Embarrassing from AMD. With the rumours around RTX5000 perf, AMD is going to have a really rough time. Maybe even feel the squeeze from Intel battlemage.
NVIDIA vs. AMD GPU Workstation Performance For Blender 4.3
Collapse
X
-
Michael
If you want a really interesting article, do a similar test using one of the newer AMD APUs, with the Blender tests running on the iGPU, one of the newer Intel processors where the Blender tests run on the iGPU and compare that to the fastest TR or EPYC system you have on hand running the Blender tests in software, i.e. on the CPU.
It would be interesting what would be faster for rendering using Blender, an iGPU or a monster TR.
Comment
-
-
I mean using optx but not hiprt makes these benchmarks kinda invalid, hiprt makes a huge difference here.
just eyeballing it and assuming that hiprt on rdna3 scales the same as on my rdna2 card nvidia would still win, but the gap would shrink quite a bit makeing the rdna cards maybe cost efficant.
Comment
-
-
Originally posted by NeoMorpheus View PostI dont think its entirely apples to apples here since those AMD gpus are based on RDNa, which we know are not great at computing loads.
But i dont know which AMD gpu uses CDNa, besides the Radeon VII.
CDNA cards are bad a hip rt, CDNA lacks ray intersection hardware that RDNA has. My mi100 is mutch slower than my rdna2 gpu in hip-rt for this reason, even though its a mutch bigger gpu.
The benchmarks in this article are invalid precisely because optix was used, which uses fixed function rt hardware found in nv gpus, but hip-rt was not enabled leaving the fixed function rt hardware in rdna cards unused.
Comment
-
-
Originally posted by DiamondAngle View Post
the Radeon VII is NOT a CDNA card, its GCN altho CDNA is just a continuation of GCN with a few more instructions.
CDNA cards are bad a hip rt, CDNA lacks ray intersection hardware that RDNA has. My mi100 is mutch slower than my rdna2 gpu in hip-rt for this reason, even though its a mutch bigger gpu.
The benchmarks in this article are invalid precisely because optix was used, which uses fixed function rt hardware found in nv gpus, but hip-rt was not enabled leaving the fixed function rt hardware in rdna cards unused.
Comment
-
-
Originally posted by DiamondAngle View PostThe benchmarks in this article are invalid
The fixed function bits on the GPU are useless if the software that can use them doesn't exist or doesn't work.
It's like those "Raspberry Pi killers" from AliExpress which have GPUs 3 times more powerful on paper but no working drivers for them.
Comment
-
Comment