't use LinX because it would overheat then.
ps. I put the computer into a case and it lower the max temp by a few degrees.

Always clean the old tim off thoroughly before applying new.

I know that it's best to do that but I ran out of TIM

 

[BonzaiDuck]

4.7GHz maybe with water I think with AIR cooling 4.4GHz should be possible or even higher provided I didn't use LinX because it would overheat then.
ps. I put the computer into a case and it lower the max temp by a few degrees.


I know that it's best to do that but I ran out of TIM

Clean it up with isopropanol; get a tube of IC Diamond; use a razor-blade or expired credit-card to coat the IHS. Even if you are compelled to "re-use" the TIM, its efficacy will not be reduced.

With all this in mind as you continue tuning up the 5820K, I think you can scrape up a few extra Celsius in temperature reduction. If you're happy with the highest feasible 24/7 over-clock settings, you can get by on air-cooling with the D14.

I'm most likely to build mine with some radiators, dual pumps and reservoir. And the paradox of it: I feel like taking my time because this old 2600K is just perfect with its best OC settings -- and the D14 is the center of my cooling system.

 

[Lepton87]

Clean it up with isopropanol; get a tube of IC Diamond; use a razor-blade or expired credit-card to coat the IHS. Even if you are compelled to "re-use" the TIM, its efficacy will not be reduced.

With all this in mind as you continue tuning up the 5820K, I think you can scrape up a few extra Celsius in temperature reduction. If you're happy with the highest feasible 24/7 over-clock settings, you can get by on air-cooling with the D14.

I'm most likely to build mine with some radiators, dual pumps and reservoir. And the paradox of it: I feel like taking my time because this old 2600K is just perfect with its best OC settings -- and the D14 is the center of my cooling system.

I'm considering building a custom water loop just for the fan of it, I never had a water cooled system, but I think it would make sense only if I included graphics cards in the loop and water blocks for graphics cards are very hard to get here not to mention very expensive. I don't think reapplying TIM will give me much, there's very little dust in there, almost like fresh application and the contact is very good I checked the imprints on both IHS and the cooler.
BTW. It makes literally no sense to invest in faster DDR4, I payed 1300PLN for my memory when I could have bought 2133MHz for 900PLN and now I see even 2400MHZ available for also 900 which wasn't available just two weeks ago, way to waste money for 1% more performance.
ps. I'm testing ~4.4GHz but I had to increase the voltage to 1.225V just to achieve initial stability in LINX, now it's time for the full tests, but so far so good, ~4.4GHZ is close to what could be possible with water.

 

[Lepton87]

I just realized that for what I over-payed for RAM I could build a water loop, what a stupid waste of money. My initial rationalization for the faster DDR4 RAM was that if I'm already overpaying for DDR4 I want it to be at least faster than DDR3, which obviously I didn't think through. WC would be immensely more needed or helpful.
ps. ~4.4 seems to be fine at 1.225 but that's the cooling limit at least with LinX.

UPDATE: 4375MHz seems to be completly fine with 1.225V and it does not overheat even in the most strenuous loads, so if I wanted I could eek out even more MHz, probably by raising BCLK slightly. Next multi jump would require to much Volts not to overheat. Also there's still cache overclocking left on the table, with PH2 it gives very significant gains, any tests comparing 2.4GHz NB to 4GHz? I got mine clocked at 3GHz probably because of 125MHz strap.

 

[BonzaiDuck]

I just realized that for what I over-payed for RAM I could build a water loop, what a stupid waste of money. My initial rationalization for the faster DDR4 RAM was that if I'm already overpaying for DDR4 I want it to be at least faster than DDR3, which obviously I didn't think through. WC would be immensely more needed or helpful.
ps. ~4.4 seems to be fine at 1.225 but that's the cooling limit at least with LinX.

Well, after turning over in my mind everything folks were saying about the 4790K/Z97 and other Haswell thermal issues with stress-testing, I could incline to the view of OCCT's designer that his "OCCT:CPU" test is faster, more accurate and a few degrees less thermally troublesome.

Do you set "affinity" to every other thread when you run LinX?

The RAM question is another reason my Hassy-E-building plan includes a deliberate deferral until the next year. Of course, I'm surprised at how many different DDR4 kits are available right now.

 

[Lepton87]

Well, after turning over in my mind everything folks were saying about the 4790K/Z97 and other Haswell thermal issues with stress-testing, I could incline to the view of OCCT's designer that his "OCCT:CPU" test is faster, more accurate and a few degrees less thermally troublesome.

Do you set "affinity" to every other thread when you run LinX?

