I have now run my 4770K in my Maximus VI Formula with EK CPU cooler, EK chipset cooler, GTX780Ti with EK fullblock cooler, EK block on MOSFETS for about 2 months. I have tested putting the CPU to be directly after the pump vs having it directly after the XTX360 radiators. I am using the EKWB 4.0 Pump which has somewhat good pressure. Everything is in one loop and the GPU is running cool after the CPU which is the direct limiting factor on heat now.
With the setup I get a stabile setup to run Realbench comfortably at 4.5 GHz and at 4.6 GHz on a cool day only. I manage to get some Benchmarking softwares to run ok on 4.9Ghz.
I delidded the CPU and put some GC extreme TIM on the CPU which helped me get from 4.5 GHz to 4.6 GHz for every day use where i can run the CPU at below 80C with the general ambient temperature being 26 - 27C where i am.
I am happy with the scores I get on various benchmarks which can be seen below
I have tried to squeeze more out of the system but seems what ever I do will crash the system due to clocks being to fast in combination with heat mainly on CPU. Thus 2 months with the same hardware and a lot of tests if I missed something to tweak it would be in minute details on memory settings and so forth.
So what is next.. What would give me the best bang for the buck to step it up to 4.7 - 4.8 as a 24/7 setup with a few benches on 5.0 GHz for example.or even reach 4.9GHz more stabile as a general running machine.
I have ordered the liquid metal so will redo the de-lidding which should give me a better result.
But then what?
Would it be to add another 360 radiator with more fans or would it be getting a mega pump to get a superflow? knowing that the GPU runs full time at 1.212 Volts fully stable in any benchmark (40C) and I cant physically get more voltages out of the card without a hardware modification I am questioning if making 2 loops really worth the extra hardware or would it be more cosmetic improvements. Am I really using the radiators to the max or would a higher flow be able to get cooler water to the CPU or is the radiator the real limiting factor rather than the flow?
At this juncture what is the real limiting factor in the system to get the CPU heat down which seems to be the stopping block in all tests I have done.
In the end I am planning to get a Hailea 500A to get some properly cool water but at this time I am more looking at the fastest system I can get into the corsair 900D for a 24/7 operation without such additional cooling system. I am not into filling the case up for the cosmetics, thus interested in just what makes the most sense performance wise.
Any ideas ? any hints on where i can read more info on such? I keep searching on the web but have no real conclusive answers yet.
It started with a VIC-20... and now it is. Maximus Formula VI Intel 4770K 4.6GHz ASUS GTX780Ti 3GB Corsair Platinum Dominator 16GB 2400MHz Raid Express 240GB Samsung EVO 512 GB SSD Corsair 900D EK watercooling Corsair RM1000 ASUS PB278Q
With the setup I get a stabile setup to run Realbench comfortably at 4.5 GHz and at 4.6 GHz on a cool day only. I manage to get some Benchmarking softwares to run ok on 4.9Ghz.
I delidded the CPU and put some GC extreme TIM on the CPU which helped me get from 4.5 GHz to 4.6 GHz for every day use where i can run the CPU at below 80C with the general ambient temperature being 26 - 27C where i am.
I am happy with the scores I get on various benchmarks which can be seen below
I have tried to squeeze more out of the system but seems what ever I do will crash the system due to clocks being to fast in combination with heat mainly on CPU. Thus 2 months with the same hardware and a lot of tests if I missed something to tweak it would be in minute details on memory settings and so forth.
So what is next.. What would give me the best bang for the buck to step it up to 4.7 - 4.8 as a 24/7 setup with a few benches on 5.0 GHz for example.or even reach 4.9GHz more stabile as a general running machine.
I have ordered the liquid metal so will redo the de-lidding which should give me a better result.
But then what?
