3 ms·
I don’t think it’s a bargain, but based on price to CUDA core counts alone, it actually is fairly priced. It’s just that their product line has a more or less
by andrewmcwatters 4y ago
I don’t think it’s a bargain, but based on price to CUDA core counts alone, it actually is fairly priced.
It’s just that their product line has a more or less linear CUDA core to price best line of fit.[1]
There’s a dip specifically at the RTX 3080 Ti, where you’re getting more cores for the money than elsewhere on the best line of fit.
Not sure what trade offs there are there except less VRAM and in one instance I read that cooling wasn’t as good.
[1]: https://docs.google.com/spreadsheets/d/1MHTVDAyU7i8oXYAlmGPEGNs8rUpukT_eXlizQfXZ2JI/edit?usp=sharing https://docs.google.com/spreadsheets/d/1MHTVDAyU7i8oXYAlmGPE...
- pixelpoet 4y agoIt's important to note that the definition of CUDA "cores" changed with Ampere due to their adding dual-issue, which is not always possible, but they count it as a flat doubling of "cores", where "cores" previously meant ALUs / vector lane units. There's also been quite a bit of inflation since the days of the 1660 Super...
- andrewmcwatters 4y agoThanks for that, I didn't know!
- BonoboIO 4y agoIs very cuda core as powerful as very other cuda core? So 10x 4090 cuda cores are as powerful as 10x 2060 cores? A graph that plots price per core and power usage per core would be awesome. Google Sheets is so amazing, it just works and I can watch how somebody edits the document and collaboration is dimensions ahead of every service I know.
- andrewmcwatters 4y agoThere does seem to be a close association when comparing GPU benchmark scores.[1] [1]: https://www.videocardbenchmark.net/high_end_gpus.html https://www.videocardbenchmark.net/high_end_gpus.html