The level2 boss lives in a junk yard of computer parts that you get sent to in person. He has helper minions that chase you, and you can only kill them in VR, so you have to search the junkyard for parts to build a VR capable PC from scratch; soldering parts, building custom cables, daisy chaining battery cells charged by a diesel engine to create a stable power supply; everything. It's a lot. Once you kill the minions in VR, you have to use the PC to code a neural network that will generate the weapon needed to kill the boss. No one ever gets enough training data for the neural network. That's why we're still at level1.
@@kalervonakki2527 sure, but if he even gets a dozen people or so into more specialized stuff it was worth the trouble. Different audiences, different presentation style, same interest, same goal Though I would prefer it if they'd publish their test data and methodology in text form somewhere... like you've got the data, you've got all the shit already written up. Just let me read instead of watch it.
@@iQKyyR3K I would be happy if they would just clearly say the product name or what the content is about in the video title. I feel like they(and many others like lts) don't really care about their viewers at all anymore. Keep the thumbnails but just give me some useful information at least. I'm sure they work for the intended audience. But sadly, they are part of the problem..
@@kalervonakki2527 absolutely this. I really don't mind the presentation style, but the clickbait titles with the waaay to over the top and non-descriptive thumbnails are just in poor taste
While I have 30 years experience building/maintaining/troubleshooting desktop PCs, I will freely admit that my knowledge of enterprise computing is microscopically small. But I still watch these kinds of uber-high-end IT videos with fascination. It's like tech porn to me. ;-p My main system is so old that I am still using SATA SSDs and HDDs for storage, yet I still feel it is plenty fast enough for my needs. ;-)
@@SirReptitious I have about 20 years doing the same for myself and friends. Videos like this have inspired me to start studying for some certifications. A+ seems pretty simple but I really want to get my CCENT and I am realizing how little I know about enterprise and networking stuff!
@@xbox360Rob That's true. My AM3+ mobo doesn't support NVMe drives(yes, I know they can be added via PCIe cards in x8 and x16 slots). But MANY people before me have said that the difference between using a system with HD and SATA SSD is much larger than between SATA SSD and NVMe. Even though NVMe 4.0 can be like 12x faster than SATA in sequential copies, that is not the normal operating pattern of a normal PC. Last year I built a new Ryzen 5 2600 system for a friend, and installing windows & office, copying over his old files, and generally configuring and optimizing it for him felt only a little faster with the NVMe in his system versus the SATA in my system.
Exciting times to be into computers! I'm starting my first job in that space soon and I can already tell that a lot is going to change and evolve over the next couple years with ever higher demand and serious competition going on now. I'm personally not that into enterprise hardware, but I like big numbers and I appreciate big computing power!
I've downloaded most of Level1Tech's videos and some other youtubers and archived in on my server so even when after WW3 or any other apocalypse ruined this g forsaken planet at least we can still preserve some knowledge
It genuinely saddens me to think back to this time when we were all so amazed by Optane, and so hopeful for how it would change everything. 3DXPoint was SUCH a good tech. But we don't get to have nice things because the marketing department at some NAND flash companies convinced the world that big numbers from Q64T64 are what consumers should care about. If only the general public knew just a little more. If only they bought a few more Optane drives. Just enough to see the price steadily drop from economics of scale. We could all have SSDs that a normal user is incapable of using up, with 10x more iops per chip, and latencies so fast that most software will just report 0.0μs.
Just a side note: Intel also created DPDK, basically the same thing for networking. Cisco built a framework based on it, called VPP and open-sourced it - you can basically get 10x better performance than regular OS networking(it runs in user-mode and gets even faster as you add more RAM), without FPGA. The downside here is that all the software needs to specifically target these APIs to take advantage of it, it’s not like a driver install and that’s it, “gains”.
