Takes like 10 years max. I myself have a dual socket serverboard from Supermicro as well, bought 2 Xeon E5 2697v2 which had a 2013 release price of 2.600$ and bought them for 60€ each
Kudos to Jake and team to actually shoot the whole video in almost 1 hour (based on the start time 12.06 to 1.11 at the end in Jake's watch) It's mind-blowing scene changes have happened in matter of minutes, attaching monitor and booting it, etc, too little time in every step very efficient guys
I did desktop support for an oil and gas energy company (before the lockdown) that specialized in deep water exploration. The geologists there processed seismic data to find oil deposits. They had monster workstations with crazy Quaddro graphics at every desk. I could definitely see this type of machine going in to some of their homes.
I actually studied some of the algorithms used for calculating interface boundaries in seismic data in my Simulation Science major and it was pretty damn interesting... But damn the math was hard xD
Really. What was gained by unplugging water lines or trying to remove a graphics card? Maybe the next video is Linus getting invoiced for the computer if he left one of the waterlines unhooked. lol.
@@the_omg3242 Showing how easy they were and that they don't leak when disconnected? Dry breaks are pretty cool. Pointing at hardware and handling it is very different content.
I used to work for an SI. I got a phone call from some lovely people at Google who needed help fixing their $60,000 Desktop that was a lot like this. It was a Thelio Massive workstation with a couple of Platinum 8280 CPUs and 3 NVIDIA A6000 GPUs. It's not quite this but it came close to it. Those guys were using it for machine learning experimentation. which would be the same application as this system. It was a stressful phone call. I know nothing about machine learning. Fortunately those Google peeps were so hyper focused in their profession that the problem was their system stopped booting because their experiments filled up all of the storage in their system so fast that they didn't even know what hit them and they needed help booting into a Live disk and clearing some of their crap out of the storage.. These super desktops are incredible, and incredibly stressful when they aren't working as expected. Imagine doing an RMA for a $60,000 to $80,000 computer. No thanks. That won't be fun for any party involved.
@@prashanthb6521Just because you know how to build AI models does not mean you know how to build a PC. My friend who works in ML couldn't boot a live USB
@@prashanthb6521most developers can't reinstall their own os Basically because web developers never really had to? (And we web developers are definitely the _most_ developers by headcount)
@@danielthedoc It really depends a lot on how you got into data science. I started my career in computational physics and built several simulation rigs from scratch--including working directly with Dell to spec and set up a $50k Beowulf cluster. This was back in the days before there were dedicated data science degree programs. Meanwhile a lot of my colleagues got masters in data science and have worked exclusively through cloud computing, where you have neither direct hardware access nor any reason not to use Windows or macOS.
Hope the sponsor is not mad cause that line made me actually watch the entire thing, which I usually always skip, I'm sure others did the same. 9001 IQ marketing
I know Jake has hosted various things before but... I think he's done exceptionally well at CES. His presenting has been spot on and his video's are really engaging. Well done that man 👏
@RandomUser So people showcasing techs' latest products is boring to you? I would suggest you're watching the wrong channel and your time is being severely wasted. Try a Mukban or sewing channel, they maybe more relevant to you.
@RandomUser The whole point of attending CES and filming in hotel rooms is so they can bring us the latest tech as fast as possible. So just because of the surroundings and video production, you would prefer to wait until just before a product is released to hear/see the details of it? I would think that 99.9% of people are grateful that companies/creator's such as LTT, Paul's Hardware and many others travel to Vegas to get us the latest information on upcoming products. I know I am.
As a mechanical engineer, this workstation is wonderful. This would make running COMSOL Multiphysics simulations actually viable at either an individual engineers desk or at their home (if they're working remotely) without having to load the small jobs onto a compute cluster at work like you do for large jobs - time on the big compute cluster at work is a premium so currently small jobs don't get much time so designs don't get optimised anywhere near as much as would be ideal. This is a game changer.
Fascinating. I imagine equipment like this would change AI and simulation workloads from a labor based model where engineers and scientists are constantly waiting for servertime (history sure rhymes a lot, doesn't it?) and companies have to pay many engineers a good salary, to a capital intensive model where large incumbent corporations with capital backing can afford to invest heavily in poaching the most productive engineers and scientists and giving them capital intensive home AI deeplearning workstations that rival the capabilities of midlevel businesses. Maybe this might spell a mass layoff of AI engineers in the future as AI research gets consolidated into the companies that can all 100 engineers one of these and do the work of 100,000?
