nvidia RTX 5080 rtx 5090 compute benchmarks

Benchmarks for a future article.

HTML result view exported from: https://openbenchmarking.org/result/2501295-PTS-NVIDIART22&grr.

NCNN

Target: Vulkan GPU - Model: FastestDet

NCNN

Target: Vulkan GPU - Model: vision_transformer

NCNN

Target: Vulkan GPU - Model: regnety_400m

NCNN

Target: Vulkan GPU - Model: squeezenet_ssd

NCNN

Target: Vulkan GPU - Model: yolov4-tiny

NCNN

Target: Vulkan GPUv2-yolov3v2-yolov3 - Model: mobilenetv2-yolov3

NCNN

Target: Vulkan GPU - Model: resnet50

NCNN

Target: Vulkan GPU - Model: alexnet

NCNN

Target: Vulkan GPU - Model: resnet18

NCNN

Target: Vulkan GPU - Model: vgg16

NCNN

Target: Vulkan GPU - Model: googlenet

NCNN

Target: Vulkan GPU - Model: blazeface

NCNN

Target: Vulkan GPU - Model: efficientnet-b0

NCNN

Target: Vulkan GPU - Model: mnasnet

NCNN

Target: Vulkan GPU - Model: shufflenet-v2

NCNN

Target: Vulkan GPU-v3-v3 - Model: mobilenet-v3

NCNN

Target: Vulkan GPU-v2-v2 - Model: mobilenet-v2

NCNN

Target: Vulkan GPU - Model: mobilenet

Llama.cpp

Backend: NVIDIA CUDA - Model: Llama-3.1-Tulu-3-8B-Q8_0 - Test: Text Generation 128

Llama.cpp

Backend: NVIDIA CUDA - Model: Mistral-7B-Instruct-v0.3-Q8_0 - Test: Text Generation 128

Llama.cpp

Backend: NVIDIA CUDA - Model: granite-3.0-3b-a800m-instruct-Q8_0 - Test: Text Generation 128

Chaos Group V-RAY

Mode: NVIDIA CUDA GPU

Chaos Group V-RAY

Mode: NVIDIA RTX GPU

IndigoBench

Acceleration: OpenCL GPU - Scene: Bedroom

IndigoBench

Acceleration: OpenCL GPU - Scene: Supercar

NAMD CUDA

ATPase Simulation - 327,506 Atoms

Blender

Blend File: Barbershop - Compute: NVIDIA CUDA

Llama.cpp

Backend: NVIDIA CUDA - Model: granite-3.0-3b-a800m-instruct-Q8_0 - Test: Prompt Processing 2048

VkResample

Upscale: 2x - Precision: Double

Blender

Blend File: Barbershop - Compute: NVIDIA OptiX

FluidX3D

Test: FP32-FP32

Blender

Blend File: Pabellon Barcelona - Compute: NVIDIA CUDA

Llama.cpp

Backend: NVIDIA CUDA - Model: Llama-3.1-Tulu-3-8B-Q8_0 - Test: Prompt Processing 2048

clpeak

OpenCL Test: Transfer Bandwidth enqueueReadBuffer

clpeak

OpenCL Test: Transfer Bandwidth enqueueWriteBuffer

Llama.cpp

Backend: NVIDIA CUDA - Model: Mistral-7B-Instruct-v0.3-Q8_0 - Test: Prompt Processing 2048

Llama.cpp

Backend: NVIDIA CUDA - Model: granite-3.0-3b-a800m-instruct-Q8_0 - Test: Prompt Processing 1024

RealSR-NCNN

Scale: 4x - TAA: Yes

NAMD CUDA

Input: STMV with 1,066,628 Atoms

FluidX3D

Test: FP32-FP16S

FluidX3D

Test: FP32-FP16C

Blender

Blend File: Fishy Cat - Compute: NVIDIA CUDA

Llama.cpp

Backend: NVIDIA CUDA - Model: Llama-3.1-Tulu-3-8B-Q8_0 - Test: Prompt Processing 1024