Would it be to add another 360 radiator with more fans or would it be getting a mega pump to get a superflow? knowing that the GPU runs full time at 1.212 Volts fully stable in any benchmark (40C) and I cant physically get more voltages out of the card without a hardware modification I am questioning if making 2 loops really worth the extra hardware or would it be more cosmetic improvements. Am I really using the radiators to the max or would a higher flow be able to get cooler water to the CPU or is the radiator the real limiting factor rather than the flow?
At this juncture what is the real limiting factor in the system to get the CPU heat down which seems to be the stopping block in all tests I have done.
In the end I am planning to get a Hailea 500A to get some properly cool water but at this time I am more looking at the fastest system I can get into the corsair 900D for a 24/7 operation without such additional cooling system. I am not into filling the case up for the cosmetics, thus interested in just what makes the most sense performance wise.
Any ideas ? any hints on where i can read more info on such? I keep searching on the web but have no real conclusive answers yet.
MatsGlobetrotter wrote:
I have now run my 4770K in my Maximus VI Formula with EK CPU cooler, EK chipset cooler, GTX780Ti with EK fullblock cooler, EK block on MOSFETS for about 2 months. I have tested putting the CPU to be directly after the pump vs having it directly after the XTX360 radiators. I am using the EKWB 4.0 Pump which has somewhat good pressure. Everything is in one loop and the GPU is running cool after the CPU which is the direct limiting factor on heat now.
I have ordered the liquid metal so will redo the de-lidding which should give me a better result.
But then what?
Would it be to add another 360 radiator with more fans or would it be getting a mega pump to get a superflow? knowing that the GPU runs full time at 1.212 Volts fully stable in any benchmark (40C) and I cant physically get more voltages out of the card without a hardware modification I am questioning if making 2 loops really worth the extra hardware or would it be more cosmetic improvements. Am I really using the radiators to the max or would a higher flow be able to get cooler water to the CPU or is the radiator the real limiting factor rather than the flow?
At this juncture what is the real limiting factor in the system to get the CPU heat down which seems to be the stopping block in all tests I have done.
In the end I am planning to get a Hailea 500A to get some properly cool water but at this time I am more looking at the fastest system I can get into the corsair 900D for a 24/7 operation without such additional cooling system. I am not into filling the case up for the cosmetics, thus interested in just what makes the most sense performance wise.
Any ideas ? any hints on where i can read more info on such? I keep searching on the web but have no real conclusive answers yet.
Mats,
I still like the looks of your rig as I said in the world cup thread. It's even better with the power cables connected - they're colorful. Your scores are high and plentiful. Your standing on the team speaks for itself. My comments are mostly 'you're good enough' with only one suggestion.
Note that if you use a chiller, you will be in the OC league at HWBOT, no longer in the enthusiast league.
You got the first thing -- liquid metal -- on the way. After or as part of delidding, that makes the biggest difference in my experience. The liquid metal is rated to conduct 9 times more watts of heat than the best of the others. (82 W/mK vs 9) I use the liquid metal between lid and chip and between lid and waterblock. Neither place gives me a thermal bottleneck. Be sure to cover that row of tiny parts near the chip with nail polish or silicon caulk to keep the electrically conductive liquid metal away. The liquid metal WILL squish out and run down the side of the chip.
As an example, for this week's world cup challenge, I ran Realbench at 4800MHz with insane 1.65Vcore. The CPU drew over 185 watts, but stayed under 83C. I have a lot of confidence in a watercooling rig with delidding and liquid metal.
Two loops, series/parallel splits and the like make very little difference as long as the pump keeps the water moving well enough, which yours should be doing.