@@RagingAura I doubt it would change much regarding AI deep learning workloads as basically anyone with an average machine can already utilise most gaming computers for that as currently one can buy or lease a dataset, so training either isn't required or is only required to fine tune the AI. To be honest as a mechanical engineer I do very little with AI - it has zero relevance or place in mechanical engineering. In Australia, engineers get paid poorly just like all top-talent. Basically in Australia businesses only want people who'll accept garbage wages whilst being top-talent. This is why many people, myself included are moving abroad where better pay and opportunities exist - a colleague and I are going to start our own engineering business / consultancy once out of Australia. As mentioned in my initial comment, this machine would be amazing as it would mean I'd be able to setup and run one multiphysics simulation every two days (runtime is about two days for meaningful results) unlike the current situation where I have to book time on the main compute cluster at work (one week of waiting) and then get results in 4 hours. Multiphysics isn't something that can be done by AI as it requires a human who understands the intent behind the design to configure and setup the simulation. The time consuming part is the actual calculation which is multi-threaded for as many CPU's as you have a licence for (the licence only cares about sockets, not individual threads as it uses as many threads as are available). There isn't any such job as AI engineer, only programmers who setup AI. Engineering is a strictly regulated industry with only a few fields (Mechanical, Civil, Chemical, Electrical and Biomedical) - beyond that people are illegally using the title engineer as there are no other legally recognised fields of engineering. The term engineer can't just be applied to any other term or role as that would be completely illegal. Currently, proportionately only a handful of people in general are employed by most large businesses (compared to pre-1980 employment statistics), pre-AI servers with automation software eliminated 98% of common jobs in the early 2000's. AI has only eliminated all remaining "office" jobs where automation software couldn't. I can tell you for a fact that one server racks worth of equipment, two mechanical engineers and one person to do the accounts and phones is all that is needed to do serious work in the space and defence industry - teams of people are no longer needed and are only found on projects that require 24/7 work to be done - i.e. rotation of teams such as a day team and a night team.
@@Blooest dog if the computers off with all variables ignored like a silent room you could hear your sisters panties dropthen you wouldnt need to refer to a None running computer outputting 0 decibels Like nigguh wat Refer to it Idling compared to it under full load
@@katrinabryce Depending on the game, it could be stellar; it wouldn't be a dog, in any case. But for five grand, you can have something that will run any game at full chat anyway. (I do have a dual-processor Xeon box a few feet away, it's a data handler, not a gamer.)
IBM, Dell etc. To be honest it's been a while since I've dealt with servers directly. But Super micro were always good. We had Dell and Tyan servers back then too.
@@isaacbejjani5116 HP Enterprise has revenue that's 5 times larger than Supermicro. Lenovo is even bigger. Yes, Supermicro isn't small, but it's hardly the largest.
Well done Jake! You totally owned this video, your excitement is infectious! It kept me engaged, I went into this video thinking ok I'll click to see without the intention of watching the whole thing. Keep it up bro!
I’m the target customer for this system, and I’m very impressed. I’ve built an AI deep learning workstation and know the industry options and this is a very compelling prebuilt workstation option.
@@Rynnakkosampyla He just told you, like Jake on the video. Its used to train AI's (Deep Neural Networks or other algo). Those algorithms are very hard to train correctly and most of them can only be trained well in machines like these or else would take years on regular hardware...
@@Rynnakkosampyla I wanna know how much Horsey this thing can spit !?! That's a $#!T LOAD of Power in 1 Box ! Like a mini Sever Farm on your Desktop. I see Intel is "ALMOST"close,(cause that is ONLY 56 cores)up to catching AMD in to the core counts. Took them LONG enough ! They had to steal 1/2 of AMD's engineers to get the job done,especially working on 10nm,i wonder if they still got that right? If you ever looked in your bug log,it would scare you. It happens ALL the time.Intel,AMD or ANY others. BUGS in Hardware & Software are SO common,it's like fleas in your backyard. You just don't notice them. The OS handles them,unless you get the the 1 nasty screen,and you know what i'm talking about,the blue or old black screen of death. Everyone JUMPED when the BIG bugs Meltdown and Spectre hit the scene. 😱 If you look back in history on both sides of the fence,them bugs crop ALL the time! They need to catch them before 1rst rollout,but that doesn't happen. More testing should be done in 3rd party,but that would negate their tight @$$ Security procedures. Just look at the burning NVIDIA plugs as an example. Sure they had alot of R&D with the consortium,but then they put the MEAT to it! I just think that's too much power for that plug,simple! As a retired EE,i call it as i see it. If it smells,it stinks! 😎
@@jonathanthomas2449 it's about as suspicious as not knowing why a door moved on it's own, there are about 1000 boring explanations none of which are as entertaining as a falsehood
I wouldn't say HOME USE but for lab use this is awesome! It's waaay faster than some supercomputers available on universities. Als as a single system is easier to upgrade and maitaing if you don't have datacenter like infrastructure! Only a 30A outlet and it's flying
Remember the regular blade servers are 1RU (1.75") high. To get air across that area in that confined a space with a small fan/blower you need high RPM. This tower can house large fans and thus run slower for the same airflow. The 12V bus bars are classic server rack. It would be very cool if Supermicro advocated a move to 40 or 60V for distribution. Much thinner wires and its the standard voltage in rack systems. It can also be less regulated as long as its clean.
