**The Evolution of Rendering and Computing Power**

I recall my old machine being around 3000, maybe 3800, with the next best machine at work being something like 1600, 1600-ish. This range is quite common for older systems that are still capable of handling certain tasks. However, when it comes to modern computing power, especially in the realm of visual effects (VFX) rendering, these numbers pale in comparison.

The top-of-the-line Mac Pro can cost upwards of $56,000 and boasts a processing speed of around 10k, with some models reaching as high as 14k. These systems are capable of handling even the most demanding tasks, but it's essential to note that the bottleneck often lies not with the hardware itself, but with the software and workflow.

**The VFX Industry and Computing Power**

One significant change in the VFX industry over the past four to five years has been the shift towards more complex and powerful computing systems. The last major milestone was when NVIDIA's Tesla K40 GPU became a standard tool for VFX rendering. This marked a significant increase in processing power, but it also highlighted the importance of software optimization.

In reality, most projects require not only immense computational power but also substantial memory and storage resources. Software like Adobe Premiere Pro and After Effects must be optimized to handle these demands, as well-known artists have discovered that simply throwing more hardware at the problem doesn't always yield results. It's often a matter of workflow optimization and careful management of resources.

**The Future of Rendering and Computing Power**

As we move forward, it's clear that the future of rendering and computing power lies in hybrid memory solutions. The ability to access GPU memory directly from the CPU would significantly reduce overhead and allow for even more efficient processing. NVIDIA's recent advancements in this area have shown tremendous promise.

One potential game-changer is thread ripper, which unlocks GPU horsepower by a factor of 4+. This technology allows artists to work with much higher resolutions and detailed models than previously thought possible. However, it also raises questions about how these new workflows will be handled on a daily basis.

**Artistic Workflow and Rendering**

The shift towards more powerful computing systems has forced artists to rethink their workflow. With the advent of real-time rendering, many are finding that they can work on multiple projects simultaneously without sacrificing performance. This is made possible by software that allows background rendering, which can be run at a low priority without affecting other tasks.

In this scenario, all machines in the workstation group can render in the background while still allowing artists to focus on their current project. It's an exciting development, one that could fundamentally change the way we approach artistic workflow and collaboration.

**Justin's Perspective**

Justin brought his own machine for testing a particularly demanding scene, which was running at 2 FPS on an older system. By comparing these results with more powerful machines, he gained valuable insights into what factors contribute to poor performance. This experience highlighted the importance of understanding one's own computational limitations and how to optimize them.

As Justin noted, most systems are not optimized for VFX workloads, leaving room for improvement through careful software selection and workflow optimization. His experience demonstrates that even with powerful machines, there can still be significant bottlenecks in terms of workflow.

**The Power of Community**

Finally, it's essential to acknowledge the power of community in driving innovation forward. The ability to share knowledge, collaborate, and learn from one another is crucial for advancing our understanding of computing power and its applications.

Whether through online forums, social media groups, or workshops, artists are continually sharing their experiences, discoveries, and insights. These connections foster a sense of camaraderie and shared purpose, driving us forward as we explore the frontiers of rendering and computing power.

