Noctua NH-U12A 90° kit ?

@Mr. Fox i have a question
i am building a gaming desktop
i bought the noctua NH-U12A but it dont have a mounting to put it at 90°. i have to put it like that to aling the air flow from bootom to top (the CM case i bought work like that) my main question is, do you know if this kit : https://www.amazon.com/gp/product/B01MTEFT52/ref=ox_sc_act_title_1?smid=A1Z5H6ZGWCMTNX&psc=1
work for that cooler ? i cant find info about that, i guess it will work but i dont know, and i want everything ready when the cpu and motherboard arrive.

my setup is this:
case: cooler master SL600M
fans: 6 corsair ML120 pro (3 on the bootom 3 on the top)
cpu: ryzen 3900x when available
cpu cooler: NOCTUA NH-U12A ( i will use TG conductonaut)
motherboard: gygabyte X570 aorus master when available
ram: patrio viper steel series 4133mhz cl19 1,35v ( can go up to 4400mhz)
ssd: patriot viper VPN100 nvme ( can go up to 3400mb/s write) it comes with heatsink
RGB: phanteks RGB combo
monitor: AW25hf 240hz
power supply: NZXT E650
keyboard: razer blackwidow ultimate ( orange switches)
mouse: corsair darkcore se RGB
headphones: Razer kraken pro 7.1 v2

 

Hello Brother @judal57 - happy Monday. I am sorry, I am not sure what the kit in the link works for. I haven't owned but one inexpensive desktop CPU air cooler (Cooler Master Hyper 212 EVO) as a temporary workaround while waiting for an RMA for a liquid cooling setup. I use only water cooling, so I do not have any knowledge about the parts you are asking about.

 

Yes, the two sets of mounting bars are included to flip the orientation of the cooler. Use the short mounting bars to achieve your desired airflow.



Skip to 0:56 to see how the mounting bars change the orientation of the cooler.

 

thanks man, i send a mail to noctua and making you an abstract, they are gonna send me the small bracket to mount it at 90°

"
Greetings from Noctua,

Thanks for sending over the requested data.

The parts will go out either today or tomorrow then. Shipping usually takes around 3-8 business days within most parts of Europe and 10-16 business days overseas.
You'll receive a confirmation e-mail containing a tracking number (when available) once it's shipped.

Kind regards,
Your Noctua Support-Team"

i liked a lot the support from noctua, but i am not happy with 10 to 16 business days ****

 

@Mr. Fox i finished the bulding. i have a really good binned ryzen 9 3900X, i overclocked the infinity fabric at 1866mhz and set the ram at 3733mhz CL16, 1.35V
https://www.3dmark.com/3dm/38055649

 

You smoked by BGA junk. I'll be following in your footstep and buy Ryzen 3 3900X or even Threadripper once the prices are sane, now they are insane and low on parts in amazon.

 

Congrats on the new desktop. Nice!

The case almost has a steampunk look to it in that photo.

 

An update for you. He is using Asus Crosshair 7 and also managed to OC Infinity fabric to same clocks as RAM to boost the performance further. He has already surpassed 60GB/s in RAM bench test from AIDA64 with only 2 sticks.
GB4 scores, IO scores are almost exceeding intel Skylake-X at fraction of the price. You won't believe the IO scores after he OC'ed the IF.

I hope he can play GTA VC at 16K with insane settings. Haha...

 

Nice! Yes, it seems memory and IF overclocking is key to getting the best performance results with Ryzen. @ajc9988 talked about that a few times in the desktop overclocking thread. Just imagine how wicked it would be if you could run the CPU cool and stable at 5.0-5.2GHz and RAM at 4000-4200 MHz.

