its already 128 gigs. Also, filling entire slot will harm the clock speed, so its just already as big with full clock.

@@azravalencia4577 they could just use dummy RAM sticks too

Fill up your ram slot.

He obviously builds for a client, not for his own. Just doing what his client asked so I wouldn't blame him.

I don't think this is a build were looks or epicness was a priority

lmao😂

try 25999

Real

3.5year that is enough, because next 2 generations hardware will make those pc obsoleted.

The watercooling-components will still have its price.

Bro that PC is more expensive that my house 💀

​@@-_-_EpicaL_-_- The price is unnecessary.

lol me too

However most people don’t need such machine

@@-_-_EpicaL_-_- wtf bro

Since the additional 4090s don't do anything for gaming (nor do 64 cores), that it basically just a standard 4090 build for gaming.

Nvidia doesn't care about consumers, they underspec other cards and increase prices of the flagship card every gen because gamers are a small spec of dust in the GPU market compared to the AI industry

The way those RAM sticks look already tells this build can still get more expensive

@@rensomusni8091 Yup. The most money in a high end server goes to RAM, a couple TB can easily cost medium five to low six figures, more than 8 Xeon Gold or AMD Epyc combined.

yes, it's humble

Can you explain in layman's terms this 1(d): you're using fully parallel terminal = so you're splitting 4L of flow to 0,8L/min with dramatic pressure And then how your solution: Use semi parallel terminal or linear with radiators in the middle of this flow. Solves this issue. And please kindly remember, if you can't explain this to an idiot, then you don't really know the issue. Thanks.

@@dreamfunction4491 First "And please kindly remember, if you can't explain this to an idiot, then you don't really know the issue." Well there are many things that idiots don't understand even when presented with all the facts and given simple presentations. And as thermal fluid dynamics are very complicated subject even for students, do not assume there's a simple explanation so any idiot can understand it.. So this argument is in fact out of the window. In fact: in fluid dynamic and thermal fluid dynamics there are that many numbers and ratios than not many teacher understands them fully. But I'll try :D You have 1 liter bottle / minute produced by a machine. If you're drinking yourself only then you have a litter for yourself every minute. But if you're making a party with four other guys you'll have only 1/5 of this flow every minute. So the pressure in all your bladders will be lower ;). or you have 1 pizza in 20 min, if you're eating alone you have 1 pizza every 20min. If you're eating with 4 friends you have 1/5 of pizza per 20 min. Jokes aside : a limiting factor of flow is the diameter of the single tube and throughout of MORA with 2 QDC in the middle of this. Now you're splitting 1 stream into 5 same diameter streams. There's no magic there, unfortunately. Your streams have to merge after terminal also into same diameter of tube. Normally the limiting factors of streams are jet plates of waterblocks, but here it will be the flow through rads, QDCs and tubes. If you'll poor 1 liter of water on the ground with an angle, your stream might divide or not. If not: it will be thick one stream, if it does: divides into smaller ones. The terminal is not multiplying water in the system. It divides and distributes it. The flow from the pump has it's volume (V) and speed (v) and from those (in very complicated way) you're getting pressure (P). If you're using parallel terminal you're splitting this Volume Vrc1 into 5 same diameter streams Vrc2 which is in fact Vrc1 x5. Volume per minute in everyone of them will drop to the maximum volume that a single tube pre and after terminal can carry in time. So when the volume per minute drops - it means the velocity of the flow dropped. And when the velocity of flow drops the pressure drops. Most of the water blocks has the jet plate - for jet impingement cooling. This plate narrows and elongates the stream into line so it will go through all the fins of cold plate. But the higher the flow and pressure of stream through this jet plate is, the more turbulent flow occurs. The more turbulent flow is the smaller stagnation zone will be between the fins of cold plate - the more water will be pushed deeper between the fins. If you'd use semi parallel terminal you're dividing stream into 2 or 3 streams -not into 5. So the flow is higher = volume per minute is higher and pressure is higher. Let's say pump+mora+QDCs -> CPU + 2x GPU -> radiator on top (cooling) -> 3x GPU-> pump. You'll then have divided the stream into maximum of 3 same diameter streams. So in every stream of this situation you'll have higher flow/min. The higher the flow the better the cooling. Did you get it? Some equations and phys.libretexts.org/Bookshelves/College_Physics/Book%3A_College_Physics_1e_(OpenStax)/12%3A_Fluid_Dynamics_and_Its_Biological_and_Medical_Applications/12.01%3A_Flow_Rate_and_Its_Relation_to_Velocity www.researchgate.net/publication/323314537_Impinging_jets_-_a_short_review_on_strategies_for_heat_transfer_enhancement

Was going to just say the same thing. He should be running at least a dual d5 in parallel. I would also recommend that he has another separate pump on his MORA.

awesome comment

4090 don’t have sli or nvlink. Cad and blender already can multitask without this.