Added radiators can make a difference. Water temperature rise depends on the difference between watts of power into the water and watts out through the radiator. Do you have a sensor for water temperature? There must always be temperature differences from CPU to waterblock to water to radiator to air. Temperature difference is what drives heat through the cooling system. Water temperature will rise to the point needed to get the watts out through the radiator. More radiator area lets the heat get out with lower water temperature. Lower temperature water will cool the CPU better, degree for degree.
http://koolance.com/radiator-2-fan-140mm-30-fpi-copper
The specification tab on the page above shows this quantified to an extent. Water flow rate has an effect and so does air flow. Note that the biggest difference is in the size of the radiator -- 1 vs 2 vs 3 fans. The same happens when radiators are added -- total size provides a path for power to get out and sets the temperature difference required. The bottom line is that a good 2x140mm radiator like mine keeps water temperature below 34C in extended benching with room air at 28C. That's cooling the CPU and M6F VRMs. Water can keep the chipset cool, but there aren't many watts there. You also cool the GPU. That can add more watts than the CPU and call for a whole second radiator. For that whole load, I would use 2 2x140mm or 2 3x120mm radiators. If that's what you have, you probably don't need more. With that huge case, a third radiator might fit easily make a contribution. It depends on water temperature.
The only suggestion I have is to cool the RAM. Some use little air fans blowing down on the sticks. From what I read about it, I have my eye on an EK RAM watercooler that's built to sit on the heat spreaders of Dominators. They also have replacement heat spreaders to work with other models like mine. That could add another 20 watts from two RAM sticks to the water -- not enough to drive increasing the radiator.
The advantage to cooling ram is that there is room for some tweaking when the memory chips stay cool -- particularly in a shorter refresh interval.
Hope this rambling helps.
Jeff
Thanks Jeff,
Yes it helps :cool: so would not call it rambling. I had attached the wrong pic in the other thread. I only have the one EK XTX360 radiator with 6x 120mm at 2400rpm fans in push pull configuration so there is plenty room for more of what would be needed.
The chipset got hot when feeling it with my fingers which is why I got the block kit for it but I am starting to think that it just add resistance to the waterflow compared to benefit for the overall system compared to the standard air one.
The EK MOSFET block somehow looks to be just a tad further away (not closer as I would have expected!!) to the MOSFETS compared to the original I am not sure it actually contributed to anything positive overall.
I could see a big difference of delidding but there as you mention some interesting drops in heat with the liquid metal I can look forward to. Yes will cover the small components in the vicinity with nailpaint. I found a video on Youtube on the whole process which was good to see before I made the attempt 🙂
I did put some temporary fan on the memory when I OCd them to the 2600 mem stage which works fine at 1.85 Volts and the configuration for Samsung in the Bios tweak. It took some time to figure it out as in the beginning it crashed full time. The memory surely get hot when running at the higher voltage so cooling would be needed to keep them in reasonable shape for longer and tougher loads.
I keep looking at your Realbench score and wondered how on earth you could get it across 90 K :cool: as that is a tough mark to pass so Congrats on that one. I have run my 4770K too hardnhot which is probably also why it is a acting up on Realbench now but needed to learn anyway.
What I can read out from your post though is that it would be an idea to go ahead with another radiator and with that I will surely need another pump as it is a lot of water to pump through the various pieces. I checked the Koolance page and understand the concept. I could see less turbulent water in the reservoir after putting in the chipset cooler and adding some 90degree angles to straighten out the tubing which seems to kink too easily otherwise. With that I do assume that the flow should be less even if the pump is strong.
Hmm the initial idea was to just watercool the system overclock a bit but hmm this overclocking seems to be addictive 😉 so that is why the cables are sleeved but not finalized yet ... Think I will remain in the enthusiast league a bit longer before going for the chiller.
It started with a VIC-20... and now it is. Maximus Formula VI Intel 4770K 4.6GHz ASUS GTX780Ti 3GB Corsair Platinum Dominator 16GB 2400MHz Raid Express 240GB Samsung EVO 512 GB SSD Corsair 900D EK watercooling Corsair RM1000 ASUS PB278Q
Yes it helps :cool: so would not call it rambling. I had attached the wrong pic in the other thread. I only have the one EK XTX360 radiator with 6x 120mm at 2400rpm fans in push pull configuration so there is plenty room for more of what would be needed.
The chipset got hot when feeling it with my fingers which is why I got the block kit for it but I am starting to think that it just add resistance to the waterflow compared to benefit for the overall system compared to the standard air one.
