[OC] Nvidia’s consumer GeForce GPUs: CUDA cores & VRAM capacity by model & generation

5 comments

1. I wanted to compare the difference between number of CUDA cores and VRAM memory in each of Nvidia’s consumer GPU (GeForce) offerings.

   Graph generated using Python and matplotlib:

   * The left Y-axis is number of CUDA cores (bar graph with different colors for each series).
   * The right Y-axis is VRAM memory (dashed blue line).
   * Some models have two options for VRAM capacity, those are represented by the greyish bars (the line chart represents the average for these models.) Unfortunately for the 3080, this bar intersects with the CUDA count bar, but for the other models with VRAM variance, they show up a bit more clearly.

   Numbers are from Nvidia’s official website:

   * 50 series [https://www.nvidia.com/en-us/geforce/graphics-cards/compare/](https://www.nvidia.com/en-us/geforce/graphics-cards/compare/)
   * 40 series [https://www.nvidia.com/en-us/geforce/graphics-cards/40-series/](https://www.nvidia.com/en-us/geforce/graphics-cards/40-series/)
   * 30 series [https://www.nvidia.com/en-us/geforce/graphics-cards/30-series/](https://www.nvidia.com/en-us/geforce/graphics-cards/30-series/)
   * 20 series [https://www.nvidia.com/en-us/geforce/graphics-cards/compare/?section=compare-20](https://www.nvidia.com/en-us/geforce/graphics-cards/compare/?section=compare-20)
   * 10 series [https://www.nvidia.com/en-us/geforce/10-series/](https://www.nvidia.com/en-us/geforce/10-series/)

2. Ordering them by release date would probably make more sense

3. CUDA core-MHz would make more sense than just cores alone. (e.g. an RTX 2080 running at 1710 MHz would have 5,034,240 CUDA-MHz)

   We care about processing performance, not the amount of parallelism possible.

4. you are missing 4070 super (7168 Cuda cores and 12GB VRAM)

5. The 3090 were 24gb of vram weren’t they?

Comments are closed.