"WEBVTTKind: captionsLanguage: enit's backwards where I boo myself right here and I walk some the quick release my boy boy you massacred my boy so what I've got for you today is something special Justin my friend we go way back he works in the VFX industry and he is a supremely gifted individual that's been working in VFX since the dawn of time and he is not satisfied with the status quo ever he is always looking to tweak and tune and mess with his applications and squeeze as much as possible out of his community hi isn't it insane it is he's bringing in his machine so he works at a pretty big studio they do a lot of really cool stuff you've heard of the stuff that they've worked on it's legit and they've got some older machines he's bringing in a machine that was last upgraded like five years ago some of the parts are even older than that some of the parts were like x79 all right let's build behind here so nobody can see anything he fit we've already done all the testing of course we've tested the $3,000 w31 775 X which technically by extension I guess means that we've tested the Mac Pro because the Mac Pro is the new socket it's a different CPU it's actually a slower CPU than this 28 core monster yeah we've also tested through first second and third generation we'll talk a little bit more about that so your old workstation had for nine 80s mm-hmm and a six core X I think it's a nexus 99 next time mounted CPUs like this Explorer whatever that model is i75 780 I don't make any sense it was from the time when pretty much all of the VFX stuff was just running on GPU only and now we've got a good mix of CP and GPU so six cores doesn't do it but it's also a really nice custom loop it was never cleaned maybe for almost ten years he's bringing in a machine that was last upgraded like five years ago some of the parts are even older than that some of the parts are like x79 but it's a full tower Corsair monster case dual radiator it's got four GTX 980 s in them they're using these for rendering and they basically gave him one because it died and the reason it died we found out because the radiators in chunk we're trying to resuscitate this thing it's almost like somebody neglected the crap out of it I was given this as a gift and it was a terrible gift but in the best way we spent how long cleaning it oh and I already used a bunch of vinegar and the blocks are full of junk and the couple of the video cards literally died because one of them has bad vram because of the heat two of them won't post and the other one is clock and really weird so for GPUs dead we we tried to reuse some fittings that was perhaps a mistake yeah started leaking that radiator sprung a leak required a little bit of soldering something something adventures it was the quick release fittings the quick release fittings they were suspect I said there were suspects from the beginning but it was like hey I won't probably be let's use them let's how does it stay yeah the quick release is not quick releasing well that's why you got new ones but luckily we caught it you were able to get it switched out without there being a complete amount soon inside but I'm here which was a huge concern for a moment I should point out too that not doing maintenance is not Justin's fault they were his work machines and it was like nobody told them they had to do maintenance on a loop I mean I feel like somebody should have known that yeah you for soldered but you've also got a quick-release now you didn't have as a fort in the secret door on the back when it comes time for cleaning or whatever you can literally just just and it's a new quick release we didn't recycle this one so the idea of this extra tubing is we come out and around and loop over and back in some of this extra space but this quick release here will make it a lot easier to work on the pump because all your returned water you can just use the pump to basically pump everything out of the system and then gravity will do the rest because you can just lower that below the like you can like just drop it in the floor and then your loop will be better before we had like the main artery running up here but we're actually using that just the the main output now to run directly into the CPU we've got some odd choices in the build like these are old server drives but they're four terabytes that's uh that's gonna be interesting you know it's also amazing I'm pretty sure this is just a cd-rom drive in here and not even a CD burner I just wanted to call that out it's nasty you got a bunch more of these machines at work which means that your studio is probably gonna buy a lot of thread Ripper you're welcome AMD mmm the other thing is PCI Express storage and so we had a unique opportunity we had these four four tire about the Intel nvme drives which were available for less than $100 a terabyte and these are not like PCI Express for SSDs are gonna be faster at least initially but these don't slow down these are triple level cells these are engineered to last a lifetime and they still have two years two and a half years of Intel warranty left and knew they were like three thousand dollars and you can still buy these on new egg about $2,000 and I could be honest with you I'm still pretty new level when it comes to PC building and when you mentioned that it was just like hitting me in the face and just going around like I had no idea what you were talking about but we're not just going into this willy-nilly because I'm some sort of crazy thread Ripper evangelist whatever although it may be that's true to you I don't know so I did a lot of homework before Justin actually got here doing what's called well what I would call probably application profiling so looking at his programs a lot of them you probably never heard of but one of them that you probably have heard of is Adobe After Effects but it'll be After Effects on its own really isn't that interesting it's the plugins that go in After Effects the after effects plugins they can use CUDA acceleration some of them use a lot of CPUs some of them don't after effects on its own it doesn't