Couple other things to consider. (I have a 3960x 24 core) 1) Zbrush loves cores, I can easily sculpt with no lag at all with 100 million polygons. If blender could handle sculpting and booleans with 20 million polys with no lag I would probably never use Zbrush again. But that is a very difficult request to make to the devs and I doubt it will ever happen. 2) You are limited by your Vram on the GPU. This is not the case with CPU you can use your system ram ... so a 3d scene with 16GB of data will not render on most graphics cards but you could render 64GB -128GB 3d scenes on CPU. 3) If you want to do game development. More cores = faster compiling as you are building a scene. 4) And lastly, if you ran those simulations on Realflow 10 the more cores the faster it is. It's just that Blender currently does not make good use of multi cores. But looking at blenders updates over the last year we could have proper multi core usage next week. Or in 3 years. :D
Its not just blender that lags on AMD's cores when simming. Nor is it an IPC/core count issue As was also mentioned in the video, most simulation software top out at 10-30 pct usage on AMD cpus while utilizing 80pct or above on intel. FumeFX, Houdini thinkingparticles, phoenix fd all of them demonstrate the same behavior. VDB solves in Houdini result in frequent crashes on AMD's TRs btw. As for GPU rendering, while limited memory can present a bottleneck but shader optimization on both octane and redshift make things faster. furthermore, the main factor in rendering performance are the CUDA cores. Also, if you are rendering a sequence that's between 64-128gbs, odds are its getting sent to gridmarkets. Also, as a zbrush artist, i am sure you realize the importance of faster ideation and look dev- which is generaly much better when you have gpu rendering. AMD's approach of slapping cores that dont scale up properly isnt any software vendor's fault.
@@sniper1pk It is not AMDs fault, but actually the software vendors: As you can see in the Cycles benchmark, the Threadripper is absolutely A LOT faster than all the other CPUs if all cores can be used. And that depends 100% on the software. Rendering is extremely intensive and uses all cores 100%. The problem is more specific that many sims and display engines or sculpt engines were originally written for single-core in mind and then extended. They are not 100% parallelized in every way (i.e. might use a max of 4 cores or have other bottlenecks due to the code having been extended after the fact) Some physics simulations actually can't be parallelized, but a lot of other stuff can. These engines (for example the sculpt engine) needs a rewrite really badly in Blender. The story is better on ZBrush, as you mentioned, so that engine was rewritten and optimized with many many many cores in mind. So it is purely a software issue really.
Indeed, but I guess he benchmarked what he had. I for one, am extremely happy with my 3900x in blender, though for more serious scenes there is no substitute for GPU
@@iftikharmalik639 new ryzen CPUs are supposed to be releasing this fall. You can wait until then and buy something with more performance brand new for potentially better value or get one of the existing models that'll come down in price due to the new CPUs being out. You can also find more CPUs on the used market once new CPUs are released
Thanks for doing an dive into blender for the creative space for these CPUs. I went with a 3950x but it’s nice to see charts like these to get a sense of single core vs multi core performance. It can be about having a good balance, in my humble opinion.
Thank you for explaining not only why the answer is "It depends", but also what it depends on. It cleared up the misconceptions I had about the newest processors on the market right now.
Any plans to make this video for 2022?? Id love to see some of the newer stuff have a go at Cycles x etc as im looking to upgrading my entire desktop. Either way great video.
I eager to see comparison 8core ryzen 5800x3d and 12core 5900x which are at very similar price here and now. What about testing some heavy modifiers and non destructive modeling?
You helped me a big time! I was deciding between 3950x and 3900x thinking whether it is worth the money and it is not. For rendering I will be using RTX 2080super (eventually will get second one as well) and for processing and physics I will go with the 3900x! I though more cores will equal quicker calculations, but you just showed me it is about frequency not the amount of cores! Thank you so much!
It is really a shame that the 32 cores are never really used fully inside Blender outside of the raytracing rendering aspect. This means that a lot of workflows are permanently limited, they will forever be slow, since we don't get cores that are that much faster (and definitely not on Laptops), i.e. you can't buy a laptop that has 4 cores that run at 8 GHz, no such thing in the world. So making things parallel so they can use many many MANY cores is key and that is currently not done.
super useful and proper comparison! definitely agree that just rendering the typical benchmarks doesn't give a good overview of the process and performance.
Oh yes, they actually do: The rendering and raytracing shows the pure performance of the CPU if software is properly written to be scalable and use all the cores and performance (so to speak, use all the engine power that the CPU has) fully. We know as a matter of fact that we get most performance not from faster cores, but more cores, many more cores. GPUs are so fast, because they have tiny stream processors, but many thousands of them and they all work parallel.
For the majority of people, the answer is simple, these CPUs are stupidly priced, ryzen CPUs with 8 cores and 16 threads like the 1800x are going to be the best thing you can get and will often be around 150€ in the second hand market. Only an extremely small percentage of people will need the higher performance that Threadripper CPUs can provide. Anyway, if you really want good blender performance, don't spend more than the previously mentioned 150 or at most 250€ on a CPU, an RTX GPU will give you far better performance for the price, hopefully the next generation of AMD GPUs will do so too, but right now? the full price of a 2070 and e-cycles is going to give you better performance than the most expensive CPU you can find.
I personally wouldn't recommend a 1st generation Ryzen at this point. Maybe the 1600AF because that's a Zen+ re-release of a Zen CPU. But if you aren't super budget constrained I would go with 3600x or 3700x. The 3800x isn't worth the extra cost. Then 3900x and 3950x if you're a baller and have the money.
