Interesting results, however they are largely academical. 20% reduced CPU FPS is very little compared to 400% resolution increase from 720p to 1440p, but it's good to know that some games have a difference at all. Cities Skylines is a notoriously poorly optimisd game, it even has needlessly well modeled teeth for NPCs, I suspect NPCs are rendered with higher LOD in higher resolution. You can use Reshade to get statistics on draw calls and vertexes in scenes, that'd also help you compare extreme resolutions (4K, 8K via DSR/VSR) without worrying about GPU bottleneck.
Teeth were fixed, it no longer renders them. I think the problem with this game’s optimization is there because of sim based simulation model where each individual som has some basic intelligence and pathfinding model. This is why it gets really laggy really quickly
Historically, the CPU has handled the initial construction of BVH structures because it already manages much of the scene data required for rasterized rendering. Ray tracing introduces additional work, as the CPU must also prepare scene data in BVH format to hand off to the GPU. In some cases, CPUs can still help manage updates when objects in a scene move, rotate, or change frequently, though this may add to the CPU’s workload. Modern APIs like DirectX Raytracing (DXR) and Vulkan Ray Tracing allow the GPU to handle BVH construction at runtime, which is far more efficient for large, complex scenes. In DXR, the GPU builds two BVH levels: 1. Bottom-Level Acceleration Structure (BLAS): Contains detailed geometry and procedural primitives, optimized for fast ray intersections, and 2. Top-Level Acceleration Structure (TLAS): Refers to instances of BLAS structures, allowing reuse of geometry for flexibility and performance. The TLAS is faster to build than the BLAS, which can be advantageous when updating scenes dynamically. To avoid performance drops, NVIDIA recommends keeping BLAS structures separate, with minimal overlap. Vulkan Ray Tracing offers developers more flexibility by allowing them to build the TLAS on either the CPU or GPU, though this isn’t the default behavior. This option lets developers leverage any idle CPU resources, but if latency becomes a concern, they can stick to the GPU. Most GPU manufacturers, including NVIDIA and Intel, are moving towards fully hardware-accelerated BVH construction, aiming to handle this on the GPU as much as possible. This trend reduces the dependency on the CPU-GPU link, improving performance and leaving the CPU free for other tasks. While the CPU still plays a role in updating BVH data, its workload is lighter as more construction is offloaded to the GPU. In Summary 1. BVH construction is shifting from CPU to GPU for efficiency. 2. DXR relies heavily on GPU for runtime BVH building, especially for BLAS. 3. Vulkan Ray Tracing allows the TLAS to be built on either the CPU or GPU. 4. Future direction is fully hardware-accelerated BVH, minimizing CPU load. 5. With the GPU taking on more of the heavy lifting, ray tracing is becoming more practical for real-time graphics, opening up new possibilities for dynamic and visually complex scenes without taxing the CPU.
even FOV can have a huge impact in some cases, but nowadays all those problems are mostly "fixed" and the impact is limited, that's why is not surprise seen that in Crysis even if is a bad port of a bad port, the PC original has its quirks too
I went through 25 years of balancing game set ups on PC, and I went through focusing on GPU first, then I focused on getting CPU first, because I realized I shorted on CPU side, and eventually I realized that you probably should get a great CPU and then match the GPU to it, according to your resolution choice
Higher res tasks the gpu more doesnt lower cpu usage tho only if the gpu is limiting it. Infact it might task the cpu more too if the game switches resources for higher quality ones
The main mistake is to assume that "CPU fps" and "GPU fps" are two completely independent things. And what do you even imply by "perfromance"? CPU load or frame time? If it's a CPU frame time, then it's no surpise, CPU and GPU frame time have never been a completely independent things. If it's a CPU load, then there's no a single clean reasonable test that shows the CPU load raise in this video. If we are talking about CPU load, then the question is which thread is getting more work? The main game thread? User-mode driver thread? Kernel-mode driver thread? The CPU load might increase, but frame time might stay the same as long the main thread stays unaffected. Or frametime might change only because CPU clocks went down due to multiple core load. There's no a single test in this video showing what, when and where is changing with resolution raise. Because of that talking about why resolution affects "CPU performance" is just a random guessing.
