Some RX 480 4GB stuff leaking out.1080Mhz Clock

I guess, you can imagine what kind of textures and how many you fit in RAW uncompressed format into 1GB.

Yes, there are games which may need 4GB and in future there will be games which will choke on PCIe as they'll need more.
But that's not question of real need. That's question of lazy developer. Or HW manufacturer pushing for higher requirement.

AB is tool you can trust.
+ Custom OSD configurable per game. Working FPS limiter per Game.
+ OC/Power Target
+ Custom Fan profile for those who can't edit vBIOS directly

 

10.6MiB for each 4K BC1 (DTX1) textures. 20.3MiB for each 4K BC3 (DTX4/5) textures. Both with mips. And we are just talking about common quality (ie: no BC7) diffuse maps.. Diffuse maps are just a piece of the cake.

Of course the requirements may vary form game to game....A total war game probably will requires more VRAM then a Fifa game..

 

Yes, now slap it on object and that right into your face (Close to viewport).
Think about per pixel precision vs Texture resolution needed (Pixel coverage of each object => texture on screen).
That's where things get really crazy.

I am all for procedural and material based object texturing. But industry goes way of making extreme large textures and just map (wrap) them around objects.
They take easy path, customers pay for it.

 

16K texture size is the maximum limit allowed by most APIs and drivers. Due some driver/architecture/runtime issues usually you never go up above 8K resolution (but that's not the point).

DooM and Wolfenstein use Virtual Texturing (tiled resources/reserved resources in DX11/12).
As I said, the texture resolution of a game is not related to the screen resolution. This means that at same quality settings the textures involved are the same despite the window rendering resolution.

What you are talking about ini/register tweaks are just settings use for internal rendering, not for color/diffuse textures (eg: shadow maps, render-to-texture - aka post processing surfaces) or the maximum mips resolution allowed (however if the mip0 texture size is below that setting, you will not see any graphics quality improvement). This mean you cannot create by magic diffuse texture details just changing a number in a text file.

So what's the point: do not choose the VRAM size of your graphics card by your screen resolution. Choose it thinking about the games you play and how much years you will stand with that card. In a long perspective, bigger VRAM pool are usually preferred.

 

Procedural materials are compute expensive, they need usually more time loading, and it is hard to obtain very good results. Moreover they still need to take place in the VRAM pool to reach decent performance. With procedural materials you usually save just some disk space.
This is why the industry prefer increasing the texture size.
Civilization IV use a lot of procedural techniques, however the VRAM pool does not take any advantage at all (just play a "huge" map and see what happen at your GPU memory ).
Finally, remember that VRAM is not used only for textures: internal render-to-texture buffers play a fundamental role too. And of course frame-resources too. More details? More cool effects? More shadows? Better lightning techniques? More VRAM! More bandwidth!.

 

Tell me more about it. What exactly is filling vRAM when you increase game map size?
And maybe you'll realize that it has nothing to do with procedurally generated textures.
Today even shapes of objects are affected by computational blending of several textures, easily creating hybrids.
And who's to say that patterns are bad if done is smart way and use pseudorandom placement?

Reusing same texture block again and again is unlikely to save memory bandwidth, but it surely reduces memory usage.

 

xfx 1328mhz 480





source: http://imgur.com/a/3iTkS#

here my overclocked 970 for comparision
http://www.3dmark.com/fs/7124050
NVIDIA GeForce GTX 970 video card benchmark result - Intel Xeon Processor W3680,Dell Inc. 09KPNV

 

Disappointing to say the least...oh well, it was to be expected, I mean assuming perfect scaling, if two of these matched a 1080 according to amd, then a single 480 would be slightly slower than a gtx 970.

 

rx 480 dx 12 perfomance quantom break 60fps 1080p :banana:

https://youtu.be/6sAAgmpU30w

 

Sorry, I was talking about Civ5 (not CIV4). Probably Civ6 will behaves similar.
In Civ 5 if you look at map tiles there are a lot of different result even for the same terrain tile type. All those information have to be stored on VRAM. Yes they are pseudo-procedural since they start from very few common patterns, but this just help in saving disk space and having a more natural terrain patterns applied to the game tiles. I guess identical tiles are stored in the same VRAM place instead of having multiple copies, however on a huge map there are a lot of different tiles for the same terrain type.
If you look at tools like GPU-Z on huge maps the VRAM occupation is very high, and it increase with the map discovery and the progress of the game (since more models are applied to the game in the modern age then in the ancient age).
I can tell you on CIV V a huge maps can easily saturate 3GB VRAM GPUs after a while... and I guess it reaches 4GB and even more in the latest phases of the game. But that's Civilization V.

 

Seems like a little slower than 970 1500mhz. Thats like stock 980 performance and it has 8GB vram. It also costs lower than both. Why is that disappoiting? It would be a huge upgrade to your 670.

 

in his other videos this guy has "his 480" running at 1025mhz.. i think its fake
https://www.youtube.com/watch?v=pQ9CiDozMMQ

 

my bad, mixed up firestrike total with graphics scores
if true it will indeed be a good upgrade for me, better wait for custom cooled versions though, I got spoiled by the total silence of the dcII cooler.

btw my estimation was based on the only (legit) benchmark so far by amd, the one were 2x480s run ashes of the singularity and therefore assumed it wouldn't perform anywhere near a 980. Here's hoping I was wrong

 

It seems in some scenarios like 3dMark and other games (i think also the witcher) it will be slower than Nvidia 970.
In other scenarios, like dx12 or heavy compute it will reach 980 performance.

In some scenario Nvidia cards are better than GCN and viceversa as always...
But assuming the future will be DX12, it could be a great buy.

 

Any DX12 game it will probably outperform even 980. Quantum Break, TW WH, AOS, Hitman and upcoming update for Star Citzen and Deus Ex HR.

Nvidia really needs an 1060 becouse 970 and 980 are not competitive if you are thinking forward.

 

DX12 does not really matter for nvidia or AMD atm, very few games on dx12. actually the only worth game on dx12, would be Battlefield 1 which will probably run better on AMD.

I think 1060 won't be successful, not as 480.

 

Yeah, I mean DX12 games are less than 1% of the games. No point on concentrating on DX12 for a handful of games. By the time DX12 matters we'll all be on different gpus anyway.

 

ROTR is actually a very note worthy DX12 title, even if just basic DX12 as it boost how well that game runs on lower hardware.

Even the Forza Beta is worth it as a beta with DX12.

Dont underestimate how many people actually are wanting DX12 to be widely adopted. Its happening alot faster then DX11 did.

Im dying for PCSX2 to get Dx12 as well, as it did absolute wonders for Dolphin emulator.

 

i want DX12 as fast as possible too, with that or Vulkan, ps3 emulation can go ahead faster, plus that could remove cpu bottleneck that we have in dx11, yet it will take several years probably, problem is the quality of PC games this days aren't that good at all.

 

yeah dx12 rly helps a alot in rise of the tomb raider on low ipc - but wide cpus.
like fx8350 or westmere hexa

https://www.youtube.com/watch?v=3j7w8XMn7IE