@@Sophistry0001 i mean, that's very relative, the 1800x is a very good choice if you can get your hands on it as it will outperform everything in its price range (goes for about 130 to 150€ second hand here), other than that, a 2700 is the next great choice, Zen 2 CPUs are still very pricey and not worth it if you aren't made of money. Funny you'd mention the 3800x as it's barely 30€ more expensive than the 3700x over here, if you're going to spend around 350€ on a CPU might as well go for the 3800X.
@@Knuckles2761 sure but it's also about 50€ more expensive, which if we were talking about the 300€ range wouldn't be such a difference, but from 130-150 to 180-210, that's a pretty hefty price difference
Agree. dual rtx2070s would be sweet spot add on for enthusiast workstation. New tensor flow cores used by denoiser help reduce number of cycle samples without compromising quality, old generation (non RTX cpu) would need to crunch more samples to get comparable quality.
hearing people recomending much older and also used stuff for the same price you can find something new and decent, like the Ryzen 3300X $120, is like saying "dude you can buy a broken TR for $20 and fix it yourself" just ridiculous.
Great video! You recorded this video months ago but I just found it recently (No thanks to RUclips recommendations!) and I should say, you made the best comparison video for the time you spent. Thank you for sharing and Good Luck.
2 года назад
I've watched this video several times since its release, and I've only just noticed the nice subtle audio overdub at 3:48! 🤯 nice job making that sooo seamless!
I agree with this but he didn't mention that your system would be unresponsive when doing those things on the Intel CPU while on the threadripper it would work perfectly because it has spare cores to work on that
No, not necessarily. See the problem with the viewport is that it is not well parallelized, the code was originally not written for multi core CPUs with many many cores. It will only use a few cores, not all of them. "It does note scale properly" is the technical term, since the intel has the few cores it has running at a faster speed, it works faster in the viewport, while the threadripper has probably around 20 cores doing nothing at all. Software has to be written to use massive parallelization. Blender unfortunately is not in many areas of viewport, physics simulation and modeling.
@@Brahvim pretty dang good! It seems like it can definitely handle a lot more than it used too, but now I’m just being held back by my gpu, because it only has 3gb of video memory, (it’s a 780ti) so renders are still a little rough, but I have noticed some speed gains, and that graphics card is next on the upgrade list
@@ittueaday May I ask, do you mean you can get better fps for animation viewport? I'm thinking to upgrade my amd ryzen 3 2200g to amd ryzen 5 3600 just so I can do animation and simulation better.
@@raofriot better FPS will mainly come down to gpu if that’s what you’re running, the cpu can help though and I did notice some speed improvement overall, but it would of been faster if I upgraded gpu, up to you
Having a little experience in programming and software development, I can tell you that unless you optimize a workload to run with multithreading, you certainly will run issues with scaling. Let’s say you have just one core, you load the data and the instructions, you make the calculation and then store the result. If you add a second core, you will have to take into consideration the resources as well as timing. You end up having downtime for example if one core needs the result of the calculation the other core is performing. On the other hand, you want the cores to share as much resources as possible to minimize RAM/cache usage as well as reduce downtime while a data is being loaded into the RAM or cache. Fun fact (but is dated already): in 2016 I watched a similar video about the Adobe Suite. The conclusion was identical that the software ran faster on a few high clock cores then a lot of slow clock cores.
Nice video I’m a new one for your channel After seeing yout tutorials i falling love with blender You’r a wizard in the Blender World Good Luck Bro. 👍🏻
Wow, you got a lot of great comments and great questions, in the comments. It is too bad you didn't answer them. A lot of people posted questions right after the video was uploaded. Viewers were really helpful and answered many of each other's questions with replies to comments. But, i always like it when a creator is present in the comments, with answers to questions and additional information. It was a really good video and interesting topic. thanks.
I wish the 3900x was on the list, I'm struggling to choose between 3900x and 10900k for the viewport performance, unfortunately I can't see benchmarks athor than rendering times.
Linux is well known for handling high multi core CPUs better than windows. I wonder how much of a difference it would make in these tests, for curiosity of nothing else
I tested windows vs linux rendering with my Ryzen 2700X and the performance is the same as long as you close every background app that used any percent of the CPU in Windows. Including the Task Manager itself.
I think there is a missing component to this test; the Ryan Master software...a fair comparison would be to adjust the core count in the software to match the new chip set. less cores, less heat, resulting in a higher boost clock. The TR would still lose, but the numbers would not be so dramatic.
Threadrippers have no memory caping, so can address up to 1TB if there as many udimm slots on mobo or module with large GB units are available and are designed for data centers to compete with Xeon. More interesting would be see comparison with 3900x/3950x Ryzen as those are direct competitors to iCore line and have comparable price. So this was a bit pro-intel review imo.
I love this video! Just bought an i9-9900, specifically for Blender. I'm happy with my purchase, so far. This is exactly what I was looking for to help me decide what to buy. Unfortunately, I bought it a couple of weeks ago. I had to make my decision without this info, but it's nice to know that I would still make the same choice. Time is money, but sometimes, I have time available while money is not. Cheers!
I don't get that takeaway, the viewport performance didn't matter because if you're getting 2 FPS then 3 FPS isn't really an upgrade and the threadripper parts actually did better in the fluid/smoke sims which I do way more often than rigid body sims. And by the way, the 3950x probably eliminates the "gaming" advantage the intel CPUs enjoy whilst still having a lot more cores. So 3950x all the way, hybrid rendering to use my 2080 ti and Ryzen.
Okay, what I want to be faster for me in Blender is speed of scene loading and compiling shaders, more fps in a polygon heavy scene. Should I go with Ryzen 5 5600x with 6 cores but faster single core speed than the 3700x or Ryzen 7 3700x with 8 cores but slower single core speed than the 5600x I won't be rendering with CPU, I just want it to be faster with scene loading and having more fps in the scene when the ram is like 90% full.