Wish there was more research done on why and how games get cpu limited, from memory and frequency speeds to cache size. I think lowering resolution and settings to be cpu limited as possible only tells us half the story.
Using an AMD set up for this test methodology as Ryzen CPU zen2 and later are really cache heavy which is what gave them such a large uplift gen over gen and closed the gap to intel, for this reason a CPU like a 10400f would be optimal if you wanted to demo this as it has basically no L3 cache and would have to reach out to memory more frequently causing slowdown and in some cases stutter. For the GPU you would need probably a 4070 or 3070ti to minimum to demo this and the core reason for this is that AMD GPU are semi autonomous, they don't rely on CPU caching or CPU draw calling to generate the frame data which is why they have typically tight 1% lows as they remove the external factors while Nvidia GPU's can and will murder low end CPU's with driver overhead.
Interesting fact about Cyberpunk 2077 - it's also CPU bound on Ultra/PT settings, cards like 4090 will not be fully utilized even with fastest CPU's today.
Yes, we have another video about Ray Tracing impact on CPU performance as well. It doesn’t only happen in Cyberpunk, it happens in plenty of games, especially those built on Unreal Engine 4 th-cam.com/video/JAotTICLst8/w-d-xo.htmlsi=uJP1GUYBMx9-46qR
i currently have an 8600k|3060ti. My games stutter quite often and my load times are pretty bad. Would a new CPU help me get more frames or just smoother experience?
It depends. It may be related to a bad ssd that’s slow when filled up or an hdd (if you’re using one). Also you should check the background tasks, maybe you have something sketchy running. Those 6 cores i5s are not best but not as bad as you’ve described
Get an SSD if you don’t have one already. If I were you I’d upgrade the CPU in a year… the 3060ti should serve you just fine at 1080p until next generation consoles come out and we see vram utilization go above 8gb at 1080p
8600k and 3070 owner here. I delayed my CPU upgrade by at least a year by overclocking the CPU to 5.0ghz at 1.35V (I got a DeepCool Assassin IV, CPU stays at less than 60°c) To check where your bottleneck is, use MSI afterburner to monitor usage, temps and whichever component is maxing out is the problem. (In my case, in most games, all CPU cores were at 100% with zero or one 1%Fps Low, after the OC it went down to 75-90% usage and I get better 1% lows)
Dude, your CPU is pretty slow for new generation of games, I had 5600x before it was not enough, you should look for something like ryzen 7500f or better it will be good vs that you have, if you have higher budget get x3d.
A lot of stuff is software limited. In a perfect scenario there is no overhead but that is almost never the case. Vulkan is a great example where the API used can make a big difference
@ Yea some games did take advantage but I noticed most of the games I play like High clock frequency speeds. Rocket League is one I play a lot and unfortunately it does not take advantage of X3D 😔
I did a test too at warzone2: ryzen 5600x make it undercooked to 3.6 all cores as idle and full load I did not allow it to get flexible Mhz, then played 1080p native lowest graphics (but texture resulotion and texture filter at normal) then I got below 70 to 80% GPU usage limited by CPU at then moved to 1440p got GPU usage at 85 to 96% but less fps becuze of CPU too be too slow. It was a test something I was wondering if higher res effect CPU usage and fps and it did on my test. Ofc I do not run normally 3.6Ghz on 5600x, was clocked at 4.85Ghz until it got unstable after a year and I decided to use PBO for 1 month untill I got my hands on 5800X3D and undervolt it by 0.012mV with -30 all cores scale with increased fps and loot and my rtx 3070 was enough so I decided to get 6950xt.
@@HardwareLab Well they also lose performance when you raise the settings. So would you consider GTA 4 and Saints Row 2 well optimized because they do the same?
