NVIDIA GeForce GTX 1050 Ti Linux Benchmarks

[johnc]

You're a complete idiot guy.

...Why?

[smitty3268]

Now this test is really interesting. With only 2.3 TFLOPS vs. 4.9 TFLOPS for the RX 470 the results show how bad the current AMDGPU drivers still perform on Linux in several games. David vs. Goliath.

Comparing TFLOPS like that isn't valid across different architectures, the same way comparing GHz on CPUs just leads you to bad conclusions.

[smitty3268]

Micahel, I assume you are copy/pasting parts of the text between articles, because for the last week or two they've all had the same typo.

You are testing the Mesa drivers with LLVM 4.0, not LLVM 3.0. (I assume)

[dungeon]

Comparing TFLOPS like that isn't valid across different architectures, the same way comparing GHz on CPUs just leads you to bad conclusions.

Why isn't valid, if i count it right 2.3 is 47% of 4.9, so...

And, if we... let say that Xonotic is first, one and only GL app on planet Earth we like to take as example ... then results are right there:

So, 79 fps on GTX 1050 is about 50% out of RX 470's 159 fps In whole, it is less then 3% difference in TFLOPS vs Xonotic perf comparasion of those two

[dungeon]

It is joke math, but in Xonotic example math somehow nearly exactly match

[smitty3268]

Why isn't valid, if i count it right 2.3 is 47% of 4.9, so...

And, if we... let say that Xonotic is first, one and only GL app on planet Earth we like to take as example ... then results are right there:

So, 79 fps on GTX 1050 is about 50% out of RX 470's 159 fps In whole, it is less then 3% difference in TFLOPS vs Xonotic perf comparasion of those two

Theoretically it could be if it was representative of real-world performance.

Numerous tests you can see everywhere (on windows and linux, and multiple drivers) show that it isn't.

It's a theoretical max that can be hit in certain perfect conditions. Standard gaming is not one of those scenarios. Typically it's only hit in certain restricted compute tests.

If you were going to draw anything from the TFLOP numbers, it should be that the hardware architectural efficiency of NVidia cards is higher than that of AMD. Not that the AMD drivers suck. (That might be part of it too, but it's impossible to pull out from the underlying architecture differences which appear to be a much greater influence.)

[dungeon]

Theoretically it could be if it was representative of real-world performance.

Xonotic is one of representatives of real-world perfromance.

Numerous tests you can see everywhere (on windows and linux, and multiple drivers) show that it isn't.

But in Xonotic it is exactly there, so it might be

I don't claim it is, but somehow numbers are exactly there

[dungeon]

It's a theoretical max that can be hit in certain perfect conditions.

So certain condition seems happened in imperfect Xonotic

[smitty3268]

Xonotic is one of representatives of real-world perfromance.

Exactly. It's the exception that proves the rule.

To be fair, things might be different in Vulkan/DX12. I haven't checked out how well perf there lines up with the TFLOP values. Perhaps async shaders and other tech lets the GPU be fully utilized.

[dungeon]

He, he, TFLOPs can't line up when you have game on top of it some special feature... let say you fire up HDR and MRS in Shadow Warrior 2

Now Playstation 4 Pro will do MSR too... AMD bought HiAlgo this year, so they are there with a feature... so might end up even as hw engine i guess

Those kind of features improve performance, so you implement HyperZ - voila large boost... you implement threaded GL in opensource driver - voila, large boost here and there... you implement MSR and voila large boost here and there...

nVidia peobably have some algo in their driver which artifically boost perf... it is not stright clean API which boost something for sure, otherwise even nouveau will already have same perf as blob