Blender 2.90 Performance With GeForce RTX 3080, 18-Way NVIDIA CUDA/OptiX Comparison

[phoronix]

Phoronix: Blender 2.90 Performance With GeForce RTX 3080, 18-Way NVIDIA CUDA/OptiX Comparison

Complementing yesterday's large GPU compute comparison with the GeForce RTX 3080 across many different workloads, today's article is looking at the Blender 2.90 render performance for this consumer Ampere $699+ graphics card.

http://www.phoronix.com/vr.php?view=29575

[qarium]

Phoronix.com has become a place for permanent advertisement for monopoles and closed standards no one but Nvidia is allowed to perform CUDA and OptiX
we watch pointless benchmarks with nvidia only no competition is allowed.

i have to agree from technical point of view 3080/3090 is great hardware but the monopole politics from this company is a absolute no go.

anyone who supports this is just a slave. yes you can become a slave by free choice you can also burn your body with gasoline wile you are alive.

i say we do Vulkan Compute as fast as possible or we can burn our future with gasoline.

[schmidtbag]

Phoronix.com has become a place for permanent advertisement for monopoles and closed standards no one but Nvidia is allowed to perform CUDA and OptiX
we watch pointless benchmarks with nvidia only no competition is allowed.

i have to agree from technical point of view 3080/3090 is great hardware but the monopole politics from this company is a absolute no go.

anyone who supports this is just a slave. yes you can become a slave by free choice you can also burn your body with gasoline wile you are alive.

i say we do Vulkan Compute as fast as possible or we can burn our future with gasoline.

Oh quit your whining. Michael always does several batches of tests when he gets new hardware, because he likes to be thorough and test as many use-cases as possible. Since Nvidia has more features like CUDA, Optix, DLSS, and (for now) raytracing, anyone who is trying to prove Nvidia's marketing is going to have more work to do.

Even if that weren't the case, you're still wrong. As far as I'm aware, Nvidia is not stopping anyone else from using CUDA, it's just that it's rather difficult for a company like AMD to write the drivers for an entire platform that they have limited architectural documentation on and is purpose-built for a different piece of hardware. So even if Nvidia was like "here ya go AMD, take all the source code to make your own drivers" AMD is still at a disadvantage by being years behind in getting something properly optimized. And even if AMD were given all the time in the world, they don't have purpose-built hardware like tensor cores. It is not within the interest of Nvidia's competitors to support CUDA.

AMD, Intel, Khronos, and others had an opportunity to beat CUDA but they all just sat back and did nothing. The reason CUDA is popular even among open-source platforms like Blender or Meshroom is because Nvidia actually cared to make it an appealing product. It's relatively easy to use, it's powerful, and there are plenty of very well-made resources. Have you tried to learn OpenCL, or better yet, attempt to use it with any language that isn't C++? Because trust me, it's not easy. An amateur programmer can write a basic CUDA program with no GPU programming background within a couple of days.

As for being a slave, don't be so dramatic. It's a goddamn API that nobody needs to have. So tell me - how long can you go without a search engine? How long can you go without a refrigerator? How long can you go without your phone? These are things people are often tethered to.


Come back when you have a real problem to gripe about.

[milkylainen]

Will it blend? (With a Blendtec, of course).

[Klassic Six]

Michael, please do a comparison with AMD GPUs.

[qarium]

Michael, please do a comparison with AMD GPUs.

how?... its not possible CUDA is nvidia only... also optix...

[andre30correia]

Phoronix.com has become a place for permanent advertisement for monopoles and closed standards no one but Nvidia is allowed to perform CUDA and OptiX
we watch pointless benchmarks with nvidia only no competition is allowed.

i have to agree from technical point of view 3080/3090 is great hardware but the monopole politics from this company is a absolute no go.

anyone who supports this is just a slave. yes you can become a slave by free choice you can also burn your body with gasoline wile you are alive.

i say we do Vulkan Compute as fast as possible or we can burn our future with gasoline.

well if the Michael only do some things because they are open we lose a lot, nvidia is the best company out there making gpu and gpu drivers, when Intel (Michael put some oneapi benchmarks some days ago) and amd put some decent drivers and techs for pros to use maybe he can put more benchmarks about them but the nvidia have a lot of things the others don't have

[piotrj3]

Michael, please do a comparison with AMD GPUs.

comparison with AMD GPUs nowadays would be AMD opencl vs Nvidia Optix (for RTX) and CUDA for non RTX. In both cases AMD would be eaten alive and such comparison would do AMD more harm then good. And OpenCL vs OpenCL benchmark is pointless since no one uses OpenCL in blender (for Nvidia users).

[oleid]

how?... its not possible CUDA is nvidia only... also optix...

Well, int theory it should be possible to port the Cuda backend via AMD's HIP to the ROCm platform. HIP code can be compiled to Cuda or ROCm. Not sure if anybody tried, yet.

https://github.com/ROCm-Developer-Tools/HIP/blob/master/docs/markdown/hip_porting_guide.md#scanning-existing-cuda-code-to-scope-the-porting-effort

[lowflyer]

Phoronix.com has become a place for permanent advertisement for monopoles and closed standards no one but Nvidia is allowed to perform CUDA and OptiX
we watch pointless benchmarks with nvidia only no competition is allowed.

i have to agree from technical point of view 3080/3090 is great hardware but the monopole politics from this company is a absolute no go.

anyone who supports this is just a slave. yes you can become a slave by free choice you can also burn your body with gasoline wile you are alive.

i say we do Vulkan Compute as fast as possible or we can burn our future with gasoline.

qarium, full support here. :+1: :thumbsup:
Let the others burn themselves.