Will the i5 6500 bottleneck the gtx 1070

[Midniteb0ne]

I have a PC with an i5 6500 and RX480 8gb rn. I want to upgrade my GPU to a gtx 1070.

I will play on a 1080p 144hz monitor, I mostly play eSports and some indie games but I also want to play some AAA games like Battlefield V.

Would the upgrade be worth it and will it be better if I upgrade to an RTX2060.

PS If I upgrade to the RTX2060 I'll be using the tensor cores for Machine learning.

[Jurrunio]

Even the 1070 will bottleneck more often than not, let alone the 2060. How much does the 1070 cost you though? Getting a overboard GPU and upgrade the CPU later is also an option.

[191x7]

GTX 1660Ti?

[martward]

16 minutes ago, Midniteb0ne said:

I have a PC with an i5 6500 and RX480 8gb rn. I want to upgrade my GPU to a gtx 1070.

I will play on a 1080p 144hz monitor, I mostly play eSports and some indie games but I also want to play some AAA games like Battlefield V.

Would the upgrade be worth it and will it be better if I upgrade to an RTX2060.

PS If I upgrade to the RTX2060 I'll be using the tensor cores for Machine learning.

I would get either the 2060 or the 1660Ti, unless you're looking at a used 1070. It kind of depends on your budget which one is the better choice and whether or not you are at all interested in Raytracing and DLSS. You don't need tensorcores for Machine Learning, GTX series cards are already way faster than your CPU. Also I'm not sure whether existing libraries already support using the tensorcores on the new cards, so I would also advice you to look into the library you plan to use and to check GPU support.

[Midniteb0ne]

1 hour ago, Jurrunio said:

Even the 1070 will bottleneck more often than not, let alone the 2060. How much does the 1070 cost you though? Getting a overboard GPU and upgrade the CPU later is also an option.

the 1070 is 230$ (used) the 1660ti is 310$ (new) and the rtx 2060 is 380$ (new)

[Settlerteo]

1 hour ago, martward said:

GTX series cards are already way faster than your CPU

LoL what? it's the other way around in fact and in a couple of years i5 6500 will hold it's ground in comparison to anything below 1070ti which will definitely bottleneck the i5.

[Teddy07]

1 hour ago, martward said:

I would get either the 2060 or the 1660Ti, unless you're looking at a used 1070.

1

Stop hating the 2060, please. It is a fine card.

[martward]

17 minutes ago, Teddy07 said:

Stop hating the 2060, please. It is a fine card.

What are you talking about? I own a 2060, I know it's a fine card and I wasn't hating on it...

I was just saying that it depends on budget and usage whether it makes sense to buy the most expensive option or go for a cheaper option.

[martward]

27 minutes ago, Settlerteo said:

LoL what? it's the other way around in fact and in a couple of years i5 6500 will hold it's ground in comparison to anything below 1070ti which will definitely bottleneck the i5.

I was talking about Machine Learning performance. As in, training a neural network on pretty much any decent GPU is faster than training a neural network on an i5. I know this because I own an i5, a Xeon E3 1231v3, a RTX 2060 and have previously used a GTX 1060 which I used for ML, the GPUs are faster.

If I wasn't talking about ML my sentence wouldn't even have made sense since with other application the CPU and GPU do completely different things...

[Settlerteo]

3 minutes ago, martward said:

I was talking about Machine Learning performance. As in, training a neural network on pretty much any decent GPU is faster than training a neural network on an i5. I know this because I own an i5, a Xeon E3 1231v3, a RTX 2060 and have previously used a GTX 1060 which I used for ML, the GPUs are faster.

If I wasn't talking about ML my sentence wouldn't even have made sense since with other application the CPU and GPU do completely different things...

If you have defined that you were talking about ML then i am sorry i might have missed that. That makes sense. I thought when you made the comparison you were talking about bottleneck because they definitely are different things and can't be compared.

[Settlerteo]

8 minutes ago, martward said:

I was talking about Machine Learning performance

And in terms of Machine Learning a 780 ti/980 ti would be much better than any 1060/1070/2060 card.

[martward]

1 minute ago, Settlerteo said:

And in terms of Machine Learning a 780 ti/980 ti would be much better than any 1060/1070/2060 card.

Do you have any sources for that? I have read an article where the 2060 has very impressive results, though it's only tested with TensorFlow and using images.

https://www.phoronix.com/scan.php?page=article&item=nvidia-rtx2060-linux&num=8

[Settlerteo]

12 minutes ago, martward said:

Do you have any sources for that? I have read an article where the 2060 has very impressive results, though it's only tested with TensorFlow and using images.

https://www.phoronix.com/scan.php?page=article&item=nvidia-rtx2060-linux&num=8

I stated that due to cuda cores. I found this article https://gist.github.com/cavinsmith/ed92fee35d44ef91e09eaa8775e3284e although it does not include any rtx cards. Really don't know if i can post a url like that here, so if i am doing it wrong please correct me and i will edit my post.

