Why isn't GTX 980 TI 37% faster than GTX 980, when it has 37% more CUDA cores?

I just got my hands on a GTX 980 TI, and when I run the BMW27_gpu.blend test, with it’s default settings I get a render time of 03:25 (205 seconds). By setting the tile size to 960x540 (100% of the image), i got 03:11 (191 seconds).

On my other machine with a GTX 980, i get 3:46 (226 seconds). With 960x540 tile size I get 3:56 so it was worse…

Since GTX 980 TI has 2816 CUDA cores, and GTX 980 has 2048 CUDA cores, I would assume the TI would be somwhere near 37% faster. And rendering would be done in 02:45 (165 seconds). Also it has more memory, higher memory bandwith, although lower clock speed. Gaming benchmarks seem to be about 30% faster.

Does anyone get a better result with their GTX 980 TI, or is it normal that the difference is so small?

Also I only tried the BWM benchmark, maybe there are bigger differences in other kind of renders?

Here’s a table comparing the specs:

[TABLE=“class: center text-90, width: 957”]
[TR=“class: yellow-background”]

Size of Power Supply **
Memory Type


Base Clock

Boost Clock


GTX 980
500 watt
256 bit
224 GB/s
1126 MHz
1216 MHz

GTX 980 Ti
600 watt
384 bit
336.5 GB/s
1000 MHz
1075 MHz


Do you use Windows 10?

Try higher clockspeed on the 980ti or lower clockspeed on the 980

@MZGarmi: Ah, interesting. I’m using Windows 7 on the GTX 980 ti and Win 10 on GTX 980. Is rendering times faster in Win 10?

@skw: About Amdahl’s law, I see there’s a bit of processing before it starts to render, is that what you mean, that isn’t faster regardless of the GPU?

@bigbad: I never tried overclocking, is there any risk to it? Why would I try lower on the 980, how would that help? :slight_smile:

more power = more heat = less lifetime

Because your 980Ti is ~11% slower than the 980 for clockspeed, so you would have to take that from the extra CUDA cores.

I get 21.5% faster if I adjust for the CUDA core and clockspeed difference.

This isn’t including the memory speed/bandwidth difference, however when rendering I’ve always noticed that the CUDA cores are at 100%, whilst the memory is not fully used, so I’m assuming that the memory may not be a bottleneck here.

No,quite the opposite. Windows 10 is slower for GPU based rendering. It is because the whole UI for the system is GPU accelerated. Great for normal use, Horrible for professionals. Not to mention that Windows 10 also uses up a rediculous amount of vram based off the total amount of vram, not a set amount.

The other factors you need to factor in, is Pre and Post Processing times, these are usually all CPU bound so if you have a faster cpu / slower cpu it will effect the results.

TL;DR Windows 10 is horrible

Thanks, then I won’t update this machine and just live with the fact that 980ti isn’t extremely faster, it’s still an awesome card for rendering I think… :slight_smile: