Eh, my boss is nerdy and often likes to learn what we do and new innovation so he can always step in if needed and be of actual use. My boss is abnormal and the absolute shit, though
If I was a hardware manufacturer I wouldn't send anything to LMG unless there was a signed agreement that Linus would be handcuffed to the wall the whole time. He's a danger to innocent hardware!
I have met people who can instantly break things as soon as they approach them, they are great for using as software testers when coding... The reverse also exists which is someone who frightens computers so much that they miraculously start working in their presence, much to the annoyance of the users, who will swear that it wasn't working until you showed up...
If you guys can get the Time Spy run to complete, your world record will likely stand for three more hardware generations. 4×1080 Tis still beat 2×3090 Tis in Fire Strike.
@@ricogol I 2nd this. I NEED to know what the score is. I can't even fathom how incredibly powerful this workstation is. A Sub 8 second BMW render?!?! I literally screamed WAHT?!? at my monitor when that happened lol
For me (in games with full SLI support), 2 SLI 1080ti cards are equal to 1 OC 3090. I love my 3090, but it only helps me with games that don't have SLI support.
@@charlesswhitlock That is more likely due to such games being optimised for SLI and not for modern hardware. They probably leave a lot of the potential performance of the 3090 untapped.
That cooling system was honestly the coolest part. I’m upset that such an implementation hasn’t been quite adopted for the consumer market. Those quick disconnects and custom water blocks were a thing of beauty.
“I don’t know Intel’s lineup wery well anymore” from Linus is a real testament to the progress AMD made in last several years, especially on server side.
Linus has reach the point where people send him Industry and Research rated 60.000$ Maschines for random "shits and giggles" review videos. I love it. :3
@@Manish-fm5iv I am one of those guys who recommends shit like this for stuff so i can play with it.. normally its just a trip to the ceo's desk asking for a signature on the purchase order, with no questions asked.. 60k for a machine is nothing in the grand scheme of things for large businesses who need the power, they drop millions on robotics
@@fitybux4664 On the off chance you're being serious, it's a thousands separator, not a decimal comma. Between that and the $ sign being put after the amount, OP is probably not from an Anglophone country, my guess being somewhere in Europe.
Well, it kinda makes sense. Employee breaks an expensive piece of equipment; they're getting their ass kicked... or fired. Boss breaks an expensive piece of equipment; Meh, cost of doing business.
@@The_Keeper Except in this case the employees are running behind their boss trying to keep him from breaking expensive stuff. Normally it is the other way around.
@@jake_ actually no, I'm pretty sure it's standard in many companies for the employees to be the competent ones, and the bosses being the ones that need supervising.
Me: "That is meant for serious work. There's no way any lowly peasant could justify the cost of that hardware" Also me: "I need to find a way to justify the cost of all that hardware so I can game and edit photos at the same time"
@@aurunemaru oh definitely. As I said, it's meant for serious work. I'm just not sure there's any task I, specifically, could come up with to make back 60k, even if I came up with a deep learning task for it
3d photogrammetry will do it. A high detail reconstruction on my 5950x/3090fe takes over an hour and loads both at 100% during various stages. Peak draw around 600W according to my UPS. 3dfzephyr has a benchmark if you want something standard.
I love how modern computers and servers are also a heater for your room so you don't need extra heating. And lets not forget the electrical bill that you will get when using things like this(imagine using so much electricity that your meter just malfunctions).
"It's professional software, and you need a professional GPU to run it. Performance on GeForce is very poor." SolidWorks: _crashes_ _on_ _Quadro_ Let's call it what it is: low quality software
A Heads Up for the LTT team: The Blender BMW benchmark is outdated (even the website doesn't exist anymore) - there's a new version, which runs 3 tests and then calculates a proper rendering score out of that. It's a huge upgrade.
There was the rendering benchmark. I always wondered why movie studios don't use realtime rendering more, but I now realize they can afford a bunch of these. What I call offline rendering is what this machine calls 4K144hz...
