You Want This. - HOLY $H!T Threadripper Pro 5995WX
ฝัง
- เผยแพร่เมื่อ 19 มิ.ย. 2024
- Power through work more efficiently and confidently with the help of Grammarly. It’s free, why not?! Go to grammarly.com/LTT to sign up for a FREE account and get 20% off Grammarly Premium!
Check out the CORSAIR VOYAGER - ORIGIN EDITION at bit.ly/3PNcvfr
Before the launch of Ryzen 7000, AMD has slipped in one more Zen 3 product launch. Or at least, they are finally letting consumers buy it. The AMD Threadripper Pro 5995WX is a beast of a CPU that likely has Intel quaking in their boots. But with a massive price leap and the death of the non-pro lineup of Threadripper CPUs, has AMD turned to the greedy dark side? Of course they have, they’re a corporation.
Discuss on the forum: linustechtips.com/topic/14546...
Buy a SuperMicro SuperWorkstation 5014A-TT: lmg.gg/yffCh
Purchases made through some store links may provide some compensation to Linus Media Group.
► GET MERCH: lttstore.com
► SUPPORT US ON FLOATPLANE: www.floatplane.com/ltt
► AFFILIATES, SPONSORS & REFERRALS: lmg.gg/sponsors
► PODCAST GEAR: lmg.gg/podcastgear
FOLLOW US
---------------------------------------------------
Twitter: / linustech
Facebook: / linustech
Instagram: / linustech
TikTok: / linustech
Twitch: / linustech
MUSIC CREDIT
---------------------------------------------------
Intro: Laszlo - Supernova
Video Link: • [Electro] - Laszlo - S...
iTunes Download Link: itunes.apple.com/us/album/sup...
Artist Link: / laszlomusic
Outro: Approaching Nirvana - Sugar High
Video Link: • Sugar High - Approachi...
Listen on Spotify: spoti.fi/UxWkUw
Artist Link: / approachingnirvana
Intro animation by MBarek Abdelwassaa / mbarek_abdel
Monitor And Keyboard by vadimmihalkevich / CC BY 4.0 geni.us/PgGWp
Mechanical RGB Keyboard by BigBrotherECE / CC BY 4.0 geni.us/mj6pHk4
Mouse Gamer free Model By Oscar Creativo / CC BY 4.0 geni.us/Ps3XfE
CHAPTERS
---------------------------------------------------
0:00 Intro
1:45 Y tho?
2:23 Benchmarks
5:24 About that price...
8:55 Epyc 7773X
10:16 Socket Support
11:03 Overclocking Support
14:18 Outro - วิทยาศาสตร์และเทคโนโลยี
Thank you for watching! We’re writing all the time at work whether it’s emails, drafting up video scripts, etc. but having a tool like Grammarly will help improve your productivity and work more efficiently! It’s FREE, why not? Sign up for a FREE account and get 20% off Grammarly Premium: grammarly.com/LTT
gramar
For god sake ive been asking this for ages...what the hell is that damn little blue thing he keeps playing with in vids ?
Sadly nothing is free; if something is free it just means that the merch is you
He really needs money to pay of his lab doesn't he...
@@darthgorthaur258 The remote for his teleprompter.
Yes Linus, I indeed want this, but you see the problem is, I'm broke. So I'm gonna continue to watch your product reviews without ever buying them on my Intel HD graphics thank you very much.
burger 🍔
Same!
Intel hd graphics gang ✊😔
If broke - you can at least rent one on AWS or Azure - for up to several minutes....
Intel HD graphics op
We got our 512gb ram 5995wx machine up and running a week and a half ago. Huge time savings for us, processes our 3D laser scans in 1 scan per 6 Seconds. The pcie lanes support is huge for us.
For our applications there are totally viable cases to use Ryzen 7000/i9 12900ks systems but the 128gb RAM and pcie lanes support limitations reduce the maximum project size we can effectively work on with those systems. The 64 core 512gb and 7 full bandwidth pcie 16 slots allow us to tackle enormous projects without slowing down. The pricing breakdown is totally justified here as when you price things out linearly you end up coming out quite a ways ahead over multiple smaller systems, even without the project size limitations.
What kind of scans are you taking?
Damn what y’all be doing, Webb telescope renders or something 🧐
@@St0RM33 3D scans of primarily oil and gas infrastructure and facilities
@@louistru8652 long story short we make oil and gas piping systems that fit like adult legos
I think the lack of substantial improvement in Adobe product based workbenches has less to do with the chip and more to do with the legacy, patchwork framework that Adobe is still carrying on, especially in Premiere. Maybe try a comparison in Da Vinci or some other more modern framework where a ground up effort has been put to juice out every available core.
Adobe software runs a lot like a game with a single mainthread that calls every asynchronous task (additional thread). Many parts of the code are still synchronous and will hold up the main thread and make their software at BEST lightly threaded. It's really just hot garbage with a familiar UI and a stranglehold on the market due to being one of the first.
It's still an important test. Ultimately you don't buy a more powerful computer to run benchmarks better, you buy it to get your work done better. If the tool of your trade is Adobe then how much better that runs is what matters to you. It may also be time to change tools if another one can do the job better.
I love seeing massively over-the-top rigs like this that are used for a very specific tasks.
Linus is like a child on Christmas when there is new tech
And so are we, which is why all our nerd-selves are here lmao
I mean I would be too if I got to fart around with that kinda power
0:00 shows
Most of us here are including me
a child on christmas with a multi million $ business to buy everything :D
64 cores and 4,5 GHz turbo.. What a monster!
This is single core only. When all cores are loaded it goes to base clock.
@@killersberg1 All depends on cooling and load, as any other Zen.
all that and Tarkov still stutters and lags
Single core turbo, the biggest misunderstanding
@@SpaceRanger187 that's because tarkov is limited to 4 threads or 8, more powerful cores > more cores for games
Compiling Unreal from scratch (and letting it compile the shaders as well when first opening it up) is a good benchmark imho of "how useful it can be for a professional".
I upgraded once from an i7-4790k to my current 3990x as I was doing a lot of freelance work remotely and I did cut my wait time for compilation drastically in many instances. (from 2h to 10-15 mins and from a pc that I couldn't even watch videos in the meantime to now being able to watch something on the side to wait)
Honestly, the TR line-up, even at "enthusiast" level was never meant for gamers but I feel like it's useful for some freelancers that can't afford to throw down close to 20k$ in a machine.
But again, it so depends on the daily workload you work with... :)
I sincerely miss SGI's workstations. Those were actually, TRUELY worth the money. UNIX, baby. I still hope to own an Indigo someday, of which I intend to actually use daily.
7:55 if you're 7x faster at financial applications than the competition then yeah you can basically price your pc however you want
7× faster Excel
@@lucidnonsense942 You're very generous with a millisecond in HFT. 😄 In that field, a microsecond is an eternity.
True, but those calculations are running in server farms across the street from the NYSE, not on your analyst's desk.
No you can't because LTT is the usual 'we don't actually know anything about professional computer users', or maybe actively pandering to AMD - the i9 isn't Intel's pro line.
Threadripper Pro is *extremely* competitive for certain applications against Intel's pro offerings, and if you're e.g. a trade-from-home type with only one PC, then its super compelling.
@@lucidnonsense942 True - especially the last part, since they actually do sell all the chips (rather: chiplets) they can make. I'd wager that the absence of a Zen3 TR non-pro is not a business decision as Linus sees it but the result of practically all Zen3 chiplets being used up by the existing lineup. That's a business decision alright, but of the "what to to with limited supply" kind.
I agree, having support many in my time.. Nothing makes an engineer happier than a speedy beast of a workstation. And you say waiting for 10 minutes.. waiting on that render or computation for 10 minutes to complete also probably means adding an extra 10 or so that they stay extra playing table tennis or whatever else 🙂
It’s amazing how slowly I’ve gotten to a point where I watch (and want to watch) every Linus media group video.
I have my doubts as to if you can even measure Threadrippers performance the standard way, that is with normal benchmarks. I'm also VERY interested how in performs in engineering fields where FEMs are abused to death. Interestingly enough, some form of FEM calculations could be a great benchmark for CPUs too. Encryption might be another field. It feels like benchmarks used in testing are pretty narrow and always come around to neighbourhood of computer graphics and 3D design.
What is FEM?
@@myselfremade Finite Element Methods, which are numerical algorithms with many variants (Continuous and Discontinuous Galerkin to name a few), that are used to solve partial differential equations in fluid mechanics (elliptic pdes) and in other fields like physics and engineering.
@@myselfremade finite element method. It is a technique where you approximate continuous functions like airflow over an object by dividing everything into small cells, the finite elements (finite because they have a finite size instead of being infinitesimal as they ideally should be), and then you use an update function to propagate a discrete (stepped) version of your function through the cells. This is basically how almost very modern physics simulation is done.
Obviously the smaller your cells, the more accurate the result (usually) but also the compute complexity usually explodes. So having more performance allows you to run rough simulations quicker, meaning less idle time, and it allows you to run more accurate simulations at all.
If I could I would send them some files from my job. I wish I could see how fast a threadripper like this could run some of my sims. Because depending on how fast it is I probably could convince my boss to spend that money.
@@jakobwest4811 I've got my disgusting python code from my undergrad thesis on a benchmark method of solving a particular elliptic convection diffusion pde using continuous Galerkin fem I could send them
I really respect the editors notes in the subtitles. Yet another example of LMG's outstanding production quality, but more importantly research of *both* the writers and the editors. Bravo.
Idk, if they really cared they woul put in seperate, real subs for the whole video
@@kkon5ti my point was more on the point that the editors catch mistakes of the writing team / presenters (as no one is perfect).
Its more like thw6 dont put to much effort on searching and then watch the video to see if they did good or need to put the notes(corrections :c) in the video :v
The real problem is they probably filmed this a month or two ago, by then new information on the AMD is already out. Timeliness in tech is important when the technology changes so quickly.
Even their own screwdriver out for that time. Makes me wonder, what has delayed this video that much.
Love the animation on the graphs so you can see what is being talked about
I have a 3955 with a 3090 in my desktop. It's good enough at playing games, but it's literally amazing for office tasks. Applying a MLA to a dataset and outputting as a Visio flow is almost instant, which is basically impossible to do on my high end business laptop. Identifying process flows from use data is an amazing tool, and a single map could save a company millions of dollars, and his computer is the difference between me charging $50 an hour or $200 an hour. I'm really excited for these chips to hit the consumer market.
Now that you have a fully fledged engineering department, I'd love to see some sort of content on the optimal CAD/CAE machines. I use Ansys Mechanical/Fluent/LS-Dyna for simulation all day everyday, and it's frankly a ballache to determine what is the best way to spend money on hardware, or if it's even worth it with the cloud compute options that are available. For example, Ansys Mechanical supposedly benefits greatly from memory bandwidth, and thus recommends Epyc due to its octa-channel memory capabilities, but would this Threadripper also provide benefit in higher clocks for less money
It's also worth mentioning that all simulation/high-end engineering software packages require HPC licenses to use this many cores i.e., £50k+ per seat (perpetual) for 64 cores.
This is not relevant to the video but may I ask how much storage do I need for a m.2 boot drive and what kind of programs should I install on it beside web browser, anti-virus,...?
🤓 hai guith leth thee thome engineering
@@bokute2020 hey, atleast 250gb, for casual stuff is enough, is you use more, then 512,, if you play big games,, rather put in on HDD,, otherwise any software can go on SSD
antivirus on windows 10/windows 11 is not needed really, as defender is quite good,
@@Callagwhan Thanks bro I think I'll go with 512g
In absolute awe of the production values of LTT videos. After watchin the staff meet and greet video its astounding you are where you are given the humble beginnings. So much respect and love for Linus and the entire team
I'm really glad for you linus. You've done alot of great things for yourself.
I may end up with one of these next year, or a related item. It would certainly save many hours of work.
This much compute in a single workstation node is unmatched elsewhere in the market.
Overkill
I'm glad you start talking about the bussiness side of tech on this channel. Very useful for future engineers watching this. Hardware is cheap, Time is expensive.
Also, List price for hardware is very negotiable, perhaps not for LTT but surely for larger corps :).
I'm just loving the look of the case... Far classier than all the RGB on a modern gaming rig. Want!
looks kind of bland tbh, but it's just there for performance.
@@DJSerpent as opposed to the average chassis that looks tacky? 🤷♂️ Different strokes for different folks
@@AntneeUK some look tacky, some look clean, some look bland like this, some actually look classy.
@@DJSerpent I really like the chassis Nvidia DGX Station. The copper designs on the professional gear is 👌
Turns out the chassis is a Supermicro CSE-GS7A-2000B, and it doesn't appear to be available separately. Shame
Can confirm it's aimed at business. Signing off anything sub $50k for hardware is very easy, so a $6.5k CPU is just a no brainer
This Episode Exploded My Head. In The End I Was Speechless. And All I Can Say LTT You Have Opened My Eyes.