@@kaojaicam I bought the Ryzen 7 3700X, great decision because it has 16 threads and not 12 threads like 5600X. Forget about the 6 cores, I get like 25% render speed increase from Ryzen 7 3700X. Blender and basically any other software are being made more compatible for more CPU cores. So more cores is more future proof.
As we all know, most rendering in blender gets handed off to our GPU's which are way faster than CPU's. So, which processor delivered the "build" information to the GPU faster might have been an interesting test. Seems like Blur team would have won.
Basically for Blender it's more useful to buy Ryzen rather than Threadripper. Would be interesting to do the same tests with X3D versions, for example 7950x vs 7950X3D.
ONE STEP YOU MISSED. Before the rendering starts the CPU has to built the Kernal to send to the GPU. Given I could get an RTX 3090 to do the rendering it would still be waiting, every frame, for that kernal calculation. As an animator this is really important. Would love your thoughts. And thanks for the content.
Considering the 3990X Threadripper stomps the heck out of the 3970x being that it's 64 cores versus 32 cores is impressive. Also consider that adding a couple high end CUDA Nvidia cards would make a world of difference with viewport EVEE / Cycles rendering. Hmm.... Depends on what you do day to day to day doesn't it? Question is what happen with Zen 4 in the next year or two and new GPU designs as well.
I think I'm missing something, wouldn't most of these tests be done with the GPU by default? If that's the case then the only thing that matters is rendering time.
If I'm not wrong Blender uses only one core to do those calculations which is slow, at least up to 2.79 I know that is the case (bake (insert name) simulation, 25% CPU in use and wait for it to end using a Quadcore). Why not use all the cores? or at least all but one so as not to leave the OS stuck.
But why? you are comparing a CPU that is obviously for multi threaded performance against one that is for single threaded performance, it would have made more sense to see a comparison especially in viewport (active work) with a $330 ryzen 7 3800X which seems to have the same single core performance as the TR3970X and/or $720 ryzen 9 3950X which has a better single core performance and still much more multi core performance than the i9s
Because this is about Blender, not about threads. Which one gets your work done faster? That’s all that matters. This is not about benchmarking for its own sake.
Great video, I would love something like this but in the laptop world... would it be better a branded laptop or one of this custom gamers laptop for real world blender usage
great testing methods but i think this only applies to blender. i think in a more optimized application (houdini) we would see more gains from threadrripper in simulation tests. but thats just my 2 cents. thanks
My father: I am bringing a new computer Me: Can you bring me a computer with i core 10 My father: Here you go son Me saw the sticker and got to know its i3!
I think threadripper is limited by its TDP and maybe heat and current. I tested fluid sim on my end which is a slightly overclocked 3950X. I disabled 1 CCX leaving it with 8 cores. With same frequency, and during heavy simulation baking, 16-core is 20% to 30% faster than 8-core. Mantaflow has some weird optimization, if you set domain resolution higher than 300, you might see CPU utilization goes over 80% and I think I set the resolution to 600, the simulation just swallowed my CPU, the CPU utilization is over 95% which is insane. Mantaflow sometimes spread the task to many cores so the boost clock speed will be much lower compared to Intel's 8 to 10 cores CPU. For a threadripper, it is like running 4GHz while Intel's 10900K might stays at 4.8GHz to 5GHz
If you compare the animation bench you can see the performance from 2.8 to 2.9 dropping. The only improvement about this was recently done by INTEL. And since I have a feeling the blender development is heading in the wrong direction, I think such comparisons should be done not only with the different processors but also with different software completing similar tasks. AMD and Blender fanboys are usually searching for that fake feeling, they are gonna achieve right about everything with their cheap machines :D
Hi Steve, just a question, or if anyone else can answer this, that would work as well, but are the $1 Patreon benefits included in the $3 benefits? And also, I don't see the $1 benefits anymore.
Are you kiding me evee is more GPU dependent and this is not a good comparision threadrippers are better for cycles rendering. And ryzen 3950x should be a better comparision. Plus the 10900k has higher power draw and can go up to 250 watts. I wonder how much Intel is paying him. Ryzen is still best for prductivity. See the Linus tech tip's video on 10900K review ruclips.net/video/mLsoJkQb0Cs/видео.html 🥵😡
Please make a Video on how to create a Dinosaur... Just like you in your Blender Course. But this time with Blender 2.8x. Really want it so badly. Including Sculpting, Texturing and Rigging and Animation... Will be the happiest Subscriber if you'd do it... Thank you So much...
Thank You!!!! so far you are the only one doing amd vs intel simulation and physics benchmarks for blender, do you think you could some for hybrid rendering
Do l need 32gb ram or 16 enough? does ram speed matter? is m2 vs sata SSD any different? if you know please help l am building a PC, l already have RTX 2060 ko, ryzen 7 2700. recommend other parts please.
Blender? 32 GB of ram, I don't know if you are fully gonna utilize it, but it fits there. A sata SSD should be more than enough for storagd., although I recommend a nvme one for your os, and the blender app itself, not out of strickt nessesity but out of comfort. HDD's are a great option too, if you need to store large, and currently not needed projects. Or if you are on a budget in generall, just shuffle the projects between the HDD and the SSD. Edit: Ram speed matters, make sure your Mainboard is compatible and go for 3200 MHz or more. That's about it. If you plan on upgrading to Ryzen 4th gen, make sure that you have a 400 motherboard or above
@@bytezero3818 keep in mind my goal specifically is add more poligons and textures without crashing, render speeds are good but l have time l can wait 4 more hours
It depends what you want to get out of it. If you're going to fully utilize Blender and make *very* high quality animated films, then you're going to need a pro card at around 32gb.