 

i have run this kit at 4400mhz CL18 BUT the IF works at 2:1 and it means a latency increase
edit: also the copy, write and read dont scale too much from 4133. 60.000 is a barrier and start growing slowly

 

@Mr. Fox the key is the infinity fabric OC, but i am learning at this moment. Too much options to learn in too little time. The oc of the IF impact directly on the uncore performance

 

@Mr. Fox i can get 4.5ghz strong using air. this chip is really good, but at that frequency the heat start growing and i can see 85°C and i dont like that temperature. i am sure that with an OP water cooling you can achieve that 5ghz stable. 16 phases works like a champ

edit: i can do 4.3ghz at 1.25V and is a reallly sweet spot to me. MAx temp is 73°C on stress

 

I see the weather is pretty cool in Colombia right now. Maybe wait until later at night for the temps to drop down to around 8°C, then open a window. Put the case right next to that window with the side for CPU/GPU access facing the window. Pull the side panel off and add a nice room fan running full blast to grab the cold outside air and shove it violently into the chassis to see how the temps are at higher overclock. Should improve from 85°C by a fair amount.

 

i have a portable air cooler lol, it is on the depot. also the weather higly depend on the city, 1 hour of travel and you can feel hell at a confort 45C, another hour or two of travel and you can feel the cold at 4C. were i live the mean of july have been 22° / 14° (max-min). Also i dont like dust, so i use the vacuum every day and dont open windows usually. but i will do something to make the benchmarks work

 

Quick note: stay at or under 3800/1900 men/IF. Focus instead on tightening timings at those clocks. You avoid the 1:2 switch, AMD said there is some issues in BIOSes to be worked out on that separator, but no real info other than that being the limit. So focus on 3733 and 3800 tight. That should lower the latency for you.

The reason is additive latency. They still have something that clocks the data down to 1:2 above that (some say the UMC) so that even manually setting the IF and clicking 1:1 isn't allowing for the Max IF frequency while doing what you want on the memory.

For LN2, overclockers are using 2933/1466 due to the cold.

It is nice getting able to use the higher mem frequencies, but it isn't worth the latency hit yet. Since the IMC is on the I/O die, your mem latency to and from the I/O die is added to the latency of the IF. And, if I've read things correctly, every time you go off CCX, you use IF to the I/O die. With windows scheduler now being CCX aware and spawning threads to stay on the same CCX four core complex, it would standardize all latency going off CCX to another CCX, helping to fix the stale data issue. So, all memory trips will go over the IF, and if something then kneecaps the IF latency, it can cause worse than just staying in those guidelines and trying to tighten timings like crazy.

 

i tried, but there is a problem with the RAM overclock, i think is realted to virgin BIOS. there is no difference (not much) from 3733mhz CL16 to 3733mhz CL14. I also tried 3800 with the IF to 1900 but is the same performance, it dont have sense to me :S. also i can note that the PBO frequency is attached to ram speed. low ram give me more GHz on pbo, and fast ram give me less Ghz (4.1ghz on all cores when benching) they have to fix that. the good thing is that Asus tend to pay attention to Bios, so i hope there is a Bios fix to that. a lot of peope is complaining about that ( all brands Mb) i think is a problem of the new AGESA ( or that ****) i cant remember lol is the 1.003AB or something like that

you can judge my photo and tell me if my score is good or bad, there is not too much out there to compar, thanks.

 

@Mr. Fox talking about another thing i want to tell you about the Noctua NF-A12x25 PWM, those fans are INSANE, the performance and noise levels are just incredible. I wish I had known before I buying those 6 corsair LM120pro :'( . i dont care paying double with that noise difference. I am really starting to like that color LOL

 

Here:

 

he is using a 3700X and have the half of the write speeds i have, due to 1 chiplet. but i think that is why they can go under 67ns, i think is not the same for 3900x

 

The principal in the video IS the same. 3800 is the goal, or 3733, then tighten it up. Keep 1:1 on the IF and UMC scaling. You'll have better writes. Latency, that takes some work, but you are also only a couple ns off of his lowest latency, and his wasn't fully stable, I'm supposing you tested and are fully stable. That means you are doing fine, just tinker as you see fit.

Then, on memory and heat, by increasing memory speed, you cause both the I/O die and infinity fabric to run hotter. That then takes thermal headroom from the core die.

To fix that, best you can do is play with lowering voltages. AMD now has a quirk with this gen where it won't crash, but you'll get lower performance if you go to low on voltage, at least on core voltage. So you want to play with SOC voltage without negatively effecting memory or I/O tests while using something like CB15 or similar looking for changes in score outside of run to run variance for core.

You can also run the core tests while lowering the SOC voltage just to check of it lowers there as well, but AIDA and Sisoft Sandra memory would be my go to from what I've read (note-i do not have my hands on one of these chips currently, but follow news closely).