The EK MOSFET block somehow looks to be just a tad further away (not closer as I would have expected!!) to the MOSFETS compared to the original I am not sure it actually contributed to anything positive overall.
I could see a big difference of delidding but there as you mention some interesting drops in heat with the liquid metal I can look forward to. Yes will cover the small components in the vicinity with nailpaint. I found a video on Youtube on the whole process which was good to see before I made the attempt 🙂
I did put some temporary fan on the memory when I OCd them to the 2600 mem stage which works fine at 1.85 Volts and the configuration for Samsung in the Bios tweak. It took some time to figure it out as in the beginning it crashed full time. The memory surely get hot when running at the higher voltage so cooling would be needed to keep them in reasonable shape for longer and tougher loads.
I keep looking at your Realbench score and wondered how on earth you could get it across 90 K :cool: as that is a tough mark to pass so Congrats on that one. I have run my 4770K too hardnhot which is probably also why it is a acting up on Realbench now but needed to learn anyway.
What I can read out from your post though is that it would be an idea to go ahead with another radiator and with that I will surely need another pump as it is a lot of water to pump through the various pieces. I checked the Koolance page and understand the concept. I could see less turbulent water in the reservoir after putting in the chipset cooler and adding some 90degree angles to straighten out the tubing which seems to kink too easily otherwise. With that I do assume that the flow should be less even if the pump is strong.
Hmm the initial idea was to just watercool the system overclock a bit but hmm this overclocking seems to be addictive 😉 so that is why the cables are sleeved but not finalized yet ... Think I will remain in the enthusiast league a bit longer before going for the chiller.
"how on earth you could get it across 90 K "
The big step to getting from the 87-89K range to over 90 for me was tweaking RAM. That was of more benefit thans OS trimming. My mixed kit of DRAM breaks all the rules, but it works with the Haswell IMC very well. Secondary timings and RAS Active came out much tighter than XMP. RAM bandwidth measured in Aida64 went from 35GB/s to nearly 37GB/s. Scores went up proportionally in RB and better than proportionally in XTU.
The guide
http://rog.asus.com/242142013/labels/rog-exclusive/maximus-vi-series-uefi-guide-for-overclocking/
goes into the RAM tweaks of BIOS. The method of tweaking within those parameters is to:
1. reduce BIOS settings one at a time to remove confusion about what went wrong. Sometimes a tweak depends on others. For example RAS Active time depend on the sum of three others. Four active window also depends on a setting that should be attended to first. Don't reduce Refresh Interval -- the really large number. You can leave it on auto. Increase it if compelled to tweak, but watch for memory errors.
2. With just one setting adjusted from a known good profile, try to boot. If it works, great. If not, the MEMOK button is your friend. When tweaking memory, MEMOK is faster and less destructive of BIOS settings that clearing CMOS.
3. Run some benchmark that checks RAM bandwidth. Memtweakit is a calculation. Aida64 is a tested bandwidth measurement. RB image processing is probably a good indicator since it's sensitive to RAM speed.
4. If bandwidth improved, keep the change. Otherwise revert back. Latency boundary worked that way for me--had sweet spots around 10-12 but was worse on either side of that range.
5. Rinse and repeat for the next setting.
The tedious part is that each tweak has to be done in BIOS with the accompanying reboot. Took all day.
90 degree angles will do it to water flow much more than an EK chipset block. I'll use angled fittings where I need a fitting anyway. There isn't much difference -- any fitting is a little constriction. 30 or 45 degree fittings work better than 90's in theory and work for my layout, to that's what I use.
Jeff
The big step to getting from the 87-89K range to over 90 for me was tweaking RAM. That was of more benefit thans OS trimming. My mixed kit of DRAM breaks all the rules, but it works with the Haswell IMC very well. Secondary timings and RAS Active came out much tighter than XMP. RAM bandwidth measured in Aida64 went from 35GB/s to nearly 37GB/s. Scores went up proportionally in RB and better than proportionally in XTU.