Great review. Just a quick question - Does pcie 4 come into play at all (more along transfer times et al? It would also be nice to get a head to head with the ryzen 3700/3800 and 3900x. Graet job - love your channel.
PCie 4 really hasn't seemed to come into play with Blender yet, as GPU's don't use up all the bandwith yet, and Nvme drives with Pcie 3 never seem to bottleneck Blender.
unfortunately, but you are probably wrong, it is not the blender that can not use TR, the problem is windows itself. Watching the tutorials one of the owners of the new TR, mentioned that the first thing he did threw windows and installed linux and simulations use the entire CPU.
I feel like this video is slightly biased towards Intel. It might just be me being delusional, but it seems like when team blue is ahead even by a little, he makes a big point about it, and when team red is ahead, he makes a point about how the Intel CPU is right behind it. Again, this might just be me being an AMD fanboy, but it does seem a little unfair.
I think intel needs all the help they can get honestly. Not that they deserve it... But people just feel bad for them at this point. When we consider that the significantly lower clockspeed threadripper ( a 3950x probably would've done better in all but cycles ) beat intel's best offering in fluid/smoke sims + rendering it's a pretty simple proposition. Use AMD for all around better blender performance for the money including top end. It's a shame the 3950x didn't get a showing though since it's still more cores than the best intel part but it also has the highest boost clocks out of the entire Ryzen lineup would should narrow if not reverse the few gains intel showed here.
Depited all Blender being a populer CPU benchmark, I never use CPU to render. GPUs are way more efficient and don't get scorching hot. Could you please do a speed comperison between a thred ripper and a high end GPU like the one in your rig?
Greetings to everyone who, like me, works on a 5-year laptop!
Greetings from an overheating macbook air, which was rendering in the background, and then crashed while I watched this video.
Greetings, my laptop is seven years old now 😅
Greetings from a 2015 MacBook pro !
Greetings from an early 2015 PC!
Greetings from my 2 year old i3 HP Laptop. (Crashed using blender like 2 minutes ago, had to reboot the laptop)
Couple other things to consider. (I have a 3960x 24 core)
1) Zbrush loves cores, I can easily sculpt with no lag at all with 100 million polygons. If blender could handle sculpting and booleans with 20 million polys with no lag I would probably never use Zbrush again. But that is a very difficult request to make to the devs and I doubt it will ever happen.
2) You are limited by your Vram on the GPU. This is not the case with CPU you can use your system ram ... so a 3d scene with 16GB of data will not render on most graphics cards but you could render 64GB -128GB 3d scenes on CPU.
3) If you want to do game development. More cores = faster compiling as you are building a scene.
4) And lastly, if you ran those simulations on Realflow 10 the more cores the faster it is. It's just that Blender currently does not make good use of multi cores.
But looking at blenders updates over the last year we could have proper multi core usage next week. Or in 3 years. :D
This comment has better insights than the Intel sponsored content.
Its not just blender that lags on AMD's cores when simming. Nor is it an IPC/core count issue As was also mentioned in the video, most simulation software top out at 10-30 pct usage on AMD cpus while utilizing 80pct or above on intel. FumeFX, Houdini thinkingparticles, phoenix fd all of them demonstrate the same behavior. VDB solves in Houdini result in frequent crashes on AMD's TRs btw.
As for GPU rendering, while limited memory can present a bottleneck but shader optimization on both octane and redshift make things faster. furthermore, the main factor in rendering performance are the CUDA cores. Also, if you are rendering a sequence that's between 64-128gbs, odds are its getting sent to gridmarkets.
Also, as a zbrush artist, i am sure you realize the importance of faster ideation and look dev- which is generaly much better when you have gpu rendering.
AMD's approach of slapping cores that dont scale up properly isnt any software vendor's fault.
finally some good info
What about material nodes? Is it better to buy more cores?
@@sniper1pk It is not AMDs fault, but actually the software vendors: As you can see in the Cycles benchmark, the Threadripper is absolutely A LOT faster than all the other CPUs if all cores can be used. And that depends 100% on the software. Rendering is extremely intensive and uses all cores 100%. The problem is more specific that many sims and display engines or sculpt engines were originally written for single-core in mind and then extended. They are not 100% parallelized in every way (i.e. might use a max of 4 cores or have other bottlenecks due to the code having been extended after the fact) Some physics simulations actually can't be parallelized, but a lot of other stuff can. These engines (for example the sculpt engine) needs a rewrite really badly in Blender. The story is better on ZBrush, as you mentioned, so that engine was rewritten and optimized with many many many cores in mind. So it is purely a software issue really.
This video is great, I think throwing the Ryzen 3900X in there would be a great comparison as it's a desktop CPU as well
Indeed, but I guess he benchmarked what he had. I for one, am extremely happy with my 3900x in blender, though for more serious scenes there is no substitute for GPU
The 3900X is what I use for Blender and it works good. It's especially handy with large particle fluid sims.
@@abdulwahabjag Are you trying to emphasize the difference between "good" and "well" or are you being a grammar nazi?
tell me bro? im planing to buy 64 core. thirdriper with 64 gb ram 2080ti nivida ssd 2tb. is it good? or any other cpu best let me know please
@@iftikharmalik639 new ryzen CPUs are supposed to be releasing this fall. You can wait until then and buy something with more performance brand new for potentially better value or get one of the existing models that'll come down in price due to the new CPUs being out. You can also find more CPUs on the used market once new CPUs are released
I use blender on ThinkPad so i can make an omelet during render.
And you don't even have to take the eggs out of the fridge
And he literally isn't joking
If you think you've got it bad, I could wait for the chicken to hatch, watch it grow up and then make an omelette when IT starts laying eggs.