Awesome information, thanks! Also for our edification, when creating the test suite mentioned at the end of your video please include RAM drive I/O comparisons. Right, with 256GB of RAM and 256GB NVMe being so fast and so cheap using the two for system or application execution environments are interesting to say the least. :) PS: I really hope said test suite is available for DL. :D
Steve Gibson is finding that flash drives can have huge performance differences depending on where the data is located in the drive as well. I'm confident that will hold true with NVMe as well. Because nothing can ever be easy.
He signed out very quickly at 14:50 and then got distracted again. It's kind of amazing. "i'mwendellthisislevelonei'msigningoutyoucanfindmeattheleveloneforums"
@@Onihikage I totally understand, I was just foolin' since he said it twice lmao I love Wendell and could totes listen to him for hours even with most of it going over my head lol All just a bit of fun!
RIP optane. Love them, use the 380GB 905P as my boot drive and have one in most of my servers. Hopefully I can upgrade to one of these in a few years lol
Spinlocks are not constrained to linux....it's a common method for handling multithreaded operations. Instead of the consumer/producer waiting and then locking for the other....they spin (do nothing) for a very small amount of time, as it's expected after the spin there will be data to consume/be able to write to write to the resource/queue/etc.
I think, I run into a somewhat comparable problems with IOPs and latency on my own potato. My hardware is a Ryzen 3 2200G; 16GB DDR4 (3000MHz); 512GB Silicon Power nvme-SSD (3400/2300MB/s). I run Ubuntu 21.04 on OpenZFS 2.0 with L1ARC
Videos like this make me realize how much I hate my organization sometimes (I love my job). I am fighting to replace a Dell R820 with dual 4617 and no real oboard storage. The 5900x I just built is so much better.
I want to see someone run a PC with as little memory as possible (like 1-2 GB; maybe less), with just a somewhat fast SSD (probably nothing crazy like this P5800X) operating for the page file and/or as a readyboost file (I'm not sure if the Readyboost gives any benefit over a page file or what). Like some of these cheap Optanes have 32 GB of memory on them themselves. I presume that's really garbo memory but still it would be interesting to see how it responds. I feel like it could potentially be a pseudo-viable solution for certain people on a budget or something (at least if they had some sort of niche case where they needed a lot of memory on a cheap system). Well actually, that's not really something I think is possible, but rather I just want to see how such a setup would perform.
Didn't the Level1Techs news mention something about a new theoretical technology called Dynamic Flash Memory? This would be great with helping to reduce latency :-)
I have an interesting test… almost funny really. Some maps on cities skylines have many tens of gigabytes of data, across many thousands of assets. When you start the game you load the map… there is a loading screen mid to gather details on what is loading when, you might have to turn that off for best performance… but seeing how long it takes for the game to become playable on initial load on such a map would be a test
3rd gen Optane Persistent Memory (Crow Pass) will be compatible with DDR5 memory slots and ship with Sapphire Rapids CPUs next year. Intel will very likely make a 3rd gen Optane PCIe 5.0 SSD as well.
Wendell enough with the mumbo jumbo, I need to know if the P5800x will make my system feel like when I went from HDD to SDD ? I have a 11900K system and I need it to be snappier and instantaneous. NVME is not doing it for me.
God Wendel is so smart.. All I can do is build a PC and optimize it and Windows 10.. LOL.. I can't do what Wendel does.. And I'm pretty Techie, I work work the Cable company.. I never got into Lenox or trying to write my own programs. Thats just way above my pay grade, but I still like watching his stuff..
If this is still level one, I'm pretty scared of the what the level 2 boss will look like.
Level 2 comes with a mullet.
You are not ready for Level2Techs ua-cam.com/video/chL2fBZUh8c/v-deo.html
The level2 boss lives in a junk yard of computer parts that you get sent to in person. He has helper minions that chase you, and you can only kill them in VR, so you have to search the junkyard for parts to build a VR capable PC from scratch; soldering parts, building custom cables, daisy chaining battery cells charged by a diesel engine to create a stable power supply; everything. It's a lot.
Once you kill the minions in VR, you have to use the PC to code a neural network that will generate the weapon needed to kill the boss.