@@HardwareLab Shame you can't download common sense because a lot of people need it. If you believe an optimization being poorly implemented has nothing to do with all the other optimizations that have been poorly implemented then I simply can't help you.
@@FatheredPuma81 wow wow wow, take it easy man haha. I only said that if the game cpu dependency scales with resolution it has nothing to do with optimization level. That’s it, nothing else
I know I'm asking a lot, but I'd love to see the Intel Ultra 265K/285K vs 7900X3D using 1440p/4K and 8K and see what happens. As far as I know, 1440p+ on any high end CPU right now the FPS is roughly the same.
FPS is not “the same on CPUs” at 4K and 8K. It’s just that you can’t become CPU limited in 8K because of how much strain is there on GPU, check the explanation in the beginning of the video and our video about bottleneck to learn more about it
Go check Hardware Unboxed, 9800x3d is 52% higher fps vs 285k in 1080p, even at 1440p diffrance will be over 20%. Even at 4k in some games will be noticeable
@@HardwareLab 4:20 I don't see it? Was it temporary or only once? It's the type of thing you want on there all-through referencing or using another's content
Interesting results, however they are largely academical. 20% reduced CPU FPS is very little compared to 400% resolution increase from 720p to 1440p, but it's good to know that some games have a difference at all.
Cities Skylines is a notoriously poorly optimisd game, it even has needlessly well modeled teeth for NPCs, I suspect NPCs are rendered with higher LOD in higher resolution.
You can use Reshade to get statistics on draw calls and vertexes in scenes, that'd also help you compare extreme resolutions (4K, 8K via DSR/VSR) without worrying about GPU bottleneck.
Teeth were fixed, it no longer renders them. I think the problem with this game’s optimization is there because of sim based simulation model where each individual som has some basic intelligence and pathfinding model. This is why it gets really laggy really quickly
I wonder if ray tracing affects cpu performance with different resolutions
Haha that’s an oddly specific test. I think it shouldn’t unless the distance or quality for bvh content is no related to resolution
Ray tracing has it's own separate resolutions independent of the rendering of the output signal. More precisely....less rays are cast
Historically, the CPU has handled the initial construction of BVH structures because it already manages much of the scene data required for rasterized rendering. Ray tracing introduces additional work, as the CPU must also prepare scene data in BVH format to hand off to the GPU. In some cases, CPUs can still help manage updates when objects in a scene move, rotate, or change frequently, though this may add to the CPU’s workload.
Modern APIs like DirectX Raytracing (DXR) and Vulkan Ray Tracing allow the GPU to handle BVH construction at runtime, which is far more efficient for large, complex scenes. In DXR, the GPU builds two BVH levels: 1. Bottom-Level Acceleration Structure (BLAS): Contains detailed geometry and procedural primitives, optimized for fast ray intersections, and 2. Top-Level Acceleration Structure (TLAS): Refers to instances of BLAS structures, allowing reuse of geometry for flexibility and performance.
The TLAS is faster to build than the BLAS, which can be advantageous when updating scenes dynamically. To avoid performance drops, NVIDIA recommends keeping BLAS structures separate, with minimal overlap.
Vulkan Ray Tracing offers developers more flexibility by allowing them to build the TLAS on either the CPU or GPU, though this isn’t the default behavior. This option lets developers leverage any idle CPU resources, but if latency becomes a concern, they can stick to the GPU.
Most GPU manufacturers, including NVIDIA and Intel, are moving towards fully hardware-accelerated BVH construction, aiming to handle this on the GPU as much as possible. This trend reduces the dependency on the CPU-GPU link, improving performance and leaving the CPU free for other tasks. While the CPU still plays a role in updating BVH data, its workload is lighter as more construction is offloaded to the GPU.
In Summary
1. BVH construction is shifting from CPU to GPU for efficiency.
2. DXR relies heavily on GPU for runtime BVH building, especially for BLAS.
3. Vulkan Ray Tracing allows the TLAS to be built on either the CPU or GPU.
4. Future direction is fully hardware-accelerated BVH, minimizing CPU load.
5. With the GPU taking on more of the heavy lifting, ray tracing is becoming more practical for real-time graphics, opening up new possibilities for dynamic and visually complex scenes without taxing the CPU.
