AMD Threadripper 3990X CPU Review | Code Compile, Render, & Compression

We threw together a 3990X review in the last few hours before leaving for factory tours in Taiwan. We've got more numbers we want to publish later, but started with this.
Sponsor: Buy Thermal Grizzly Conductonaut on Amazon (geni.us/1D8Eeq) or Hydronaut paste (geni.us/Fsray)
Find the Threadripper 3990X on Amazon: geni.us/xvKS5rE
Threadripper 3970X on Amazon: geni.us/uB7SHnO
AMD R9 3950X on Amazon: geni.us/NAE6f
Intel 9900K on Amazon: geni.us/qsJO
Won't have the article up for this one due to active travel in Taiwan for the factory tours. In this video, we're reviewing the AMD Threadripper 3990X CPU vs. the 3970X, 3960X, Intel Xeon W-3175X, Intel 10980XE, and more. Benchmarks include programming/code compile of Chromium, Adobe Premiere, Blender, V-Ray, Adobe Photoshop, Compression, and Decompression. We also have additional benchmarks for gaming and power, but didn't include them for sake of timing. We'll try to produce additional content around this CPU later.
The best way to support our work is through our store: store.gamersnexus.net/
Like our content? Please consider becoming our Patron to support us: / gamersnexus
** Please like, comment, and subscribe for more! **
Links to Amazon and Newegg are typically monetized on our channel (affiliate links) and may return a commission of sales to us from the retailer. This is unrelated to the product manufacturer. Any advertisements or sponsorships are disclosed within the video ("this video is brought to you by") and above the fold in the description. We do not ever produce paid content or "sponsored content" (meaning that the content is our idea and is not funded externally aside from whatever ad placement is in the beginning) and we do not ever charge manufacturers for coverage.
Follow us in these locations for more gaming and hardware updates:
t: / gamersnexus
f: / gamersnexus
w: www.gamersnexus.net/
Editorial, Testing: Steve Burke
Testing: Patrick Lathan
Video: Andrew Coleman, Keegan Gallick

[@GamersNexus]

We have a few other tests we still want to do with this CPU, but those will have to wait until we get home. We threw this one together just before leaving for Taiwan (in the final hours!) for factory tours, but it's still pretty complete. New code compile stuff will be fun as we expand it. If you want to watch us lap/sand a Threadripper 3970X CPU, check out this video: kzbin.info/www/bejne/bYjZdWCVd7iDptU

[@teaser6089]

I bought the Australia T-Shirt, I received it last Thursday(I'm from the Netherlands) And I am very pleased! It's good quality and it is very comfy to wear!

[@brettanderson6743]

Watch out for the human malware!

[@BeardedHardware]

Whats up slackers!!!! Need a lap job?

[@anonymous_coward]

If when you come back from Taiwan you lose your voice, please have some one else read the script.

[@lucidnonsense942]

You really need to start doing dual tasks - eg. Premiere and Blender renders at the same time. That's a more realistic scenario for someone looking for a 64 core chip - 2 virtualized workstations, or a situation where you have multiple tasks going on at the same time.

[@junkerzn7312]

We do bulk compiles of open-source applications. This isn't just compiling chromium or blender or open-office, this is compiling EVERYTHING in the open-source world to generate binary ports for distribution. Over 100 project slots compile in parallel at any given moment, each one typically -j 24, for 30,000+ applications (5000 of which are huge, like chromium for example, which alone has 30000 source files). Millions of source files in total. Our 2990WX could do the x86 run in about, oh, 16 hours or so. We've been burning in a 3990X and it does the same workload in half the time.. 8 hours, in roughly the same power envelope (only roughly 30W more). We have the 3990X capped at around 300W at the wall for our work, on air of course (its kinda stupid to overclock it and equally dumb to water-cool a box that has to run flawlessly 24x7 for years. Water is fine for enthusiasts... not so fine for real continuous workloads). The word insane doesn't even begin to describe the performance. Looking at the statistics, the kernel is allocating and freeing roughly 10 GBytes/sec of anonymous memory on behalf of the build (make, sh, c++ compiles mostly). Continuously for 8 hours straight. System call rate peaks at over 40 million system calls per second and average in the 3-4M/sec range. VM page faults are also in the millions per second, continuously for 8 hours. The CPU load exceeds 400 during the build and the system is still responsive under those conditions. Of course, this is NOT running windows which would fall over under these conditions. To use the full concurrency of 128 threads doing c++ compiles you need around 2 GBytes of memory per thread... 256GB total. Even 64GB isn't even remotely enough memory (maybe for your benchmark, but not for any real-world bulk compile). And certainly not less... that would be insane. The last thing you want to have happen is to force the system to page heavily while running concurrent compiles, nor do you want the system to be forced to constantly re-IO the same data files over and over again (project dependencies) because it doesn't have enough memory to cache most of it. -Matt

[@TheShoelaceExpress]

For the workload you described above, would you have seen meaningful gains going to an Epyc CPU in order to get more than 256GB into the system? I know the Epyc "equivalent" is much more expensive, not even talking about how much the RAM would end up costing on it's own, just a thought exercise I'm wondering if you explored.

[@junkerzn7312]

@@TheShoelaceExpress The workload does not scale with memory once you have 'enough'. You just need to have sufficient memory so that it can utilize all 128 threads doing concurrent c++ compiles without having to page to swap much or do excessive amounts of normal filesystem disk I/O during the bulk run. So a single-socket EPYC with more memory would not be any faster. Also, I kinda wonder what GN's setup was... if they were just compiling one application (like blender for example), they were probably doing it on a normal filesystem and not on a TMPFS filesystem. The big bulk compiles done on Linux and BSD systems all use TMPFS to stage the work. Because its kinda dumb to waste cycles writing (and reading) temporary object files to a storage-based filesystem that you are only going to wipe when the build is done. Only the source dist is read and only the final binary distfile is written out to a normal storage-based filesystem. So in terms of c++ compiles, 256GB memory is fine. In terms of other industrial workloads... maybe not. For the concurrent compiles you could probably get away with 128GB but its pushing it. 1GB per concurrent C++ compile is not a whole lot of memory (all the memory is shared of course, but the aggregate memory load is still up against the wall and will force some paging). GN's workload is not really reflective of the reality... they are just compiling one project, not tens of thousands of projects. So it might seem that it fits in a smaller memory footprint but that would only be true for that particular project and not for a real multi-project bulk compile load. Of course, a dual-EPYC ... oooh, 128-core/256-threads. Now THAT would be faster. It would probably be able to do the bulk in 4 hours. But it is also way out of our budget (we're talking 3x to 4x as expensive as a 3990X rig with 256GB of ram). We only have the funds for one 3990X :-), which we use for the fast-turnaround bulks. Because the open-source application set is constantly undergoing updates (literally hundreds of updates a day across the entire set), projects have to run their bulks continuously to find and debug problems. Once the issue is resolved maintenance builds can be shuffled off to slower machines. In our case, two 3900X's (each with 64G of ram and can do a bulk in 26 hours), and a dual-socket Xeon (which has 128G of ram and can do a bulk in around 31 hours). In terms of price efficiency, its really hard to beat a 3900X or 3950X (as long as you don't mind the extra build time). But having that fast-turnaround with the 3990X is worth its weight in gold for developers. -Matt

[@ashm3406]

It can’t even game though 😎 Intel is gaming KING! AMD isn’t for gaming or anything really you’d be far better off with an Intel CPU, only a poor person would ever buy AMD. All those cores yet it can’t even come close to the 9900KS GODLIKE performance epic gaming. AMD peasants are bowing before Intel master race as always

[@Verpal]

@@ashm3406 I hope you are trolling because not even Intel fan can be this dumb.... right?

[@TheShoelaceExpress]

@@junkerzn7312 Got it! Wasn't sure if you were 256gb was 'enough' between the described workload and the compile benchmark in the video illustrating that 32gb wasn't enough for a single chromium compile. I've got no experience and not much knowledge on code compilation, so I appreciate the quick and informative response!

[@chandlerreynolds40]

Gonna get one of these so my dad can play solitaire

[@teaser6089]

He could play 10.000+ Solitaires at ones!

[@MafiaboysWorld]

Imagine the size of the minesweeper game you could make. 🤔

[@jasonlisonbee]

@@MafiaboysWorld You'll destroy it after nearly finishing such a level and losing.

[@MafiaboysWorld]

@@jasonlisonbee I wouldn't make it past square two. It would be first click and I see 8. 😬

[@jasonlisonbee]

@@MafiaboysWorld At some point you might be bored enough to get good at it. It would take months to get through the largest size it could handle. If you get that far and lose you might be too po'd to remember how much you paid for the hardware until after you wreck it.

[@Weaver_Games]

I like how the Cute Pet case has just become a permanent arm rest for Steve lol

[@insomnyuk]

That's the most important development here, really.

[@samtime2711]

Now if you could get case in custsom colors, I know a few cat lovers out there who would be game for case if only in a different color.

[@TheKazragore]

The fact that this single desktop CPU approached the render performance of dual Tesla V100 compute cards is insane.

