1070 vs 980ti

[Devin92]

First of all, I am pretty new to GPU usage, so sorry if I am asking nonsense questions.

 

So my question is one of my friend is considering between 1070 and 980ti. He will be using this GPU for machine learning and AI development. Hence using the cuda and parallel computing power. I know 980ti have more cuda cores, but i think 1070 can boost to faster speed. Is the speed gonna affect cuda core performance or not?

And what is comparison of pure computing power between the two other than gaming?

 

[AresKrieger]

Well pascal has better performance in an deep learning applications so if he were to use it in those sort of applications the 1070 should be better, generally the 980ti is a little better for cuda utilization but as I said if the use case is ai the 1070 should perform significantly better.

[Lukas1337]

The 980TI has much better performance and a better price.

[DocSwag]

1 minute ago, Lukas1337 said:

The 980TI has much better performance and a better price.

Better price? Much better performance?

 

I'm pretty sure that's not the case...

[Enderman]

4 minutes ago, DocSwag said:

Better price? Much better performance?

 

I'm pretty sure that's not the case...

The 980ti has more cores which offers a bigger improvement in parallel processing power than higher clock speeds.

That's why the 980ti performs better than the 1070 in certain rendering tests and benchmarks.

 

It also costs about $300-350 which is cheaper than most 1070s going for $400.

[Lukas1337]

I live in Sweden and I can get my hands on a GTX980TI for $456 vs the 1070 which costs $585 and the 980ti performs better in real world applications eg games.

[scottyseng]

12 minutes ago, Devin92 said:

First of all, I am pretty new to GPU usage, so sorry if I am asking nonsense questions.

 

So my question is one of my friend is considering between 1070 and 980ti. He will be using this GPU for machine learning and AI development. Hence using the cuda and parallel computing power. I know 980ti have more cuda cores, but i think 1070 can boost to faster speed. Is the speed gonna affect cuda core performance or not?

And what is comparison of pure computing power between the two other than gaming?

 

They're pretty much neck and neck, but the 1070 is more power efficient. I myself went with a used 980 Ti because it was $100 cheaper than a new 1070 at the time (Even if the 1070 is 5-10% faster), it can't cover the cost of being 1/3 more (Got the 980 Ti used for $340...new 1070 are floating at 430ish). It's very tough because the used market is very up and down since everyone's selling their 9 series GPU for a 10 series.

[AresKrieger]

3 minutes ago, DocSwag said:

snip

Especially if were talking about ai performance, considering that was one of the focuses of pascal besides gaming

 

Though price tends to be cheaper

Just now, Enderman said:

The 980ti has more cores which offers a bigger improvement in parallel processing power than higher clock speeds.

That's why the 980ti performs better than the 1070 in certain rendering tests and benchmarks.

 

It also costs about $300-350 which is cheaper than most 1070s going for $400.

It wouldn't be stronger in ai development though, most other use cases yes, but ai is one of pascals specialties

[Enderman]

Just now, AresKrieger said:

Though price tends to be cheaper

It wouldn't be stronger in ai development though, most other use cases yes, but ai is one of pascals specialties

AI and deep learning is heavily multithreaded. That's why people use GPUs for that, not CPUs.

GPUs have thousands of cores you can use, and the 980ti has more than the 2070, which is why it is better.

[AresKrieger]

Just now, Enderman said:

AI and deep learning is heavily multithreaded. That's why people use GPUs for that, not CPUs.

GPUs have thousands of cores you can use, and the 980ti has more than the 2070, which is why it is better.

I would agree if nvidia didn't design the architecture in a manner to boost deep learning applications, unless all the data Nvidia gave on the subject is inaccurate I don't see how the 980ti could win in this specific use case. (otherwise I would agree with you)

[Enderman]

Just now, AresKrieger said:

I would agree if nvidia didn't design the architecture in a manner to boost deep learning applications, unless all the data Nvidia gave on the subject is inaccurate I don't see how the 980ti could win in this specific use case. (otherwise I would agree with you)

Just look at benchmarks of the 980ti vs 1070 that aren't games.

In rendering and stuff like that which is parallel processing, the 980ti comes out on top.

[AresKrieger]

4 minutes ago, Enderman said:

Just look at benchmarks of the 980ti vs 1070 that aren't games.

In rendering and stuff like that which is parallel processing, the 980ti comes out on top.

Running Caffe AlexNet as a deep learning benchmark, the GTX 1070 performance was right around the speed of a GTX 980 Ti and GTX TITAN X.

[mlalcevski]

here is a comparison base on user submitet benchamrks http://gpu.userbenchmark.com/Compare/Nvidia-GTX-980-Ti-vs-Nvidia-GTX-1070/3439vs3609.

i say go for the 1070 since it is cheaper and you can add another down the road for a much lower price. also note the 1070 is more power efficent there for cooler :).

[Devin92]

39 minutes ago, AresKrieger said:

Running Caffe AlexNet as a deep learning benchmark, the GTX 1070 performance was right around the speed of a GTX 980 Ti and GTX TITAN X.

Can you explain a bit more about this benchmark or provide a link of describtion?

[AresKrieger]

3 minutes ago, Devin92 said:

Can you explain a bit more about this benchmark or provide a link of describtion?

It's just a couple of deep learning related benchmarks I could find through reddit, as deep learning and ai are essentially the same thing I was just showing enderman in an ai use case the 980ti doesn't have an advantage.

 

Ai is a very niche use case so finding a lot of information on it is proving difficult

 

Anyway this is where I found those benches

https://www.reddit.com/r/nvidia/comments/4s0h07/gtx_980ti_1060_1070_or_1080_for_deep_learning/

[xAcid9]

1070, because Deep Learning is the only thing that came out of Huang's mouth when he revealed Pascal arch late last year.

[Devin92]

16 minutes ago, AresKrieger said:

It's just a couple of deep learning related benchmarks I could find through reddit, as deep learning and ai are essentially the same thing I was just showing enderman in an ai use case the 980ti doesn't have an advantage.

 

Ai is a very niche use case so finding a lot of information on it is proving difficult

 

Anyway this is where I found those benches

https://www.reddit.com/r/nvidia/comments/4s0h07/gtx_980ti_1060_1070_or_1080_for_deep_learning/

okay thx! I will past this info to him. pretty useful thread you found