 

i can do this, better than him. but is not stable as you say. i think the best spot is 3733mhz, also is only a 2ns difference :S and i have to increase the voltage from 1.35V (relaxed) to 1.48V (i dont like that)

 

This is why I'm saying, from what you have shown, you seem to be doing just fine. Work with what your hardware can do.

For mem stability testing, I start with TM5 v0.12, then I do the TM5 with 1usmus modded file from overclock.net (look up the Ryzen DRAM Calculator, go to the thread in OCN, then look for the cfg file tagged in 1usmus's signature at the bottom).

After that, I hit it with HCI memtest overnight. With all that, you should be stable. Karhu's memtest is pretty decent as well, but is paid. The DRAM Calculator now has HCI memtest incorporated on a separate tab, or you can run it on your own, etc.

 

i fixed a little bit the timmings, 3733mhz 1.35v is the way to go. I am really happy with that (100% stable)

 

Now that you have your memory performance down, you should check the overall performance in other tests, as memory effects some of those (including games).

 

firestrike: https://www.3dmark.com/3dm/38229719?
the cpu have PBO active and is at 1.243V on bios

 

@ajc9988 @Mr. Fox I can do all cores 4.2ghz at 1.2v. i think that is too little
https://valid.x86.fr/yqhisw



I also tried 4.4ghz all cores at 1.35V i dont know if i can go lower, i just shoot and it work https://valid.x86.fr/yqhisw

 

@Mr. Fox @Vasudev @VICKYGAMEBOY
FIRESTRIKE
https://www.3dmark.com/3dm/38245451?
FIRESTRIKE EXTREME
https://www.3dmark.com/3dm/38245902?
TIME SPY
https://www.3dmark.com/3dm/38246038?
TIME SPY EXTREME
https://www.3dmark.com/3dm/38246189?

 

4.3ghz stable at 1.2v https://valid.x86.fr/wjrg1e

@ajc9988 it helped to reduce the latency

 

You are decently close to my scores there. My 1950X is in the 3600-3700 area at 4.2. Best score was like 3740s around 4.275 for CB15.

 

@Mr. Fox I am sure that using a liquid cooling solution should allow me to achieve 4.5ghz or more. I am only using an air cooler. I don't know which voltage is safe for this CPU, do you think 1.2v is safe for 24/7 ? If it is, what do you think is the max ? I don't want to brake this beauty :3

 

Voltage depends on heat. For example, I try to stay under 1.4V for daily use. So I run [email protected] on my 1950X and my temps are around the low 60s under load, under 70 for the heaviest loads I throw at it.

So voltage depends on your cooling system. I have 3x480s in my loop, so I am pretty capped on air. That and my VRM get to the 70s to low 80s under load at these frequencies and voltages.

So, I'd say try to limit temps to around 70C and see what you can do while keeping voltage low. Also, some have noted that if you drop the voltage too much using precision boost, your score will go down, but it may not crash. That means you need to bench different apps while lowering the voltage at any given frequency.

What helps is to write down and plot out your voltage at each frequency interval, which on Ryzen, at minimum, I'd say is every 50MHz. Then you can find where power jumps up significantly for the next clock ratio.

You may be able to go higher in temp than that, but that generally is a safe guideline.

I forgot the temps and voltages AMD gave, but it might have been like 1.325V for daily use for the new chips.

But you seem to be doing well. One person took 4.45GHz to get to 3489 in CB15, as a reference.

Edit: hell, even the 7920X takes around 4.9-5.1GHz to be where your score is!

 

i will try to break your friend record today lol as @Mr. Fox say, today i am going to open the windows to the benefit of science lol

 

@ajc9988 i have found many things:
1) i can do 1.125v 4.2ghz all cores fully stable https://valid.x86.fr/zufebw

2) i can do 4.4ghz at 1.32V, need to validate it
3) the curve frequency vs voltage becomes really bad because 1.125V to 4.2 - 1.2V to 4.3 - 1.32V to 4.4 - i can imagine 4.5 ~ 1.42V maybe ??
4) i love my cooler, but i plan to change it when new gpus arrive like nvidia 3000, and convert all to custom loop.