@Soyel if using a plumbing fitting for plumbing is “repurposing” then I must need to brush up on my English. To be clear there were no automotive fuel rails in the PC in the video.
@Soyel what I’m saying is that using a pipe thread to barb fitting to connect a hose to a threaded port when the system is designed for exactly that is not repurposing anything. It’s using the stuff for its original purpose. Another way to put it: if that company pulled washer fluid reservoirs out of junkyard Toyotas and put them into an $80k PC, then yeah, that’s repurposed. But all I saw was general-purpose fittings and hoses, and building things like cooling loops is what they were originally intended for. Anyway, I’ll give up first on the semantics argument, because that’s all this is.
I couldn't hear the cooling on this thing over the sound of my WFH laptop, which I use to connect to an equally expensive server which is nowhere near as powerful. Actually impressive.
Jake is one of my favorite people from this channel, he talks about everything with a clear passion. He loves what he does and it makes me love watching it even more
@@mikwit I still have free access to more powerful GPU servers (4x MI250) via university. Lamda are quite expensive. I hope the A100's will show up on ebay for a couple hundred bucks in a decade or so :)
@@ProjectPhysX You can get a P100 on eBay cheap as chips ($300) and for none AI work they are ~60-70% the speed of an A100. We just brought a whole bunch because the A100's were being tied up with MD work causing issues for those wanting to do AI work. Yes we benchmarked it before get the P100's in (we had a single P100 for available for that). The V100's are still too pricey.
@@jonathanbuzzard1376 yep I've seen them on ebay :) For my purpose, I actually don't need FP64, but large VRAM capacity instead. There is loads of Tesla M40/P40 24GB for ~140/230 Euro on ebay right now, which is super cheap. I just don't have a server to put in 8 of these :D
The red coolant is the same color as what my truck uses. I started to use it for my DIY cooling in my system. I dilute it down a bit more due to the smaller pump I use, but the diesel coolant is the best one can use. Great wetting, freeze, metal compatibly, and anti-alga. Also is warranted for three or more years. Unlike car motors, commercial truck motors need good cooling and manufactures can not skimp out on QA due to large fleets have the power to switch coolant if one causes breakdowns that cost $$$$$.
As someone that has been assigned with a similar workstation recently, I really hope for Jake's confidence in removing those tubes scrub onto me somehow.
Same! I want to see how fast this beast able to do, when it runs all the benchmarks on Blender, VRay, Octane Render, Cinebench, etc. 😃 A $80000 USD Machine should be blazing fast! ⚡
I know everybody is busy but please add CES videos to the playlist as you upload them. They can always be adjusted if changes are made afterward. Also secret shopper playlist is only a placeholder. Thx Linux
i will never understand why there is no PC case ventures like this, with prebuilt radiator, intergrated pumps, resorvoir, quick-disconnects, swapable harddrives and PSU but just for the high end "gaming" segment...and the best thing would be NO RGB
My 8 year old daughter just took a polaroid of me sitting at my desk bc she got one for christmas, and jake is in the background. forever in my family photobook jake
@@pneumonoultramicroscopicsi4065 I just like my Ultra high end products to be released on time and not nearly 2 years late and half the cores. Remind me again who had the CES Keynote this year? certainly wasn't intel showing off their renamed platform.
@@tim3172 Intel only leads on a select few library, and in reality most users who are looking to do actual ML or DL work are using GPU or dedicated accelerator, like they literally show in this video. But go ahead and give me an incoherent caps lock response.
I work in IT (Linux/Open Source) and have used the Meiko Computing Surface and Unisys ES7000's. Living in a financially crippled/energy restricted country (United Kingdom), it is interesting to consider what computer has the Wow! factor for me these days. This TensorFlow machine is impressive and amazing. However, almost nobody in the UK could dream of running a computer which takes 4kW (aside from needing a 16A socket). Home server is a RPi CM4 (8GB RAM/32GB eMMC boot) in an Axzez Interceptor board mounted in a Mini-ITX NAS case (32TB SATA disk) - as it sips tiny amounts of power. My Dev PC is a 4yo Asus ROG running Linux Mint - again it uses little power. Must be a different world in Canada with relatively cheap power!
There’s nothing like the static electricity from peeling off that plastic, frying your GPU, from peeling it off after the PC is built! Wincing every time you removed one! Haha.
Makes a lot of sense for something like structural biology. You can run Teams on Linux (and I do), but you'd probably have a system like this maxed out with real work and have a second computer for things like email and calls.
Someone made an interesting test were if you limit the RTX 4090 to 144 fps in 1440p and medium settings it actually uses less Watts overall than lower power hardware. Would be interesting if you could test that out. Like limit all to the same settings with all hitting the same fps and see the power usage.