I need to sell my kidney for that 💀

the day when either AGI is released or immersive VR sets - all this trash with fans, liquids and CPUs becomes obsolete and costs nothing

Art retains or increases in value over time.

Yeah the moment I saw 4x 4090 I started worrying if that cooling will even be enough. I hope all of those 4090's will be undervolted at least somewhat.

Flow meter and temp sensor are avsolutely essential and their restriction is negligible in this setup.

@@dazeen9591 The 4090 doesn't support more than one card without Nvlink or SLI, and that was finally dropped by nvidia after the 30 series cards. They are just expensive paperweights.

@@gargolgaming8101 having gpus act independently can be beneficial

@@traestephen7276 But you can only use one gpu at a time, can you not?

When I saw him building the radiatior I was just thinking wow an entire side panel of fans(cause I was skipping the video) and then i see him keep the radiator outside. Maybe that is a flex who knows. I would keep it near my ac vent. If it was outside but making the radiator the side panel is a cool idea.

Because case restrict airflow. Radiators and fans has best effiency out of the case just like that.

this aged wewll. I'm assuming you've seen the new compensator from ltt

And the entire PC costs as much as a Honda Accord

What’s bro on about

Who let this man cook?

What do you work as ?) I also want 4090 :)

i got a 7870 damn

@@Ruwrize I used my old gtx1080 for like 5 years, then upgraded to rtx3080ti for 500$ and I am more than happy about it.. don't see any reason to buy 4090 for crazy amount of money when 99% of games run 150-200fps on 3080ti for third of price..

@@benjaminshtark5977 cool and where did you buy an rtx 3080 ti for $500 ?

@@Ruwrize a lot of people sell them 2nd hand after crypto market collapsed. just replace thermal paste and you are good to go.

4 киловата😊 В НЁМ ДВА БЛОКА

@@user-bp1cc3cs2o Это запас вторые два киловатта когда отопление отключат :)

Интересно, для чего такая сборка? Какое-то безумие

@@apelsin1980 Рендеринг скорее всего, видео обработка

Тетрис потянет?

It's probably not a conductive liquid, and the components seem very high quality. Air cooling a setup like this is also impossible because there is not enough clearance in between the cards.

He's probably going to disassemble the build right after the video was completed. Just for videos , not real.

This guy is not a random gamer dude, but a professional machine builder for institutions or businesses. His blog introduces a lot of examples like this, delivered to Research Centers and Hospitals.

@@technocracy90 You are missing the point that Mamdouh is making. Read his comment again and try to understand what he is saying. He is right.

@@dreamfunction4491 I'm not talking to OP but you. My point is they don't disassemble it. Custom water cooling system is one of their sales point.

Well don't expect gaming benchmark though, but blender's cycle viewport performance must be sick or rendering time.

Gaming performance on this would not be that good tbh. Its not a gaming system

@@thelespauldude3283 it would actually be really good, what? Zen 3 and a 4090. If the cooling could handle it, it would game like a beast, but more than half of the hardware would be idle.

@@LtdJorge can you actually play with 4 4090 in sli ? and threadrippers if i remember fall a bit behind for gaming

@@wilihey1425 SLI is pretty much dead nowadays so no

You don’t have to see them to know they’re shit.

Instant throttle while opening file explorer

Doesn't matter wich resoltion because you will always be in a giant CPU bottleneck even on 8K. It will always stutter as hell because that CPU is absolutely not fit for Gaming especially not for CPU heavy games like Minecraft.

@@melon3210 wait you don’t game on 2 epycs for 192 cores?

Yeah… deep learning…

If he had 15 4090s. And you're exactly right in Nvidia's terms. Deep learning is more important than gamers to them. That's why they increase prices and lower specs compared to their true capabilities

To learn how deeply you hate its price.

