The Current Radeon RX Vega 64 / Radeon VII Linux OpenCL Performance Against NVIDIA

[bridgman]

Yeah, some of the results are really odd. I'd expect latency to be lower than any kind of non-HBM solution - but its higher. Either I have a wrong assumption, or something is broken.

Latency in this case does not refer to memory latency but rather to kernel launch, ie the combination of SW code paths and HW responses when an app has loaded a shader program along with all the source data then triggers a "dispatch" command to the GPU to execute an instance of the shader program (aka kernel program) on every element in the destination array.

[nuetzel]

@Michael

Stay tuned for some fresh tests with ROCm 2.5 as soon as I'm back in the office.

ROCm 2.6

[hotaru]

how do those Nvidia cards do with open source drivers? that would be a much more interesting comparison.

[max0x7ba]

These synthetic benchmarks are pretty much meaningless

Benchmark the latest TensorFlow: https://www.tensorflow.org/guide/performance/benchmarks

[schmidtbag]

The only place where you're seeing AMD beating Nvidia is on your memory bandwidth benchmarks, where obviously HBM2 is going to beat GDDR6.

I don't disagree that most of these tests aren't very meaningful, but, you are so obviously butthurt and biased that you didn't even realize the few cases where Nvidia is crushing AMD in memory tests...

[dylo]

Summary please and if i need great gpu for video editing with kdenlive, what is better for me, amd or nvidia?

[vegabook]

I don't disagree that most of these tests aren't very meaningful, but, you are so obviously butthurt and biased that you didn't even realize the few cases where Nvidia is crushing AMD in memory tests...

Ooooh it's schmidtbag again! I won't go into answering your vacant point because there's nothing there to comment on other than insults, though had you expanded a bit more with your usual idiocy I might have had the opportunity to crush you again!

[schmidtbag]

Ooooh it's schmidtbag again! I won't go into answering your vacant point because there's nothing there to comment on other than insults, though had you expanded a bit more with your usual idiocy I might have had the opportunity to crush you again!

Dude... it's been nearly 4 months. If you haven't been able to "crush" me now, you won't ever (that being said, not sure what you meant by "again", seeing as I don't recall you ever doing it before). If there was so little to comment on about my point, then why did you? And what is there to expand on? Just look at the graphs in the article; no point in me elaborating when Michael did the work for me.
You insist you're right, but I'm calling your bluff.

[vegabook]

Dude... it's been nearly 4 months. If you haven't been able to "crush" me now, you won't ever (that being said, not sure what you meant by "again", seeing as I don't recall you ever doing it before). If there was so little to comment on about my point, then why did you? And what is there to expand on? Just look at the graphs in the article; no point in me elaborating when Michael did the work for me.
You insist you're right, but I'm calling your bluff.

You're betraying your poor grasp of computer science again, not to mention a generally tragic lack of intelligence, sadly and all too predictably characterised by a negative correlation with the amount of eye-rolling verbiage you emit.

O(higher order than N) algos don't care about a few percentage points of perf, even 50%. But orders of magnitude matter.

https://techgage.com/article/testing...l-performance/

Now take a good look at that, let it sink in over the next few months. haha

[schmidtbag]

You're betraying your poor grasp of computer science again, not to mention a generally tragic lack of intelligence, sadly and all too predictably characterised by a negative correlation with the amount of eye-rolling verbiage you emit.

Lots of words with no substance at all. Elaborate, please.

O(higher order than N) algos don't care about a few percentage points of perf, even 50%. But orders of magnitude matter.[/qoute]

Not sure what your point is here. Where am I disagreeing with that?