The NRZ N2 Waterblock Beta Thread

[thenrz]

OCCT Linpack, 24 threads, 13500MiB memory (which left me with like 400-500MiB available) @ 10 minutes:

View attachment 209485


So, coming from a guy who had his head in the sand for years when it comes to pretty much everything computer enthusiast related, are these temps competitive? I'm pretty sure 145W is a limit set by the CPU because I can't seem to cross anything more than 145 and change. It is really cool to see the temps shoot up and then the liquid loop reach a level of homeostasis and everything levels out almost completely. The temps in the OCCT and IBT tests were surprisingly level except during the punch out phase or what I was seeing for about 20 seconds every 3-4 minutes where utilization went to 100% but wattage load went down under 100W. I want to understand these tests, but I want to understand how I am getting lower temps with a bigger chip.

I seriously think it's a package topology/topography thing. My coldplate geometry is more effective with the 3900X than the 3600X. This leads me to think that there's a really deep layer of modularity that could be tapped into when it comes to selling blocks. It's something that one would need to try to perceive how far to take it and in what direction. MCM CPU packages certainly have introduced some interesting variable.

 

[EarthDog]

You're definately going to want to set a static overclock and voltage. You'll likely end up around 4.2-4.3 ghz all c/t and 1.3V or so.

I'm also not so sure you've saturated the loop with such a brief test. For typical loops, about 1L of liquid, the prior wisdom was 30 mins or so. But hell, I've run and slowly went up for hours, but my room temp went up a couple of degrees C during that time too.

I'd also use coretemp to double check temps.

 

[trents]

Good point made by Blaylock about how Ryzen will accommodate itself to the cooling to prevent overheating.

But better cooling will allow the cores to spend more time in the higher frequency ranges under stress so if you were using HWINfo64 instead of HWMonitor it will give you an average figure over time for the core frequencies. It has a fourth column for average temp, frequency, voltage, whatever and gives a lot more information than does HWMonitor. Better cooling should give you higher average core frequencies and lover average temps.

I think AMD reserves their best silicon for their highest end chips. If I'm correct, they would need lower voltage which would produce lower temps perhaps than lesser chips with fewer cores. Perhaps that explains, in part at least, why chips with more cores are given the same TDP rating as their lower core count brothers.

 

[thenrz]

Anyone have advice on a nice temperature probe hub? I'm going to order the inline temperature modules that earthdog linked and I want something with a screen and preferably an interface to pipe the data into the PC and be able to log/graph it. I'm going to out a pair around the block and a pair around the radiator.

May or may not try to find a flow meter, too.


EDIT: Found an amazing deal on the inline temp probes. Got 4 of them for less than the price of 2. Thinking I'm going to order a 5th to have a spare. I can't find a good way to display the temps at all. I know my motherboard has a temperature probe input, but I highly doubt I can multiplex them. I haven't found the equivalent of a fan hub for just temperature probes... Perhaps I will end up having to pick up another Arduino clone and just building something. The advantage here is that there are a ton of pre-written libraries for temperature probe polling, and I could probably just program a little screen to output the current temps. It would be easy to log everything, too. I think I could just utilize the internal USB 2 header for connection as well. The problem is this is all a lot of time just to solve one problem when I have so many other things to do.

 

[thenrz]

Learned a lot this evening. Forgive my ignorance up to this point regarding what exactly I have in terms of a CPU and Ryzen in general. I've been pretty out of it considering I haven't owned a modern computer in quite a while now, and I had no idea of the incredible technology I was dealing with. It's not just clock speed anymore. There's a lot going on with something as state of the art as Ryzen and the engineering behind it.

I went ahead and did a stock settings run of OCCT again for just under an hour. The loop and system reached a plateau and things were stable in terms of temperatures, power usage, and general behavior. While this was happening, I was doing some reading on my phone. You know what I discovered? PBO isn't enabled by default. Go ahead and laugh and mock as should happen... I'll wait and take it gracefully... ... ... ... ... OK, so after diving into some technical literature, I realized that yes, I had reached and held with ease, the stock power limits. I think with a first beta revision that the block did well and would be able to handle any real world scenario with much aplomb. 145W held continuously at homeostasis for almost an hour at 60C.