@@RobCLynch legend 🖖
HAHAHAHA I can relate with my MacBook Air
* me searching 3970x's price *
Wow the price are equal to my 150 cc motorcycle
and you can get the i9 10900f which can also be unlocked for 360$ usd right now
You should have added ryzen 5 3600x in the benchmarks for budget pc builders.
Damn.. It's crazy what is considered "budget" for the US.
@@TheCivilProtector But it is considered expensive here in India. I used the word budget referring to the US
@@TheCivilProtector yeah US is crazy as an Indonesian everything in us seems fairly cheap for them but not for us
@@tilentgamer8758 yeah agree, setuju gan
xCivil_Protection might not be budget anymore due to economic depression
You should make another video about GPU’s
Thanks for doing an dive into blender for the creative space for these CPUs. I went with a 3950x but it’s nice to see charts like these to get a sense of single core vs multi core performance. It can be about having a good balance, in my humble opinion.
I'm on amd team as well. Don't plan to go back anytime soon. Especially with Ryzen 4000 lineup coming soon
Thank you for explaining not only why the answer is "It depends", but also what it depends on. It cleared up the misconceptions I had about the newest processors on the market right now.
Any plans to make this video for 2022?? Id love to see some of the newer stuff have a go at Cycles x etc as im looking to upgrading my entire desktop. Either way great video.
2022 update would be awesome
I eager to see comparison 8core ryzen 5800x3d and 12core 5900x which are at very similar price here and now.
What about testing some heavy modifiers and non destructive modeling?
i dont know anything about blender but my unity knowlege and ASPECIALLY you helped me make an animation
Thanks cg geek this what i'm looking for a cpu benchmark
You helped me a big time! I was deciding between 3950x and 3900x thinking whether it is worth the money and it is not. For rendering I will be using RTX 2080super (eventually will get second one as well) and for processing and physics I will go with the 3900x! I though more cores will equal quicker calculations, but you just showed me it is about frequency not the amount of cores! Thank you so much!
It is really a shame that the 32 cores are never really used fully inside Blender outside of the raytracing rendering aspect. This means that a lot of workflows are permanently limited, they will forever be slow, since we don't get cores that are that much faster (and definitely not on Laptops), i.e. you can't buy a laptop that has 4 cores that run at 8 GHz, no such thing in the world. So making things parallel so they can use many many MANY cores is key and that is currently not done.
"We're not exactly comparing Apples to Apples in this video."
Yeah, but I don't see Linux here either. ;)
iii see what u did there..
@@CGGeek do you think that a r5 3600 and 5700xt is good for blender. I will game on it as well.
@@sbc2954 what im trying to know too
super useful and proper comparison! definitely agree that just rendering the typical benchmarks doesn't give a good overview of the process and performance.
Oh yes, they actually do: The rendering and raytracing shows the pure performance of the CPU if software is properly written to be scalable and use all the cores and performance (so to speak, use all the engine power that the CPU has) fully. We know as a matter of fact that we get most performance not from faster cores, but more cores, many more cores. GPUs are so fast, because they have tiny stream processors, but many thousands of them and they all work parallel.
would love a 2022 version with a high end AMD CPU as well and maybe the newest generations CPUs as well :)
For the majority of people, the answer is simple, these CPUs are stupidly priced, ryzen CPUs with 8 cores and 16 threads like the 1800x are going to be the best thing you can get and will often be around 150€ in the second hand market. Only an extremely small percentage of people will need the higher performance that Threadripper CPUs can provide.
Anyway, if you really want good blender performance, don't spend more than the previously mentioned 150 or at most 250€ on a CPU, an RTX GPU will give you far better performance for the price, hopefully the next generation of AMD GPUs will do so too, but right now? the full price of a 2070 and e-cycles is going to give you better performance than the most expensive CPU you can find.
I personally wouldn't recommend a 1st generation Ryzen at this point. Maybe the 1600AF because that's a Zen+ re-release of a Zen CPU. But if you aren't super budget constrained I would go with 3600x or 3700x. The 3800x isn't worth the extra cost. Then 3900x and 3950x if you're a baller and have the money.
@@Sophistry0001 i mean, that's very relative, the 1800x is a very good choice if you can get your hands on it as it will outperform everything in its price range (goes for about 130 to 150€ second hand here), other than that, a 2700 is the next great choice, Zen 2 CPUs are still very pricey and not worth it if you aren't made of money.
Funny you'd mention the 3800x as it's barely 30€ more expensive than the 3700x over here, if you're going to spend around 350€ on a CPU might as well go for the 3800X.
@@Knuckles2761 sure but it's also about 50€ more expensive, which if we were talking about the 300€ range wouldn't be such a difference, but from 130-150 to 180-210, that's a pretty hefty price difference
Agree. dual rtx2070s would be sweet spot add on for enthusiast workstation. New tensor flow cores used by denoiser help reduce number of cycle samples without compromising quality, old generation (non RTX cpu) would need to crunch more samples to get comparable quality.
hearing people recomending much older and also used stuff for the same price you can find something new and decent, like the Ryzen 3300X $120, is like saying "dude you can buy a broken TR for $20 and fix it yourself" just ridiculous.
Is that a RTX 8000 in his threadripper PC?
WOW!!!
:O
You deserve more views, your content not only helps but it also entertains!
Thanks so much! Very clear and good video(:
I was just wondering this thanks for this video!
Would be awesome to get an updated video like this! Xoxo awesome have the best day
fional solution : two computers , one for creating senses,one for rendering
this what actually i was looking for, thanks
Great video! You recorded this video months ago but I just found it recently (No thanks to RUclips recommendations!) and I should say, you made the best comparison video for the time you spent. Thank you for sharing and Good Luck.
I've watched this video several times since its release, and I've only just noticed the nice subtle audio overdub at 3:48! 🤯 nice job making that sooo seamless!