Blender

Blend File: Junkshop - Compute: NVIDIA CUDA

Blender

Blend File: Classroom - Compute: NVIDIA CUDA

Llama.cpp

Backend: NVIDIA CUDA - Model: Mistral-7B-Instruct-v0.3-Q8_0 - Test: Prompt Processing 1024

Hashcat

Benchmark: MD5

Llama.cpp

Backend: NVIDIA CUDA - Model: granite-3.0-3b-a800m-instruct-Q8_0 - Test: Prompt Processing 512

Blender

Blend File: Pabellon Barcelona - Compute: NVIDIA OptiX

Hashcat

Benchmark: SHA-512

Hashcat

Benchmark: SHA1

Blender

Blend File: Classroom - Compute: NVIDIA OptiX

ProjectPhysX OpenCL-Benchmark

Operation: Memory Bandwidth Coalesced Write

ProjectPhysX OpenCL-Benchmark

Operation: Memory Bandwidth Coalesced Read

ProjectPhysX OpenCL-Benchmark

Operation: INT8 Compute

ProjectPhysX OpenCL-Benchmark

Operation: INT16 Compute

ProjectPhysX OpenCL-Benchmark

Operation: INT32 Compute

ProjectPhysX OpenCL-Benchmark

Operation: INT64 Compute

ProjectPhysX OpenCL-Benchmark

Operation: FP16 Compute

ProjectPhysX OpenCL-Benchmark

Operation: FP32 Compute

ProjectPhysX OpenCL-Benchmark

Operation: FP64 Compute

VkResample

Upscale: 2x - Precision: Single

clpeak

OpenCL Test: Double-Precision Compute

Blender

Blend File: Junkshop - Compute: NVIDIA OptiX

Hashcat

Benchmark: 7-Zip

Blender

Blend File: BMW27 - Compute: NVIDIA CUDA

Blender

Blend File: Fishy Cat - Compute: NVIDIA OptiX

Llama.cpp

Backend: NVIDIA CUDA - Model: Llama-3.1-Tulu-3-8B-Q8_0 - Test: Prompt Processing 512

Llama.cpp

Backend: NVIDIA CUDA - Model: Mistral-7B-Instruct-v0.3-Q8_0 - Test: Prompt Processing 512

SHOC Scalable HeterOgeneous Computing

Target: OpenCL - Benchmark: Texture Read Bandwidth

Hashcat

Benchmark: TrueCrypt RIPEMD160 + XTS

NAMD CUDA

Input: ATPase with 327,506 Atoms

RealSR-NCNN

Scale: 4x - TAA: No

Blender

Blend File: BMW27 - Compute: NVIDIA OptiX

Waifu2x-NCNN Vulkan

Scale: 2x - Denoise: 3 - TAA: Yes

clpeak

OpenCL Test: Global Memory Bandwidth

SHOC Scalable HeterOgeneous Computing

Target: OpenCL - Benchmark: FFT SP

SHOC Scalable HeterOgeneous Computing

Target: OpenCL - Benchmark: GEMM SGEMM_N

SHOC Scalable HeterOgeneous Computing

Target: OpenCL - Benchmark: Triad

SHOC Scalable HeterOgeneous Computing

Target: OpenCL - Benchmark: Bus Speed Readback

SHOC Scalable HeterOgeneous Computing

Target: OpenCL - Benchmark: Bus Speed Download

SHOC Scalable HeterOgeneous Computing

Target: OpenCL - Benchmark: S3D

SHOC Scalable HeterOgeneous Computing

Target: OpenCL - Benchmark: Reduction

clpeak

OpenCL Test: Single-Precision Compute

clpeak

OpenCL Test: Integer Compute

clpeak

OpenCL Test: Integer 24-bit Compute

SHOC Scalable HeterOgeneous Computing

Target: OpenCL - Benchmark: MD5 Hash

clpeak

OpenCL Test: Kernel Latency

Phoronix Test Suite v10.8.5