Ray tracing especially path tracing is the key thing no one talk about. Maybe lack of games
@@ksks6802 tay tracing resolution is usually related to render resolution. It is even measured in “½ of screen res” or “¼ of screen res”
even FOV can have a huge impact in some cases, but nowadays all those problems are mostly "fixed" and the impact is limited, that's why is not surprise seen that in Crysis even if is a bad port of a bad port, the PC original has its quirks too
FOV does have an impact obviously because the amount of objects in the frame with higher FOV is higher. It shouldn’t be “fixed”, it’s not a problem
I went through 25 years of balancing game set ups on PC, and I went through focusing on GPU first, then I focused on getting CPU first, because I realized I shorted on CPU side, and eventually I realized that you probably should get a great CPU and then match the GPU to it, according to your resolution choice
All the CPU reviewers test at 1080p max settings, and when asked why not even lower resolution I recall some mention of complications at
Higher res tasks the gpu more doesnt lower cpu usage tho only if the gpu is limiting it.
Infact it might task the cpu more too if the game switches resources for higher quality ones
The main mistake is to assume that "CPU fps" and "GPU fps" are two completely independent things. And what do you even imply by "perfromance"? CPU load or frame time? If it's a CPU frame time, then it's no surpise, CPU and GPU frame time have never been a completely independent things. If it's a CPU load, then there's no a single clean reasonable test that shows the CPU load raise in this video. If we are talking about CPU load, then the question is which thread is getting more work? The main game thread? User-mode driver thread? Kernel-mode driver thread? The CPU load might increase, but frame time might stay the same as long the main thread stays unaffected. Or frametime might change only because CPU clocks went down due to multiple core load. There's no a single test in this video showing what, when and where is changing with resolution raise. Because of that talking about why resolution affects "CPU performance" is just a random guessing.
Wish there was more research done on why and how games get cpu limited, from memory and frequency speeds to cache size. I think lowering resolution and settings to be cpu limited as possible only tells us half the story.
Using an AMD set up for this test methodology as Ryzen CPU zen2 and later are really cache heavy which is what gave them such a large uplift gen over gen and closed the gap to intel, for this reason a CPU like a 10400f would be optimal if you wanted to demo this as it has basically no L3 cache and would have to reach out to memory more frequently causing slowdown and in some cases stutter.
For the GPU you would need probably a 4070 or 3070ti to minimum to demo this and the core reason for this is that AMD GPU are semi autonomous, they don't rely on CPU caching or CPU draw calling to generate the frame data which is why they have typically tight 1% lows as they remove the external factors while Nvidia GPU's can and will murder low end CPU's with driver overhead.
Interesting fact about Cyberpunk 2077 - it's also CPU bound on Ultra/PT settings, cards like 4090 will not be fully utilized even with fastest CPU's today.
Yes, we have another video about Ray Tracing impact on CPU performance as well. It doesn’t only happen in Cyberpunk, it happens in plenty of games, especially those built on Unreal Engine 4 th-cam.com/video/JAotTICLst8/w-d-xo.htmlsi=uJP1GUYBMx9-46qR
i currently have an 8600k|3060ti. My games stutter quite often and my load times are pretty bad. Would a new CPU help me get more frames or just smoother experience?
It depends. It may be related to a bad ssd that’s slow when filled up or an hdd (if you’re using one). Also you should check the background tasks, maybe you have something sketchy running. Those 6 cores i5s are not best but not as bad as you’ve described
Get an SSD if you don’t have one already. If I were you I’d upgrade the CPU in a year… the 3060ti should serve you just fine at 1080p until next generation consoles come out and we see vram utilization go above 8gb at 1080p
Thanks guys!
8600k and 3070 owner here.
