AMD Beats NVIDIA's Performance in the Battlefield V Closed Alpha

[jabbadap]

That's a load of hosrecrap, none of this happened and Oxide challenged NVIDIA and refused their involvement in any way in their demos which are heavily subsided by AMD, and remain so to this day.

No developer implements Async Compute because it's a pain in the ass to get it working and supported on most architectures, and the gains are limited most of the time anyway.

Afaik even nvidia implements directx async compute on their Nvidia FleX particle based simulations. But it's not the silver bullet to anything. That Oxides AotS is the best case scenario for async compute, some different type of game and performance benefit from async compute is much much less.

 

[x86overclock]

That's a load of hosrecrap, none of this happened and Oxide challenged NVIDIA and refused their involvement in any way in their demos which are heavily subsided by AMD, and remain so to this day.

No developer implements Async Compute because it's a pain in the ass to get it working and supported on most architectures, and the gains are limited most of the time anyway.

It did happen https://tech4gamers.com/nvidia-acti...-games-to-implement-directx-12-async-compute/
In Asynchronous Compute games 1080ti gets it's butt handed to it like Forza 7 and Wolfenstein 2 The New Colossus by both the Vega 56 and 64. In Nvidia DX12 titles Nvidia has the upper hand because they do not use DX12's Asynchronous Compute instead they use Nvidia's Asynch Compute the two are not to be confused. Asynchronous is a pain in 7the butt to implement but when developers develop DX12 titles using Cuda, it is automatically implemented.
https://digiworthy.com/2017/11/03/wolfenstein-2-benchmarks-amd-vs-nvidia/ and this https://www.guru3d.com/news-story/forza-7-pc-graphics-performance-benchmarks.html

 

[londiste]

Oxide has been taken help from both/all vendors. While AotS's Nitrous engine got its start as a showcase for Mantle - massive amount of draw calls as well as async shaders being the main selling point - it did get worked over for both DX12 and later, Vulkan. It is a fairly objective take on engine development with new lower level APIs in mind. Nvidia cards have a different take on Async Shaders, whether that is better or worse, real or fake, is somewhat irrelevant as there is a way to achieve pretty much the same result. Oxide implemented that at some point. It is worth noting that while the engine is a good showcase at least initially it did not use some of the usual optimizations for example to reduce reliance on excessive draw calls.

AMD/Nvidia have gone back and forth over time when it comes to AotS performance with Nvidia cards barely edging out the comparative AMD cards right now. And by that, I mean DX12 (and Vulkan). AMD's DX11 performance in AotS has been outright atrocious all the time allowing them to claim huge performance increase as DX12 benefit. At the same time, Nvidia cards run DX11 AotS with respectable enough results.

x86overclock, Forza 7 had problems on Nvidia cards that were resolved rather quickly with a driver update. Wolfenstein 2 does have a nice boost for Vega (not all AMD) architecture. The main cause is that iteration of idTech6 using Rapid Packed Math (2xFP16 instead of FP32) for some shaders, leading to some performance benefit. The implementation of some AA modes also tend to favor AMD cards.

DX12 games are not implementing CUDA nor have they ever.

 

[x86overclock]

Oxide has been taken help from both/all vendors. While AotS's Nitrous engine got its start as a showcase for Mantle - massive amount of draw calls as well as async shaders being the main selling point - it did get worked over for both DX12 and later, Vulkan. It is a fairly objective take on engine development with new lower level APIs in mind. Nvidia cards have a different take on Async Shaders, whether that is better or worse, real or fake, is somewhat irrelevant as there is a way to achieve pretty much the same result. Oxide implemented that at some point. It is worth noting that while the engine is a good showcase at least initially it did not use some of the usual optimizations for example to reduce reliance on excessive draw calls.

AMD/Nvidia have gone back and forth over time when it comes to AotS performance with Nvidia cards barely edging out the comparative AMD cards right now. And by that, I mean DX12 (and Vulkan). AMD's DX11 performance in AotS has been outright atrocious all the time allowing them to claim huge performance increase as DX12 benefit. At the same time, Nvidia cards run DX11 AotS with respectable enough results.

x86overclock, Forza 7 had problems on Nvidia cards that were resolved rather quickly with a driver update. Wolfenstein 2 does have a nice boost for Vega (not all AMD) architecture. The main cause is that iteration of idTech6 using Rapid Packed Math (2xFP16 instead of FP32) for some shaders, leading to some performance benefit. The implementation of some AA modes also tend to favor AMD cards.

DX12 games are not implementing CUDA nor have they ever.

CUDA is the software developer tool that is used by developers when they use Nvidia Hardware to develop a title. Wehat I was referring to was Nvidia's Asynch Compute which is different than DX12 Asynchronous Compute. https://developer.nvidia.com/cuda-zone

 

[londiste]

CUDA is Nvidia's compute API. Games are using a graphics API - in this context DX12 and Vulkan - both of which allow doing asynchronous compute.
It is still DX12 asynchronous compute, vendors just want this to be set up differently for optimal use.

