@@rd_626what the hell are you talking about?.The satisfying OP said isn't about “owning and training”...its about how fast it will train LLMs. Try renting tpuv4 or even the new ones tpuv5s....smooth as butter. you will never ever run anything on gpus again lol
@@rd_626 “satisfying” is literally OPs point of contention. you can't take it out of OPs context....any ML engineer only gets satisfied/orgasms on how fast his large data is being trained before his eyes....not by owning costly shit just to see it train mid compared to tpus
@@Mr.Not_Sure A100 on AWS can use upto 450W. Still, in most local setups people will limit its usage to 350-ish because even a good liquid cooling won't cut it. Fyi, the instance i use is p4d.24xlarge so they are 40Gb versions. not 80Gb. i don't know if 80Gb ones work in lower watt or not.
@@MustafaKhan-hg8lu I have my own experience that I have made over the last decade on many different computers and laptops. But I'm not going to argue just do whatever you think is right.
@@Turboriderthen you should use Kryonaut Extreme or a Kryosheet. I myself work on a lot of stuff and MX-4 is completely fine for everything you could want as a "normal" person. It also really depends on what you want to do with it. For example I've had a bunch of issues with NT-H2 squishing away from my GPU die due pump out caused by the extreme heat and power I was pushing through that card (3090 with an XOC BIOS). On my 7900X, its another story. No direct die setup. NT-H2 was fine and pretty okay, one of the better pastes that I used. But with Kryonaut Extreme, I got way more stable clock speeds while pushing the same amount of power and thermals. Heat transfer is simply just better with this one. However, I would not recommend it to anyone that doesn't want to push for the absolute best results. It's simply too expensive. But for everything I use personally, I use not much else anymore.
I am not a fan of liquid cooled servers, but we have air cooled units with similar specs (mainly Supermicro) and they are way more crowded and noisy as hell. That one looks really nice and neat, and you should be able to sit next to it without going deaf. Also, in the long run it won't end so full of dust and crap. This one of yours is a custom unit though, so the client probably had certain requirements in mind, but not others. I see he/she decided to skip things like plug and play redundant PSUs, for example. Also, any leak could cause catastrophic failure and down time, which is not always acceptable. Anyway, very nice.
I completely agree, liquid cooling in the server environment is a complete nonsense. It doesn't matter how qualitative is the cooling system, it will leak sooner or lately.
@@noJobProgrammer I know some servers use liquid cooling but it's probably more like AC, with heat exchanger close to the server racks rather than using water blocks directly on components
@@noJobProgrammerif it leaks, you haven't done it right. Does your house plumbing leak every few years for example? If you do it correctly, there is absolutely no problem to be had there
I just had to ask... "The classic question, "But can it run Crysis?" has been a humorous way to ask if a computer is truly powerful, given the notoriously demanding nature of the original "Crysis" game when it was released. The server workstation you've described, with its dual Intel Xeon Scalable 5220 CPUs, 256GB of DDR4 ECC/REG RAM, and especially with four NVIDIA A100 80GB GPUs, is vastly more powerful than what's required to run the original "Crysis" game or even its remastered versions. However, there are a few caveats to consider: GPU Compatibility: The NVIDIA A100 GPUs are designed for compute tasks, AI workloads, and data center applications rather than gaming. While they possess more than enough raw power to handle a game like "Crysis," their architecture and drivers are not optimized for gaming performance. Therefore, without proper gaming drivers, the experience might not be as smooth as expected on gaming-optimized GPUs. Software Compatibility: Running a game on a server-grade operating system might require additional configuration, as these systems are not typically set up for gaming out of the box. Compatibility layers or virtualization with a gaming-optimized OS might be necessary. Overkill Configuration: This setup is overkill for just about any current gaming requirement, including "Crysis." The cost and capabilities of this system far exceed what's necessary for gaming, making it more suitable for professional and industrial applications mentioned previously. In summary, while this server workstation could technically run "Crysis" and virtually any other game at the highest settings, it's akin to using a space shuttle to commute to work-far beyond what's necessary and not optimized for the task. The real value of this setup lies in its ability to perform high-level computational tasks, AI and ML workloads, and handle massive data processing challenges."