[@yeetusfetus8687]

Wait what?!?

[@cold9574]

what is the most powerful cpu i can buy right now

[@Gusto3791]

Even something like a 5% scaling over the course of a single project could justify the hardware price increase. Considering I bill anywhere from $100-$300 an hour as a contract, so average of $200, with 2080 baseline work hours in a year, the increase in cost would justify itself in about 6 months.

[@GamersNexus]

Assuming it gets more work in the door, yes, definitely. As long as you have more work piling up than your computer allows you to complete, it makes sense as a professional.

[@Mike_Hogsheart]

@@GamersNexus I mean, even if it means you just get to finish early and go home to enjoy life it might be worth it. $4k really isn't that much of an expense for professional work, so arguing for a slower part at a lower price would seem hard to justify for those use cases.

[@Alexlfm]

Vally123 In most workload situations this is somewhat doubtful. Even for a compile, you typically are not compiling that many times a day where that small time difference would make that much of a difference because you’re background compiling. From a graphics perspective this CPU is way overkill so the only real application is going to be very specific compression tasks and clean compiles. Programmers typically don’t do that many clean compiles a day for this to be worth the cost difference, IMO. Besides you can always run clean compiles when you aren’t there. It’s not like you have to do them to move onto the next step like photo/video post production work.

[@excitedbox5705]

Those people that truly need this CPU will get more than 5% scaling out of it. Above someone wrote about doing huge concurrent compiles where they max one of these 24/7 with 256GB of RAM and get a 3.25x improvement over the 3900x (8 instead of 26.hours). That CPU will pay for itself many times over for those that need it. This compile job was also done on WIN with 1/4 the RAM it should have had and no optimizations to the file system or compile job. Or even the I/O. Even a Ferrari is wasted on the Highway if you drive it in 2nd gear. Same with a 18 Wheeler that is only half loaded. The problem is with the Benchmarking not the CPU.

[@Alexlfm]

excited box Those are VERY specific use cases and not the norm for developers by a long shot. Those are clean compiles IE building everything from scratch. This is done for distribution not typically very often during development where you are only doing one relatively small piece at a time. In his case absolutely it makes a difference but that is not typical. Even for a clean build, I tend to do these at the end of the day or at lunch so time saved is not really important so long as it is done by the time I get back (which the cheaper parts accomplish). Some users will find value, of course, my point is that most people are not likely to either A, actually book more jobs (as Steve mentioned) with that faster time since that’s not the real limiting factor there and or B, have it give them more personal time owing to the fact that full builds are generally done unsupervised. Clean compiles are nice but not required and writing code isn’t like GFX where you have to wait for it to finish to continue work. With many IDEs code compiles in the background as well further reducing the requirement for this. The value proposition for an average development shop isn’t there, unless it’s handling multiple full compiles for extremely large projects.

[@Helios.vfx.]

Do simulations tests, fluids and so on with Houdini FX please!

[@theeskimo9875]

agree

[@Helios.vfx.]

@@theeskimo9875 Yeah that must be amazing to see how that processor handles all those things.

[@sahdikann5947]

Fluid dynamics modeling. oh to have one of these when I was a graduate student.

[@Helios.vfx.]

@@sahdikann5947 I think for that, you'd need a Ryzen 9 3950x, it's more than enough. Or its equivalent in the moment you'd buy a processor. It supports 128gb ram.

[@montazownianr1]

Starmaker There is no one cpu that is enough for this. And event in next 10-20 years one cpu will be bullshit for fluid sim and particle sim. The only hope is in graphite cpus up to 400ghz.

[@michaelgerrard8878]

One of the few tech tubers that get that much in depth with 0 bias

[@clansome]

0:26 "Tough to benchmark. At $4,000 it's an uncommon purchase for most of our audience." Steve, I think you might mean ALL of our audience.

[@Triplechomending]

that audience is going to find steve now tho

[@fuckoff565]

Not all.

[@erichall090909]

There are people who would and should buy this. And I’m sure a couple of them watch this

[@heliosfa]

Not all of the audience. We just bought a 3970x based system for some research. The only reason it was not a 3990x is that none of our suppliers could get it in reasonable time...

[@Zarcondeegrissom]

if I had the funds, I would have the 3990x already, 3delight and RenderMan would love that 64c CPU as can be clearly seen on the GN monkey head tests. My electric bill probably wouldn't like me as much with that CPU, however that is another matter. lol. Agreed, Gamers that watch Gamers Nexus, probably shouldn't waste their money on the 3990x or the 10980xe or W-3175X for a gaming computer.

