![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
It's because nVidia can't raster. You turn all thier voodoo magic off you get abysmal raster performance. It truly means there is no real power behind thirr silicon except the kind that generates massive heat. I'm not fan shilling for AMD, I own nVidia and AMD stuff. To be fair I don't find these Gpus impressive at all. We are buying a pack of cheap parlor tricks at tye end of the day. At least our CPUs generate raw performance but GPUs are based on trickery and fairy dust. I miss the good old days of actual silicon performance. Not hat tricks. But those days are forever gone.
Well, your password isn't on your GPU, either.
I'm really hoping we get some better UE5 games down the road. So far the performance has been really poor even if the visuals are good.
There’s lots it can do, it’s a solid engine, but it’s also one you can throw shit at, and it will gladly gold plate.
Modern GPU's aren't that simple. With or without Ray-Tracing? With or without upscaling? With at most 8GB of VRAM or more? With or without sponsorship? These factors make a difference if AMD or Nvidia is faster. But yea Nvidia is generally faster*, with caveats. Assuming you're using DLSS and Ray-Tracing because of course you would you paid for these features. Where as AMD with FSR and Ray-Tracing is just not as good. Turn off upscaling and Ray-Tracing and crank up the settings past 8GB of VRAM, then AMD is generally faster. Then it comes down to do you value Ray-Tracing? Do you value DLSS over FSR, over TAA? It's not that easy anymore.
DLSS and FSR both look as good or better than TAA at the very least, either will do a better job native to native than TAA will.
Pure raster 4k, 3080 vs 6800xt with drivers-game when the 4070 was released:
Yea but I'm in the camp that I don't like TAA. I'd rather use SMAA, but very few games come with this option. FSR and DLSS are better than TAA, but that just says a lot about how bad TAA is.
Problem is MSAA doesn’t work with modern rendering setups, MSAA does not work on textures only polygons, and it struggles with any objects that use opacity or programmable shaders. It also incurs a higher overhead as it causes each frame to essentially get rendered 1.5 times over.
Reading the link I do not get what the FSR 3 source being open has to do with it.
They were able to see that FSR3 was the DXIL function which apparently was very poorly optimized.... This is also used by DX.
Like Enigma already mentioned, DXIL is used by FSR3, which is used by VKD3D-Proton's MSAD instruction. The person was able to optimize MSAD for use with VKD3D-Proton, ACO compiler and NIR, because these are also open source projects. This is why in some cases there are games that run faster in Linux compared to Windows because these kind of optimizations can be done, despite that Linux has to run translation layers which slow things down. If Microsoft were to implement FSR3 natively into Windows DirectX then even further optimizations can be done. Apple was able to implement FSR into Metal and call it MetalFX, so when it Microsoft going to do the same?
Microsoft can't just implement it without a few hoops, if they did it would likely be a DX12.5 Ultimate Extreme thing, or possibly even a DX13 feature as a form of native AA/Upscaling, but if they did I highly suspect they would make Intel, AMD, and Nvidia jump through validation tests for it and give all time to implement their own hardware accelerations for it for a future hardware release so they could choose how to accelerate it on a driver side to be compliant with the new DX version.
That's what I totally expect them to do.
Microsoft certainly has the market position but like with Direct3D 12 they just don't care. It took AMD to create Mantle to push the industry towards newer API's closer to metal design. FSR is probably not something Microsoft would use directly, but like Mantle it can be used for reference, assuming Microsoft cares enough to do so. Something like frame generation could be enough of a reason for Microsoft to make it a Direct3D 13 or maybe 12.5, or even worse make it limited to Windows 12. I could also see it requiring AI accelerated hardware, which could push older GPU's out.
You can do all of that if FSR 3 code was closed, if I understand the situation correctly.
Can you tell me how the software pipeline works for Intel XeSS? or DLSS? Can you find the inefficiencies in the way the tech works as it relates to other libraries?
Wilt multiple open source project having being changed here, there is no need.
Usually there interaction with external tools will be catchable yes.
Could you? Maybe, but would you? All you gotta do is look at DLSS-FG and XeSS support for Linux to figure out how well closed source is working. Popular emulators like Yuzu and RPCS3 only support AMD FidelityFX for a reason. AMD did the optimization themselves so they would have access to the code either way, but now Valve can step in and further optimize if they wish, as they have done in the past. Seeing as AMD is powering Valve's Steam Deck, they have a lot of incentives to further optimize FSR 3. There are no upsides to closed source either way, but plenty of upsides by keeping it open source.
Like the value of a dollar
Not disagreeing with you just pointing out that Microsoft had been working on DX12 as a low level API for years before Mantle was a thing, Work on DX12 started internally during DX10’s active life cycle and before DX11 had officially launched.