I agree with this but he didn't mention that your system would be unresponsive when doing those things on the Intel CPU while on the threadripper it would work perfectly because it has spare cores to work on that
No, not necessarily. See the problem with the viewport is that it is not well parallelized, the code was originally not written for multi core CPUs with many many cores. It will only use a few cores, not all of them. "It does note scale properly" is the technical term, since the intel has the few cores it has running at a faster speed, it works faster in the viewport, while the threadripper has probably around 20 cores doing nothing at all. Software has to be written to use massive parallelization. Blender unfortunately is not in many areas of viewport, physics simulation and modeling.
Finally, I be waiting for video like that. Thank you!
Thank you for your comparing, like it very much !
I've had a ryzen 3 1300x for the longest time, finally bought a ryzen 7 2700x for only $200, hope it's good 🤞
It HAS to be!
Well its been 4 months, how is it performing? :D?
@@Brahvim pretty dang good! It seems like it can definitely handle a lot more than it used too, but now I’m just being held back by my gpu, because it only has 3gb of video memory, (it’s a 780ti) so renders are still a little rough, but I have noticed some speed gains, and that graphics card is next on the upgrade list
@@ittueaday May I ask, do you mean you can get better fps for animation viewport? I'm thinking to upgrade my amd ryzen 3 2200g to amd ryzen 5 3600 just so I can do animation and simulation better.
@@raofriot better FPS will mainly come down to gpu if that’s what you’re running, the cpu can help though and I did notice some speed improvement overall, but it would of been faster if I upgraded gpu, up to you
Really its a Real-World Benchmarks.. Many many thanks
Thanks man
Thank you very much for the info bro 😍🙏
CPU video: Done
Now we just need a video for GPU, monitor, and fans. And...keyboard I guess?
Having a little experience in programming and software development, I can tell you that unless you optimize a workload to run with multithreading, you certainly will run issues with scaling. Let’s say you have just one core, you load the data and the instructions, you make the calculation and then store the result. If you add a second core, you will have to take into consideration the resources as well as timing. You end up having downtime for example if one core needs the result of the calculation the other core is performing. On the other hand, you want the cores to share as much resources as possible to minimize RAM/cache usage as well as reduce downtime while a data is being loaded into the RAM or cache.
Fun fact (but is dated already): in 2016 I watched a similar video about the Adobe Suite. The conclusion was identical that the software ran faster on a few high clock cores then a lot of slow clock cores.
Excellent vid! Thanks for the work. Saved me some time and money!
I wonder is this video up to date? Have they advanced multi-threading in any way in the past two years?
Nice video
I’m a new one for your channel
After seeing yout tutorials i falling love with blender
You’r a wizard in the Blender World
Good Luck Bro.
👍🏻
💖
Wow, you got a lot of great comments and great questions, in the comments. It is too bad you didn't answer them. A lot of people posted questions right after the video was uploaded.
Viewers were really helpful and answered many of each other's questions with replies to comments. But, i always like it when a creator is present in the comments, with answers to questions and additional information.
It was a really good video and interesting topic. thanks.
time for another video
But what if you enable the CUDA cores in Blender and use GPU renderer example with the RTX 3090? Would you consider this a crucial consideration?
I wish the 3900x was on the list, I'm struggling to choose between 3900x and 10900k for the viewport performance, unfortunately I can't see benchmarks athor than rendering times.
Please do an update to this video with the new R7000's, 5000 threadrippers, and i9-13XXX with Blender 3.3.
To really compare CPUs, you should run blender on linux, to minimize the windows randomisations. But, the vid is cool, as usual !!
Also compare the beta vs stable to catch any performance *regressions* that might effect results.
Linux is well known for handling high multi core CPUs better than windows. I wonder how much of a difference it would make in these tests, for curiosity of nothing else
I tested windows vs linux rendering with my Ryzen 2700X and the performance is the same as long as you close every background app that used any percent of the CPU in Windows. Including the Task Manager itself.
Can you make tutorial how to riging car?
Watch skyline studios channel
@@Imaginestud thanks for you recommend
@@Imaginestud just saw comment you replied to, was looking for something similar, so thanks from me too 👍
I think there is a missing component to this test; the Ryan Master software...a fair comparison would be to adjust the core count in the software to match the new chip set. less cores, less heat, resulting in a higher boost clock. The TR would still lose, but the numbers would not be so dramatic.
Or the 3950x with power restrictions removed via AMD's software.
Great video, such good information, now you should do the same but with graphics cards, showing what kind of performance you could get with each one
GPUs are not used at all to speed up workspace / GUI / modelling or physics simulations, except in displaying it, which isn't much.
It’s true benchmarking is not only rendering..
Thank you cg geek 👍
Nice man you have a cool channel
haha! I love the idea that one can plonk down 2,300 on a cpu alone. I would wager some of those systems cost more than my last two cars combined!
Threadrippers have no memory caping, so can address up to 1TB if there as many udimm slots on mobo or module with large GB units are available and are designed for data centers to compete with Xeon. More interesting would be see comparison with 3900x/3950x Ryzen as those are direct competitors to iCore line and have comparable price. So this was a bit pro-intel review imo.
Sir please do a similar video with Ryzen 7000 and Intel 13th. Planning to build a system without GPU
intel: steve, have some $1000 processors for free to do a video on
me: *sad i3 noises*
I wish I could subdivide the monkey head that many times, my pc sucks! 😂
I love this video! Just bought an i9-9900, specifically for Blender. I'm happy with my purchase, so far. This is exactly what I was looking for to help me decide what to buy. Unfortunately, I bought it a couple of weeks ago. I had to make my decision without this info, but it's nice to know that I would still make the same choice.