After stopping that test and downloading Ryzen Master, I selected the PBO option with everything the way is it by default. What I really wanted to see was how much more power would be drawn by the processor. And trents, the HWiNFO tip was solid. I had tried it before, but it seemed like a data overload at the time. Now, it has just what I need. With PBO active, I started the test again without letting the loop return to a ground state (within 2 minutes). As the test ramped up, the system hit and held at ~180.951W which is a substantial 25% increase. Everything held just under 70C for the duration of the 14 minute test. I stopped it early as the VRM on the motherboard was hot at 70C-72C... maybe it can run that hot, but I felt the need to exercise more control over what was happening and there isn't any active cooling right there at the moment. I was coming off a one hour test at stock maximums and the system was already carrying additional heat. I need to consider my options when it comes to rectifying the VRM heat issues... I may look at taking the heatsink off and seeing what I could do about cutting a simple coldplate and acrylic block for that area. Something easy buy vastly superior to passive cooling. It would also be good practice for adding mosfet blocks to my product line.

I think it's pretty safe to say that I am approaching the maximums of potential heat production within the scope of a vast majority of cases. Obviously those working to truly take the CPU clocks to the clouds could generate a good deal more heat energy that must be moved from the package, but I think trying to appeal to the edge cases would require a different approach to block design. The N3 block is probably something that would be suited more towards that realm as it is a fair bit larger and utilizes a different design. I believe the biggest issue is the flow pattern is too open. I think I need to do some more flow direction with the pressure plate while trying to avoid unnecessary restriction. Now that I have some hands on experience and better understand the technology with which I am working, I can apply this knowledge to a completely performance oriented revision. See and logging the die vs die temps was interesting, and will be a challenge as I am sure there will always be variation and that CCD1 will not always be cooler than CCD2 and vice versa.

What I do know is the coldplate revision I have going right now will direct all the liquid to the areas over the CCDs first before the flow is shaped for the remainder of flow through the conduction zone. It is sure going to push my current manufacturing capabilities... The coldplate will now have 4 or 5 setup operations with the final two being quick programs but it is still a lot of additional time per piece. What I really need to do is find a way to be able to quickly setup and tear down the small rotary table for the one machine. They rarely use it due to the large amount of space it consumes and the fact that we rarely are cutting anything small or the desirable geometry to cut with 3+2 or 5-axis. This way I could do my modified irregular fin and slot cutting without the need for manual setups.


The beginning of a new cycle when temps spike for about a second
View attachment 209496


Done and the return to idle temps in under 7 seconds
View attachment 209497


Starting stress test with PBO, wattage still going up
View attachment 209495


5 minutes in, system is pretty linear at this point
View attachment 209498



So, all in all, it's been an interesting night and I really glad to be developing a much more thorough understanding of my new system and how to most effectively generate the best data to test this block and those to come. I am proud of the fact that my design was able to hold that kind of wattage level in check without any real issues. I really wish I had an off the shelf block with which to compare right now. I have a couple of old school ones, but no way to mount them... and considering this is a new system, I'm really hesitant to do any winging it at this point. There will be no ziptie mounts today. I don't know if there's any data available on a similar stress test with other blocks, but I'd be curious to see just how I stack up so I know what I need to do. Again, let me stress this, if this design doesn't cut it, and revisions within the package will not be able to compete, then I will move on to the next design. It is what it is... The amount of information and data on design principles, manufacturing experience, product testing, mounting strategies, aesthetics, materials... I could go on but I digress... that I have ascertained from this experience will be invaluable moving forward. What I wouldn't give for some machine time so I can get the two test units out... uuuuuugggggghhhhhhhhh...

Thursday is payday for me, so I went ahead and bought what I need to be able to keep track of various physical properties of the loop. I grabbed the last aquaero 5 on Amazon ($2.50 more than elsewhere, but free shipping with delivery on Saturday!), same brand flow meter, and 5 inline temp probes. Also grabbed some more 1/2" ID barbs and some rotary 90s. I dislike the 5.25" drive cage in this case (but not mad at it since I paid less than half of retail new in box), but it will come in really handy to mount the controller and flow meter. Need to add a fill port as well, so will be adjusting the reservoir's position down a bit to facilitate that.

