LTT should include AI benchmarks in their reviews of GPUs

[AusCyberman]

i'm only just getting into AI but i would really appreciate if LTT included AI training of models in their GPU comparisons. For example, training a Transformer model or Computer Vision model and comparing the times between the GPUs. Testing performance between using Tensor Cores and no Tensor Cores. This would really help people that use AI or just people in general to see how affective Tensor Cores are.

 

Thanks

[Bombastinator]

43 minutes ago, AusCyberman said:

i'm only just getting into AI but i would really appreciate if LTT included AI training of models in their GPU comparisons. For example, training a Transformer model or Computer Vision model and comparing the times between the GPUs. Testing performance between using Tensor Cores and no Tensor Cores. This would really help people that use AI or just people in general to see how affective Tensor Cores are.

 

Thanks

How?  Different apps would work in different ways.  Would have to be ( and might already be) included in some benchmark or other.

[BTGbullseye]

13 hours ago, AusCyberman said:

i'm only just getting into AI but i would really appreciate if LTT included AI training of models in their GPU comparisons. For example, training a Transformer model or Computer Vision model and comparing the times between the GPUs. Testing performance between using Tensor Cores and no Tensor Cores. This would really help people that use AI or just people in general to see how affective Tensor Cores are.

 

Thanks

Considering howw niche that is, even among the niche that is the people that care about GPUs and also watch Linus, it's probably worth at most a single video. It would still be interesting to see.

[Mark Kaine]

Yeah, but only if Kizuna AI does the tests!

 

Linus goes anime, I personally can't wait!

 

[AusCyberman]

On 9/27/2020 at 8:04 PM, Bombastinator said:

How?  Different apps would work in different ways.  Would have to be ( and might already be) included in some benchmark or other.

i doubt that normal benchmarks include AI benchmarks. A PyTorch or Tensorflow example model could be run using Tensor cores  and compare the training times

[AusCyberman]

14 hours ago, BTGbullseye said:

Considering howw niche that is, even among the niche that is the people that care about GPUs and also watch Linus, it's probably worth at most a single video. It would still be interesting to see.

it is pretty niche but also a clear defined difference and advantage that tensor cores have. and the difference. like a lot of people would be interested

[Bombastinator]

57 minutes ago, AusCyberman said:

it is pretty niche but also a clear defined difference and advantage that tensor cores have. and the difference. like a lot of people would be interested

A lot must be measured in percentages of viewership.  If “a lot” is a tiny percentage it would hurt viewership rather than help it.  If “a lot” is a large percentage it would possibly help though.  One would assume there are other you tubers doing these kinds of tests.  If there is not, perhaps there is a market niche. How big of one I don’t know.  If it’s too small it might explain why there isn’t one.