I’m pretty happy with my 3960x. All cores at 4GHz all the time. It smokes a couple of the compute clusters I use.
ok FREUD
imagine a whole cluster full of these
@@electroflame6188 imagine a cluster full of COME BACK COME BACK COME BACK COME BACK COME BACK
i really like that ever since wendell has split off tek syndicate you do more stuff with him. definitely would like to see more.
Back when I was a self-employed Graphics Artist Professional and I was building websites using Dreamweaver and Photoshop, time was money. In fact, it was so much so that I would build custom machines with enough RAM that I could create a RAM DISK where I would load Windows, DreamWeaver and Photoshop into RAM. (I also used RAID 10, dual GPU's and cooled it using an independent window AC unit.) It took about 10 minutes for my machine to boot up, but that was perfect to get a muffin and some coffee. For the rest of the day, even the most intensive tasks were completed in seconds. This level of responsiveness paid for itself quickly and made the work easy and fun - waiting is energy-sucking.
The refurb market (even on Amazon) is a ripe one for server nerds and it might make for some interesting content.
Consider that you can get a refurbed 1u system with 64gb ram + 2x Xeon e5-2670 (total of 16 core/32 thread), and 2x512gb SAS drives for $200 USD.
I think you guys should start compiling Unreal Engine 5 instead of Firefox, I went up to 72 logicas cores getting ~linear time scaling.
I have a similar experience about CPUs performing similarly to the GPU at 3:04, rendering in Arnold with a 5900X and a 1660Ti, they perform about the same, if not better, on the 5900X. Less passes and slightly less time needed.
What kind of sick person pairs a 1660Ti with a 5900X?
@@jsVfPe3 garbage pairing huh i have a 5900x but i paired it with something more reasonable. (6800xt) beasst for 1440p gaming got it hooked up to the asus pg279qm 1440p 240hz :)
My R7 2700 also renders at about the same speed as my GTX 770 in Blender Cycles, so... But I think he means that in the way it's a 3090. He's ignoring the part the CPU costs a lot more and that the 3090 won't be able to render anything that takes more than its VRAM can hold (something I suffer with so so much so...). Oh, how much I want GPUs with upgradable VRAM, specially when Nvidia is drip feeding the VRAM sizes on the consumer cards this dirty. I would be okay with even as low as 2060 performance if that meant having upwards of 48 GB to use.
@@jsVfPe3 I paired my 3900X with a rx570, because gaming comes after compile times. It was tough tho.
@@jsVfPe3 computers are not only for gaming, ya know
The Zen4 chips have more than enough compute for me, but not enough PCIe lanes. I want to be able to put a graphics card from each of the major GPU vendors in my machine and there just aren't enough lanes for that.
The AEC firm I work for has 3 5965WX systems on order (128GB RAM, RTX A5000), not top spec systems so “only” about $10k a piece. Will be used for CPU VRay rendering through 3DS Max, Leica Cyclone/3DR, and some Tableau and ArcGIS Pro. They’ll replace some old 9900K based systems so should blow the doors off what we have now, I’m stoked to try them out.
AMD : I heard you like cores, so we put cores on your cores
You know it's a good product when the community starts arguing about whether the benchmarks are sophisticated enough to meaningfully measure it's performance capabilities. If you've gotta start measuring things in a whole new way just to do it justice it's definitely moved the game along.
I am building a high performance multithreaded Java application doing fintech and would love to get my hands on one of these. When markets are moving, you want as many threads crunching numbers for you.
Used to work in an HPC (high performance computing) environment in the pharma industry. Money for them matters little. I remember when covid started we couldn't get cpus so we ended up having to rent HPC servers in aws, OCI, GCP etc. Some times we would get bills o er 20k a day
Whats the specs of your work computer?
I could asbolutely use this at work. I run full 3D EM (electromagnetic) sims and FEM for a living and I bet this would really help. I use an older Xeon and 128GB of RAM in my daily workstation right now, but commonly use our networked machines with 128 cores and 2TB ram. I really wish I could send you guys some HFSS or AWR files to test on rather than Blender or 7zip.
This is what they're missing I think. It's been a hown a bunch of times that only a handful of content workloads benefit from these things, but I've heard from people in other industries saying they absolutely slay.
STH comes closest I feel to being able to properly benchmark these, aside from L1 techs on Dev workflow stuff, but it's till hard to get a clear vision of where they make the most sense.