Anyway, thanks for reading everyone, and I hope there's a lot of input on all the data posted above. Whatever you can give me pertaining to somewhat comparable performance of my block at this power level to others would be greatly appreciated. I just need to know if I'm on the right track...



EDIT: Missed 200W by 0.796W.

View attachment 209501

View attachment 209500

 

[Blaylock]

Martin's liquid lab and Skinee Labs were the last two "full review" sites pertaining to liquid cooling that I can recall. Sadly Martin shut down in 2012 and Skinnee two years later, if memory serves me correctly.

As far as extreme water cooling, I think Trents mentioned it earlier, you will likely reach a maximum amount of cooling before hitting a wall. The top manufacturers are there now , or damn near there, anyhow. This is because of the limiting factor of ambient cooling. To get any better than the ambient wall you need to go sub ambient. That discussion is far longer than the water cooling discussion and truly deserves it's own thread.

 

[thenrz]

Martin's liquid lab and Skinee Labs were the last two "full review" sites pertaining to liquid cooling that I can recall. Sadly Martin shut down in 2012 and Skinnee two years later, if memory serves me correctly.

As far as extreme water cooling, I think Trents mentioned it earlier, you will likely reach a maximum amount of cooling before hitting a wall. The top manufacturers are there now , or damn near there, anyhow. This is because of the limiting factor of ambient cooling. To get any better than the ambient wall you need to go sub ambient. That discussion is far longer than the water cooling discussion and truly deserves it's own thread.

I've done sub-ambient... Direct die, chillers, and ln2. The amount of work and maintenance it takes to keep a system running in danger of condensation is just far too much for me. There are also a ton of permutations with radiators, pumps, fans, tubing type... I feel that in my case the monsta 240mm radiator has been a real boon to the setup. It's thicker than double the average 240mm unit, and some reviews put the performance on par with high tier 360mm products. I guess what I am getting at is I need you to test on your setup and see how conduction changes based on changing basically every extraneous variable.

Handling a 200W load and capping at 72.3C seems like a rather interesting milestone. I wish I could have had enough machine time to have put a block in your hands by now. I feel like once I get the flow pattern sorted out and finalize the fins, this thing will be able to handle a 250W load without much issue. The fast return to idle temps is also motivating as that implies that conduction is happening rather efficiently. I am going to get my revisions programmed and try to figure out a way to avoid all the setups on the coldplate slotting ops. The faster you have one, the faster I understand if I'm throwing good money at bad money or if this design is worth finalizing.

 

[trents]

I found this article helpful in understanding the power managing technologies built into Ryzen 2: https://www.gamersnexus.net/guides/3491-explaining-precision-boost-overdrive-benchmarks-auto-oc

You are not alone. I don't think anyone has a complete handle on how these things work in Ryzen 2 apart from AMD engineers. It is indeed a technological masterpiece.


Edit: By the way, you shouldn't have to worry about mosfet temps until you reach at least 100c.


Edit 2: threnz, can you create a Signature for us so we don't have to try and remember what are your system components, I mean the ones that are staying constant during our testing.

 

[EarthDog]

Maybe it's just me... but 200W load @ 72C with a 2x120mm (double depth) radiator feels like an unbelievable result. Maybe it's just my lack of 3900x knowledge, but my 3x120mm rad and kyros next block cant come close to that with a 200w load on my 7960x. Different chips and such, I get it...but...

I'd like to see another block in the same loop to see if its actually the block.

Temps dropping back to idle in 7 seconds is also unbelieveable. If you test for 30+ mins, what is the water temp starting and ending (you'll find out, lol)? Do you know how much energy it takes to cool 1L of water 10C) in 7s? I couldnt do that in a deep freezer with the rad inside.

Regarding your loop temps, you'll just want 2 sensors. In a properly flowing and radded loop temperatures should only be around 1-3C different (say from rad intake to rad exhaust, or block in to block out.

Will you confirm temperatures using coretemp, plz?

