Originally posted by skeevy420
View Post
Announcement
Collapse
No announcement yet.
Radeon Vulkan Driver Adds Option Of Rendering Less For ~30% Greater Performance
Collapse
X
-
- Likes 1
-
Originally posted by phoronix_is_awesome View PostSo, after using all the nasty tactics such as forcing GA104 8GB to swap at 1440P, pricing a 192bit GPU at 256bit competitor's price, forcing media to not talk about raytracing performance or lack of dedicated matrixmul units, or lack of ROCm compute support, it still doesn't solve the problem that the 3060's MSRP is nearly 30% less. So now the shift of 30% burden needs to come somewhere. I guess 30% less image quality is a lot harder to detect. Great work David and Scott. It's a shame that all the 16Gb Samsung GDDR6 16Gbps dies are wasted on this piece of shit lineup.
If Nvidia would realize that ECC/ P2P RDMA is the barrier to professional AI computing, and start putting the samsung 16Gb dies on GA104/GA102 chips, and make those cards unavailable to miners and available to university CUDA compute people, it would do the greatest good to society.
About VRAM : wierd part of current GPU market is that Nvidia who is superior at computing for desktop GPUs, have less VRAM that you need to a lot of compute workloads. At the same time AMD who is worse at computing workloads (at least with RDNA2) suddenly has more VRAM. Nvidia would benefit a lot from more VRAM, while AMD would benefit a lot from wider memory bus but small VRAM modules.
The only card that does both compute and high vram is 3090, that is priced way it is because it competes with workstation solutions in a lot of cases.
It also makes me wonder why Samsung or Micron doesn't make 1.5GB VRAM chips, like 6800 or 6800XT cheaper with 12GB would make so much more sense, same as 3070 with 12GB slighty more expensive also would make a lot more sense. Is there some technical limitations for that? Skipping totally GDDR6X because for those 2GB modules do not exist at all.Last edited by piotrj3; 10 April 2021, 01:29 PM.
Comment
-
Originally posted by Quackdoc View PostAhh AMD, always so far behind... oh well, still better than the other options. just need a DLSS competitior and actually decent RT preformance
- Likes 3
Comment
-
Originally posted by r1348 View Post
Still, the premise sort of invalidate the usefulness of this feature, since nobody but miners own RDNA2 cards.
Sure, the cards are difficult to get. You can't just go online and click "Buy Now". But they're not impossible to pick up.
Comment
-
maybe not very useful for top-tier cards, but are there or will there be low-range gpus from the RDNA2 lineup? how about rdna2 igpus for laptops?
if yes, then this feature will eventually find use in low-power cards regardless of being hardware-limited to a very recent lineup, and laptops would probably also benefit, not just due to having weak igpus, but due to potentially reducing power consumption / thermal dissipation when this is a concern (eg: something feral gamemode could leverage)
while this is something games would better use selectively, and while enabling such a setting globally comes at a probably more noticeable expense of image quality, it's nice to have tweaks you can enable from the system side even for games that don't have it built-in on their video settings... some game devs are very careful in providing low-resource modes for their games, others not so much, so it'd be nice to have
- Likes 1
Comment
-
Originally posted by piotrj3 View Post
I think it is simply there because Vulkan finally finalized variable rate shading which RDNA1 never supported in first place even on DX12. Adding such variable is easy, although fundamentally useless for RDNA2 GPUs as all of them are too overpowered to on purpose reduce image quality like that.
About VRAM : wierd part of current GPU market is that Nvidia who is superior at computing for desktop GPUs, have less VRAM that you need to a lot of compute workloads. At the same time AMD who is worse at computing workloads (at least with RDNA2) suddenly has more VRAM. Nvidia would benefit a lot from more VRAM, while AMD would benefit a lot from wider memory bus but small VRAM modules.
The only card that does both compute and high vram is 3090, that is priced way it is because it competes with workstation solutions in a lot of cases.
It also makes me wonder why Samsung or Micron doesn't make 1.5GB VRAM chips, like 6800 or 6800XT cheaper with 12GB would make so much more sense, same as 3070 with 12GB slighty more expensive also would make a lot more sense. Is there some technical limitations for that? Skipping totally GDDR6X because for those 2GB modules do not exist at all.
Now, adding more traces is the limiting factor for higher bus width, and I don't know what design issue AMD was facing, but currently traces are becoming a bottleneck for the entire industry, from motherboards to GPUs.
Comment
-
Originally posted by r1348 View PostStill, the premise sort of invalidate the usefulness of this feature, since nobody but miners own RDNA2 cards.
And I don't understand the thinking that having any feature developed before you can utilize it is useless. Did you consider that having the feature in advance of when you can use it might give developers more time to refine and optimize it, so that it works even better, when you finally can? Software development is a fundamentally iterative process, and this is one of those features that seems like it's going to need some tuning and refinement -- both on the part of RADV and games, themselves!
- Likes 4
Comment
Comment