[@tgdxdrakoon7019]

Well done and informed video. The rap up at the end, will help a lot of people choosing savely what to buy. Thanks for your passion

[@inkysteve]

I think it was Wendell who lent you the chip.

[@olo398]

yea, i cant imagine who else it would be.

[@programmier]

why cant they tell who sent it? Is there a reason?

[@materialburst983]

@@programmier amd mad

[@andrewb6]

BPS or Bearded Hardware

[@lucasgillard-goode1947]

I reckon it’s jays

[@bonuspretzel3347]

07:45 "...44 more"

[@nathanc4504]

been waiting a while for this vid!! Great vid Steve!!

[@drizztcat1]

I love that the kitty cat case has been on the workbench for the last few vids.

[@chocopie099]

It's a nice review, I own 2990wx a and a 3990x, and usually work with 2970wx and 3970x as well. and there is something in this review that you didn't mention. There is apps that scale better and can use the full potential of the 3990x like Vray and Blender, and then there is the ones that won't scale that nicely because there is just SSD bottleneck or any other bottleneck due to software like Premiere and so ... BUT: I think here there is another very important thing to mention. There is times where the bottleneck is the windows Scheduler. ( yeah I know you will do a Linux review ) but i'm talking about windows to windows here ( and yes I know they kinda patched windows pro, and also the statements of AMD about it ) but windows pro separate the threads in two groups in some applications we are noticing the CPU just uses at 100% power only, and exactly, half of the core counts. Yes this would be a "software bottleneck" but it's the SO so I think is worth mention, Looks like windows pro separate the cores in 2 groups and that the scheduler fails terribly into send jobs to both of them ( unless the software is properly set to do so like Vray or Blender) Looks like windows pro Enterprise/Workstation doesn't separate the threads in two groups so this specifics apps that are not programmed exactly to use all the cores, and that rely on the windows schedulers could get benefit from this... again, I can stress my CPU max power but just on half of it. If i could get the other half I would double the power. For instance I can open another instance of the same program, run it and will utilise the other half of the cores available and works like charm also. But simetimes you want more power in one instance of the program, not to run 2 instance of the program I think this would be worth testing cos that means 1, the scheduler of windows pro is shit. 2 if you own a 3990x and this enterprise version just can run soft in all cares and not like the windows pro then is worth the upgrade to windows enterprise/workstation. Sadly.. very little information about this topic...

[@Nothinglastsforever]

>Sadly.. very little information about this topic... What do you mean? Anandtech covered this extensively.

[@chocopie099]

@@Nothinglastsforever yes and no, Anandtech exposed the problem then AMD deny it... and we are here. Unless Anandtech said something else that im not aware of.

[@Nothinglastsforever]

@@chocopie099 I did not know AMD denied it. Why would AMD deny it when they can blame it on Microsoft? And not just Microsoft, but the applications that don't know how to prioritise separate CPUs.

[@drescherjm]

As a software engineer I am happy to see compile benchmarks. I have spent so many hours of my life waiting for large code bases to build.

[@Noman1000]

EVGA making an AMD motherboard, never thought I'd see the day lol.

[@denanhajrovic4183]

Really?

[@inferno72229]

They will go bankrupt if they only supporting shintel

[@Raptr121]

EVGA originally made boards for AMD.... But where in the video did Steve say they are making AMD boards again?

[@Noman1000]

@@Raptr121 Sorry I misinterpreted the SR-3 shown as a TRX motherboard. Doing two things at once when watching the video originally lol didn't know they made AMD motherboards, must have been way way back because I started with PC hardware around when the SR2 came out and I don't recall anything AMD since then.

[@Raptr121]

@@Noman1000 no worries, just how the internet is it's hard to tell if someone just got confused or they're a troll. Lol

[@nO_d3N1AL]

Good on you for mentioning Phoronix. Michael does great work on testing Linux distros for performance with recent hardware.

[@whiskeymike7364]

Great review, some fascinating results. Solidified the 3950X for me as a great choice if AMDing. Now to wait a bit more for prices to drop.

[@johnpaulbacon8320]

Thank's for this very informative video.

[@CR055H41RZ]

If anyone wasn't in on it already: it was Linus who sent the CPU, he just probably can't say because they got their's maybe not through AMD

[@BeardedHardware]

CR055H41RZ so sure?

[@icosaphilia6342]

CR055H41RZ I’d be tempted to agree given the recent compensator build on LTT

[@CrowCZ]

I'm glad that you are doing compile benchmarks. I'm a software developer and every time I'm in market for new CPU I try to look for compile benchmarks, but nobody does them...