@@ChristianStout this korean bro is running his pc assembly service company for at least 7 yrs and he is professional. if your claim was a serious debate and his client's server just broke down, his company would lose its reputation and he has to discharge tens of thousands of bucks, and he would have already had to close his company a few years ago, whereas he is still running with the reputation of many past years. And let me guess your biggest achivement is that you built your own small gaming pc whereas this korean guy has been building hundreds of pcs and servers.
@@qpakrkdhw7966 This thermal paste application will cause the GPU to perform worse and fail within 4 years. His clients' machines don't break because they are liquidated after only 3 years. With a proper amount of thermal paste, they would last 8 years. Ask an Nvidia engineer, they'll tell you the same thing.
Y'all have been marketed to too hard. The thermal paste is maybe the least important part of cooling this over here as long as the paste is applied evenly. You don't actually need thermal grizzly unless you wanna overclock or if you just really like UA-cam sponsors
why would you need one ? It's mostly for data centers and stuff, unless you're doing something to use that 80gb of vram, because for gaming it would be pretty uderwhelming, as the driver suport wouldn't be the best.
@@cooooooooooooooool3 because ai applications in general, aswell as LLMs run like shit on the RTX 4070. Training obj detection nets also take quite a while even with pytorch+cuda :(
@reapiu8316 That makes sense, you have pretty vram intensive workloads. My dayjob is video editing, so I never exceed 4gb of vram use even with 4k editing
Strange setup IMO. These GPUs are built to run in servers, passive cooled by server hi speed fans, connected between each other with nvlink for rapid data transfer. Water cooling is quieter of course, but who cares about noise in datacenters? BTW nvidia produce a100 in sxm credit card size format, you can easily fit 8x GPUs into 2U server. Oh, the second PSU would be nice for some redundancy.
That has so much glue, latency and bottlenecking. Plus, it's overprice AF. You think that 7/9980XE asking price of US$2K was nuts, AMD's line-up takes the cake. Thanks Evil Su!
Maybe it has to do with some compatibility issues on AMD side. Although the Intel/AMD CPUs share the same instruction sets for the most part, the software that is going to be running this hardware might be very picky.
Doesn't look good :((( water block doesn't look like covers all die and HBM memory , 4x a100 maxed out would be 1600w plus cpu$ram 300w plus the rest: water pump disks etc would't fit into 2kw PSU. I'd Expect overheating and restarts on load :( and I would go 6 channel memory
Yes enjoy this ! the future of the domestic hardware.... Ready for pay at maximum!!!! Embrace Nvidia and you ll get for your money. Hope you get a lot of money !
It doesn't matter (whatever was available/affordable in terms of CPU+mobo). Its a GPU node , the CPUs are only there to provide the PCI lanes, run the OS and handle the IO. Everything else will be handled by the GPUs on application level. That is why he didn't put much RAM in it because , it doesn't need it.
This is great but I am confused. I have question: ASUS WS C621E SAGE motherboard has PCIe 3.0 x16 support which means it can be transferred in speed 16GB, but the GPU NVIDIA A100 support PCIE 4.0 x16 which means 32 GB. Doesn't it mean that you can not use max performance of the GPU NVIDIA A100 ? It can use only 16gb in via PCIe 3.0 x16 slot.
No way that doesn't run hot with the cooling solution they chose. I mean it is probably all within specs, but would run with less power if they used more in terms of cooling.
Y se dobla un pin y ya no prende 😅 Excelente armado por qué hay que tener mucho cuidado ⚠️ Por qué están ENORMES esos procesadores un mega server Saludos
Ох и дибилизм. 1. Помазано мх4 карту которая стоит овердохера. 2. Коннекторы 8 на 8+8 зачем ? 3. Термопрокладки накидали от прикола. Как можно было доверить подобное таким людям ?
All of those and it just operate the same as one RTX1050 with 1000000 watts and 1 Fps on any game on the lowest possible setting. Devalue isn't an acceptable thing.
if i would be building a server like this,i would idk what would i do but i won't be happy when i get to the cable managment,in fact,cable managment is only for personal computers,but i just wont see where the cable is
can we all take a minute to thank nvidia for ruining computer builds with their hideous adapter cables. we all know normal PCI pins could have stayed in use, or a more sensible board mounted standard could have been implemented, but no, lets go with a molex adapter style, because what fits better in my $4000 PC
Nice, but terrible airflow. This could have been executed better if better cable management was used and the distribution block wasn't blocking the radiators' airflow.
