| Scisca said: Well. No. That's not how AMD GPUs work. Compare Fury X and Fury, which are somewhat similar to Vega. 4096 CUs vs 3584 CUs (14% difference), 640 BG/s vs 512 GB/s bandwidth (25% difference), 5% less fillrate - yet after setting the same clocks, the difference in performance is about 3%. Compare the RX 470 and RX 480, or regular R9 cards and the "+" cards - same story. AMD GPUs don't scale so well when close to the top performance. Unless they intentionally held the memory back, expect Vega 56 to be only a hair behind Vega 64. |
Well no. I don't think you comprehend how GPU's work.
If you are limited in another part of the pipeline, then of course the difference will be reduced as that would be your bottleneck.
The Fury and Fury X had the same ROP capabilities and had the same Geometry capabilities... And was extremely limited in VRAM.
Which means... If the game was ROP, Geometry or Memory intensive, then performance would equalise.
But throw a compute-only task that only stressed the shaders, then you will see performance between the two start to spread apart.
Vega however has the same Shader count as Fury, but dials up the clockrate, has overhauled ROPS, Geometry capabilities, doubling of RAM and the introduction of new techniques which further increases memory bandwidth and Geometry/ROP throughput. It's not going to be apples to apples.
In short. It's a difference beast and we should wait on benchmarks from the likes of AMD and then possibly revisit this discussion if you so desire.
| Chazore said: |
You still don't get it. - I don't care about Ancedotals.
If you think I will blindly believe someones claim over the internet. Think again. - That would be moronic for me or anyone to do.
And I am more than happy to change my view and even apologise if I am incorrect, if something is substantiated.
www.youtube.com/@Pemalite