[NineEyeRon]

I know the best bottleneck test, you have to buy the card thought...

 

Do you get the FPS you want in the games you play at your desired resolution and detail level?

 

If “yes” then bottleneck is not an issue.

 

If “no” then bottleneck is an issue.

 

Then, if issues...

 

Can you lower detail to achieve framerates wanted and still deliver the experience you want?

 

If “yes” then no issue.

 

if “no” then spend money to buy better components.

[Settlerteo]

2 minutes ago, NineEyeRon said:

I know the best bottleneck test, you have to buy the card thought...

 

I made some changes so that debugging is easier.  

 

Do you get the FPS you want in the games you play at your desired resolution and detail level?

Case

When “yes” Then bottleneck is not an issue.

When “no” Then bottleneck is an issue.

Else ...

END;

UNION

Can you lower detail to achieve framerates wanted and still deliver the experience you want?

Case

When “yes” Then no issue.

Else “no” then spend money to buy better components.

END;

[NineEyeRon]

Just now, Settlerteo said:

I made some changes so that debugging is easier.  

 

Do you get the FPS you want in the games you play at your desired resolution and detail level?

Case

When “yes” Then bottleneck is not an issue.

When “no” Then bottleneck is an issue.

Else ...

END;

UNION

Can you lower detail to achieve framerates wanted and still deliver the experience you want?

Case

When “yes” Then no issue.

Else “no” then spend money to buy better components.

END;

Now make it IoT enabled.

[martward]

45 minutes ago, Settlerteo said:

I stated that due to cuda cores. I found this article https://gist.github.com/cavinsmith/ed92fee35d44ef91e09eaa8775e3284e although it does not include any rtx cards. Really don't know if i can post a url like that here, so if i am doing it wrong please correct me and i will edit my post.

Just counting cuda cores is not really a valid way of benchmarking performance though. I mean the 660Ti has more cuda cores than the 1060, but I'm willing to bet that in games as well as machine learning the 1060 is going to be significantly faster. 

[NineEyeRon]

My point regarding bottlenecks is there are two sorts of end users for a system.

 

 1. Benchmarking enthusiasts who need to eek out every point

 

2. Experience focussed user, who has an experience in mind and wants to deliver that for their use cases.

 

Yes you can be both and so can a system and it is a scale between but assessing a bottleneck you are one or the other. Either you get the experience you wanted so the bottleneck isn’t an issue OR you miss out on benchmark points due to one component holding another back.

 

Even a 10-15% bottleneck of a top end graphics card can still make the card value for money if it delivers the experience required at the framerates and resolution. Yes it might not be the best value for money but dropping down to a lower card may tip you the wrong side of that experience or not provide enough buffer for high demand situations.

[Settlerteo]

10 minutes ago, martward said:

Just counting cuda cores is not really a valid way of benchmarking performance though. I mean the 660Ti has more cuda cores than the 1060, but I'm willing to bet that in games as well as machine learning the 1060 is going to be significantly faster. 

Sure cause it's a very old card right now. But a 780 ti which is almost on par in terms of general performance with a 1060 depending to the drivers also, and with that much more cuda cores than the 1060 would be a better choice for ML. 

[martward]

7 minutes ago, Settlerteo said:

Sure cause it's a very old card right now. But a 780 ti which is almost on par in terms of general performance with a 1060 depending to the drivers also, and with that much more cuda cores than the 1060 would be a better choice for ML. 

No it wouldn't...

1) 780Ti only has 3gb of memory

2) 780Ti has no tensor cores

3) 780Ti has half the clock speed

4) 780Ti has lower compute performance

5) 2060 can use FP16 to boost performance

 

I have had at least one network that exceeded 4GBs, that would be a problem with only 3GBs of VRAM. Have you even taken a look at the link I shared?

Or how about this link?

https://towardsdatascience.com/rtx-2060-vs-gtx-1080ti-in-deep-learning-gpu-benchmarks-cheapest-rtx-vs-most-expensive-gtx-card-cd47cd9931d2

 

[Settlerteo]

3 minutes ago, martward said:

No it wouldn't...

Yup i was wrong about the rtx. Due to tensor cores and fp16 yeah the rtx card would probably be better.

[Jurrunio]

6 hours ago, Midniteb0ne said:

the 1070 is 230$ (used) the 1660ti is 310$ (new) and the rtx 2060 is 380$ (new)

Go 2060 if you can, it's much better in machine learning and is also the fastest in games.