I would be too scared to wear gloves with fibers like this one while installing a CPU because a stray fiber that you don’t see may get caught on a pin on the socket, and pull it, and I really don’t want to do socket repairs. This is why I wear nitrile gloves instead.
При сборке компа за 90 к парень мажет термопасту GD900, ещё и точками... Надеюсь тот кто будет обслуживать сервер купит нормальную пасту и размажет термопасту
Через 5лет, процессор на АлиЭкспресс 90$ видеокарта 900$ 🤔Термопасту нанёс я считаю плохо точками, (если представить как эти точки превращается в круги лепешки которые могут создать воздушные карманы при соприкосновение друг с другом ;) Сежу на х99 5лет уже на 2696v3, a когда то стоил 2700$ а сейчас 30$ на али ;)
I would never ever place the thermal compound on the bare GPU chip in that way. The CPU has an heat spreader so it can work well even if some part of the spreader ends up without thermal compound be the bare chips of the GPUs don't have so much luck. I would spread the thermal compound with a plastic tool so as to make sure every part of the gpu chips is covered.
2 questions: 1- Why not fill the remaining memory slots with 4x32GB? 2- Would custom length cables have removed some clutter and added to the airflow in the U4 chassis? Nice job, I've already got a perfect spot picked out for it.
You know shit gonna be real expensive when you don't even recognize half of the hardware's names.
Could not be for cheap one
But all of them are easily recognizable?
tell them that you get nvidia A100 to some average gamer and see their reaction@@Lainyyyyy
We r talking about kinda above average pc user, not tech professionals@@Lainyyyyy
Jelous 😂😂😂😂
Gloves +socket pins= pain
Why are you wearing gardening gloves? 😂😂
Finally a build that can run Crysis in 1080p 30fps,nice 👍🏼
640p 20 fps Maximum
@@Porsche_x my results are using Frame Generation and low settings
@@des7royerz So yes, I said 640p at 20 fps using graphics very low without dlss and without frame generator
LTT ran Crysis on AMD Server CPU with software rendering and got like 10 or so
with dlss 3, fsr 3 and low seting 🤣
This PC consumed more power than my entire house electronics combined
about 2kw when all processores is max-loaded
Your water kettle would easily consume more power than this machine.
@@NowhereNear42 we talking about electricity, not heat
@@thenasiudk1337 You talking about power. This PC uses electricity to produce heat. So does your water kettle.
Doesn't use more than 2kw.
Finally, a server for running Latest version of minecraft in the future.
😂😂😂😂😂 Minecraft
Tetris
still only get 45fps in vanilla MC
Vastly underrated comment. 😂
А ты в теме😎
finnaly a pc to compile gentoo
You would have the fastest emerge times in the world
my i5 13600kf took 4h to emerge -avq
this thing would do that in 10 sec xD
would not like a coolant leak in that
I can't imagine how satisfying it must be to train LLM´s on that thing.
you can easily rent set up like this for around $4 per hour
@@AmanUrumbekov Renting is not satisfying as owing and running and training LLM's locallly
@@rd_626what the hell are you talking about?.The satisfying OP said isn't about “owning and training”...its about how fast it will train LLMs.
Try renting tpuv4 or even the new ones tpuv5s....smooth as butter. you will never ever run anything on gpus again lol
@@sohailmd123 Just FYI..... I wasn't saying that to the op
@@rd_626 “satisfying” is literally OPs point of contention. you can't take it out of OPs context....any ML engineer only gets satisfied/orgasms on how fast his large data is being trained before his eyes....not by owning costly shit just to see it train mid compared to tpus
-Для чего тебе такая сборка.
-Чтобы Excel работал без лагов.
Наконец-то я смогу открыть 4 вкладки в хроме))
В перерыве можно и пасьянс разложить по-человечески.
LOL
I'm afraid to ask this question and rather just look at this beauty but "Does it even run..?"
Crysis....
Used for AI backend stuff. Mostly for training. Runs only on linux
😂👏... it's top of the line ... Runner 🏃♂️ it seems
Is that 2kw PSU enough to run two Xeons and 4 A100s?
Yes
If you assume 500W per component at full power (it could happen), then it's 1000W short.