I can potentially run something for you on our machine
I want videos to keep coming out where Linus says the screwdrivers are 'coming soon,' implying an absolutely enormous backlog of like months and months of videos. Maybe even a gag where it's got something that clearly happened after the screwdrivers came out, like someone's watching Andor in the background as he says, "Screwdrivers should be soon, guys!"
I suspect that the next major step will be in some special purpose hardware rather than in the main CPU path. Things like AI could really benefit from a lot of parallel low resolution computations. Scientific computing often wants an FFT done on a whole mess of data.
7:03, the fact that Linus believes Apple will update the Mac Pro with anything but Apple Sillicon is laughable, great vid in all other regards!!
Would LOVE to see you build the ULTIMATE water-cooling PC, using a Cooler Master HAF 700 EVO with FOUR 420mm radiators at the SAME TIME, as a single water loop, with water cooling blocks on the: CPU, GPU, M.2 drive and on the RAM! This would be the ultimate dream setup for next gen hardware (like the intel 13th gen CPU, RTX4090ti, PCIe gen 5 M.2 drive and even hot DDR5 RAM). With full RGB it would also probably be the best-looking PC setup ever!
I am planning on this setup myself (to ensure I never suffer from thermal throttling, and to allow me to stably overclock as much as I want), but I am nervous as I never done a custom water loop before, so I, & I am sure many others who have saved ready to build a new top of the range setup with the release of the next gen hardware, would really appreciate it if you could do this as a step by step tutorial. Thanks. :)
RTX4090ti doesn't even exist yet you know that right?
@Eden of the East yes I know; I thought I made it clear that my point was new & upcoming hardware is just going to keep getting hotter & hotter, requiring larger & larger heatsinks & fans, OR peoplecould wake up to the common sense solutionof water cooling. Sorry if I was unclear. :)
it's crazy how many cores there and how fast they can make them
Can we just appreciate the amount of cores amd puts in their cpus
Its like BMW with their cylinders. The more the better 🤣
@@ArniesTech BMW? Most are 4 or 6 cylinders.
@@LtdJorge compares to other German cars BMWs golden era (E32, 34, 36, 38, 46, 39, 60) Put more cylinders in their engine.
While most 2L engines were 4 cylinders, BMW went straight 6. While most 3.2 liters were 6 cylinders, BMW went V8. Then there was the E60 M5 with a V10. The 5L 7-Series E38 was a V12. And of course scary Experiments like the E32 7-Series "Goldfisch" with a V16 🤣🤣🤣
@@ArniesTech but more cylinders in a less displacement engine cause the torque lowered..
Jeez, this thing is probably powerful enough to sculpt in Zbrush with sculptris pro mode on 100m tris model and do several textures bakes in same time and load less then 50%. Dat raw power
Haven't seen any setus builds for heavy streaming. Including all the equipment needed for it + software and super fast "where to begin" :)
Can you guys start to do audio processing and DAW render/ live audio processing latency benchmarks too? I feel like it would be helpful for lots of people looking to build a music production focused machine
Even if i got this one for free, I'd sell it again.
Sweet chip, but i couldn't come up with enough things to do at the same time on my computer for this CPU to be used anywhere close to a way that makes it shine.
And then buy a new PCIe 5.0 AMD PC lol
My family's on a farm, and I feel like getting an LTT screwdriver is similar to getting high speed internet - constantly led on by the phrase "should be available very soon!"
I used to run used workstations with gpus shoved in them for my gaming rigs. Never had more stable systems. High grade psus, ecc memory, heavy built cases, 6 core 12 thread xeons when my friends had 4 core 4 thread i5s, solid caps 90% of the time.
I can't imagine how AMD Genoa-X would be like 😵💫🤯
Linus remains a brillant host for the videos on LTT. I hope he never stops doing this ^_^
I just ordered mine, thanks for showing me this sweet, sweet techno-candy!
I have an older Ryzen 7 3700X system, with a small overclock on it so that it runs rock stable at 4.06 ghz on all cores and that's fine by me because I've only ever managed to push the cpu while gaming without closing down browsers with multiple tabs open, and a ton of other active program. In such insane moments I occasionally see the cpu go above 70% usage except for the occasional single thread spike that bottoms out while something takes a turn at it in the background. It's insane, so why would I need more, except as bragging rights?
The cpu does what I need it to in abundance, I would like a more powerful gpu, but that will have to wait due to market prices and such bullshit stabilizes. I've found that I'm actually really happy about the choice of cpu, for the first time in a long while.
