NVIDIA GeForce GTX 980 Through GeForce RTX 5080/5090 GPU Compute Performance

Written by Michael Larabel in Graphics Cards on 17 February 2025 at 09:18 AM EST. Page 5 of 6. 90 Comments.

Simply put, there are massive gains in performance and power efficiency when comparing Maxwell to Blackwell for Linux GPU compute.

👁 ProjectPhysX OpenCL-Benchmark benchmark with settings of Operation: FP64 Compute. RTX 5090 was the fastest.

👁 ProjectPhysX OpenCL-Benchmark benchmark with settings of Operation: FP16 Compute. RTX 5090 was the fastest.

👁 ProjectPhysX OpenCL-Benchmark benchmark with settings of Operation: INT8 Compute. RTX 5090 was the fastest.

👁 ProjectPhysX OpenCL-Benchmark benchmark with settings of Operation: Memory Bandwidth Coalesced Write. RTX 5090 was the fastest.

A fun round of Linux GPU benchmarking in the books.

👁 Llama.cpp benchmark with settings of Backend: NVIDIA CUDA, Model: Llama-3.1-Tulu-3-8B-Q8_0, Test: Text Generation 128. RTX 5090 was the fastest.

I was only able to get Llama.cpp running on the NVIDIA GPUs back to the GeForce GTX TITAN X, but even there is an interesting 6.4x performance difference from the GTX TITAN X to RTX 5090.

👁 Llama.cpp benchmark with settings of Backend: NVIDIA CUDA, Model: Llama-3.1-Tulu-3-8B-Q8_0, Test: Prompt Processing 2048. RTX 5090 was the fastest.

Or with Llama/cpp prompt processing, going from the GeForce GTX TITAN X to RTX 5090 was 39.8x the performance!

90 Comments - Next Page

Page: 1 2 3 4 5 6 Next Page

URL: https://www.phoronix.com/review/nvidia-maxwell-to-blackwell/5

⇱ NVIDIA GeForce GTX 980 Through GeForce RTX 5080/5090 GPU Compute Performance Review - Phoronix

NVIDIA GeForce GTX 980 Through GeForce RTX 5080/5090 GPU Compute Performance