 

[thenrz]

I found this article helpful in understanding the power managing technologies built into Ryzen 2: https://www.gamersnexus.net/guides/3491-explaining-precision-boost-overdrive-benchmarks-auto-oc

You are not alone. I don't think anyone has a complete handle on how these things work in Ryzen 2 apart from AMD engineers. It is indeed a technological masterpiece.


Edit: By the way, you shouldn't have to worry about mosfet temps until you reach at least 100c.


Edit 2: threnz, can you create a Signature for us so we don't have to try and remember what are your system components, I mean the ones that are staying constant during our testing.

One, the signature is done, but isn't showing up...

Two, 10-4 on the mosfet temp issue. I used to work in electronics and robotics manufacturing, and our mosfet h-bridge motor controllers would get hot when being beat on, but I'm a lot less inclined to accept that kind of heat 2 inches away from my shiny new CPU. Will have to be something I learn to accept... But I still want to make a waterblock for it!

Thanks for the link!

- - - Auto-Merged Double Post - - -

Maybe it's just me... but 200W load @ 72C with a 2x120mm (double depth) radiator feels like an unbelievable result. Maybe it's just my lack of 3900x knowledge, but my 3x120mm rad and kyros next block cant come close to that with a 200w load on my 7960x. Different chips and such, I get it...but...

I'd like to see another block in the same loop to see if its actually the block or what...

Will you confirm temperatures using coretemp, plz?

I will do that this evening. What is more surprising to me is that the loop was already well above ground state. I am 99% sure the temps are accurate as hwinfo and occt were always the same during the hour test... but, obviously occt wasn't running during the intel burn test up to 200W. What I can tell you is that my screenshots tell the whole story there. I couldn't believe that I was able to get things to 200W and have something not get upset with me. The VRM was quite hot, but capped below 76C.

EDIT: It looks like my orientation is absolutely ideal for the 2x chiplet design. My inlet liquid is swept right over that side of the chip upon entrance into the conduction zone.

 

[EarthDog]

Your sig is there...

Note my edits/addition above..

Edit: If your loop has 1L of water and flows at 1.5 GPM that means the liter of water goes through the loop once every 10s. If water temps go up 10C, how are they dropping 7C in 7a when it hasnt completed a cycle withing the loop?

Edit2: temp sensor on rad I take air is good as well....giving you a true ambient air temp just before the rad

 

[thenrz]

Your sig is there...

Note my edits/addition above..

Edit: If your loop has 1L of water and flows at 1.5 GPM that means the liter of water goes through the loop once every 10s. If water temps go up 10C, how are they dropping 7C in 7a when it hasnt completed a cycle withing the loop?

Edit2: temp sensor on rad I take air is good as well....giving you a true ambient air temp just before the rad

It's not about the loop, it's about flow through the block. Once the 200W load drops dramatically, it's not the loop temperatures that matter at that point, it's removing the excess heat from the coldplate. True idle last night after about 30 minutes of simple web browsing and stuff was floating around 29C. I was defining idle as simply not loaded after testing. I was returning to the 30-34C stable range in about 7 seconds after the high load was removed. I think that's a testament more to the design of the CPU and 7nm than to the liquid loop. I'm pretty sure any high flowing custom loop would do the same thing at that point.

 

[EarthDog]

It's not about the loop, it's about flow through the block. Once the 200W load drops dramatically, it's not the loop temperatures that matter at that point, it's removing the excess heat from the coldplate. True idle last night after about 30 minutes of simple web browsing and stuff was floating around 29C. I was defining idle as simply not loaded after testing. I was returning to the 30-34C stable range in about 7 seconds after the high load was removed. I think that's a testament more to the design of the CPU and 7nm than to the liquid loop. I'm pretty sure any high flowing custom loop would do the same thing at that point.

If you start at 30C idle with 25C water temps...then test for AT LEAST 30 mins... water goes up to 30C (numbers made up, nkte). The water temp doesnt drop instantly... it will take a few/several minutes to do so. The coldplate is only as cool as your water.

That said, the quick drops are normal be it air or water. My son's pc does that with stock amd cooler, for example. So do my systems, all intel. But when we are saying idle, we mean power on, leave idle for 20 mins to normalize, take temp data point. A return to idle is a return to that exact temperature you started with, not 1-4C above it.