[@GeoffSeeley]

@8:00 Nice Picasso plot GN! Beauty from chaos!

[@Zosu22]

AMD pushing testing methodology forward lol.

[@excitedbox5705]

no just pushing youtubers past the Best Buy sales rep level of knowledge they have.

[@Dennzer1]

Why is that funny? I need to know. Please tell me/

[@MarioCRO]

Great video and analysis. Even though these CPUs are best suited for corporate (Cloud computing) environments, where virtualization is the key, for this they offer great value and support through PCI-E lanes, cores, core cache, thermal and power consumption and also value in terms cores vs price. Thanks GN for these benchmarks since they give a lot of insight when compared to regular home or HDT processors.

[@suzukaze8461]

I could hardly afford my 3600 and yet I'm still watching this whole video knowing I'll never have this.

[@mattsmechanicalssi5833]

Good thing to know that Wendell's 3990X survived the LN2 session!

[@handlewithart]

It's definitely a beneficial CPU when you a lot of tile base renderings as you said as I do on Vray and Arnold! It's a must-have!!!

[@SeanLynchXY]

Cute Pet needs a 3990 treat!

[@d00dEEE]

Yah, but the 3990 alone is as big as the Cute Pet's mATX tray...

[@thomasdarby6084]

I hope you do get to go to Taiwan. A lot of people are being scared away and some events cancelled, especially in Asia. Good luck to you.

[@the360gaming5]

Amazing video

[@oraz.]

This is awesome. So technical. I assume the compiler is MSVC?

[@KenetorTV]

I know you guys might not see this but something to consider for rendering benchmarks is bucket/tile size. too big and you dont take enough advantage of many cores, too little and you create alot of bucket overhead, you might want to do some testing around this and find a sweet spot for your new methodology. Also I know blender is free but its not exactly the most robust 3D app on the market, testing in 3ds max with vray might prove to use the resources better. The stand alone benchmark is a bit outdated at this point. Keep up the good work guys,!

[@Apollo-Computers]

All I said when I seen it delidded, PRETTY

[@Mattia_98]

I'd wish Phoronix made KZbin videos. Would you consider doing a collab with him? Make sure to have some Bavarian beer at hand ;)

[@kalmtraveler]

Glad you covered this - I got a lot of flack for not upgrading my HEDT to a 3970x (stuck with the 'budget' option of a 10980xe because it is faster in anything that doesn't use more than 36 threads which is 100% of anything I use a computer for). 3990x would have just been even worse for me.

[@Collin0]

lol at the 64 line graph 7:48

[@michelvanbriemen3459]

This is the cutest 3990X review ever made

[@tomhsia4354]

I'm only 16 seconds late this time, never been this early to any video before. This CPU is an absolute monster, if your workload benefits from a boatload of cores. Efficient, too. Can't wait for 7nm EUV to come out.

[@gudenau]

When can someone do a direct die thing with this? That's sounds like a crazy setup since this has several dies.

[@UndercoverFerret404]

Is the GN logo blender test available somewhere so I can run it myself?

[@LordAthos18]

Hey, amazing video! Thanks for the help! And thanks so much for the rich details. I would like to ask for some advice that I could learn more about the topic. I am trying to a new computer with this CPU. Could you give some advice having the following context? I hope you have the patience to read it. Lol. -> Doubt Context My actual computer does not turn anymore. I have it for around 8-9 years. So, I am trying to build a computer for the first time. I am a software engineer who will start to learn how to build games with the Unreal Engine. Since the code until the art using Blender and other heavy graphic tools that I may found to create the best game and game art that I can. I do not change computer parts if it is not really needed. Therefore, I am willing to get this 3990X CPU even though it may be extra for doing this gaming development. In the video, it was focused on being as economical as possible. However, I would like to spend a little more on this computer build. -> Parts selected - CPU: AMD Threadripper 3990X CPU - Motherboard: Asus ROG ZENITH II EXTREME ALPHA -> Doubts if it is compatible - GPU: Question!!!! Would this CPU and Motherboard work with the new Nvidia RTX 3090? - Memory: I would like to get 2x 32 memories initially. Aiming one day to achieve 256 GB. What is the fastest Ram that 3990X can handle? Its specification says DDR4 3200. However, the website "ca.pcpartpicker.com/" which check this peaces compatibility, shows in his search this memory: It says that I could use Corsair Vengeance LPX 64 GB (2 x 32 GB) DDR4-4000 CL18 Memory. I know that memories can be overclocked. In the case that this computer can hold this DDR4-4000. Would it be able to be overclocked? - Case - How would I know if a case would be able to fit these computer parts and water-cooling. And maybe a second or third Nvidia RTX 3090. Not that I would have money for that, but would I be able to do it? I want to buy this powerful computer parts. But I will buy initially the minimum. Aiming in some years to add more RAM, maybe another GPU if I think that I needed it. I do not have the intention to build a new powerful computer with a new motherboard and CPU in the following 5-10 years. Unless the computer stop working. Lol. So, thanks so much for your time and attention. Especially patience if you were able to read until this point. Lol. Sincerely, Kelvin Firmino

[@jeffm2787]

I'm still very happy with my 3950X. Bit of overkill for most stuff however. The 3990X, would be good for a Hypervisor.