I've got a pair of HP DL165's that i have to use a prybar on to the get the lids off. The push points just don't provide enough leverage to move the bloody things more than a few cm, enough to get a prybar in. I stopped putting the lids back on a few years ago.
You should let these rip on Folding@home or Boinc Manager projects to just see how many you get through in 15min. It'd be cool to see as a bench mark and you'd be helping Scientists so it's a win win.
I was just thinking about Folding@home and SETI@home, but wasn't sure if they were still a thing. I see that the latter stopped in March 2020, but could start up again using data from other radio telescopes.
When I was in my undergrad degree, we did a lot atomic crystal structure modeling on a gpu server. We used Quantum Espresso to research quantum sensing materials. Some of our lattice structures would take several hours to run. I wonder if that would make a good GPU test?
@Nightrogue Designs UK it's basically looking at materials that could be used for things like quantum computing. You basically model something like a diamond crystal by building a 3D model of each atom location in a modeling software. Then you take the coordinates of the atoms you model and put them in an input file. The input file is then feed in our case to the GPU server. The server is then used to calculate/estimate the forces and energies between each atom in the crystal structure. You can then use that data to things like ZPL (zero phonon line emission), band gap, etc.
4:02 The "very durable lttstore" water bottle has a dent on the bottom of it from falling off a table. It will now wobble anytime it is placed on a surface.
9:43 " you just have an aura " reminds me of my boss in the eighties, if he came into the production area he had to put his hands in his pockets because he would point at machines and they would stop working !
"It's been stuck at the border due to some stupid COVID delay." Border guard: Hey guys, check out this cool PC - I bet we could play Cyberpunk on Ultra!😀 Let's just hang on to it for another week or so, we'll just give them the standard COVID excuse 😊... To be honest though: If they had gamers over there and knew what it was, you'd most likely never receive it - it would be _lost_ in the mail 😉.
i feel like linus has ideas that his kids would have. Problem: big wooden box with thousands and thousands of dollars of tech inside. Solution: break it with a big hammer.
The infinity gauntlet was able to run 5 gem SLI. Quin-SLI has never been supported, but his glove made it work! And when you are capable of fully supporting quinSLI, you can literally rewrite reality on a universal scale. That's why you never see anyone try more than 4
after 5 yrs , everyone of LTT will have this as a media center station for the dogs to watch movies. Next gen LTT employees will show this in a what awkwardly slow systems they had back in 2020's running a sub 8 second renders.
@@JohnCiaccio Between buggy driver support (Nvidia does not officially support running games on workstation GPUs) and the lower CPU frequency and worse memory latency it is quite likely even a modest gaming PC will beat it significantly in most games at a fraction of the cost per person.
Good job shooing Linus away, Jake!! God he was damn near giving me a full blown panic attack with his hands in there, the crowbar, and sledge hammer. LOL
As far as I know, for many machine learning / deep learning purposes, those A6000s won’t scale well - they scale up almost linearly when using 2 of them (since NVlink is only for pairs of GPUs). To scale well, you probably need SMX NVswitch capable GPUs (like A100) with Mellanox (for multi-system scaling). Bandwidth (between GPUs) is a b***h nowadays.
The best strategy is to just wait for GPUs with more VRAM, and not waste any time on multi-GPU and the countless problems that come with it. The AMD MI250 already have 128GB at 3.2 TB/s. Should be way faster than 2 A100. Next week I can benchmark them at JSC, then I'll know :)
It's much cheaper than the DGX and even performs better. Crazy to have this much power be affordable now. They also make them smaller which is nice, almost attainable to have a Datacenter at your desk. Some of the marketing is for "forensic computing" which is essentially breaking encryption.
Armoury Crate sucks. It does not detect my mobo or fans. I contatced, support attached all the details and they say reinstall of os is the only way to get it working again. I tried openrgb and it works without any problem.
It took me hours to get RGB between armory crate and corsair software to properly talk with each other. I still have to run two pieces of bloatware software, and every time armory crate has an update, it intentionally stops working until I open the software (and it tells me that there is an update). However, I refuse to update the software because I'm afraid that it'll break what I've previously set up. I'm definitely going to have to try one of these 3rd party RGB controllers.