No one ever gets enough training data for the neural network. That's why we're still at level1.
When Linus has a crazy project, you call in "Special Server Forces": Wendell. 😉
Linus is just trash level clickbait these days. Forget him.
@@kalervonakki2527 yeah, his recent click bait is getting rather obnoxious these days.
@@kalervonakki2527 sure, but if he even gets a dozen people or so into more specialized stuff it was worth the trouble.
Different audiences, different presentation style, same interest, same goal
Though I would prefer it if they'd publish their test data and methodology in text form somewhere... like you've got the data, you've got all the shit already written up. Just let me read instead of watch it.
@@iQKyyR3K I would be happy if they would just clearly say the product name or what the content is about in the video title.
I feel like they(and many others like lts) don't really care about their viewers at all anymore. Keep the thumbnails but just give me some useful information at least. I'm sure they work for the intended audience. But sadly, they are part of the problem..
@@kalervonakki2527 absolutely this. I really don't mind the presentation style, but the clickbait titles with the waaay to over the top and non-descriptive thumbnails are just in poor taste
Another great video from Wendell!
This is video is one of those that make you actually think if you really know computers lol
While I have 30 years experience building/maintaining/troubleshooting desktop PCs, I will freely admit that my knowledge of enterprise computing is microscopically small. But I still watch these kinds of uber-high-end IT videos with fascination. It's like tech porn to me. ;-p My main system is so old that I am still using SATA SSDs and HDDs for storage, yet I still feel it is plenty fast enough for my needs. ;-)
@@SirReptitious I have about 20 years doing the same for myself and friends. Videos like this have inspired me to start studying for some certifications. A+ seems pretty simple but I really want to get my CCENT and I am realizing how little I know about enterprise and networking stuff!
@@SirReptitious you haven’t tried one that’s faster
@@xbox360Rob That's true. My AM3+ mobo doesn't support NVMe drives(yes, I know they can be added via PCIe cards in x8 and x16 slots). But MANY people before me have said that the difference between using a system with HD and SATA SSD is much larger than between SATA SSD and NVMe. Even though NVMe 4.0 can be like 12x faster than SATA in sequential copies, that is not the normal operating pattern of a normal PC. Last year I built a new Ryzen 5 2600 system for a friend, and installing windows & office, copying over his old files, and generally configuring and optimizing it for him felt only a little faster with the NVMe in his system versus the SATA in my system.
Wendell is on an uploading spree. I love it!
Loving the deep dives. Thanks for sharing!
Exciting times to be into computers! I'm starting my first job in that space soon and I can already tell that a lot is going to change and evolve over the next couple years with ever higher demand and serious competition going on now. I'm personally not that into enterprise hardware, but I like big numbers and I appreciate big computing power!
This is phenomenally exciting aaa
I've downloaded most of Level1Tech's videos and some other youtubers and archived in on my server so even when after WW3 or any other apocalypse ruined this g forsaken planet at least we can still preserve some knowledge
Pair it with some basic research papers and you're good to go
@@monsterous289 good idea... imma start hoarding some research papers and pdf books
I'd love to hear much more about SPDK - hope you'll make a deep dive on that soon. :)
I think I love you, Wendell
It genuinely saddens me to think back to this time when we were all so amazed by Optane, and so hopeful for how it would change everything.
3DXPoint was SUCH a good tech. But we don't get to have nice things because the marketing department at some NAND flash companies convinced the world that big numbers from Q64T64 are what consumers should care about.
If only the general public knew just a little more. If only they bought a few more Optane drives. Just enough to see the price steadily drop from economics of scale.
We could all have SSDs that a normal user is incapable of using up, with 10x more iops per chip, and latencies so fast that most software will just report 0.0μs.
Just a side note: Intel also created DPDK, basically the same thing for networking. Cisco built a framework based on it, called VPP and open-sourced it - you can basically get 10x better performance than regular OS networking(it runs in user-mode and gets even faster as you add more RAM), without FPGA. The downside here is that all the software needs to specifically target these APIs to take advantage of it, it’s not like a driver install and that’s it, “gains”.