 

Sounds like you are having some fun. Also, you have potentially a chip toward the higher end of the scale there, similar to the one that Bearded Hardware (Steponzi) has for his 3700X. Granted, you didn't use the exact same qualification tests as Silicon Lottery, but their top bin is [email protected]. That's running P95, realbench looping, and a couple others, often with an AIO.

Also, you can see where the efficiency curve takes a dump and voltage goes up significantly. But the fact you can do 4.4 all core at 1.32V says you are doing just fine and you have a good chip in hand (always nice to find out you got a good chip).

I'd say next year, even though Nvidia will likely have the top card by going to 7nm, keep an open mind. You have Intel coming out with their new GPUs and there is always the chance second gen NAVI winds up as a good choice. But, as mentioned, it will likely be the Nvidia card as the one to get.

 

You may also find you can use lower CPU voltage on a custom water loop. Lower temperatures often enable the use of lower voltage and as temps go up the need for more voltage often does as well. It is a vicious circle.

Also remember that you can always buy a few more fittings and some extra tubing later on to add a GPU into your custom loop later. In the meanwhile, you could be enjoying better temps for the CPU on water and that might even lower the temperatures in the case and result in the GPU running a little bit cooler (on air) without the CPU air cooler radiating heat.

The nice thing about a custom loop is you can do it in stages as time and cash flow allows, or in phases when you are not set on your final specs. The only thing that is generally hardware specific is a GPU water block.

I saw an article yesterday where ASROCK is using the Intel LGA-1151 socket bolt spacing on at least one AMD motherboard to allow universal fit of coolers and water blocks regardless of whether the customer is using an Intel (non-HEDT) or AMD CPU. That is really smart. It would be good if all of them did that on all mobos. Things like that should be standardized as much as possible. It's kind of stupid to have to buy another cooler or water block to fit if you make a change like CPU brand. Really no different than mobo form factors like ATX and mATX.

 

the thing is that i always want the best, so i will not be satisfied with only custom loop on cpu, i want all ( cpu, gpu, vrm) vrm because it have the fittings, but is not a problem. Also i dont know too much about that, if you can suggest me the best pump, tube, and everything let me know. also this case Coller master SL600m is capable of 2 (360) radiators.
i am thinking on something like this:

 

I really like the Photon D5 pumps with the large reservoir. See info in my signature. I really like the quality of the XSPC water cooling parts. Almost everything except my tubing is XSPC cooling parts.

Rigid tubing looks super nice. It's not really better unless you value aesthetics more than functionality and ease of use. Rigid is more prone to leaking if you move the case around frequently for cleaning general housekeeping, and it is more of a pain in the back side to deal with if you need to take things apart. For those reasons (especially the latter) I have made an intentional decision to stick with flexible tubing and I use quick disconnect fittings between end connectors. That makes everything super easy to take apart with no draining of the system necessary. Literally "plug and play" on the loop. At best, you might lose a teaspoon or two of coolant with the quick disconnect fittings. All you need is a paper towel to catch any stray drips with the quick disconnect fittings.

The nicer aesthetics of the rigid tubing is not worth the extra hassle to me. If you do a nice job of routing the flexible tubing it doesn't look bad. You can always use compression fitting inline elbows and 90° fittings if you need to make any hard turns inside of the case.

 

@Mr. Fox what abput double pump ?
https://www.amazon.com/XSPC-Twin-Re...rds=xspc+pump&qid=1565037750&s=gateway&sr=8-5
2 (360) of this rads
https://www.amazon.com/XSPC-Perform...ds=xspc+pump&qid=1565037750&s=gateway&sr=8-19
do you know the name of this particular tube ?

 

That is an excellent radiator. I have it.

I do not know what that tube is. It looks really nice. It might just be black plastic, but it could be metal tubing. I cannot tell from the photo what kind of fittings are on the ends.

Given you have limited space inside of the case, the dual D5 pump would probably be really good.

I have an inline D5 on the water chiller. It runs from its own AC adapter and I can use it to circulate water through the system when the computer and the other pumps are turned off. I can totally disconnect and bypass the chiller setup and inline D5 from the cooling loop using the quick disconnect fittings in a matter of seconds with no tools or fussing with anything.