Oh, sure I can bottom out the cpu by doing transcoding or video encoding, but even now with 3 browsers running and all of them used more than 10 tabs, several background apps running, a video on pause since I'm watching this video, two audiobooks on pause and a few ebooks just because I like to read and the danged cpu is puttering along at 2% use. What I have up now is my everyday run while I'm goofing about online or relaxing.
Memory though and ram is lacking even with a 32 GB setup, but who could have figured my need changing this much, eh?
Love to see such insane tech as you show in the video, it makes it more fun to wait and see what the new launches are going to offer us, it might be lies or it might be cool or even drool worthy, we will know by next month and your video makes the wait easier. Thanks
It's crazy to me that Gigabytes worth of cache is coming closer and closer in the server space
This CPU has more L3 cache than my first PC twenty years ago had RAM
Insane
1:06 Maaaan i'm applauding you ! The way you put the sponsor is so clever 😂
Tricep flex on 6:08 was peak! Lukin good linus!!!
Maan... And I remember when Linus was going insane about the e5-v3 18 core Xeons.
Technology evolves so goddamn fast!
real? -thhardglump
Just wait to see Zen 4 based TR with 3D-cache too and DDR5 along even more cores. It will be a monstrosity.
I wonder what TDP will be though.
@@michaelsemyanovsky9638 Well it's not like even Threadripper 3xxx likes Air cooling that much either.
I cant help but absolutely love Flagship models but, for mow im still happy with my 2990wx its amazing tbh. Thank you Linus and Lisa!
Those workstation specs are the same specs a very very big database server a few years ago used to have. Except for a lot more money. So for running a database or fileserver it's actually very cheap.
I just love how excited you always get about new stuff :D
If he doesnt then he cant expect anyone else to
His excitement is so infectious, always love it
"Ridonculous" - Linus 2022
I still remember the Abit B6 (?) board with dual sockets... I'd really love to one day own a dual socket PC, just for the bucket list for it ... :-)
Abit BP6 dual Celeron FTW! I overclocked my two 366 MHz Celerons to 550 MHz!
Lifescience and engineering have lots of applications that can use a "infinite" number of threads, i guess you could run complex database searches on that type of hardware and datacenters could host host tons of homepages or cloudservices on these these machines.
I would very much like to see a new line of "enthusiast grade" chips on a modern DDR5 platform. I'm still hanging onto my 6950x since there's no real reason to upgrade, and I get that these new CPUs are more powerful than anything on the X99 or even X299 platform, but from an enthusiast perspective, something that supports quad channel memory and a whack-ton of SATA/ PCI-E expansion on a modern architecture would be very cool to see. Don't get me wrong, I'm thrilled with how far CPUs have come since X99, and I'm very excited for the upcoming Ryzen 7000/ Raptor Lake launches, but I miss the days of having a more feature rich platform to go alongside the consumer grade ones, not counting the server/ enterprise level stuff. Seeing consumer Threadripper get canned was very disheartening.
i'm using a "new DDR5" platform for more than a year with Intel already....
@@s.i.m.c.a When I say new, I mean relative to enthusiast platforms. The last consumer grade Threadripper was released on DDR4, and it's the same story with X299
These days server/enterprise equipment is much cheaper than a workstation. No org with any IT/budgetary sense would allow workstations to be purchased over Epyc servers.
the only quad channel consumer chip i can remember is some old intel
maybe around 4th gen
@@LeLe-pm2pr X299 was the last from intel I think, and that woulda been 7th gen. I have no personal experience with X299, but I can at least attest to X99 (5th/ 6th gen) having quad channel. Either way, it's certainly been a few years 😂
My favorite part of this was actually learning about Gramerly (the sponsor spot at very end).
I can sometimes spend close to an hour editing and re editing 50 times an important email before sending it, because i try to make it as concise and clear as possible. I'm going to try it out.
Thanks Linus!
Would be interesting to see the results if someone could benchmark Mixed Integer Programs on this.
Should probably point out some of the Epyc chips are workload optimized, the 7763 is optimized for anlytics, erm/scm and high density vm's
love how amd was so interested in profiting I didn't even know about the chip until now
Youre not their demographic then.
The problem is even at this price they can't make them fast enough. There's a huge backlog of orders for TR Pro, so why would they cannibalize that market to make a lower end SKU?
