>this is the FP performance of a 7900 XTXHow in the everloving fuck is Radeon not smoking NVIDIA GPUs all these years with these numbers? What if they were able to run the same software as though NVIDIA made them? AMD GPUs have always had more raw teraflops over the years (especially double precision, such as on the Radeon VII) and yet this is never reflected in most real world tasks. Seriously, on paper this should handily beat an RTX 4080. Can someone explain WTF is going on here? Why is this calculator not out-calculating the other calculator despite being able to do more calculations? Are the calculations just retarded and less valuable than the other calculations from the other calculators and thus not directly comparable or what?
in dx12/vulkan stuff like timespy, resident evil 4, ghosts of tsushima, dxvk etc the 7900xtx matches or beats the 5080. when it comes to RT or DX11 and older Nvidia scales better. there are ofc some DX12 titles that just like Nvidia better like Silent Hill 2 UE5 or God of War Ragnarok but that's just Nvidia partnerships.
>>107881043But this should surely go beyond just games and extend into workstation applications as well, should it not?
>>107881081workstation apps tend to favor cuda which nvidia has and amd doesn't. also nvidia nvenc tends to output better quality at a faster speed than amd's encoder. amd has to rely on apps that explicitly take advantage of afmf or their rocm architecture which are just getting off the ground while nvidia has had cuda built up for nearly 2 decades now.
>>107881006Dual issue instructions are never used so cut the ALU throughput in half
How fast is it in video generation? Is the 9 series cards better than 7? Asked grok/gpt yesterday but the information returned is not something I understand
>>107881233In your opinion is rocm something that will continue to grow and get better? As in is it worth getting a 9 series card or the gap is to big/amd shows signs of a clown show?
>>107881417>>107881445Just buy nvidiaAmd had over a decade to do anything and they chose to fumble instead.
>>107881445First figure out exactly why are you using a PC.If you plan to use Windows then get Nvidia it's just better objectively.If you plan to use Linux then get AMD as RADV is the defacto mesa driver built into every kernel for graphics. Nvidia on Linux is primarily for headless servers as in no 3D/2D graphics ROPs working just CUDA and encoder working.
>>107881445Theyre all TSMC PCIE cards at the end of the day. Get the one that matches your budget and needs.
>>107881006everyone is an Nvidia sister and this has no CUDA
>>107881574why can't it have CUDA, it's just software
>>107881006those are bloated flops, realistically it can't double issue oftenwhich is why it isn't much faster than a 48tflops RDNA4 chip
>>107881006theoretical max, only happens under certain conditions using certain instructions, requires hand tuning to take full advantage of generally, though amd's shader compiler has gotten better since launch at doing some optimizations automatically.
>>107883397Welcome to clown worldEnjoy your stay