Way better for shallow learning

Air in the radiators shouldn't be an issue I think. I'm only aware of that being a problem in AIOs if the air accumulates in the pump itself.

Air can accumulate in one of the waterblocks, and the liquid bypasses it through the other.

my critique is the 512 gb nvme...

I can't even imagine a business case where a system like this would be preferred over a good computer and a separate render server in another room

its built badly and with very little care, clearly can see it in the video.

@@beeseechurger Clearly? Except for a couple of cables and hoses, that aren't organized perfectly, I haven't seen any "clearly" bad thing. But: Your opinion, of course. 👍

@@Mathias-RetroFutureTech temperature reading unit is in a nearly useless spot (the outlet), the hoses are managed onto a huge block with no definite way to manage the flow, pump is a measly d5 which will barely get any of that water moving, half of the hoses are shoddily secured with little screw-in clamps there's no NV-LINK so that person using it is limited to learning ONLY which is not a good idea since, this being in a desktop case, is clearly going to go to someone's house and not in a rack. Meaning it very well could be used for things other than learning, which is a big realization for the next person. theres no PSU shroud, and the intake on the PSU is on the wrong side. The water lines are 100% going to get hot with that slow flow-rate, not to mention those are managed like spaghetti. The cables are managed like ass, which is just a silly way to do things when you're building a 20k PC. Plus, dont think I forgot, but theres only 4 sticks of RAM in a board clearly made for 8 sticks. If you're going to spend so much, you might as well actually get enough RAM for these learning tasks or AI or whatever. I mean come on you could atleast have double for 256gb. IDK man. This PC seems like it was built by some rando and this guy just threw it together. Silly build.

@@beeseechurger Damn, that's an eye opener! To be honest, I'm really not that deep into custom water-cooling and believed that nvlink isn't a thing anymore (had actually asked myself if one really doesn't need it for connecting these cards nowadays)... So I guess my uneducated point of view made me believe this is a thought trough system. Thanks for elaborating these flaws! 👌👍

​​@@Mathias-RetroFutureTechThen you know nothing about computers. This system will overheat as soon as you try to do any actual work with it.

What? The end goal is flow, not pressure. Parallel will provide better cooling. By the way, he installs four of gpu so they can work in parallel. He could supply same amount of task to one gpu in serial but in parallel they do the same job faster. Different areas - same principle. Parallel splits the task

@@eee8662 Low pressure => low flow. Water will flow fast to the last block (due to pressure by gravity) and almost none to the uper block of the splitter.

@@rodrigocunha34 1. The rule "low pressure leads to low flow" is correct when resistance is constant. I.e. if you have only one gpu. You can achieve higher flow by increasing pressure. When configuration changes - not one gpu but 4(!) Of them - you can not use that the same measure stick (pressure) that was in Internet references use for one gpu. Consider this: you can decrease the pipes diameter (or better yet plug them off completely to see the point) - the flow will approach zero but yea, at least you will get the high pressure. The lesson is: yes you increase the pressure to achieve the high flow, not the other way around. Going serial - is that other way around - akin plugging pipes off - you will kill flow but skyrocket pressure. 2. Role of gravity. My intuition tells me it will be negligible here. If any. The easy way to prove - to put meter on each gpu and see. 3. Quick thought tells me - gravity will play no role at all - the amount of water that goes down is equal amount of water that goes up. I may be wrong about this however. 4. You may be correct that there may be different amount of water going to each gpu but that is not due to gravity but rather due to distance from main radiator/pump ports. The way to fix it - system must be symmetrical.. Bottom line - you want it symmetrical and parallel. Serial - will kill it.

@@rodrigocunha34 And. Don't forget the temperature. Symmetric-parallel will ensure each gpu gets fresh cold water from pump. Serial - the last gpu in line will get hot exhaust water warmed by three of those gpu before it.

@@rodrigocunha34 if resulting water flow is less than desirable - he must use more powerful pump. Or even several pumps. But. No way GPUs be connected serially.

ale chujowa

Teraz żeby zrobił 8 rtx 4090

that would be a fat azz no, this is made for whole home whole damn office / neighborhood etc networking from single point machine closed away in the shed for heat

@@mybsite5697 homie just explained a server

This machine is not a gaming rig, it's a deep-learning AI workstation.

think he'll need to fill all ram slots to run chrome!

@@chris9517 firefox still freezes up though.