[@julianangell342]

So basically: it's amazing, but don't buy it

[@julianangell342]

@Im.a.tea.addict Wouldn't they need more than 256gb of ram? The lower clocked single-socket 64 core epic is similarly priced, but supports LRDIMMs and registered ECC ram

[@DuvJones]

@@julianangell342 They didn't... kzbin.info/www/bejne/op2VlJymir-KhrM

[@cm01]

@@julianangell342 yes the 256GB RAM is a bit of a kneecap but it could change with a firmware update at any time. That's basically the whole difference between Threadripper and Epyc, firmware.

[@yummyramen2821]

@@cm01 i think epyc is a better option if you really need all of those cores

[@cm01]

@@yummyramen2821 just depends on whether your workload needs more than 256 gigs or not. And the amount of ram supported could change with a firmware update.

[@fredEVOIX]

tip from someone who has a 3960x, even 24 cores is too much for some games, they stutter, fix: task manager->details->right click on game process->set affinity->select 4-8-16 threads (8 to simulate a 4 cores cpu recommended) select threads at the end, not the 0 nor any of the 1st ones as those are the most used by the OS and generic software, you will have to do this everytime unless you install utilities like PROCESS LASSO bitsum.com/ where you can setup profiles, we already did that with old games that had trouble running more than 2-4 cores when 8c was a brand new HEDT-only feature back in the days, still works as good in the few games I tried, I confirmed this using rivatuner on-screen fps and frametimes graphs like you can see in this unrelated demo kzbin.info/www/bejne/jpqVmYtsdqmXmMk lines will go from "choppy" to smooth after setting affinity

[@pegasusted2504]

Interesting stuff. Youmentioned in a recent video wanting questions for the pro overclockers. It is not so much about the actual aspect of what they are doing and why it is more about the actual physics of the situation. Absolute zero is the abscence of any movement, I think, and the point of overclocking the cpu is to reduce temperatures to possibly increase the performance out of it. I am wondering what the lowest temperature you could get down to before the speed the electricity can move across transistors becomes negatively impacted and stops functioning. Also why liquid nitrogen? Is that because it only gets temps down to the point it can and using something colder like liquid helium goes too far and past the point nitrogen becomes a solid?

[@FreeThePenguins]

Did you try 128gb of ram as well?

[@catalystguitarguy]

I don’t think anyone buying that cpu is running 32Gb of ram. I suspected most who buy it run at least 128GB or maxing it out at 256GB

[@catalystguitarguy]

Erotikstudio Winkler GmbH I guess, but I’ve been running 32GB in my mostly gaming computers for 5 ish years. Just because of multitasking, multi monitor, and generally caching.

[@Amberion]

Good luck in Taiwan! Don't die!

[@Joeteck]

I currently have the 1950x, bought it at launch. The IPC is amazing on the 3960x and 3970x. That said, i use it mainly for video editing and gaming from time to time. $1399 vs $1899. Which would you do? Serious question.

[@hanswurst8317]

As you introduced a surface-levelness Test for CPU coolers already could it be possible to do the same for CPU Heatspreaders?

[@JimboJamboYT]

you could setup 16 x 4c/8t VMs on a server captured by stand alone screens as work station desktops. the possibilities are endless!

[@Cal94]

13:40 [donut media] "is more cores more better?"

[@Timmys_Dent]

MO POWA BABEH!!!

[@gblandrr]

HOW MANY HURSPERS?

[@gblandrr]

LIGHTENING

[@1blackice1]

Those guys must have hit the algorithm lottery lately, because they have been showing up in my recommended and I don't even watch car videos lol

[@Cal94]

@@1blackice1 never dug into their backstory... and maybe i'm paranoid, but i feel like Donut Media is a bunch of youtubers first, or some production company, that was like [you know what market is underrepresented in mainstream? cars]. like for a bunch of people that look old enough to drive, their meme game is too on point. they're too good at what they do. they give me this LMG vibe, like there's a hidden professionalism behind their passion. or a half decent writing team And I'm all for it.