@@NegativeROGA100's are 250 W.
@@Mr.Not_Sure lul
@@Mr.Not_Sure A100 on AWS can use upto 450W. Still, in most local setups people will limit its usage to 350-ish because even a good liquid cooling won't cut it.
Fyi, the instance i use is p4d.24xlarge so they are 40Gb versions. not 80Gb. i don't know if 80Gb ones work in lower watt or not.
It must be so scary to work on gpu's that expensive
У меня дома такой же пк, но на 90тыс долларов дешевле)))
Kinda want to see the thermal performance of this build. The fans don’t seem like powerful enough.
Its not about the fans. Its about shitty MX-4 that he put on the everything
Yeah, the most overrated thermal compound. Noctua NT-H2 is way better for example.
@@Turboriderthermal compound isn't making or breaking this build lol. Maybe the most unimportant part of cooling as long as it's applied right.
@@MustafaKhan-hg8lu I have my own experience that I have made over the last decade on many different computers and laptops. But I'm not going to argue just do whatever you think is right.
@@Turboriderthen you should use Kryonaut Extreme or a Kryosheet. I myself work on a lot of stuff and MX-4 is completely fine for everything you could want as a "normal" person.
It also really depends on what you want to do with it. For example I've had a bunch of issues with NT-H2 squishing away from my GPU die due pump out caused by the extreme heat and power I was pushing through that card (3090 with an XOC BIOS).
On my 7900X, its another story. No direct die setup. NT-H2 was fine and pretty okay, one of the better pastes that I used. But with Kryonaut Extreme, I got way more stable clock speeds while pushing the same amount of power and thermals. Heat transfer is simply just better with this one.
However, I would not recommend it to anyone that doesn't want to push for the absolute best results. It's simply too expensive. But for everything I use personally, I use not much else anymore.
i guess a single cooling fan is worth my entire house but okok 👍
I am not a fan of liquid cooled servers, but we have air cooled units with similar specs (mainly Supermicro) and they are way more crowded and noisy as hell. That one looks really nice and neat, and you should be able to sit next to it without going deaf. Also, in the long run it won't end so full of dust and crap. This one of yours is a custom unit though, so the client probably had certain requirements in mind, but not others. I see he/she decided to skip things like plug and play redundant PSUs, for example. Also, any leak could cause catastrophic failure and down time, which is not always acceptable. Anyway, very nice.
I completely agree, liquid cooling in the server environment is a complete nonsense. It doesn't matter how qualitative is the cooling system, it will leak sooner or lately.
@@noJobProgrammer I know some servers use liquid cooling but it's probably more like AC, with heat exchanger close to the server racks rather than using water blocks directly on components
In high density applications such as HPC or AI training liquid to chip can be used, particularly in the 80kW/rack+ range
IBM Mainframe can be liquid cooled. Since its reputation for reliability, I’m sure reliability isn’t an issue if do it correctly.
@@noJobProgrammerif it leaks, you haven't done it right. Does your house plumbing leak every few years for example? If you do it correctly, there is absolutely no problem to be had there
5 year old cpus :(
Why don't you do an actual review of the power it has? Some benchmarks. It would be awesome.
I just had to ask...
"The classic question, "But can it run Crysis?" has been a humorous way to ask if a computer is truly powerful, given the notoriously demanding nature of the original "Crysis" game when it was released. The server workstation you've described, with its dual Intel Xeon Scalable 5220 CPUs, 256GB of DDR4 ECC/REG RAM, and especially with four NVIDIA A100 80GB GPUs, is vastly more powerful than what's required to run the original "Crysis" game or even its remastered versions.
However, there are a few caveats to consider:
GPU Compatibility: The NVIDIA A100 GPUs are designed for compute tasks, AI workloads, and data center applications rather than gaming. While they possess more than enough raw power to handle a game like "Crysis," their architecture and drivers are not optimized for gaming performance. Therefore, without proper gaming drivers, the experience might not be as smooth as expected on gaming-optimized GPUs.
Software Compatibility: Running a game on a server-grade operating system might require additional configuration, as these systems are not typically set up for gaming out of the box. Compatibility layers or virtualization with a gaming-optimized OS might be necessary.