The guide
http://rog.asus.com/242142013/labels/rog-exclusive/maximus-vi-series-uefi-guide-for-overclocking/
goes into the RAM tweaks of BIOS. The method of tweaking within those parameters is to:
1. reduce BIOS settings one at a time to remove confusion about what went wrong. Sometimes a tweak depends on others. For example RAS Active time depend on the sum of three others. Four active window also depends on a setting that should be attended to first. Don't reduce Refresh Interval -- the really large number. You can leave it on auto. Increase it if compelled to tweak, but watch for memory errors.
2. With just one setting adjusted from a known good profile, try to boot. If it works, great. If not, the MEMOK button is your friend. When tweaking memory, MEMOK is faster and less destructive of BIOS settings that clearing CMOS.
3. Run some benchmark that checks RAM bandwidth. Memtweakit is a calculation. Aida64 is a tested bandwidth measurement. RB image processing is probably a good indicator since it's sensitive to RAM speed.
4. If bandwidth improved, keep the change. Otherwise revert back. Latency boundary worked that way for me--had sweet spots around 10-12 but was worse on either side of that range.
5. Rinse and repeat for the next setting.
The tedious part is that each tweak has to be done in BIOS with the accompanying reboot. Took all day.
90 degree angles will do it to water flow much more than an EK chipset block. I'll use angled fittings where I need a fitting anyway. There isn't much difference -- any fitting is a little constriction. 30 or 45 degree fittings work better than 90's in theory and work for my layout, to that's what I use.
Jeff
Thanks Madcratebuilder,
hmm Delta T . you are talking greek to me 🙂 my ambient temp is normally 26-28 C and at about 1.488 Volt and pushing the CPU hard on clocks I get about 81 -85 C. If I run the process on 4.5GHz at 1.392Volt the machine is perfectly stable and I get around 74-78 C on all cores with Realtemp reporting a peak at 83C on one core after 30 minutes of stress test in Realbench. if the machine is idle or im browsing the web with Windows loaded it stays around 33-39C with what 1- 2.5% load. So not sure what that means in Delta T.
Jeff,
I also did tweak my memory when I got my 87k score. Despite the proposed (dont use memtweaker as it gives wrong info and score not relevant) I used it for checking settings and the DRAM efficiency speed. The standard score with my 2x8GB 2400 Dominators were reported art 37800 points or around there. I found a tweaked profile for the same type mem on this site ..hmm have no clue where anymore, which had settings giving 42224 and I tried it and it worked fine enough to push my score.
Right now I have the memory running at 1301 MHz XMP 2600 with a score of 41690 at 1.85Volt instead of the normal setting. which works out fine also. Since I cant get realbench to work on 4600 Mhz any more on my machine I cant see if there would be a difference in the score except to say that I have not been able to match my score on even the GIMP.
I had for those runs a completely clean Windows, and turned of just about everything on the Formula also.
Including deinstallation of the bloatware from Nvidia, Arne´s windows bacthfile to stop various services, turned of just about everything else manually in task manager and put up realbench to high priority. even put realbench on a ramdisk-moved the cachefile there too. I did not get the top score though until I got the 337.88 Nvidia drivers. That one kicked up the score a bit. On the Realbench score list I only see one! guy Folex with a clockspeed of just above 4600 Mhz reaching 90K Everyone else is on 4700 - 5100 Mhz and that on a Maximus VI Extreme which might have a few component being superior to Formula? I tried the BCLK then also but with the mem tweaks on timings that was futile. So
I have gone through the link you sent also and see a few more thing I could try but see that I have touched most of them already. None of the hints there gives me stability on 4600 Mhz for Realbecnh though so I believe I have done the damage I can on that score.