Time is money, but sometimes, I have time available while money is not.
Cheers!
smart move for not jumping on the amd hype bandwagon. i did and regretted it HEAVILY
@@sniper1pk which cpu you choosed?
@@saugatpoudel4088 two machines, first and 2nd gen threadrippers to use alongside our mostly Intel x79 and x299 based vfx shop
@@sniper1pk What don't you like about the Threadripper? I was thinking about getting one...
@@imhk1000 the video made it pretty clear why the Threadripper is overkill, it only beat the 6, 8, 10-core CPUs during Cycles rendering.
To summarize: Create your scenes with the i9 10900k but render them with the TR3970X
I don't get that takeaway, the viewport performance didn't matter because if you're getting 2 FPS then 3 FPS isn't really an upgrade and the threadripper parts actually did better in the fluid/smoke sims which I do way more often than rigid body sims.
And by the way, the 3950x probably eliminates the "gaming" advantage the intel CPUs enjoy whilst still having a lot more cores.
So 3950x all the way, hybrid rendering to use my 2080 ti and Ryzen.
i9-10900K good in mesh simulations and posing characters
3070X good in heavy simulations and renders
Okay, what I want to be faster for me in Blender is speed of scene loading and compiling shaders, more fps in a polygon heavy scene.
Should I go with Ryzen 5 5600x with 6 cores but faster single core speed than the 3700x
or
Ryzen 7 3700x with 8 cores but slower single core speed than the 5600x
I won't be rendering with CPU, I just want it to be faster with scene loading and having more fps in the scene when the ram is like 90% full.
Same problem with me, all comparison is about rendering time either gpu or cpu.
hit me if you find the answer.
Man I have this exact question lol
@@kaojaicam I bought the Ryzen 7 3700X, great decision because it has 16 threads and not 12 threads like 5600X. Forget about the 6 cores, I get like 25% render speed increase from Ryzen 7 3700X. Blender and basically any other software are being made more compatible for more CPU cores. So more cores is more future proof.
As we all know, most rendering in blender gets handed off to our GPU's which are way faster than CPU's. So, which processor delivered the "build" information to the GPU faster might have been an interesting test. Seems like Blur team would have won.
We need updated video please
Basically for Blender it's more useful to buy Ryzen rather than Threadripper. Would be interesting to do the same tests with X3D versions, for example 7950x vs 7950X3D.
I don't mean to be that guy, but 155% faster does not mean that the other one is 155% slower.
ONE STEP YOU MISSED. Before the rendering starts the CPU has to built the Kernal to send to the GPU. Given I could get an RTX 3090 to do the rendering it would still be waiting, every frame, for that kernal calculation. As an animator this is really important. Would love your thoughts. And thanks for the content.
Considering the 3990X Threadripper stomps the heck out of the 3970x being that it's 64 cores versus 32 cores is impressive. Also consider that adding a couple high end CUDA Nvidia cards would make a world of difference with viewport EVEE / Cycles rendering. Hmm.... Depends on what you do day to day to day doesn't it? Question is what happen with Zen 4 in the next year or two and new GPU designs as well.
comparing a $500-600 cpu to a $2000ish cpu isn't quite fair... but i guess the premise is speed vs cores
Yup exactly, the premise was to determine if the extra cost for more cores vs faster cores is worth it.
Fx 6300 is the best for sure.
Ps: just kidding
bruh I still got a FX 4130
STILL better than the 9590. That thing wasn't a CPU, it was a miniaturized George Foreman grill/calculator combo built for socket AM3+
@@plapbandit hahaha
@@slavko5666 old but gold haha
the voice cracks are real in the video lol
I think I'm missing something, wouldn't most of these tests be done with the GPU by default? If that's the case then the only thing that matters is rendering time.
If I'm not wrong Blender uses only one core to do those calculations which is slow, at least up to 2.79 I know that is the case (bake (insert name) simulation, 25% CPU in use and wait for it to end using a Quadcore). Why not use all the cores? or at least all but one so as not to leave the OS stuck.
I have the 3900x and that thing is awesome
But why? you are comparing a CPU that is obviously for multi threaded performance against one that is for single threaded performance, it would have made more sense to see a comparison especially in viewport (active work) with a $330 ryzen 7 3800X which seems to have the same single core performance as the TR3970X and/or $720 ryzen 9 3950X which has a better single core performance and still much more multi core performance than the i9s
Because this is about Blender, not about threads. Which one gets your work done faster? That’s all that matters. This is not about benchmarking for its own sake.
@@lawrencedoliveiro9104 You make no sense, and i won't explain, sorry you are just like so far off...
Great video, I would love something like this but in the laptop world... would it be better a branded laptop or one of this custom gamers laptop for real world blender usage
Thank you for the useful tips.
6:26 linus ran into this too, so he split the cores? (sorry i'm not a geek) and THEN it used all the cores. and blew all the benchmarks away
Sauce please
@@ExcelonTheFourthAvalonHeirs did you mean to say sauce?
@@shakaama
Can you give me link to Linus video you are talking about?
split the cores? That is not a valid thing that you can do. Do you still have the video of Linus where he does improve performance?
great breakdown.. thank you
I saw your low poly man making tutorial can you make it once more i am not understanding the tutorial
Great video, I believe it would be a good idea to download the files so that we can test our hardware and see if it pays for an upgrade.
great testing methods but i think this only applies to blender. i think in a more optimized application (houdini) we would see more gains from threadrripper in simulation tests. but thats just my 2 cents. thanks
My father: I am bringing a new computer
Me: Can you bring me a computer with i core 10
My father: Here you go son
Me saw the sticker and got to know its i3!
Does GPU help with viewport playback performance?
Is a 13 Inch Arm-based Macbook enough?