I delayed my CPU upgrade by at least a year by overclocking the CPU to 5.0ghz at 1.35V (I got a DeepCool Assassin IV, CPU stays at less than 60°c)
To check where your bottleneck is, use MSI afterburner to monitor usage, temps and whichever component is maxing out is the problem.
(In my case, in most games, all CPU cores were at 100% with zero or one 1%Fps Low, after the OC it went down to 75-90% usage and I get better 1% lows)
Dude, your CPU is pretty slow for new generation of games, I had 5600x before it was not enough, you should look for something like ryzen 7500f or better it will be good vs that you have, if you have higher budget get x3d.
A lot of stuff is software limited. In a perfect scenario there is no overhead but that is almost never the case. Vulkan is a great example where the API used can make a big difference
I swapped back to the 7700X because X3D doesn’t really benefit with AD102
definitely benefits if you play CPU intensive titles, particularly with ray tracing
@ Yea some games did take advantage but I noticed most of the games I play like High clock frequency speeds. Rocket League is one I play a lot and unfortunately it does not take advantage of X3D 😔
I play bo6,cs6,mmo and x3d did huge diff for me.
@@Kapono5150Rocket league is one of the exceptions where X3D doesn’t perform outstandingly good but it’s just one game
I did a test too at warzone2: ryzen 5600x make it undercooked to 3.6 all cores as idle and full load I did not allow it to get flexible Mhz, then played 1080p native lowest graphics (but texture resulotion and texture filter at normal) then I got below 70 to 80% GPU usage limited by CPU at then moved to 1440p got GPU usage at 85 to 96% but less fps becuze of CPU too be too slow.
It was a test something I was wondering if higher res effect CPU usage and fps and it did on my test.
Ofc I do not run normally 3.6Ghz on 5600x, was clocked at 4.85Ghz until it got unstable after a year and I decided to use PBO for 1 month untill I got my hands on 5800X3D and undervolt it by 0.012mV with -30 all cores scale with increased fps and loot and my rtx 3070 was enough so I decided to get 6950xt.
Cities Skylines 2 is such an unoptimized mess that it's not really surprising so much performance is being held back by the resolution.
It has nothing to do with optimization. Would you consider Forza Horizon 5 on release date or Talos Principle 2 unoptimized?
@@HardwareLab Well they also lose performance when you raise the settings. So would you consider GTA 4 and Saints Row 2 well optimized because they do the same?
@@FatheredPuma81 no, I’m saying there’s no relation
@@HardwareLab Shame you can't download common sense because a lot of people need it. If you believe an optimization being poorly implemented has nothing to do with all the other optimizations that have been poorly implemented then I simply can't help you.
@@FatheredPuma81 wow wow wow, take it easy man haha. I only said that if the game cpu dependency scales with resolution it has nothing to do with optimization level. That’s it, nothing else
I know I'm asking a lot, but I'd love to see the Intel Ultra 265K/285K vs 7900X3D using 1440p/4K and 8K and see what happens.
As far as I know, 1440p+ on any high end CPU right now the FPS is roughly the same.
FPS is not “the same on CPUs” at 4K and 8K. It’s just that you can’t become CPU limited in 8K because of how much strain is there on GPU, check the explanation in the beginning of the video and our video about bottleneck to learn more about it
Go check Hardware Unboxed, 9800x3d is 52% higher fps vs 285k in 1080p, even at 1440p diffrance will be over 20%.
Even at 4k in some games will be noticeable
Please visually credit people like digital foundry; as otherwise people may not go to them for their work if they like what they see/hear.
We did. Maybe next time it should be larger, but the text bar is there
@@HardwareLab 4:20 I don't see it? Was it temporary or only once? It's the type of thing you want on there all-through referencing or using another's content
Checked; yeah it was very brief is why i never even noticed until i did want to look for it and saw none
Yes, my bad, next time will make it larger and longer
Introducing DF @3:11 followed by Alex’s voiceover it’s quite obvious and seems nitpicky imo I think HL did an excellent job