render fast it just doesn't you have to use something else to accelerate it because it's super multi-threaded with After Effects you really want to render one frame at a time but you can take a sequence that's you know 600 frames long and break that up into jobs and that's what we did with red ripper to test the 24 and 32 core thread Ripper well we tend to do which is a little bit more unique you know differentiating it from other professionals is our workload is a three to four minute all CG piece so we needed a solution that gave us the ability to render that out faster because Trinita After Effects you're looking at like hour and a half two hours I mean there have been times we'd hit the render button it would say like eight hours of this and it got out of control and render garden was the first opportunity that we had to speed that process up render garden is the name of the software that does that now render Gardens website advertises a two to three X speed up but by systematically looking at where the bottlenecks word memory i/o vram because when you've run out of VRAM things get shuffled from your GPU to main memory now that might be something that can be optimized in the driver I'm not sure I actually reached out to Nvidia about this but for this test setup for the system that we're using now we are just using a GTX 1080 Ti but we found the shelf but would three minutes is what you can do the gtx ten atti with about five to ten instances of render garden depending on how complex your After Effects project is and how much CUDA your After Effects project needs and it depends it's it is very very project specific so it depends on the plugins you're using and everything else well now I remember us working on some really old projects way back in a day and there was a checkbox in After Effects which has rendered multiple frames at once that's gone that was gone before almost before I left so that's how long it's been like kind of gone so we've got a we've got a we've got a system to build but everything is in balance in this system based on your specific workload like your specific projects I mean with render garden that 50 minute job immediately was like 20 minutes on thread Ripper and then with render garden it was like what three and a half yeah and we never really found out what the what the chef was on oh we found the Shelf in later testing time down to a minute and 22 seconds with 16 instances of render garden with a GPU that had 24 gigs of VRAM that GPU the titan r-tx this is not always going to be the case like it depends on your project and the plugins that you're using I mean even a mid-grade GPU may be sufficient for doing stuff you might remember some of the testing that we did with a post Vox DaVinci Resolve scaled perfectly to two GPUs yet in that third GPU you weren't really getting scaling after more analysis it turns out that more scaling meant that we needed more CPUs when we did that testing thread Ripper 3000 wasn't available but now da Vinci with three GPUs actually does scale a little better what a day it was a day it only rained like five times that last one was really bad it was pretty intense well the crazy thing is that you know as you know with your old workstations it was super important I mean you've got a three hundred and no it's like a 420 and a 480 millimeter radiator or something crazy it's like triple 140 s up here and quadruple 120 s down there and and right now it's all going to the CPU yeah nuts but we routed a cable so you'll be able to insert your GPUs between the CPU goes directly to the top radiator and the top radiator goes to the bottom radiator now but you'll be able to insert your GPUs when you get your blocks just with this cable so this cable runs down you'll be able to put a quick release in that and you know one or two GPUs Titan RT X's but for now 20 atti but the thing that blows my mind is we did the tests and on this loop with the old 980 s this system is running circles around for 980 s if we're not ATS really they're not a slouch but it just goes to show you if you don't do maintenance on your loop you're gonna have worse performance than even probably a single 980 yeah it was pretty much hobbled and this is the the 24 core just imagine - 32 core and 64 core came and fathom through what all did we run through crystal disk mark yeah crystal disk mark but he also did octane I did octane bench did crystal diss mark and then we did the new Puget bench which is still in beta but still gave me some good numbers compared to any machine that we had ring was it payments it was like over three thousand yeah it was it a three 3800 or something like that somewhere in that range and the next best machine that you got at work which is not super old was like six hundred a 1600 1600 something like that yeah so around that range I mean it's invaded and those numbers might be flush motive but still well cinnamon char 20 right well yeah the new $56,000 Mac Pro the best score you're gonna get on the Mac Pro is like 9600 let's call it 10 K will be generous and say 10k this thing out of the box was 14 K stock stock I love working on real-world projects like this it's so it's great and there are so few people out there that will actually sit down and look at what your problem is it's like let's just keep throwing hardware at it until the problem is solved it's like no it's not necessarily a case I mean with these kinds of projects especially when you're talking about like VFX rendering the bottom that could be anywhere it could be GPU it's usually not could be memory storage could be software like your software it's like okay this effect that you're doing let's just bake that into the render and then re-import that footage because every time you do anything with his footage its recalculating that this is never changing it doesn't need to be done that way this is a workflow problem this is not really a problem with your hardware or anything else but it's also interesting because the VFX industry has changed dramatically at least in terms of hardware the last four or five years and one