Overkill Configuration: This setup is overkill for just about any current gaming requirement, including "Crysis." The cost and capabilities of this system far exceed what's necessary for gaming, making it more suitable for professional and industrial applications mentioned previously.
In summary, while this server workstation could technically run "Crysis" and virtually any other game at the highest settings, it's akin to using a space shuttle to commute to work-far beyond what's necessary and not optimized for the task. The real value of this setup lies in its ability to perform high-level computational tasks, AI and ML workloads, and handle massive data processing challenges."
simple example would be movie making, games, large scale 3d render/preview, realtime weather or anything 'ai' like water simulation etc
there are 'desktop supercomputer' if you keen on having server cpu to run crysis
W ai comment
That's not enough thermal paste on the GPUs. Anything less than 100% coverate will create hot spots that will kill parts of the chip.
u know pc better than him?
@@qpakrkdhw7966 I know at least one thing he doesn't.
@@ChristianStout this korean bro is running his pc assembly service company for at least 7 yrs and he is professional.
if your claim was a serious debate and his client's server just broke down, his company would lose its reputation and he has to discharge tens of thousands of bucks, and he would have already had to close his company a few years ago, whereas he is still running with the reputation of many past years.
And let me guess your biggest achivement is that you built your own small gaming pc whereas this korean guy has been building hundreds of pcs and servers.
@@qpakrkdhw7966 This thermal paste application will cause the GPU to perform worse and fail within 4 years. His clients' machines don't break because they are liquidated after only 3 years. With a proper amount of thermal paste, they would last 8 years.
Ask an Nvidia engineer, they'll tell you the same thing.
MX-4? Seriously???? Why didnt you use Noctua NT-H2 thermal paste? Or Thermal Grizzly Kryonaut Extreme?
Y'all have been marketed to too hard. The thermal paste is maybe the least important part of cooling this over here as long as the paste is applied evenly. You don't actually need thermal grizzly unless you wanna overclock or if you just really like UA-cam sponsors
@@MustafaKhan-hg8lu you know its A100 we are talking about, right?.. its not your average 4060
там и GD900 справится
I'm waiting for the benchmark test, do you have it? how long does it takes to train nanoGPT medium size model?
He just wanted to show off and prove he could afford one.
@@shidongxu3410 They are an actual computer shop! This is for their client
@@shidongxu3410 This is probably his job and he's just showing the process
@@shidongxu3410 no way someone can own 4 A100's and use it for personal usage
@@rd_626 a lot of people can but why would these people do it is the proper question.
Can it Run Crysis?
Can it run Google Chrome with 10 tabs?
Just one A100 would be my dream frfr
why would you need one ? It's mostly for data centers and stuff, unless you're doing something to use that 80gb of vram, because for gaming it would be pretty uderwhelming, as the driver suport wouldn't be the best.
@@cooooooooooooooool3 because ai applications in general, aswell as LLMs run like shit on the RTX 4070. Training obj detection nets also take quite a while even with pytorch+cuda :(
@reapiu8316 That makes sense, you have pretty vram intensive workloads. My dayjob is video editing, so I never exceed 4gb of vram use even with 4k editing
@@cooooooooooooooool3my Firefox uses 11 gb vram :(
Depending on the task. Nvidia Tesla P40 is not a bad entry for 24gb of vram in the used market. Runs RVC smooth as butter in my experience
Strange setup IMO. These GPUs are built to run in servers, passive cooled by server hi speed fans, connected between each other with nvlink for rapid data transfer. Water cooling is quieter of course, but who cares about noise in datacenters? BTW nvidia produce a100 in sxm credit card size format, you can easily fit 8x GPUs into 2U server. Oh, the second PSU would be nice for some redundancy.
smx is not the size of a credit card ,but more like a phone like s23 Ultra ,
GTA 6 recommended requirements:
whos ready to play minecraft :D
Хули опера не во всех слотах? Не жир!.
out of curisousty...why 2 Xeons and not the 7995WX Pro Threadripper?
sweet server build though!
That has so much glue, latency and bottlenecking. Plus, it's overprice AF. You think that 7/9980XE asking price of US$2K was nuts, AMD's line-up takes the cake. Thanks Evil Su!