It only runs on physical cores, as can be seen in performance monitor.
I always leave virtual cores out of LinX as that's the way it should be done.
ps. that OCCT:CPU really leaves up to those claims, I had to increase voltage to 1.25 to be stable in it, when it was LinX stable at 1.225. And the temp is 20C lower in it at 1.25 than it was at 1.225V wiht LINX, massive! I think real world temp will be around that with OCCT a nowhere near LINX.
Power Consumption dropped from 380W to just 270W, that reduction alone is close to the CPU TDP!!! Few degrees less is a massive understatement it's probably close to half the power consumption in LINX.
20C less, 110W less on a 140W TDP CPU and catches errors faster, who would have thought?

 

[BonzaiDuck]

It only runs on physical cores, as can be seen in performance monitor.
I always leave virtual cores out of LinX as that's the way it should be done.
ps. that OCCT:CPU really leaves up to those claims, I had to increase voltage to 1.25 to be stable in it, when it was LinX stable at 1.225. And the temp is 20C lower in it at 1.25 than it was at 1.225V wiht LINX, massive! I think real world temp will be around that with OCCT a nowhere near LINX.
Power Consumption dropped from 380W to just 270W, that reduction alone is close to the CPU TDP!!! Few degrees less is a massive understatement it's probably close to half the power consumption in LINX.
20C less, 110W less on a 140W TDP CPU and catches errors faster, who would have thought?

There is a Dan Brown novel entitled "Digital Fortress," the plot involving NSA, and a scene in which a computer runs malicious code, heats up without restraint and explodes. Or so I vaguely recall . . .

I suspect that code can be written which might be called "computer abuse." Now we have this new instruction set component AVX2, but that aspect of the tests can be turned off in some programs. At least with IBT there is a scale of stress: Maximum, High, Standard or Normal -- I forget the precise labels, since I usually choose Maximum or High.

So one has to wonder, indeed, whether certain established stress-tests are feasible standards or needlessly excessive. Of course, I remember that IBT was first touted for its thermal stress.

I'm throwing out these thoughts and suggestions just to stimulate more exchange about it. I cannot be sure that I am completely correct, and it's something I'd like to see clarified.

But if it's true -- that a test like OCCT:CPU is a high standard without any excessive or needless thermal stress -- it makes an E-system seem more worth building.

I'm still inclined to find a couple 140x280x40mm radiators and a reservoir with two D5 pumps. And before I order the parts, I need to settle on either a mid-tower case-mod, or a new case -- possibly full-tower. I have more bizarre ideas, maybe for a bigger radiator . . . a little sheet-metal work. Time and trouble . . . time and trouble . . .

 

[BonzaiDuck]

It only runs on physical cores, as can be seen in performance monitor.
I always leave virtual cores out of LinX as that's the way it should be done.
ps. that OCCT:CPU really leaves up to those claims, I had to increase voltage to 1.25 to be stable in it, when it was LinX stable at 1.225. And the temp is 20C lower in it at 1.25 than it was at 1.225V wiht LINX, massive! I think real world temp will be around that with OCCT a nowhere near LINX.
Power Consumption dropped from 380W to just 270W, that reduction alone is close to the CPU TDP!!! Few degrees less is a massive understatement it's probably close to half the power consumption in LINX.
20C less, 110W less on a 140W TDP CPU and catches errors faster, who would have thought?

There's a Dan Brown novel entitled "Digital Fortress," with a plot that includes NSA and a scene in which malware running on a computer causes it to heat up and explode -- if I remember correctly.

The recent issues we'd seen here with the 4790K and Haswell cores in general seem to raise this question about heat, and whether or not traditional stress-tests are a "necessary" standard. I imagine that it's possible to write code which could be called "computer-abuse." So if the OCCT:CPU test reveals errors more robustly at lower temperatures, I could see the possibility that sufficient stress-tests could be written to produce less heat. We're also beneficiaries of a new instruction-set addition with "AVX2" and some of the other stress-test programs may offer a chance to turn off the AVX2 features.

IBT in particular had gained an early reputation for thermal stress.

Howsoever this may be, I submit these thoughts in hope if a more vigorous discussion: I for one want more "clarification." In the meantime, I am definitely looking at a case and watercooling selection as a first priority. But it's looking like the 5820K could be the popular candidate for over-clocking, unless newer production-runs of the other processors can prove greater over-clock-ability.

 

[Lepton87]

I don't think that newer 5820K will be better OCers, unless they want to release a new stepping which is very doubtful, but such a scenario happened with both Q6600 and i7 920 where new steppings overclocked much better, however we didn't see a repeat of that with both SB, IB and now HW. Also 22nm is very mature now and quite frankly even out-dated, there are first 14nm CPUs on the market, only low-power but still.