Hey man! I have a question...
How much can I sell a human kidney for? It's nothing I'm just asking for a friend.
I think threadripper is limited by its TDP and maybe heat and current. I tested fluid sim on my end which is a slightly overclocked 3950X. I disabled 1 CCX leaving it with 8 cores. With same frequency, and during heavy simulation baking, 16-core is 20% to 30% faster than 8-core. Mantaflow has some weird optimization, if you set domain resolution higher than 300, you might see CPU utilization goes over 80% and I think I set the resolution to 600, the simulation just swallowed my CPU, the CPU utilization is over 95% which is insane. Mantaflow sometimes spread the task to many cores so the boost clock speed will be much lower compared to Intel's 8 to 10 cores CPU. For a threadripper, it is like running 4GHz while Intel's 10900K might stays at 4.8GHz to 5GHz
3950x seems like the best option out of everything mentioned.
Go for amd ryzen thread ripper 128 cores.
If you compare the animation bench you can see the performance from 2.8 to 2.9 dropping. The only improvement about this was recently done by INTEL. And since I have a feeling the blender development is heading in the wrong direction, I think such comparisons should be done not only with the different processors but also with different software completing similar tasks. AMD and Blender fanboys are usually searching for that fake feeling, they are gonna achieve right about everything with their cheap machines :D
imagine being smug about paying more for the same performance
Hi Steve, just a question, or if anyone else can answer this, that would work as well, but are the $1 Patreon benefits included in the $3 benefits? And also, I don't see the $1 benefits anymore.
Hey Man, I had to un-list the $1 perks, as it was too much to keep up on, but you're right every level includes the previous perks!
Are you kiding me evee is more GPU dependent and this is not a good comparision threadrippers are better for cycles rendering. And ryzen 3950x should be a better comparision. Plus the 10900k has higher power draw and can go up to 250 watts. I wonder how much Intel is paying him. Ryzen is still best for prductivity.
See the Linus tech tip's video on 10900K review
ruclips.net/video/mLsoJkQb0Cs/видео.html
🥵😡
People even sell their soul for sweet Intel money. Many "tech" youtubers do that kind of stuff.
Love from 🇮🇳India
Please make a Video on how to create a Dinosaur... Just like you in your Blender Course. But this time with Blender 2.8x. Really want it so badly. Including Sculpting, Texturing and Rigging and Animation... Will be the happiest Subscriber if you'd do it... Thank you So much...
Thank You!!!! so far you are the only one doing amd vs intel simulation and physics benchmarks for blender, do you think you could some for hybrid rendering
Do l need 32gb ram or 16 enough? does ram speed matter? is m2 vs sata SSD any different? if you know please help l am building a PC, l already have RTX 2060 ko, ryzen 7 2700. recommend other parts please.
am nvme m.2 ssd. for start 2x8gb, more can be add later, and with a speed of around 3200mhz can be found cheap and is enough.
Blender? 32 GB of ram, I don't know if you are fully gonna utilize it, but it fits there.
A sata SSD should be more than enough for storagd., although I recommend a nvme one for your os, and the blender app itself, not out of strickt nessesity but out of comfort.
HDD's are a great option too, if you need to store large, and currently not needed projects. Or if you are on a budget in generall, just shuffle the projects between the HDD and the SSD.
Edit:
Ram speed matters, make sure your Mainboard is compatible and go for 3200 MHz or more. That's about it.
If you plan on upgrading to Ryzen 4th gen, make sure that you have a 400 motherboard or above
@@bytezero3818 keep in mind my goal specifically is add more poligons and textures without crashing, render speeds are good but l have time l can wait 4 more hours
@@cicciograziani6697 is 1x16gb okay coz most mobos have two slots
It depends what you want to get out of it. If you're going to fully utilize Blender and make *very* high quality animated films, then you're going to need a pro card at around 32gb.
Great review. Just a quick question - Does pcie 4 come into play at all (more along transfer times et al? It would also be nice to get a head to head with the ryzen 3700/3800 and 3900x. Graet job - love your channel.
PCie 4 really hasn't seemed to come into play with Blender yet, as GPU's don't use up all the bandwith yet, and Nvme drives with Pcie 3 never seem to bottleneck Blender.
Great now make for gpu
unfortunately, but you are probably wrong, it is not the blender that can not use TR, the problem is windows itself.
Watching the tutorials one of the owners of the new TR, mentioned that the first thing he did threw windows and installed linux and simulations use the entire CPU.
I feel like this video is slightly biased towards Intel. It might just be me being delusional, but it seems like when team blue is ahead even by a little, he makes a big point about it, and when team red is ahead, he makes a point about how the Intel CPU is right behind it. Again, this might just be me being an AMD fanboy, but it does seem a little unfair.
I think intel needs all the help they can get honestly. Not that they deserve it... But people just feel bad for them at this point.
When we consider that the significantly lower clockspeed threadripper ( a 3950x probably would've done better in all but cycles ) beat intel's best offering in fluid/smoke sims + rendering it's a pretty simple proposition.
Use AMD for all around better blender performance for the money including top end. It's a shame the 3950x didn't get a showing though since it's still more cores than the best intel part but it also has the highest boost clocks out of the entire Ryzen lineup would should narrow if not reverse the few gains intel showed here.
@@OGPatriot03 Very well formed reply.
How does rendering with the Titan compare to the Intels and Threadripper?
Depited all Blender being a populer CPU benchmark, I never use CPU to render. GPUs are way more efficient and don't get scorching hot. Could you please do a speed comperison between a thred ripper and a high end GPU like the one in your rig?
Please Make a video for "Camera rig" IN BLENDER
I am your fan from Bangladesh.
Team Red wins again
Thanks ig sir
Next video - Which *pro* graphics card