Maybe it has to do with some compatibility issues on AMD side. Although the Intel/AMD CPUs share the same instruction sets for the most part, the software that is going to be running this hardware might be very picky.
budget system with last generation nvidia ampere GPUs even v3 or v4 haswell or broadwell xeons would be ebough
For machine learning, this is what the purpose of this build is for. For ml task, use Xeons due to Intel’s math library
Amd threadripper is for workstation
Intel xeon and amd epyc is for server
Doesn't look good :((( water block doesn't look like covers all die and HBM memory , 4x a100 maxed out would be 1600w plus cpu$ram 300w plus the rest: water pump disks etc would't fit into 2kw PSU. I'd Expect overheating and restarts on load :( and I would go 6 channel memory
looks like there's mistame in wikipedia, a100 80gb pcie has 300w TDP, so should be enough and I hope cooling will be fine with those pads.
Yes enjoy this ! the future of the domestic hardware.... Ready for pay at maximum!!!! Embrace Nvidia and you ll get for your money. Hope you get a lot of money !
Ok I can build my setup I have the Alphacool A100 Carbon, now I need a A100 80GB.
Yes 320gb vram 256 gb dram 😂😂 3:06
The server is cool and all but why it doesn't have the SLI on the gpus?
Is there like a reason why
A build like that, and not even using redundant PSU? 😮
ur crazy
@@lastlyhi Thanks
Question, why Xeon and not Epyc platform? If i recognize correctly, Epyc is made for servers, and its powerful
It doesn't matter (whatever was available/affordable in terms of CPU+mobo). Its a GPU node , the CPUs are only there to provide the PCI lanes, run the OS and handle the IO. Everything else will be handled by the GPUs on application level. That is why he didn't put much RAM in it because , it doesn't need it.
What the hell IS bro cooking
His CPU's most likely with that thermal paste application.
Bro gonna create another dimension
finally nobody whining like a baby about memory
Nice build but 2TB ain't going to hold anything.. and you have more GPU memory than actual memory😂
it is not a data server so it does not need a lot of storage
Can it run 5 chrome tabs?
This is great but I am confused. I have question: ASUS WS C621E SAGE motherboard has PCIe 3.0 x16 support which means it can be transferred in speed 16GB, but the GPU NVIDIA A100 support PCIE 4.0 x16 which means 32 GB. Doesn't it mean that you can not use max performance of the GPU NVIDIA A100 ? It can use only 16gb in via PCIe 3.0 x16 slot.
More vram then ram in this lmao
No way that doesn't run hot with the cooling solution they chose. I mean it is probably all within specs, but would run with less power if they used more in terms of cooling.
Y se dobla un pin y ya no prende 😅
Excelente armado por qué hay que tener mucho cuidado ⚠️
Por qué están ENORMES esos procesadores un mega server
Saludos
Ох и дибилизм.
1. Помазано мх4 карту которая стоит овердохера.
2. Коннекторы 8 на 8+8 зачем ?
3. Термопрокладки накидали от прикола.
Как можно было доверить подобное таким людям ?
Da para minerar quando em criptomoedas por dia kkkkkkkkk
Estão sendo usadas para criar máquinas com consciência, que inclusive podem aprender até quebrar criptografias de várias criptomoedas.
Any reason it should be mounted this orientation? Why not upside down, so leaking coolant will not damage the mobo?
Nice. Do similar water cooling adapters fitting V100 SXM2 modules exist?
What is this server for?
now im ready to play minecraft
Ready to open google chrome with 10 tabs)))
So thats 4 A100s but there was no NV link or any thing that means that system has 4 80 GB gpus Dope for running you own GPTs
90000 USD in single pic .
Хотелось бы узнать сколько выдаст ФПС на этом монстре в киберпанке 2077 в 8к на ультра настройках.😮
Нисколько. У карт A100 нет вывода для монитора
@@exhaust4685 обида обидная.
All of those and it just operate the same as one RTX1050 with 1000000 watts and 1 Fps on any game on the lowest possible setting. Devalue isn't an acceptable thing.
if i would be building a server like this,i would idk what would i do but i won't be happy when i get to the cable managment,in fact,cable managment is only for personal computers,but i just wont see where the cable is
can we all take a minute to thank nvidia for ruining computer builds with their hideous adapter cables. we all know normal PCI pins could have stayed in use, or a more sensible board mounted standard could have been implemented, but no, lets go with a molex adapter style, because what fits better in my $4000 PC
90K for the PC, only uses 4 of 6 Memory Channels. Wtf, thats only like 200$ extra...