I have an XSPC Photon D5 and 270mm reservoir on the push side of my loop and a second XSPC Photon D5 and 270mm reservoir on the pull side with dual 360mm XSPC rads. The more water capacity you have, the longer it will take for the water to get warm under extended periods of heavy load.

 

So, I use primarily XSPC black chrome throughout my build and soft tubing as well. I used the photon res (170mm) with a swiftech DDC pump (I have two pumps, but one got gunked up with the dye of the first fluid I used, but the second fluid, a mayhem clear fluid, is doing fine, although I have not yet replaced the faulty pump yet). But if going with the res like the one you showed, you should be fine. I went with DDC just because of using 3 of these: http://hardwarelabs.com/nemesis/gtx/480gtx/

Just double check that the two spaces for the rads have clearance after you add 25mm for the 120mm fans on top of the thickness of the rad.

 

Here is a closer look at that setup...

 

its a soft tubing, but looks insane good

can i use the double pump and add a reservior to le loop?
i really like the idea of double pump lol looks strong
edit: look how sexy the EK water block looks on my motherboard :O_____

 

@Mr. Fox your build looks insane :O___ i love those mini fans for VRM lol but what about the noise ??? i insist, do you have to try those noctua fans, the difference is night and day, they also sell mini fans.

 

Sure, you can (and should) add a reservoir.

Thanks. When the time comes to replace my fans I will definitely consider them. I do not need to spend money on fans at the moment. My radiators are push/pull, so there are 12 120mm fans. The X299 Dark has 2 VRM fans (stock) and a PCH/M.2 cooling fan (stock). The 3 stock X299 Dark fans are essentially silent unless I max them out. I run them at 50% speed just as I do the 120mm fans.

The 4 small fans over the RAM are added. I don't really need them and if they wear out or go bad I probably will not replace them. My RAM modules have never been warm to touch, even running the sticks at 1.650V 4200. These 4 little fans are stupid loud running full blast, but I only give them enough juice to turn them on. So they run pretty slow and make only a faint rushing air sound.

There is not much noise unless I run them full blast and there is no need for that. When I use the chiller they are turned off because they warm the water and fight against the chiller. When I am benching I disconnect the radiators completely because they also warm the water and work against the water chiller. The chiller is like a mini refrigerator or air conditioner and it make more noise than the fans running at normal speed, but it is not super loud. Noise doesn't matter much when benching anyhow. All you care about is the scores when benching. When gaming and during normal use, the chiller is turned off and the fans are run at medium speed and are pretty quiet. There are rarely ever any conditions I need to run the fans more than 50% speed.

 

@Mr. Fox @ajc9988 how much time i have to run prime 95 to consider my cpu stable ?

 

Well, that's kind of a loaded question that will return a variety of answers. My answer is zero. I don't use it for stability testing myself. If my CPU passes Cinebench and wPrime 1024M tests and never crashes when doing ordinary things I call it good to go. The only thing I have really used it for is to melt the Indigo Xtreme metal pad TIM. To me it is kind of like running Furmark GPU core burner. The difference being the GPU might die (literally) from Furmark GPU core burner and the CPU will just be subjected senseless amounts of abuse and unnecessary stress from Prime95.

 

nice to know that comming from you
i found that my cpu run at ~ 32W while web browsing

 

I mentioned P95 because that is how Silicon Lottery tests. Personally, I do not use it either. I stick with blender benchmark or a blender render for stability testing and sisoft Sandra for peak temp due to the financial and scientific tests. CB R20 runs pretty hard as well and under the drop down, you can now set a time value and run it in a loop. CB R20 does stress a bit more than CB15 did, but unsure if CB15 extreme mod runs harder than it or not.

wprime 1024 looping is pretty good as well. I just finding rendering to be about the heaviest workload I do.

I also check with V-ray and Corona. Basically, if it can do heavy render, it can do just about every other task you can throw at it. P95 is designed to hit it hard on purpose, harder than any other load, but it can leave a lot of performance on the table if you personally never use something that hard.

Realbench can be put into stress test mode, but I find that too easy to pass.

But all that really matters is that it is stable and temps are good for your workloads (including any obscure one you may run occasionally).