It Will be enough that's 2 120mm space for each 4090 and 1 120mm space for the CPU that's thinking they would all be hit at the same time Wich is unlikely and also they're running the system at stock speeds so the power draw won't be that high

AFAIK RAM cooling is not for practical functionality but for aesthetics

RAM doesn't really need cooling. A guy called buildzoid pushed a DDR4 stick without radiator to 5000MT/s. He was using a fan to blow air on it but still, radiators are only for the looks

think derbauer did that test heatsink indeed drop temps, but yeah prob adequent airflow and slot gap are sufficient

8 pack PC uses old graphic card and old CPU. so no way to beat this monster PC.

4x RTX 4090 > 2x RTX 3090 duh!

@@luckys1evin Except for gaming where two 3090 can work in SLI which the 4090 can't, so the dual 3090 may outperform the single 4090 in many cases (and the other three just idle).

@@magicmulder a. SLI is pretty much dead - either does not work, or has loads of issues, unless you're all about breaking benchmarks - it's nearly useless. b. this ain't a gaming PC, it is used for professional work where each GPU can work individually.

It will be used for deep learning bro

@@SuperMetal4Eva Do 4090 even support sli? How would it work?

@@TurboGroszek1 No. The 3090 does though.

@@TurboGroszek1 SLI doesnt matter as it will be used for Deep Learning.. more Graphics Power = more power for deep learning

@@TurboGroszek1 Deep Learning requires multi Graphics processing power... in SLI GPu memory is mirrored not stacked so not relevant for deep learning

Мы пронаблюдали как из компьютерных комплектующих можно сделать домашний обогреватель на 2 киловатта или если подключить к нему водопровод то бойлер на два киловатта :)

@@igorshustov4591, мы пронаблюдали за копи-пастом.

Yeah, and the loop is configured for seemingly the lowest flow rate possible, check the flow meter in the last couple scenes, slow as hell. Everything in that is gonna throttle. Shame. Should’ve had some internal rads, plus the cpu first in the loop. As well as that, an additional d5 pump would have done wonders.

idk man this is almost 2kw of power... even in the best case scenario I think that would lead to some throttling

Even the shown setup is not cool enough, he would need another pump and double on fans to set up a push pull.

lol

that poor D5 doing its best

Client gets what client wants

Its actually best to give all of the radiator surface area to all of the parts, though you are right about the pumps. The pressure and flow rate are going to suffer due to being pushed through a parallel distribution block and a huge amount of tubing to an external triple rad.

@@ZeroX252 i was talking about basically giving each part its own loop and having the rads outside of the case.

How are you going to fit a 360 meter radiator into that tiny little case

i was talking about external cooling @@thomastakeshita4979

well he uses 4 sticks DDR4-3200 mhz without a heatspreader wich is kinda old and bad for a build like this

@@Managarm1999 it's ECC memory they can't go as high because they made for 0 tolerance and in rigs that need to be on 24/7

@@Managarm1999 Heatsinks on memory sticks are there for aesthetics and branding. They are literally not needed whatsoever since DDR3 came out.

@@user-tb2uz5mu8l tell that to ddr5

I dunno dude, how are you gonna use 4 4090's, sli is dead, the only thing i can see this thing being useful for is machine learning, like the only other reason to have more than one gpu is because you are running linux and want to have a dedicated gpu for windows, that's why i have a 6950x and a 4090, and the only reason why i even need a 6950x for linux is because i got 4 4k screens which just use too much vram when i'm even just on the desktop, other than those 2 reasons there is no point in having more than 1 gpu, let alone 4 gpu's lol

@@MrTurbo_ I completely agree that it's overkill and almost useless, cool nontheless.

​@@MrTurbo_ Well like you said, this is probably going to be used for machine learning. In which case it's an absolute beast of a machine. I highly doubt someone bought 4 4090's to play games.

After effects, blender, Unreal Engine 5, Premiere Pro, Cinema 4D, And houdini All at the same time under max load lol@@MrTurbo_

В наилучшем? Взято дорого и богато, а по факту умрёт через пару дней

But there was rgb, all set to rainbow puke 😅

Budget friendly? That setup is probably worth more than my life

this is what i want to know to my understanding its impossible to use them all at once on one task but maybe you can run four different tasks at once one on each GPU idk though

distributed processing

@@ANONYM0US Its possible to use all GPUS at once for deep learning