I could see it if they had another fab capacity to saturate that market and then some, but it's a bit more nuanced than they just want more money.
If you don't know about it, you're not their target demographic and weren't ever going to buy it anyway. And AMD doesn't need to do any marketing at all for these, they sell like hotcakes anyway
This machine is made for calculating mold flow simulation? Nice follow up after the LTT screwdriver video at the injection molding machine!
We use these at Makinarium, Pinewood studios UK in our workstations for 3D work in the film industry. They cost a lot but our work in demanding.
LTT is a staple now and in the 90's they would have had their own show on cable tv
There should be a Core Grouping feature that groups all of these cores to 32 threads so that it can also rip in single core
There is SOME grouping in the speed of getting the cached data from the same chiplet vs a different chiplet.
Like pixel binning on camera sensor
Super late to the party but IBM has been doing that for a while in mainframes. The issue you run into is latency. Intel I think is also looking into it for their diamond rapids server parts.
@@jonathanjones7751 If they're all on the same die you just group the ones that are closer to eachother; in fact apple has been doing similar things with the m2 ultra, even though the threads aren't united, there's 2 cpus joined and acting as one big cpu
Had to laugh at the fact that they suggested heading over to L1T to buy some merch; I literally bought some from them the other day. Shipping was a bitch, but after adding a few other things to bring the difference between the month shipping time and '7-10 day' shipping. It arrived from the other side of the world in 3 days (4 if you take the fact that the order was placed the day before in the US, but just keeping single time zone I think it was 3). Much respect for Wendell and the team over there, both for the content and the merch. Hopefully they get the next generation of the DP Repeater + Splitter soon; 1440p144 on DP1.2 is good, but 1440p240 on DP1.4 would be awesome, and while I just bought the 144Hz one, I wouldn't hesitate to buy the DP1.4 version to get that extra boost...and buy even more merch to get that fast shipping price difference down ;)
We have one where I work. We use it to slice for out 3d metal printer using velo flow software.
I wish my company understood the long term costs of giving their engineers slow computers. I had to wait six months to get an extra stick of RAM for my previous workstation so that it could run all the corporate spyware without having to swap everything to the pagefile on the spinning hard disk. They easily spent 5 figures on me sitting waiting for things to become responsive in that time, and I was far from the only one in that boat.
This. Our development teams have largely become an Intel wasteland in favour of Apple, because for what we do the M1 slays the shit corporate laptops the business buys.
The IT guy lamented to me the other day as we picked up yet another M1 Mac for a Dev, I said to him the minute we get a Ryzen 7 or 12th gen performance oriented laptop for engineers I'll be right back on that bandwagon.
There are a couple of contractors in the team that have the current Dell alternative and the amount of time they spend waiting to be able to run compiles, multiple docker containers for DBs and servers, 100 chrome tabs, etc. They sound like they're taking off, while the M1 rips through most of it AND gets legit all day battery life.
@@morosis82 My m1 macbook pro concurs with your experience. x86 has yet to meaningfully catch up.
Yes, I want this.
For which Tasks do you need this Power?
@@gabomasterbp8170 Solitaire
@@kevinmorgan7085 understandable
You are right, i absolutely want this, yet i am watching most of your uploads on my 2010 windows 7 laptop
For CPU reviews, can you do a benchmark that compiles a large amount of code? I think it would be good to know for developers looking for the right hardware.
That's the Firefox compile that they have as a benchmarks already in the video
Gonna become the fastest graphics designer in the west with this thing 😎
more TOKENS. MORE TOKENS
I'm curious if this CPU is fast enough to make rendering animations on GPU pointless. With that many cores I'm curious if this thing can render a blender animation faster than a high-end GPU
@@marcogenovesi8570 Right but those cores function entirely differently. 4 CPU cores and four graphics card cores are not equal
I looked at this, but I don't plan to move to threadripper until it's on the Zen4 platform. The 5995WX is way more than I need, but it's neat.
At about 8:40 he said that the screwdriver was coming soon, but it's already up for sale in case anyone wants to order and get in line.
AMD is still cheaper per core than Intel. They are selling for about $100/core which is a steal for rendering.
"I DISARGEE're -Ministry of Magic: Department of Mysteries" -thhardglump
While I am AMD fan, I am impressed by intels top consumer tech being just a factor 4 slower than the best performing extreme die size server cpu on the market.
That's... not something to be impressed by
Intel is faster per core. That could matter more. Not all jobs can be shared to lots of cores.