Another issue is that cpu temp sensors tend to be less accurate on idle....

 

[thenrz]

If you start at 30C idle with 25C water temps...then test for AT LEAST 30 mins... water goes up to 30C (numbers made up, nkte). The water temp doesnt drop instantly... it will take a few/several minutes to do so. The coldplate is only as cool as your water.

That said, the quick drops are normal be it air or water. My son's pc does that with stock amd cooler, for example. So do my systems, all intel. But when we are saying idle, we mean power on, leave idle for 20 mins to normalize, take temp data point. A return to idle is a return to that exact temperature you started with, not 1-4C above it.

Another issue is that cpu temp sensors tend to be less accurate on idle....

My usage of the term idle was flawed, and this is where the confusion is borne. Idle meaning system idle and temps returning back down to the unloaded state. So, let me correct this: CPU temp returned back to the steady state of the liquid after about 7 seconds (will be confirmed when measurement devices arrive and are installed). By noting this, I am simply saying that flow remains high and the block is not restricting the removal of that additional heat from the CPU/IHS. I will be more aware of my choices within this lexicon in the future...

 

[Blaylock]

Great discussion here. I love it.

My only contributing comment, CPU temperature drop to an idle "like" condition is primarily due to the lack of load (i.e. heat source) while a water temperature drop back to idle/pre-saturization is more in part to the radiator, fans, and ambient temperature. None of the above mentioned are primarily caused by the CPU water block.

What needs to be tested and measured for the water block are the rise in CPU and water temps under full load as well as the duration it takes to get to full saturation. This data then needs to be compared to other water blocks under exacting (or as close to as possible) conditions to be able to claim one better than the other. Using the same build with only the water block changed we can chart a temperature curve and compare the fastest time to saturation indicating which water block is extracting the heat with the most efficiency.

 

[thenrz]

Great discussion here. I love it.

My only contributing comment, CPU temperature drop to an idle "like" condition is primarily due to the lack of load (i.e. heat source) while a water temperature drop back to idle/pre-saturization is more in part to the radiator, fans, and ambient temperature. None of the above mentioned are primarily caused by the CPU water block.

What needs to be tested and measured for the water block are the rise in CPU and water temps under full load as well as the duration it takes to get to full saturation. This data then needs to be compared to other water blocks under exacting (or as close to as possible) conditions to be able to claim one better than the other. Using the same build with only the water block changed we can chart a temperature curve and compare the fastest time to saturation indicating which water block is extracting the heat with the most efficiency.

You are correct, but that heat will stay unless ferried off somewhere. Obviously there's an order of magnitude less heat to carry away since it is whatever the heat energy is of the small amount of silicon, metal substrate, IHS, PCB, and various surface mount components at that moment without any additional heat being added at a high level like during a stress test. I think my point was being misconstrued. I was insinuating that the rebound back to temperatures within a few degrees of a consistent idle temperature based on ambient was pretty fast and this means that there isn't any residual heat build up within the block that would add to the thermal mass of the area. This is really a testament to the efficiency of liquid cooling in general and not specific to any part in my loop. A quick rebound also implies the efficiency of the CPU itself to essentially burn off minimal heat at low loads.


And yes, this is great discussion taking place right now. Looking forward to extracting a ton of pertinent information from all of this to apply moving forward. And Blaylock, as soon as I have machine time and can get the last thing cut, I'll get you one of these. It's unfortunate that I haven't had a single moment to do what I want to do lately, but it is what it is with everything going on in the world.

 

[trents]

Yesterday you talked about making design mods that will do a better job of pushing the water down into the cut grooves. Be careful you don't create too much resistance. That will make the pump work harder and run hotter and might offset the benefit of the redesign.

 

[thenrz]

Yesterday you talked about making design mods that will do a better job of pushing the water down into the cut grooves. Be careful you don't create too much resistance. That will make the pump work harder and run hotter and might offset the benefit of the redesign.