The recent issues we'd seen here with the 4790K and Haswell cores in general seem to raise this question about heat, and whether or not traditional stress-tests are a "necessary" standard.

I think LinX is totally unnecessary, even if your system throttles in it it won't everywhere else. On the bright side, my heater just broke down, so to heat up the room I just run LinX and Furmark, 1KW is the power that my heater had, my computer matches that Of course, it's only temporary better to buy a new heater then to needlessly stress the components.

 

[BonzaiDuck]

I don't think that newer 5820K will be better OCers, unless they want to release a new stepping which is very doubtful, but such a scenario happened with both Q6600 and i7 920 where new steppings overclocked much better, however we didn't see a repeat of that with both SB, IB and now HW. Also 22nm is very mature now and quite frankly even out-dated, there are first 14nm CPUs on the market, only low-power but still.

I think LinX is totally unnecessary, even if your system throttles in it it won't everywhere else. On the bright side, my heater just broke down, so to heat up the room I just run LinX and Furmark, 1KW is the power that my heater had, my computer matches that Of course, it's only temporary better to buy a new heater then to needlessly stress the components.

I had a college roommate from New Jersey -- a factory trained Chevy mechanic. One day, we drove by a house with a rusting 1955 BelAir on blocks. He bought it; eliminated any rust pits; primered it; had an engine shipped from NJ to CA with special parts. Hurst transmission -- all new. Black imitation satin button-tuck upholstery. All new drive train; beautiful chrome wheels; solid racing tires.

Somebody could build something like that and take it to Fontana or some other current part of a racing circuit -- just to tear it up.

Or, driving it on the street, and but for the occasional drag-race on Palmyrita Avenue in the wee hours -- well -- it was like a brand new ride, for a long, long time.

UPDATE: THIS LOOKS LIKE A GOOD ROG GUIDE

http://rog.asus.com/365052014/overclocking/rog-overclocking-guide-core-for-5960x-5930k-5820k/

 

[Lepton87]

One thing I haven't noticed before that negatively affects performance in some cases. With HW-E you have two memory multipliers either 1 or 4/3 don't use the latter. To run memory at 2666MHz the board automatically sets the second multiplier but for optimal performance it should be a multiplier of BCKL, with 125MHZ strap it should be either 2500 or 2750, 2666MHz will be slower than 2500. I increased the voltage to 1.35 and set the memory at 2750MHz, AFAIK every module beyond 2666MHz has the default voltage set at 1.35V so it should be perfectly safe, as for the effects of 4/3 multiplier:
http://pclab.pl/art59968-4.html

The effect isn't drastic in any way but it's worth it to set the memory at 1:1 multi.

 

[BonzaiDuck]

One thing I haven't noticed before that negatively affects performance in some cases. With HW-E you have two memory multipliers either 1 or 4/3 don't use the latter. To run memory at 2666MHz the board automatically sets the second multiplier but for optimal performance it should be a multiplier of BCKL, with 125MHZ strap it should be either 2500 or 2750, 2666MHz will be slower than 2500. I increased the voltage to 1.35 and set the memory at 2750MHz, AFAIK every module beyond 2666MHz has the default voltage set at 1.35V so it should be perfectly safe, as for the effects of 4/3 multiplier:
http://pclab.pl/art59968-4.html

The effect isn't drastic in any way but it's worth it to set the memory at 1:1 multi.

Lotta hard disks and SSDs there . . . . What computer case are you using?

 

[Lepton87]

Lotta hard disks and SSDs there . . . . What computer case are you using?

Honestly I don't remember. Fractal something.

 

[BonzaiDuck]

Fractal Designs . . .

Several folks have embraced compact or mATX cases for their high-powered builds:

http://forums.anandtech.com/showthread.php?t=2403342

The OP of the linked thread was actually able to get a 60mm-thick radiator in that little case, with another 2x 120 radiator in the top. It proves what can be done, but also proves the compromise for fewer drive bays.

Are both your rear and top fans used for exhaust? I'm also wondering if you plan to keep the D14 cooling strategy, or if this is just an interim test configuration before you upgrade to water-cooling. The D14 seems to be working for you, nevertheless.

 

[Lepton87]

I wanted to edit my post but made a new response instead sorry

 

[Lepton87]

Fractal Designs . . .

Several folks have embraced compact or mATX cases for their high-powered builds:

http://forums.anandtech.com/showthread.php?t=2403342

The OP of the linked thread was actually able to get a 60mm-thick radiator in that little case, with another 2x 120 radiator in the top. It proves what can be done, but also proves the compromise for fewer drive bays.