I used to live there (in Birkirkara, Malta, visited Comino once), in 2018 when I started to follow LTT. Fond memories of that place. I cracked when I saw the reference in the intro :D
Forget that... put it to work on SETI@home and we'd find our cosmic neighbors by the end of the week, then they can tell us how to cure all diseases, completely negating the need for protein folding research in the first place.
Noob: why would you even need that? Pro: (slaps case) this bad boy can fit so much machine learning Hacker: I could brute force crack the shadow government with that thing.
It's fun to watch common folk break down different tech made and conceived by geniuses. Questions should be.... what can I invent? What would a genius do?....stuff like that. I think a lot of geniuses do not have common sense and that's why we have sites like LTT & others, but sometimes I wonder if "common sense" has been completely shoved under the rug.
3:10 that's a really cool commodore linus 3:10 PLEASE RUN 3DMARK ON THIS BEAST!!! 3:10 EDIT: 16:30 YES THAT IS AWESOME!!!!! 3:10 double edit: 18:50 NOOOO!!!! (RUN IT WITHOUT 4WAY I HAVE TO KNOW THE SCORE!!!!) PS. 3:10
this might be a farfetched video idea, but it would be really cool to have an video on the incredible technology behind the signaling systems that me fully automated trains and metro's possible. the easiest way to make a video on the topic might be to work together with tales ( the company that made the signaling equipment used on the skytrain) and TransLink/BCRTC/InTransitBC (the operators of the skytrain) to make a video about the fully automated Vancouver skytrain light metro system. But if that is not possible it wouldn't surpice me if some other compagny that makes etcs equipment would be more than willing to get you to talk about it. the might even sponsor the video.
I'm just waiting for the day that LTT suddenly goes bankrupt cause Linus takes the "I break stuff" skit too far and accidentally goes full ham with the crowbar taking out a $1m server that is on loan at this point...
I wouldn’t be happy with Linus’ review if I were the manufacturer of that PC. Very random tests and very rough handling of the machine. It kept crashing because nothing was upgraded (autocorrect unrated). I’d be furious
Nothing more inspiring than two guys getting extremely expensive equipment they don't have use for and they don't understand. If you had trashed it in some maniac way like throwing it from the roof of the office it would have been peak youtuber moment.
As much as I love LTT, I know I will be dissapointed when it comes to machine learning hardware testing. Who in the right mind will use Windows OS to test multi-gpu workstation/server which intended for machine learning. Just hire some CS grad student already to run some ViT/ResNet/EffNet on ImageNet
Well, that would be extremely boring to watch... Remember, boring watching experience means less people watching. I.E. less money for the channel. To put it bluntly, LTT can't afford to do it the "proper" way.
@@ssl3546 as they stated in the early video, the market would be for machine learning/scientific purpose. So, it would be more proper to test machine learning performance isn't it.
@@The_Keeper boring? I don't think so, with proper ML test you can utilize all capabilities this machine has to offer. As you can see in the video, they have difficulty on pushing gpu/cpu to its limit in some test. It would be interesting to see this machine pushed to its maximum capabilities. And I am sure LTT can explain the technical ML terms for general audience.
How exactly would that tech support call to the manufacturer go? "We tried to start this $60,000 computer without looking at any of the documentation then we tried to take it apart before we got it working. Now we can't figure out why it doesn't turn on." Manufacturer hangs up.
"Hey boss, you said you wanted us to put some dummy GPUs into the slots to show how easy it is to swap 'em out, but we don't know what to use! "Oh, just use chuck few of those spare 3090 Tis we have laying around in there...."
I'm surprised every time you guys mention anything like needing to be on two circuits, limits are so low on your side of the ocean... We have 3600 watts as a maximum per circuit here in Europe (well at least in the Netherlands but I think it's pretty universal). Glad we have the higher voltage so that's possible without insanely thick cables :)
GotDamn... You have blown all benchmarks out of the water! This thing makes you need to have a set of "render system" very big and heavy benchmark setup. Then you can start to rate each of these massive systems out against each other I think this one is king of the hill.