[@TransformX]

OK KZbin, I'll watch it!

[@kommander9638]

Wasnt your last compile-benchmark accidentally fully cache dependent?

[@brianmccullough4578]

Hey, level 1 techs benchmarked it too!

[@jonathanlebon9705]

@16:10 It's crazy to think that something this small can calculate at a rate of 354 BILLION instructions PER SECOND... Mind blowing!

[@AlvaroLR]

~ 20:30 , is that an Renoir APU? Can't wait to see benchmarks of these

[@hardroosterlabs3814]

I'd be interested in seeing how cores scale in Blender with Mantaflow fluid simulations, which is stock in Blender 2.82 and beyond. I've noticed that Mantaflow hits all cores, but doesn't hit them very hard, usually around 40% use in my case, so it would be useful to know how well Mantaflow scales in blender across multi-core CPUs.

[@kevinidleman7048]

Lol Steve loves the Yeston Cute Pet stuff

[@sulphurous2656]

You guys not doing After Effects benchmarks with things like Trapcode Suite?

[@yungdookie6836]

id like to see virtualization as a benchmark im curious to see how many virtual machines you can run with 64 cores

[@happygimp0]

You also need a program that uses the CPU to run in the VM.

[@farzinbi]

You need to start including cfd applications fluent in your benchmark tests, am sure the benefits will be more visible

[@cm01]

Some perspective on the 3990X is that for the same price you can get eight 3900X's, and the 3900X is a monster by itself.

[@Celician83]

I havent seen Wendall from Level 1 Tech's do a full review of this chip, my guess is him

[@Marc_Wolfe]

My first and only guess would be Wendell from Level1Techs. i imagine I would have seen their videos on it by now if he still had it.

[@k34561]

I'm a coder. I bought a 3970X. A 3970X brought my debug compiles down to 50 seconds. Release is only 68 seconds. I doubt a 3990X would help much. Portions of my build are single threaded and portions of my build are multi threaded. Less then 10 seconds of the debug build is now multi threaded. More on the release. Only a very small portion would have a chance of being sped up by more cores. It makes release builds feel almost as fast as debug builds. So the chasing down of Heisenbug might become almost enjoyable. Or at least not a grind. The $2000 saved by buying the 3970X over a 3990X, might be better spent on an $2200 Intel Optane SSD 905P. I might hold off on that purchase and stick with my Intel 750 SSD. Wait for a next generation with a PCIe NVMe 4.0 x4 over the current PCIe NVMe 3.0 x4. It might take awhile, I'm not expecting it before Intel supports NVMe 4.0 on their motherboards. I was early adopter of SSD(Intel x-25s). This current generation of 3D flash looks slower for building. Builds are file intensive with many files. The number is mind boggling. My current build area is over 18GB in size containing over 16,000 files in 1000 directors. Each version of the Qt library is 12GB and 282,0000 files. Add on to that VS 2017. etc. I figure something like 200 GB and 1,000,000+ files. So IOPS are the numbers i look at. The Optane seems to be the first real speed up since stuff like the Intel 750. There are some Samsung pro stuff worth honorable mention. -Kurt

[@NBWDOUGHBOY]

Bro! The Premiere charts just have me in Awe!!! The 3950X Is the closest thing to a HEDT Chip and its still basically holding its own. A 4K60p Render in 7.9 Overclocked and 8.2 at Stock! Thats insane!!! I wonder how long the video is. Either way, Imagine having to render 10 4k60p clips for KZbin or Production on a website. Thats literally Like an hour and half (Actually a little less) for 10 videos! THATS IN-FUCKING-SANE BRO! And to imagine that its only going to get better overtime. Zen 3 coming this Christmas. 5nm over the next two years. Higher IPC, Higher Clocks. Shit is getting real. Its so exciting.

[@asahe8392]

makes wonder if you used two instances of the same program, or 2 vm to run the benchmarks on each would it be as good as a 3970x, or would it run into memory bandwidth limitation

[@tangoseal1]

That cat case and kitten gpu had me cracking the F up

[@aluovideoanimacijainnovime9873]

Could you do some more tests on 3D rendering and simulations?

[@Nothinglastsforever]

3D rendering is predominantly done on GPUs, particularly multiples of them. The preference of most 3d artists is two 2070 supers with NVLink to stack the Vram. Blender is an AMD marketing machine and not nearly as much of an industry-standard as Maya or 3DSmax is and I have no idea why Blender is the standard for reviewers.