I'm going to be using this for rendering mandelbrot zooms.
I like the threadrippers, but the problem 1 with them is that their existence is the reason for the 128GB RAM cap on the normal desktop AM4/5 CPUs by the marketing division. The problem 2 is that their release lags much after the desktop CPUs: the very recent 5995X uses Zen3 and we're expecting Zen4 this month to make obsolete Zen3. Shouldn't premium pricing give you the latest arch before the lower end gets it ? I agree that it competes with GPUs as the core count approaches that of the GPUs so it is interesting how computational frameworks like pyTorch behave on such CPU compared to GPU.
I honestly thought that the CPU would be way more expensive than that.
He's underselling it a bit. They're retailing for 6500USD, assuming you can find stock of them.
Those clock reductions with many cores in use, oof. As ridiculously expensive as these things are, and being that the customers are almost exclusively businesses that have to justify every cent, I'm surprised CPU manufacturers haven't made more of an effort to optimize for *actually properly cooled* use cases to keep those clocks high. Performance is likely about halved vs its theoretical potential (even assuming they can "only" reach 4.5 ghz across all cores), so as long as the cost of a suitable cooler and power delivery is less than a second CPU (and second entire system to put it in...), that should come out ahead financially.
yeah i wish reviewers would show thermal throttling as part of the comparison
To be fair, that 2.7ghz clock is likely a worst-case scenario, with all cores being hit hard, AVX on, and a cooler without massive headroom. You'll probably get better clock speeds than that most of the time
They'd need liquid nitrogen, at 4.5ghz it's pulling 400w and that's not on all core? The heat is too much for most coolers, and not because of radiator limits but the block transfer and IHS. Direct die would help, but pushing much further isn't gonna be "properly cooled" no matter what, unless you are talking single core boost I suppose. As far as I can tell, I don't have a chip on hand myself. It's already pushing the limit, the difference is just how you want to when OCing it seems.
They have started doing that for some EPYC being used in water cooled datacenter applications, but I don't think they're off the shelf SKUs
They actually are optimized, and designed to scale with cooling capacity. That's what PBO offers. At some point you'd be expecting too much from 64 cores in a single socket.
I'll get excited about this when it has more than 2 threads per core. Base performance so far is still looking good though. Thank you.
I do machine learning and would make use of every little core of this beast 🔥
Linus : "Let's talk about who and why and about our sponsor...... "
Me : give this guy a medal for that
LTT's sponsor segways are just the best!
After watching this video, I somehow got reminded of AMD and Nvidia GPU launch and how their GPUs only start selling after a month or so after the event. It would be so good if every GPU is released at the same time and is available just a week after their event or maybe even days.
My gripe with these benchmark scores is they don't take into account the actual demographic for multi-threaded systems like this. For Example Adobe's Creative Cloud is an absolute abominable suite for testing cores, and while Blender is closer to the demographic you should really be using V-Ray (Non-GPU), PhoenixFD, Render Man, Octane, Houdini, DaVinci Resolve and other Top Tier Pro packages. The only reason "Adobe" is though of as a "Standard" is because of its (Very smart) position in the Market, but (we) professional Animators, VFX, Simulation Artists, use the Adobe packages to a limited extent. (AE and Premier and notoriously buggy.)
This core is meant for one thing and one thing only. Raw (non GPU) Computational Power.
So, I work at a AAA game development studio. We're in the process of switching people to Threadripper PROs (not this one). It's saving us time on our recompiles and map building. It's more than enough to cover the cost, especially considering I had just gotten a new computer already this year.
I have watched this man's videos enough to recognize a segue to his sponsor coming a few seconds ahead.
It would have been nice to see a comparison between this and the Threadripper 3000 equivalent.
It isn't an apples to apples comparison otherwise. I guess there is a reason this wasn't included.
If you want this comparison, Hardware Unboxed did a video featuring both the Threadripper 3000 and 5000 64-core models a few weeks ago.
If you want to compare the threadripper pro 3000 to threadripper pro 5000, then it's basically a 20% on average performance increase in single-thread and multi-thread.
Watching the title of this video change over the past few days was more exciting than the actual video.
"If you think the price is an issue, you're probably not the target market"
Well, if I win the lottery I'll buy one to edit my TH-cam videos on how to waste your money as a lottery winner. If I'm lucky my TH-cam will stave off bankruptcy for a few extra years.