http://videocardz.com/amd/radeon-rx-400/radeon-rx-490x
Radeon RX 490X
TEN THOUSAND GIGAFLOPS GUYS
>>54983460
Yeah, it's a flop alright
>>54983485
PAJEET MY SON
>Gigaflop
Perfect for measuring theoretical performance of HW and nothing else.
>>54983566
depends on application m8.
For high-level shit like vidya then yeah it's not a good measurement
>>54983626
what if I want some high-level memes breh???
>>54983460
>16GB of memory
>dont even need 8GB for 4k
Why are they doing this?
>>54983771
Consumer fucktards who think
>MOAR RAM --> MORE BETTER FPS
>this kills novidya (the way it's meant to be gimped^tm)
>despite all AMD cards having more raw power than their NVIDIA brothers, they perform worse in all real-world scenarios
why can't they go to teraflops?
>>54983460
Yours for the cheap price of just $999.
>>54983771
Because AMD wants, and I quote, "To bring VR to the first ten million consumers". I guess they think they need that much for a good VR experience. I know dick all about VR, so eh.
>>54983771
>dont even need 8GB for 4k
How are you measuring that? Rise of the Tomb Raider uses ~10GB of VRAM at 4K, so there's one example of how you're wrong.
>>54983460
>inb4 the ''''overcloker dream''' bullshit
Is nobody going to point out the memory bus is absurdly high? 4096 bit bus sounds fake as shit. The 480 has what...512?
>>54983974
Its hbm you twat
>>54983974
HBM
>>54983895
>lets load everything into vram!
And still run poorly. A good example of shitty developers.
>>54983818
Different architectures; each of their shaders can perform a mad+mul per cycle, but due to design choices AMD has less shaders proportionally used per cycle. AMD can fit more shaders in a smaller area than Nvidia can however.
This usually balances out to AMD needing more shaders to perform similarly in an average load.
Nvidia did take a back step recently though, 512 Fermi shaders matched 1344 Kepler shaders at equal clocks. Maxwell was generally stagnant, but it scaled better on the high end. Pascal seems to have gotten worse. At equal clocks you'd need 1536 to match 512 Fermi, the Pascal ones just clock way way way higher.
My guess is they deepened the pipeline even more. Probably mid-20 stage now.