It is definitely a point at the intersection of two functions. There will be some sort of area under the curve where there is too much resistance, as well as an area where there isn't enough. This will all intersect with the interaction between pump pressure, flow rate, and ultimately the volume of water within the convection zone at any given moment. Right now, I have a clearance of .003" above the grooves. This exact value may be a few thousandths more or less as error builds while the parts are assembled. This "slop" was designed into the beta part to facilitate a good seal upon compression. After another round of careful examination, the inlet flow metrics are basically perfect for the two core-filled chiplets on the 3900X. The flow enters from a round orifice (haha, so immature...) and is shaped into a wider flat and narrow channel and then directed into the grooves. The flow shaping decreases resistance through the block as the liquid is not abruptly hitting a flat and perpendicular wall and having to spread and fill the flat areas around simultaneously. The flow itself strikes the areas directly above the outer edge of the core chiplet and flows across them and through the grooves from there.

While all of this is great, copper is such a fantastic conductor of heat that I'm probably only gaining a couple of percent by this strategy... but I guess that's the point isn't it? To maximize this specific metric? All I know is this is enjoyable. You guys bring some amazing points to the table and truly make me think and examine things from perspectives that I may not have explored without. Keep calling me on my BS, too. Someone when all you have is data from your own head, it's easy to commit the fallacy of false application...

So, at this point, I'm going to leave it. An Intel-centric block will most definitely benefit from this modification, though. Shaping the inlet to the middle, collecting flow and directing around the circumference and to the outlet seems to be the best way to maximize things.

 

[thenrz]

OK, so in a little bit here I am going to run a 200W test for 30 minutes and then 1 hour with a ~15 minute break in between. I will be using Intel Burn Test since it gets me to 200W and just about holds there. Now, as per EarthDog, I will have coretemp (identical readings so far) running as well as hwinfo64. There will be a timer visible and I will be taking screenshots at specific time intervals.

Anything else I need to include here? I am really liking the idea of being able to dial in a consistent wattage. Totally removes the need to go through with my power resistor testbed. I had no idea they had such data driven tools like hwinfo that were available to consumers... and free no less. An easy choice to buy the dev a beer on that one. The aquaero and my usb header splitter come on Sunday, and the rest of the stuff show up Monday. Running this test with that much hard data would be incredible. It would answer most if not all of the questions on here as it would enumerate loop temps, flow rate, etc.

If I'm missing something let me know!


EDIT/Update: Stopped the test after 10 minutes. The VRM was over 80C and the PCB around that area was getting far too hot to the touch for my liking. I need to come up with a solution ASAP for this. I know that MOSFETs can run hot, etc etc etc, but it's the system's longevity that I'm really worried about. The added heat from the CPU running at 72C and the heat burn from the FETs... The heat from those are undoubtedly crossing over to the CPU via the physical substrate. But, I guess, even if it is, my loop/block seems to be handling it in stride. It is pretty wild the heat I can feel coming off the radiator; it is a pretty substantial difference.

By the way, coretemp reads the CPU frequencies incorrectly. Other than that, the temperature matched up sans rounding 100%. I also didn't reset the averages/min/max for hwinfo before the test. I will do that the next time.


View attachment 209522
View attachment 209521
View attachment 209520
View attachment 209523

 

[EarthDog]

with a ~15 minute break in between.

You ahould wait until you reach your established idle temperature. Or when you get your temp sensors, when the water goes back to the starting temp...

EDIT/Update: Stopped the test after 10 minutes. The VRM was over 80C and the PCB around that area was getting far too hot to the touch for my liking. I need to come up with a solution ASAP for this. I know that MOSFETs can run hot, etc etc etc, but it's the system's longevity that I'm really worried about. The added heat from the CPU running at 72C and the heat burn from the FETs... The heat from those are undoubtedly crossing over to the CPU via the physical substrate. But, I guess, even if it is, my loop/block seems to be handling it in stride.

I think you need to move past your 'liking' and adhere to what the specifications tell you is safe. 80C for the MOSFETs on that board are fine.

As far as heat soak... welcome to modern big boy(read: many cores and threads) computing! All blocks deal with that...the level varies by board chip and load.

I still think you need to set a static voltage and clockspeed just due to how AMD chips work.