Regards,
It started with a VIC-20... and now it is. Maximus Formula VI Intel 4770K 4.6GHz ASUS GTX780Ti 3GB Corsair Platinum Dominator 16GB 2400MHz Raid Express 240GB Samsung EVO 512 GB SSD Corsair 900D EK watercooling Corsair RM1000 ASUS PB278Q
hmm Delta T . you are talking greek to me 🙂 my ambient temp is normally 26-28 C and at about 1.488 Volt and pushing the CPU hard on clocks I get about 81 -85 C. If I run the process on 4.5GHz at 1.392Volt the machine is perfectly stable and I get around 74-78 C on all cores with Realtemp reporting a peak at 83C on one core after 30 minutes of stress test in Realbench. if the machine is idle or im browsing the web with Windows loaded it stays around 33-39C with what 1- 2.5% load. So not sure what that means in Delta T.
Jeff,
I also did tweak my memory when I got my 87k score. Despite the proposed (dont use memtweaker as it gives wrong info and score not relevant) I used it for checking settings and the DRAM efficiency speed. The standard score with my 2x8GB 2400 Dominators were reported art 37800 points or around there. I found a tweaked profile for the same type mem on this site ..hmm have no clue where anymore, which had settings giving 42224 and I tried it and it worked fine enough to push my score.
Right now I have the memory running at 1301 MHz XMP 2600 with a score of 41690 at 1.85Volt instead of the normal setting. which works out fine also. Since I cant get realbench to work on 4600 Mhz any more on my machine I cant see if there would be a difference in the score except to say that I have not been able to match my score on even the GIMP.
I had for those runs a completely clean Windows, and turned of just about everything on the Formula also.
Including deinstallation of the bloatware from Nvidia, Arne´s windows bacthfile to stop various services, turned of just about everything else manually in task manager and put up realbench to high priority. even put realbench on a ramdisk-moved the cachefile there too. I did not get the top score though until I got the 337.88 Nvidia drivers. That one kicked up the score a bit. On the Realbench score list I only see one! guy Folex with a clockspeed of just above 4600 Mhz reaching 90K Everyone else is on 4700 - 5100 Mhz and that on a Maximus VI Extreme which might have a few component being superior to Formula? I tried the BCLK then also but with the mem tweaks on timings that was futile. So
I have gone through the link you sent also and see a few more thing I could try but see that I have touched most of them already. None of the hints there gives me stability on 4600 Mhz for Realbecnh though so I believe I have done the damage I can on that score.
Regards,
Using a waterchiller wont put you in the OC league, half of the people in enthusiast are using chillers 🙂
I don't think there's much you can really do about you water loop setup to get better cpu temps. The limiting factor there is the TIM, and replacing it with LM is a good idea. I got great temps with LM on my Ivy bridge cpu's.
Adding another rad will help your temps during longer sessions of loading the system. For example if you're gaming for hours. For benching which only puts load on the system for a short period of time it won't make much of a difference.
I don't think there's much you can really do about you water loop setup to get better cpu temps. The limiting factor there is the TIM, and replacing it with LM is a good idea. I got great temps with LM on my Ivy bridge cpu's.
Adding another rad will help your temps during longer sessions of loading the system. For example if you're gaming for hours. For benching which only puts load on the system for a short period of time it won't make much of a difference.
Ok Madcratebuilder thanks for the input. I did infact get hold of an inline water temperature sensor and that was a sound investment;)
I bought a relatively cheap phobya sensor with the attached display.
I did not want a fancy automatic controller box.. Have a Bitfenix Recon thingy that keep beeping everytime I look at my computer as there is some sensor somewhere triggering.
I installed the sensor itself after the last GTX780Ti as that is where the hottest water would come out. At idle the temperature is close to room temperature. Think about 1 or 2 C higher, which I consider good.
The main question I initially had was, what next I have thrown in a watercooled CPU, VRM cooler, Chipset cooler and a Watercooled GTX 780Ti which I had ran flat out on 1.212 Volt as minimum GPU voltage just to see when I potentially would need to do something different in the fans/Radiator setup. Despite running the CPU at 4600 Mhz I had not seen the general Core temperatures go up and the GPU reported a max of 40 - 41 C. Nothing I would potentialy gain with another radiator and more fans would give me more headroom on the CPU for example. The issue on the 4770K would rather be the TIM. I de-lidded that one and could run it faster.