Okay, you have officially reached God tier I.T. Wow.
These storage engineers at Intel are doing gods work right now.
And I don't even have the PCIe lanes to keep up!
Awesome information, thanks!
Also for our edification, when creating the test suite mentioned at the end of your video please include RAM drive I/O comparisons. Right, with 256GB of RAM and 256GB NVMe being so fast and so cheap using the two for system or application execution environments are interesting to say the least. :)
PS: I really hope said test suite is available for DL. :D
Excited for Wendell to become the new Allyn Malventano! Definitely missing good storage reviews
Great video and analogy with zero copy. Hats off to Intel for messing with the kernel.
Fascinating stuff!
I really am excited for optaine. I hope Intel can keep it going and get it adopted more. Someone needs to compete with Samsung
Steve Gibson is finding that flash drives can have huge performance differences depending on where the data is located in the drive as well. I'm confident that will hold true with NVMe as well. Because nothing can ever be easy.
Do a read and write test at the same time, to see how this device performance under that condition/scenerio.
I have a 905p as my boot drive. It’s not the fastest boot, but jeezus everything launches like it was nobody’s business
I didn't catch it: Did Wendell ever sign out? Also, where can I find him?
He signed out very quickly at 14:50 and then got distracted again. It's kind of amazing.
"i'mwendellthisislevelonei'msigningoutyoucanfindmeattheleveloneforums"
@@Onihikage I totally understand, I was just foolin' since he said it twice lmao
I love Wendell and could totes listen to him for hours even with most of it going over my head lol All just a bit of fun!
RIP optane. Love them, use the 380GB 905P as my boot drive and have one in most of my servers. Hopefully I can upgrade to one of these in a few years lol
Spinlocks are not constrained to linux....it's a common method for handling multithreaded operations. Instead of the consumer/producer waiting and then locking for the other....they spin (do nothing) for a very small amount of time, as it's expected after the spin there will be data to consume/be able to write to write to the resource/queue/etc.
Does DragonFly BSD not try to address such issues?
Nice to see these developments, basically a necessity for CXL working at some point
Whats above a hyper-scaler a Hyper-ulta-super? I had nothing to add to this video but it was great :)
unpossible! unpossible!
@@auturgicflosculator2183 I like the cut of your jib! lol
Ludicrous-Scaler
@@mathyoooo2 HA! I love it! :)
THIS is good material!
Wow this is nuts! 🤩
I wonder if there will be a shift to main memory having speed instead of capacity, and the main working capacity would be 3d Xpoint
How do you do no locks when you are doing writes? I obviously have some reading to do.
Zoned Name Spaces is one method iirc.
I have to really read up on it though. It is still very new.
Look at the lockless google group for actual code samples.
as long as you have enough memory you can ensure that all new writes don't go in the same place as any old writes.
Nice, had not heard about spdk
THE BADGER??
Any chance of working with Linus again and seeing if you can get that server going with SPDK?
I think, I run into a somewhat comparable problems with IOPs and latency on my own potato. My hardware is a Ryzen 3 2200G; 16GB DDR4 (3000MHz); 512GB Silicon Power nvme-SSD (3400/2300MB/s). I run Ubuntu 21.04 on OpenZFS 2.0 with L1ARC
I don't understand much but I am commenting so I get notified if someone reply's to this 🙂
Woot!
Videos like this make me realize how much I hate my organization sometimes (I love my job). I am fighting to replace a Dell R820 with dual 4617 and no real oboard storage. The 5900x I just built is so much better.
Trying to get a dual epyc and 4-8tb of NVME because delivering rasters takes a lot of space.
I want to see someone run a PC with as little memory as possible (like 1-2 GB; maybe less), with just a somewhat fast SSD (probably nothing crazy like this P5800X) operating for the page file and/or as a readyboost file (I'm not sure if the Readyboost gives any benefit over a page file or what).