thing that I keep realizing is that I probably ought to be doing a lot more rendering and working at the same time which I kind of have completely like gave up on that reality well you can't do that oh no what now you can it sounds like I probably can and so I'm not the kind of change that or my part of my design invades I think we want to see in a year to the you know in groups of artists that are working together probably will be able to do the network rendering and just run it at a low priority with no ill effects that would be awesome so it's like during the work day when you know bob needs to use your machine for rendering if you're not otherwise rendering something all the machines in the work group are rendering but if you've got a job you need to render it you know your machine's gonna render in the background as well as everybody else's and if everybody's got something irregular in the background everybody's gonna get background rendering but the software for that is is pretty much there but it hasn't really like like you people trying to do that workflow found that it didn't work but now I can so like what's going on on the screen over here where you're at I have a really really crazy heavy scene with a lot of like layers and glows and plugins and different like something that's like basically really heavy high raster I want to test that file because that's gonna give me all the information I need cuz I'm really good at running a computer down pretty quickly and this is a really good you know work case example so it looks pretty fluid to me yeah what this was this the file on your other computer that was like two fps yeah it is a huge ever-changing ecosystem and it is very difficult to say yes this computer is always going to run really well for this particular application I think that's probably true that when your old machine was built or so GPU heavy it was true that GPU acceleration was all rage and that works really well but now it's a mix of CPU and GPU I forget that most have not no bottleneck in your workflow you have to have an insane CPU and and then same GPU set up and that you can even do this with the you know relatively consumer-grade 2080 Ti is just kind of bananas right the next big thing honestly is probably hybrid memory where the CPU can access the GPUs memory more directly with less overhead and vice versa so thank you again Justin for joining me thank you for any messages for anybody in internet land don't recycle wrap don't try to recycle read and don't try to siphon a hose when it's clean I like I like the taste of my head you can smell it it smells like we're making pickles thread report is so insane it's such an incredible like it's not just thread Ripper it's like trigger 4 plus it's really unlocked the GPU horsepower it's just out of I don't know I'm gonna stop rambling out I'm Windell this is level 1 signing out I'll see you later it's like a butt flat yeah it's a buff flap to make cleaning easier then it's a terrible mental picture absolutely terrible that's all about terrible mental picturesit's backwards where I boo myself right here and I walk some the quick release my boy boy you massacred my boy so what I've got for you today is something special Justin my friend we go way back he works in the VFX industry and he is a supremely gifted individual that's been working in VFX since the dawn of time and he is not satisfied with the status quo ever he is always looking to tweak and tune and mess with his applications and squeeze as much as possible out of his community hi isn't it insane it is he's bringing in his machine so he works at a pretty big studio they do a lot of really cool stuff you've heard of the stuff that they've worked on it's legit and they've got some older machines he's bringing in a machine that was last upgraded like five years ago some of the parts are even older than that some of the parts were like x79 all right let's build behind here so nobody can see anything he fit we've already done all the testing of course we've tested the $3,000 w31 775 X which technically by extension I guess means that we've tested the Mac Pro because the Mac Pro is the new socket it's a different CPU it's actually a slower CPU than this 28 core monster yeah we've also tested through first second and third generation we'll talk a little bit more about that so your old workstation had for nine 80s mm-hmm and a six core X I think it's a nexus 99 next time mounted CPUs like this Explorer whatever that model is i75 780 I don't make any sense it was from the time when pretty much all of the VFX stuff was just running on GPU only and now we've got a good mix of CP and GPU so six cores doesn't do it but it's also a really nice custom loop it was never cleaned maybe for almost ten years he's bringing in a machine that was last upgraded like five years ago some of the parts are even older than that some of the parts are like x79 but it's a full tower Corsair monster case dual radiator it's got four GTX 980 s in them they're using these for rendering and they basically gave him one because it died and the reason it died we found out because the radiators in chunk we're trying to resuscitate this thing it's almost like somebody neglected the crap out of it I was given this as a gift and it was a terrible gift but in the best way we spent how long cleaning it oh and I already used a bunch of vinegar and the blocks are full of junk and the couple of the video cards literally died because one of them has bad vram because of the heat two of them won't post and the other one is clock and really weird so for GPUs dead we we tried to reuse some fittings that was perhaps a mistake yeah started leaking that radiator sprung a leak required a little bit of soldering something something adventures it was the quick release fittings the quick release fittings they were suspect I said there were suspects from the beginning but it was like hey I won't probably be let's use them let's how does it stay yeah the quick release is not quick releasing well that's why you got new ones but luckily we caught it you were