Are both your rear and top fans used for exhaust? I'm also wondering if you plan to keep the D14 cooling strategy, or if this is just an interim test configuration before you upgrade to water-cooling. The D14 seems to be working for you, nevertheless.

Yes both are used for exhausting the air and the one on the back is used as an intake fan.(I'm not sure which fan orientation is optimal for that fan)I still have to install two drives, an SSD and an HDD which were previously unused because ROG MAXIMUS IV(former mobo, it made me really like the ROG series, I think the series is top notch, also it was necessary back then because of a NF200 chip, a PCI-E splitter, I needed that. At first I had QuadFire made by 6990+2x6970, now the decision was harder because a simple mobo would suffice) only had 8 SATA ports. I don't think I will go much farther than 4.4GHz which I can do with AIR, so maybe 4.6 at 1.35V? Is that a safe voltage for a HW-E? Nonetheless I'm considering water because I will be able to reuse it for a few generations, provided I buy a block that will get a new mounting system if Intel once again changes the mounting holes. Why do they do that? I hope that the mounting system stays the same for a few years even if they change the socket. Unfortunately people that reuse their old cooling on a new system are a minority so Intel does not care
ps. Is that a good case in general? Does it make a good case for a water cooled rig? Cases are the area I'm the least knowledgeable. I remember that I chose that case because it was the only case with immediate availability that could take an 6990 , other cases couldn't take it as it was too long a card.

 

[Lepton87]

I've got a strange issue with chrome which was my browser of choice until now. When my computer boots into windows after some time the idle power draw stabilizes around 140W but when I run chrome it goes up to 280W and it never goes back to around 140, even after I quit the app. I'm currently running IE and the computer draws 132W right now, the same thing in chrome results in 280W power draw. I got used to chrome and was more comfortable browsing the Internet with it. What could be the case? I suspect the CPU is unable to go to C state and/or the graphic card is also unable to go to lowest performance state that is P8.

 

[BonzaiDuck]

I've got a strange issue with chrome which was my browser of choice until now. When my computer boots into windows after some time the idle power draw stabilizes around 140W but when I run chrome it goes up to 280W and it never goes back to around 140, even after I quit the app. I'm currently running IE and the computer draws 132W right now, the same thing in chrome results in 280W power draw. I got used to chrome and was more comfortable browsing the Internet with it. What could be the case? I suspect the CPU is unable to go to C state and/or the graphic card is also unable to go to lowest performance state that is P8.

Something doesn't seem right, unless you're talking about the OVERALL power draw for the CPU, mobo, RAM, SSD/HDDs. Somebody has to have an answer for this. What do you use for reporting total power draw? It seems the only thing I used for that was the application for my UPS system.

Look at task manager, see if one or more cores are loading up to 95 to 100%.

If it is your CPU and with EIST enabled, your processor power consumption should be below 30W. My 2600K system often shows something in excess of 14W.

What you're describing has me puzzled.

On the case and fan choices, there are several ways to do things. Usually, with the D14 configured to face the back panel, the rear fan could serve almost exclusively as exhaust. The way you have it set up, you might use the top and rear exhaust vents to exhaust air, or you could use the rear fan as intake and the top as exhaust. In the first possibility, you would want really sufficient -- ample -- intake airflow from other fans. In the second case, you could get really good CPU cooling, but I could wonder how much intake airflow you're getting, or whether the cooling for other components and the motherboard are ample.

I can describe my own D14 setup.

I have a HAF 922 midtower case:

http://www.newegg.com/Product/Produc...-197-_-Product

There was more than one variation of the 922, and I'm pretty sure this is the one. There are three vents which can handle fans up to 200mm diameter: The left side-panel; the top; and the front. I blocked off the top vent with foam-art-board secured with nylon screws and nuts, removing the top fan, and I blocked off the only 140mm fan vent on the case-bottom. The front LED fan of 200mm is comparatively limp among 200mm options but should guarantee up to 90 CFM of airflow. I installed an NZXT 200mm fan for the side-panel, rated at 166 CFM. Of course, the ratings don't guarantee that sort of airflow: the vents are a sieve of holes, so there's some amount of obstruction -- all vents are like that. But the two 200mm fans pressurize the case.

All unused vents or "extra" holes in the case are sealed and blocked off; air only comes in through fans, and goes out through the single exhaust fan.

The rear exhaust fan is a Gentle Typhoon AP-30 (120x25mm), which has a reputation for dBA noise level. However, the noise components are air-turbulence or white-noise, and a slight motor noise mostly noticeable in the range from 2,500 to 4,200 (top-end) range. The AP-30 is "rated" at about 116 CFM, but had been measured in tests to prove ~144 CFM.

The fan is ducted to the rear of my D14 cooler with a neatly constructed foam-board duct box padded inside and out (4 layers on the outside, one on the inside) of Spire acoustic foam rubber. The fan itself is also wrapped in four layers of Spire around its square 120mm frame. So the motor noise is so muffled at speeds up to 3,600 RPM that it's hardly noticeable. The duct draws all the pressurized air in the case through the D14 -- which has a center "round" fan or an AKASA Viper 140R -- 140mm with 120mm mounts. So air is drawn from within the case from all three sides of the cooler, with a fan behind each of the two towers. This lowered my load temperatures by about 5C degrees -- below simply having a good case exhaust fan and the two stock Noctua cooler fans. In other words, I only have a single fan hanging in the middle of the D14 and the case rear exhaust fan, instead of the original three fans.


All fans are thermally controlled from the motherboard. Idle RPMs are:

Front 200mm -- 350 RPM
Side 200mm -- 600 RPM
Center (Akasa) 140mm CPU fan -- 800 RPM
Rear exhaust AP-30 -- 1,300 RPM

At load:

Front 650 RPM
Side 1,300 RPM
Akasa 1,600 RPM
AP-30 3,200 to 3,600 RPM, over thermal range of 60 to 70+C

That should give you an idea about cases and airflow. I think that finding an appropriate case for an ample water-cooling system is the more difficult match to make, and some folks have made it apparent that you need to cut sheet-metal or make some case-mods to fit radiators of a desired size to chosen cases. Folks may not like the HAF cases because they're flat-black and seem almost ugly. Or others may have an aversion to large fans with lower static pressure (but lower noise). But I made these choices together with the NH-D14 cooler as part of a strategy, and it turned out better than I'd hoped.

 

[BonzaiDuck]

Gotta correct my last post . .

My processor "package" wattage is more like 40+W at idle. With a lot of hard disks or peripherals, OVERALL system wattage could easily be 140W at idle. But your system should exhibit the same total wattage when it's in an idle state, so what you described is . . . puzzling.

 

[Lepton87]

I report the power draw of the whole system at the WALL so you have to take the PSU efficiency into account and it tops out at 92% according to the manufacturer. I don't have the means to isolate the CPU power draw. I measure the power draw with a power meter that is sometimes called a KILL-A-WATT device.
BTW. The high power draw in chrome is the result of both Titans going into P0 performance states while they stay at P8 performance states at idle, the difference between the power draw of P0 and P8 performance states is huge, at the P8 the GPU is clocked at just 328MHz, memory is also severely downclocked voltage is also appropriately lower. P0 is a fully clocked card, there are intermediate performance states like P5 and it's strange that chrome goes straight to P0 for GPU acceleration, one of the intermediate performance states would suffice.
ps. can you take a picture of your case? I'm specifically interested in the ducting, because you said it lowered the temp by 5C.

 

[BonzaiDuck]

Well . . . you have a picture of the 922 case in an earlier link.

The computer is sandwiched between my couch and my server, with a zillion USB and other connections, so I did "quick and dirty" with the flash attachment. First off, the flash really shows the dust: I had blown the kruft from the case just last month, and I can see I'll need to do it soon again. Not a polished and "bling-y" promotion, but the object of interest is clearly shown -- albeit at an angle.

The project itself -- is in this thread:

http://forums.anandtech.com/showthread.php?t=2370973&highlight=gentle+typhoon

Not presented all that well. Look for my posts and pictures: there are three pages in the thread, and I'd suggest you scan through, find my posts with photos and diagrams.

There is an additional use of the Spire foam-rubber on the foam-board panel which blocks the case-top fan-vent. In the picture I just took, it's hard to see much of the "collar" which fits over the AP-30 fan, but the diagram in the linked thread explains it. It is easy to remove the side-panel, which holds the NZXT 200mm fan (also pictured in the thread, but there is only so much clearance to mount the panel and slide it forward into locking position -- not any trouble, really.

The duct is almost a snug fit, but it pulls out of the case sideways, and otherwise has a baffle that rests on top of the D14 cooler. It all had to be done with precision cutting, and you have to gently wiggle it here and there to remove it or install it, but again -- fairly easy.

There is nothing "bling-y" about this, but it works. All air that goes into the case, goes through the D14 towers and out the back. All I can say -- it works. Won't take any "Bling" awards, though:

 

[Lepton87]

thanks, how much time did you invest in it? Tomorrow or rather much later today I'm going to drive to a shop to buy some TIM and clean both the cooler and the IHS and apply it fresh. I'm still not sure if I should buy some more IC diamond or some of that liquid metal stuff.