Like some of these cheap Optanes have 32 GB of memory on them themselves. I presume that's really garbo memory but still it would be interesting to see how it responds.
I feel like it could potentially be a pseudo-viable solution for certain people on a budget or something (at least if they had some sort of niche case where they needed a lot of memory on a cheap system). Well actually, that's not really something I think is possible, but rather I just want to see how such a setup would perform.
Didn't the Level1Techs news mention something about a new theoretical technology called Dynamic Flash Memory? This would be great with helping to reduce latency :-)
I have an interesting test… almost funny really. Some maps on cities skylines have many tens of gigabytes of data, across many thousands of assets. When you start the game you load the map… there is a loading screen mid to gather details on what is loading when, you might have to turn that off for best performance… but seeing how long it takes for the game to become playable on initial load on such a map would be a test
basically city skylines with mods on steroid .
This didn't pertain to me but it was pretty interesting to watch
So now that this is likely the best 3D Xpoint drives we'll ever see, what workloads/scenarios do you see these best utilized when scooped up off eBay?
We need more Cat Tax!!!! @TechTechPotato knows how to do it!!
0:10 He said "share", but what he meant was "brag".
I'm not jealous. You're jealous!
So much for Optane :(
Neat!
the smartest guy i have ever met
I wish you still shot your videos at 60FPS
I wonder what 3DXpoint on PCIe 5.0 paired with HBM2 can do. And Perstistent Memory of course 😁
3rd gen Optane Persistent Memory (Crow Pass) will be compatible with DDR5 memory slots and ship with Sapphire Rapids CPUs next year. Intel will very likely make a 3rd gen Optane PCIe 5.0 SSD as well.
@@Patrick73787 Unfortunately not anymore.
@@exilonone Yes, Optane is dead unfortunately.
How do you fit the p5800x in the pcie gen 4 slot below the gpu slot on my motherboard?
What frameless monitor is showing the 15 million IOPS ?? 👍😎
Nvme _engagement_
C'mon Intel, gimme that P5800X!!
Can you get SPDK to work on regular Samsung NVME SSD's ?
Spdk is meant to be open source, so it's a matter of time it became mainstream.
if you want to test an ssd that does bog the system down, try a kioxia exercia 480gb 2.5". by far the worst ssd i've ever encountered.
How does 80 mil iops compare to mainframe performance?
Wendell enough with the mumbo jumbo, I need to know if the P5800x will make my system feel like when I went from HDD to SDD ? I have a 11900K system and I need it to be snappier and instantaneous. NVME is not doing it for me.
thats bonkers.
Did you tell Linus?
God Wendel is so smart.. All I can do is build a PC and optimize it and Windows 10.. LOL.. I can't do what Wendel does.. And I'm pretty Techie, I work work the Cable company.. I never got into Lenox or trying to write my own programs. Thats just way above my pay grade, but I still like watching his stuff..
SPDK will be available on desktops or only on server hardware?
It is Linux and ARM exclusive for now (being desktop or server).
Work more on this stuff intel and leave the CPUs to the people who know what they are doing :)
Intel should just support io_uring instead of making there own shit SPDK fork
It's funny because they did/do and even send Jens (the io uring dev) some optane drives like the ones from this video lol
Super delicious wonderful extra terrific meme dream technology, here.
"I just wanted to brag".. lets be honest..
I know he speaks English but that's it ...
Why would storage ever need to be faster than main memory?
Because main memory doesn't persist?
So you don't need RAM.
So....When are you gonna jerryrig that P5800X into a PS5? 😁
thanks for doing a terrible job so well.
How to get rid of freaking Covid 19 YT popup .
3rd
128
Intel will find a way to detect AMD/ARM servers and then throttle performance down to floppy speeds :D
they won't get away with it with L1T and STH making it public as soon as they get their hands on the stuff
First
this video only got 124 views yet
lots to digest... I like to hear the state of the art, though!