Programming across architectures

Hi nVidia!

I am supporting scientists at the university i work at and i have a question:
Is the coding language for Matlab (or just the cards) the same for a RTX 2080TI and a Titan V card? (I notice that the architectures are different on the cards). I am asking, since the scientists wants to test CUDA programming, before purchasing a much more expensive card.
I am asking since i have googled this and found no results, but i can see, that the architecture is different
Also, are we able to use a 2080ti for larger datasets?

The Titan V is a datacenter/scientist-oriented card, while RTX20xx are the consumer-level cards. You will have to fully read the documentation of each to identify what they have in common, what is a full instruction/feature set in Titan V and what is a subset in RTX, and what is not present in RTX at all.

Rule of thumb is that you can expect stuff to perform faster on Titan V than on 2080Ti (which doesn’t mean the 2080 is slow by any means). One notable example is the memory bus of the 2080Ti, 352-bit, versus the 3072-bit on the Titan, which is ridiculously large. I have access to Titan V at work, and coding on a 1080Ti at home, I can tell you the thing is serious. I wish I had a Titan V at home but can’t justify the cost. Some guys here invested on a RTX and can provide better feedback on them.

Regarding large datasets, 2080Ti has 11GB while Titan V has 12GB, so you have similar room and if more memory is needed, then you have to look at the Teslas or multi-GPU setups.

The moderators and some very experienced contributors will provide much deeper feedback on recommendation.

Hi Saulocpp!

Thanks for your reply, much appreciated!

Is the coding language the same? Or is there some change in Syntax?

Syntax is exactly the same accross cards, and to some extent is also backwards compatible with older families. If you go to the CUDA home page, you will see that multiple languages are supported:

However, keep in mind that newer families might (and generally do) have instructions not present in previous generations, or work at reduced performance between cards of the same generation. Volta and Turing are good examples. There are discussions about it going on here.
That’s why it is important for your guys to read the full documentation of these cards so they know what each of them has to offer and what they need.