However now with a de-lidded CPU and a second GTX780TI in the box I start seeing that I would need another 360 radiator atleast. If I run A hot SLI run with 3DMark Extreme or Catzilla for example the run will start at 26-27 C and finish at around 35- 35 C which when running nice speeds on the bencher will sometimes crash the run. Most of the time its the GPU driver that will stop working but I can also see that I am coming to hot on the CPU. So with 2x GTX780 Ti getting scores of 10,786 on Firestrike Extreme (1347/1828 on the GTX's) or catzilla1440p up to scores of 18,000 (1271/1824) and a I7 4790K running at 4700Mhz in both cases will see me hit the roof as the water get too warm. Thus the 5 minute runs or so will heat up the water and system enough to go above stable limits.
Again do note that I am in a warm country with ambient temp being around 25-27C all the time.
With what I see now on how the GPU's heat up I am most likely to put in a separate loop for the CPU, VRM's and another loop for the GPUS and chipset. Im still holding off on a chiller as I want to still see how much I can get out of this machine with everything in the case itself. (thus no external gadgets for now even though its tempting to chase after Menthol and his scores 🙂
Regards,
It started with a VIC-20... and now it is. Maximus Formula VI Intel 4770K 4.6GHz ASUS GTX780Ti 3GB Corsair Platinum Dominator 16GB 2400MHz Raid Express 240GB Samsung EVO 512 GB SSD Corsair 900D EK watercooling Corsair RM1000 ASUS PB278Q
I bought a relatively cheap phobya sensor with the attached display.
I did not want a fancy automatic controller box.. Have a Bitfenix Recon thingy that keep beeping everytime I look at my computer as there is some sensor somewhere triggering.
I installed the sensor itself after the last GTX780Ti as that is where the hottest water would come out. At idle the temperature is close to room temperature. Think about 1 or 2 C higher, which I consider good.
The main question I initially had was, what next I have thrown in a watercooled CPU, VRM cooler, Chipset cooler and a Watercooled GTX 780Ti which I had ran flat out on 1.212 Volt as minimum GPU voltage just to see when I potentially would need to do something different in the fans/Radiator setup. Despite running the CPU at 4600 Mhz I had not seen the general Core temperatures go up and the GPU reported a max of 40 - 41 C. Nothing I would potentialy gain with another radiator and more fans would give me more headroom on the CPU for example. The issue on the 4770K would rather be the TIM. I de-lidded that one and could run it faster.
However now with a de-lidded CPU and a second GTX780TI in the box I start seeing that I would need another 360 radiator atleast. If I run A hot SLI run with 3DMark Extreme or Catzilla for example the run will start at 26-27 C and finish at around 35- 35 C which when running nice speeds on the bencher will sometimes crash the run. Most of the time its the GPU driver that will stop working but I can also see that I am coming to hot on the CPU. So with 2x GTX780 Ti getting scores of 10,786 on Firestrike Extreme (1347/1828 on the GTX's) or catzilla1440p up to scores of 18,000 (1271/1824) and a I7 4790K running at 4700Mhz in both cases will see me hit the roof as the water get too warm. Thus the 5 minute runs or so will heat up the water and system enough to go above stable limits.
Again do note that I am in a warm country with ambient temp being around 25-27C all the time.
With what I see now on how the GPU's heat up I am most likely to put in a separate loop for the CPU, VRM's and another loop for the GPUS and chipset. Im still holding off on a chiller as I want to still see how much I can get out of this machine with everything in the case itself. (thus no external gadgets for now even though its tempting to chase after Menthol and his scores 🙂
Regards,
MatsGlobetrotter your rig is looking good bro. I say the only thing left to do is lower your coolant temp down. you did see my cool keg I put together? or more rad surface area.