 

[jabbadap]

CUDA is the software developer tool that is used by developers when they use Nvidia Hardware to develop a title. Wehat I was referring to was Nvidia's Asynch Compute which is different than DX12 Asynchronous Compute. https://developer.nvidia.com/cuda-zone

No cuda is not used for that. Nvidia uses standard Directx DirectCompute for async compute.

 

[MuhammedAbdo]

In Asynchronous Compute games 1080ti gets it's butt handed to it like Forza 7 and Wolfenstein 2 The New Colossus by both the Vega 56 and 64. In Nvidia DX12 titles Nvidia has the upper hand because they do not use DX12's Asynchronous Compute instead they use Nvidia's Asynch Compute the two are not to be confused. Asynchronous is a pain in 7the butt to implement but when developers develop DX12 titles using Cuda, it is automatically implemented.

Nope. The regular 1080 and Vega 64 are close in both FC5 and W2 after driver and game updates.

https://www.patreon.com/posts/radeon-rx-vega-20791677

 

[x86overclock]

14 fps and 17 fps are not close.

 

[MuhammedAbdo]

14 fps and 17 fps are not close.

Are you blind? The OC'ed 1080 is faster than OC'ed Vega 64 in both games.
At stock the Vega 64 is ahead by 3 or 6 fps. A very minor difference.

 

[x86overclock]

Are you blind? The OC'ed 1080 is faster than OC'ed Vega 64 in both games.
At stock the Vega 64 is ahead by 3 or 6 fps. A very minor difference.

Oced not stock. And the Vega 64 is 1 FPS faster in Farcry 5. For that matter you might as well compare it the Vega 64 Liquid Edition. The Liquid Edition has a stock clock of 1750/945 and can be overclocked to 1850/1000. The air cooled Vega 64s do not do well with overclocks because of the high voltage which causes higher temperatures which also causes the clocks to throttle lower than their default because they hit their thermal limit. The liquid Editions do not have this issue

 

[StrayKAT]

The difference is negligible imo. Barring optimization/software issues, Vega and 1080 seem roughly equivalent. Ignore anyone who says otherwise. AMD should take criticism seriously when it comes to 1080Ti and above though... they have no answer for it (but perhaps it was intentional.. I guess? I wish they had larger GPU ambitions).

 

[hat]

The difference is negligible imo. Barring optimization/software issues, Vega and 1080 seem roughly equivalent. Ignore anyone who says otherwise. AMD should take criticism seriously when it comes to 1080Ti and above though... they have no answer for it (but perhaps it was intentional.. I guess? I wish they had larger GPU ambitions).

Intentional? That would be silly. If you were running a company, why would you just let your competitor have a superior product?

 

[StrayKAT]

Intentional? That would be silly. If you were running a company, why would you just let your competitor have a superior product?

It makes no sense to me either I'm just kind of repeating what a lot of people say: That AMD intentionally targets the midrange market. I'd prefer they aimed higher, but whatever.

edit: Actually Zen breaks this mold CPU-wise.

 

[hat]

Meh, I think they just do what they can with what they currently have. Hopefully Navi will be a much better architecture, and with 7nm, they can fit more of it in a smaller space.

 

[MuhammedAbdo]

Oced not stock. And the Vega 64 is 1 FPS faster in Farcry 5. For that matter you might as well compare it the Vega 64 Liquid Edition. The Liquid Edition has a stock clock of 1750/945 and can be overclocked to 1850/1000. The air cooled Vega 64s do not do well with overclocks because of the high voltage which causes higher temperatures which also causes the clocks to throttle lower than their default because they hit their thermal limit. The liquid Editions do not have this issue

You might as well compare that to 1080 Liquid cooled cards, they hit 2100Mhz easily as well.

 

[londiste]

Intentional? That would be silly. If you were running a company, why would you just let your competitor have a superior product?

$700 GPU is a high end, if not ultra high end product. Historically the margins there are excellent but volume just isn't there.
When money is tight, they may not want to design and manufacture a large and expensive GPU.

 

[MuhammedAbdo]

And the 580 is a 6fps faster than 1060 in BFV beta. So much for the hyperbolic trash in this thread

 

[x86overclock]

You might as well compare that to 1080 Liquid cooled cards, they hit 2100Mhz easily as well.

 

[MuhammedAbdo]

These are OLD benchmarks dating back to several months, the ones I posted are only 10 days old, with latest drivers and game patches.

 

[x86overclock]

You might as well compare that to 1080 Liquid cooled cards, they hit 2100Mhz easily as well.

That would be a fair comparison although the the liquid cooled 1080s are not reference. I would like to see how they would compare considering a stock Vega 64 with a 1200 MHz/1536 boost beats a stock 1080 with 1607/1733 boost.