able to get it switched out without there being a complete amount soon inside but I'm here which was a huge concern for a moment I should point out too that not doing maintenance is not Justin's fault they were his work machines and it was like nobody told them they had to do maintenance on a loop I mean I feel like somebody should have known that yeah you for soldered but you've also got a quick-release now you didn't have as a fort in the secret door on the back when it comes time for cleaning or whatever you can literally just just and it's a new quick release we didn't recycle this one so the idea of this extra tubing is we come out and around and loop over and back in some of this extra space but this quick release here will make it a lot easier to work on the pump because all your returned water you can just use the pump to basically pump everything out of the system and then gravity will do the rest because you can just lower that below the like you can like just drop it in the floor and then your loop will be better before we had like the main artery running up here but we're actually using that just the the main output now to run directly into the CPU we've got some odd choices in the build like these are old server drives but they're four terabytes that's uh that's gonna be interesting you know it's also amazing I'm pretty sure this is just a cd-rom drive in here and not even a CD burner I just wanted to call that out it's nasty you got a bunch more of these machines at work which means that your studio is probably gonna buy a lot of thread Ripper you're welcome AMD mmm the other thing is PCI Express storage and so we had a unique opportunity we had these four four tire about the Intel nvme drives which were available for less than $100 a terabyte and these are not like PCI Express for SSDs are gonna be faster at least initially but these don't slow down these are triple level cells these are engineered to last a lifetime and they still have two years two and a half years of Intel warranty left and knew they were like three thousand dollars and you can still buy these on new egg about $2,000 and I could be honest with you I'm still pretty new level when it comes to PC building and when you mentioned that it was just like hitting me in the face and just going around like I had no idea what you were talking about but we're not just going into this willy-nilly because I'm some sort of crazy thread Ripper evangelist whatever although it may be that's true to you I don't know so I did a lot of homework before Justin actually got here doing what's called well what I would call probably application profiling so looking at his programs a lot of them you probably never heard of but one of them that you probably have heard of is Adobe After Effects but it'll be After Effects on its own really isn't that interesting it's the plugins that go in After Effects the after effects plugins they can use CUDA acceleration some of them use a lot of CPUs some of them don't after effects on its own it doesn't render fast it just doesn't you have to use something else to accelerate it because it's super multi-threaded with After Effects you really want to render one frame at a time but you can take a sequence that's you know 600 frames long and break that up into jobs and that's what we did with red ripper to test the 24 and 32 core thread Ripper well we tend to do which is a little bit more unique you know differentiating it from other professionals is our workload is a three to four minute all CG piece so we needed a solution that gave us the ability to render that out faster because Trinita After Effects you're looking at like hour and a half two hours I mean there have been times we'd hit the render button it would say like eight hours of this and it got out of control and render garden was the first opportunity that we had to speed that process up render garden is the name of the software that does that now render Gardens website advertises a two to three X speed up but by systematically looking at where the bottlenecks word memory i/o vram because when you've run out of VRAM things get shuffled from your GPU to main memory now that might be something that can be optimized in the driver I'm not sure I actually reached out to Nvidia about this but for this test setup for the system that we're using now we are just using a GTX 1080 Ti but we found the shelf but would three minutes is what you can do the gtx ten atti with about five to ten instances of render garden depending on how complex your After Effects project is and how much CUDA your After Effects project needs and it depends it's it is very very project specific so it depends on the plugins you're using and everything else well now I remember us working on some really old projects way back in a day and there was a checkbox in After Effects which has rendered multiple frames at once that's gone that was gone before almost before I left so that's how long it's been like kind of gone so we've got a we've got a we've got a system to build but everything is in balance in this system based on your specific workload like your specific projects I mean with render garden that 50 minute job immediately was like 20 minutes on thread Ripper and then with render garden it was like what three and a half yeah and we never really found out what the what the chef was on oh we found the Shelf in later testing time down to a minute and 22 seconds with 16 instances of render garden with a GPU that had 24 gigs of VRAM that GPU the titan r-tx this is not always going to be the case like it depends on your project and the plugins that you're using I mean even a mid-grade GPU may be sufficient for doing stuff you might remember some of the testing that we did with a post Vox DaVinci Resolve scaled perfectly to two GPUs yet in that third GPU you weren't really getting scaling after more analysis it turns out that more scaling meant that we needed more CPUs when we did that testing thread Ripper 3000 wasn't available but now da Vinci with three GPUs actually does scale a little better what a day it