[@aluovideoanimacijainnovime9873]

@@Nothinglastsforever Thank you for the reply. The 2070 sounds the best offer price/performance. I guess Blender is free, and the new version is becoming more and more popular. Specially young artist, who would like to develop their skills, Blender offers a lot at no price. Personally, I grew up on Maya and I'm findng it difficult to switch, but I understand why a lot of young students choose Blender. Thnx for the info, stay safe.

[@iansutton5042]

I love watching you guys but I really wanna know if Steve is miserable or that's just his face lol

[@Florian.K.]

You guys HAVE TO take a look at Der8auer's latest truly open loop water cooling video. Imho that thing would be a perfect addition to your 'Cute Pet' system, and it definetly falls in the 'different' category.

[@mmmuck]

Based on 3d rendering, my 3900x was the optimal upper mid range budget choice for my workload.

[@SuperFriendBFG]

Most people who would need this CPU likely already know whether they need it at the moment. These are medium to large sized companies typically. Most are able to perform hardware tests of their own in order to inform future purchases.

[@vistazeng2413]

Every time a cover with 'pink cat case' and 'WANGWANG Graphic card', I click in.

[@alterbr33d]

I wonder if E-on Vue can use all of those cores. It is tiled rendering and designed to render across many computers so I think it might.

[@Napster60]

But can you do all this testing while on hover boards.

[@NoobTube06]

Just curious, why is there tape over the model number? Hope you guys are enjoying Taiwan!

[@LazyBunnyKiera]

Yeah, but can it compile Crysis? Just kidding, Great review, i like that you go into more detail than other reviewers.

[@Zellonous]

It can run it entirely in software mode. I imagine compiling it was way less intensive than running it. XD

[@QOTSAPT]

besides core number the big "thing" with the 3990 is the cache layout and it's impressive tdp.

[@berndeckenfels]

You don't do Server benchmarks, but the multi core CPUs all ask for virtualization, containerisations and multi-user use. That would be hard to benchmark but I think it is what all users of those chips are interested.

[@PookaBot]

Hope they let you back into the country without quarantine after your trip :/

[@hra4242]

Steve how about testing Machine Learning Algorhythms, there are bunch out there who scales extreme with CPU core and a lot more that scales with GPUs. Would be a nice video I guess!

[@Thedrunkautist]

I love how Wonwon has simply become a staple. ALL BOW TO WONWON!!!

[@paulvancyber1979]

the cpu of the future!!!

[@QOTSAPT]

on compression if you are going to compare the threadrippers with the w3175x you might aswell note that epyc doesn't have the same problem.

[@the_socompsp]

you should do benchmarks in different OS's both in linux and windows where possible, im certain we will see performance differences.

[@DavidMiddleton101]

CCD needs to be mentioned more

[@tmilker]

Please add a chart with a blend demo file we can use to compare results to. Video card testing too. I know it's not going to be 1:1 but even being able to ballpark stuff like your 2060 KO results or against an older CPU you don't chart would be nice.

[@mimireich]

Might be useful for MD in miliseconds and beat the Xeon out :D

[@Termiux]

I'm calling it, it was Wendell

[@micaiahflores1592]

I will buy this cpu 🙂 in ten years 🙃

[@evrythingis1]

Would it be crazy to recommend this CPU for running TWO or more instances of any processes that do not utilize 64 cores? Split your video in half and merge afterwords it to render it twice as fast? It would definitely make for an interesting test. Workflow woes would obviously be situational but for the independent contractor it has the potential to seriously increase earnings I would think.

[@Beitouzi]

The best review

[@GreekDude]

I just got a r3 1200 because my old laptop died and I cried a little watching this video.

[@churblefurbles]

You need a thousands of chrome tabs test like linus did :3

[@user-xr3rb6pn9m]

If you remember that test, the main problem was not the CPU load or RAM usage but the limitations of Windows as an OS (it just cannot handle too many processes simultaneously, no matter what hardware is used)

[@allyourfuturebelongstochina]

For the compile test would be interested to know if there's any diffference between 64gb and 128gb of ram

[@sulphurous2656]

The CPU every freelance artist wants and deserves but can never afford any time soon. Maybe in 10 years assuming we don't get stuck on silicon limitations permanently. Until then, we have the 2700/3700X and co. And hope that software developers will finally start to optimize.

[@shakaama]

QUESTION: to general AAA blender renders, do i go threadripper or a 5950x?

[@rRefuseToSelfCensor]

If you dont need the extra pcie lanes or memory then 5950x is gonna give you amazing single core performance. Really depends how much money you want to spend. 5950x can run great on a 170-300 dollar motherboard