Nice, but terrible airflow. This could have been executed better if better cable management was used and the distribution block wasn't blocking the radiators' airflow.
I would be too scared to wear gloves with fibers like this one while installing a CPU because a stray fiber that you don’t see may get caught on a pin on the socket, and pull it, and I really don’t want to do socket repairs. This is why I wear nitrile gloves instead.
Finally PC to run Yandere Simulator.
Why leave 4 memory channels unused? Is the client buying more DIMMs later?
...waterblock conduits attached manually, with no tool?
Ok, where's the trick? This can't be safe.
#HELP CPU: Intel Xeon Scalable 5220 (2EA)
whats 2EA mean?
Kinda pointless watwrcooling when you block the airflow with the exchange block
Why Xeon? Epyc is so much Better
huh
Ultra HD Raytracing DLSS NoWayFu*k CyberPunk 2077 with this one ?
4x A100 equals 1000W, this is nonsens! By the way what liquid is used and how do you fill it in? How do you bleed it?!
I have a question: with such a huge configuration, what will you use it for?
The case cover doesn't look like it'll go back on, your water cooling pipes are standing too tall
При сборке компа за 90 к парень мажет термопасту GD900, ещё и точками... Надеюсь тот кто будет обслуживать сервер купит нормальную пасту и размажет термопасту
You know it’s powerful when they use HP instead of watts ☠️☠️
Чел, там паста нужна другого уровня, мх-4 уже не подходит ни как! Хотя бы Honeywell ptm7950-SP А это полный пздц!
no ram, vrm, eth cooling, different length of tubes lead to uneven water flow, correct me if I'm wrong
Uh.. I have the same setup. It was not 90,000. More like 55k. This video title is trash.
Через 5лет, процессор на АлиЭкспресс 90$ видеокарта 900$
🤔Термопасту нанёс я считаю плохо точками, (если представить как эти точки превращается в круги лепешки которые могут создать воздушные карманы при соприкосновение друг с другом ;)
Сежу на х99 5лет уже на 2696v3, a когда то стоил 2700$ а сейчас 30$ на али ;)
I would never ever place the thermal compound on the bare GPU chip in that way. The CPU has an heat spreader so it can work well even if some part of the spreader ends up without thermal compound be the bare chips of the GPUs don't have so much luck. I would spread the thermal compound with a plastic tool so as to make sure every part of the gpu chips is covered.
yeah if pasting on a bare die it's definitely a better idea to manually spread to cover the whole die
Nvidia A100 and using intel xeon ? why not AMD epyc 96 core * 2 ?
可以同時跑4個踩地雷...
Love to see the performance stats on it once it's powered up.
2 questions: 1- Why not fill the remaining memory slots with 4x32GB?
2- Would custom length cables have removed some clutter and added to the airflow in the U4 chassis?
Nice job, I've already got a perfect spot picked out for it.
Min specs for video playback on Adobe After Effects 2023
But u just accidently face reveal while cleaning the gpu a100
I ll send to you a number. Didnt gave you the right to film my workstation
only fittingsa and tubes cost are more than my desktop's
I am LADA Niva Legend, is exchange possible?
Can it do anything good to humankind ?
Or only aggravate the global climate change
Finally a PC that could seamlessly run Plant and Zombies..
А я, щоб закіп'ятити воду, використовую звичайний чайник
finally a PC that can run MS word at 240 FPS👍
This can ruh roblox and minecraft up to 100k fps
Why you didn't put ram in all slots? I think you crippled the memory speed quite a bit. 🤷♂️
Still can't handle modded Minecraft server at 20tps
Does this computer lag when playing FreeCell?
I actually have 2 of those SK hynix M.2 2tb for my pc
Какой то странный сервер, не понятно зачем ему четыре видеокарты, когда на сервере нужны ссд
2 киловатта могут легко разогреть такой объем воды
Impressive build but its the memory kind of slow?
imagine play gta 6 with rtx in 8k 60fps with this pc 💀🗣🔥
Если так намазывать термопасту то мастер еще тот иногда она по всему чипу не растекается.
I think rust will run on medium settings at 40 fps :)