was a day it only rained like five times that last one was really bad it was pretty intense well the crazy thing is that you know as you know with your old workstations it was super important I mean you've got a three hundred and no it's like a 420 and a 480 millimeter radiator or something crazy it's like triple 140 s up here and quadruple 120 s down there and and right now it's all going to the CPU yeah nuts but we routed a cable so you'll be able to insert your GPUs between the CPU goes directly to the top radiator and the top radiator goes to the bottom radiator now but you'll be able to insert your GPUs when you get your blocks just with this cable so this cable runs down you'll be able to put a quick release in that and you know one or two GPUs Titan RT X's but for now 20 atti but the thing that blows my mind is we did the tests and on this loop with the old 980 s this system is running circles around for 980 s if we're not ATS really they're not a slouch but it just goes to show you if you don't do maintenance on your loop you're gonna have worse performance than even probably a single 980 yeah it was pretty much hobbled and this is the the 24 core just imagine - 32 core and 64 core came and fathom through what all did we run through crystal disk mark yeah crystal disk mark but he also did octane I did octane bench did crystal diss mark and then we did the new Puget bench which is still in beta but still gave me some good numbers compared to any machine that we had ring was it payments it was like over three thousand yeah it was it a three 3800 or something like that somewhere in that range and the next best machine that you got at work which is not super old was like six hundred a 1600 1600 something like that yeah so around that range I mean it's invaded and those numbers might be flush motive but still well cinnamon char 20 right well yeah the new $56,000 Mac Pro the best score you're gonna get on the Mac Pro is like 9600 let's call it 10 K will be generous and say 10k this thing out of the box was 14 K stock stock I love working on real-world projects like this it's so it's great and there are so few people out there that will actually sit down and look at what your problem is it's like let's just keep throwing hardware at it until the problem is solved it's like no it's not necessarily a case I mean with these kinds of projects especially when you're talking about like VFX rendering the bottom that could be anywhere it could be GPU it's usually not could be memory storage could be software like your software it's like okay this effect that you're doing let's just bake that into the render and then re-import that footage because every time you do anything with his footage its recalculating that this is never changing it doesn't need to be done that way this is a workflow problem this is not really a problem with your hardware or anything else but it's also interesting because the VFX industry has changed dramatically at least in terms of hardware the last four or five years and one thing that I keep realizing is that I probably ought to be doing a lot more rendering and working at the same time which I kind of have completely like gave up on that reality well you can't do that oh no what now you can it sounds like I probably can and so I'm not the kind of change that or my part of my design invades I think we want to see in a year to the you know in groups of artists that are working together probably will be able to do the network rendering and just run it at a low priority with no ill effects that would be awesome so it's like during the work day when you know bob needs to use your machine for rendering if you're not otherwise rendering something all the machines in the work group are rendering but if you've got a job you need to render it you know your machine's gonna render in the background as well as everybody else's and if everybody's got something irregular in the background everybody's gonna get background rendering but the software for that is is pretty much there but it hasn't really like like you people trying to do that workflow found that it didn't work but now I can so like what's going on on the screen over here where you're at I have a really really crazy heavy scene with a lot of like layers and glows and plugins and different like something that's like basically really heavy high raster I want to test that file because that's gonna give me all the information I need cuz I'm really good at running a computer down pretty quickly and this is a really good you know work case example so it looks pretty fluid to me yeah what this was this the file on your other computer that was like two fps yeah it is a huge ever-changing ecosystem and it is very difficult to say yes this computer is always going to run really well for this particular application I think that's probably true that when your old machine was built or so GPU heavy it was true that GPU acceleration was all rage and that works really well but now it's a mix of CPU and GPU I forget that most have not no bottleneck in your workflow you have to have an insane CPU and and then same GPU set up and that you can even do this with the you know relatively consumer-grade 2080 Ti is just kind of bananas right the next big thing honestly is probably hybrid memory where the CPU can access the GPUs memory more directly with less overhead and vice versa so thank you again Justin for joining me thank you for any messages for anybody in internet land don't recycle wrap don't try to recycle read and don't try to siphon a hose when it's clean I like I like the taste of my head you can smell it it smells like we're making pickles thread report is so insane it's such an incredible like it's not just thread Ripper it's like trigger 4 plus it's really unlocked the GPU horsepower it's just out of I don't know I'm gonna stop rambling out I'm Windell this is level 1 signing out I'll see you later it's like a butt flat yeah it's a buff flap to make cleaning easier then it's a terrible mental picture absolutely terrible that's all about terrible mental pictures\n"