SemiAccurate Forums  

 
Go Back   SemiAccurate Forums > Main Category > GPUs

GPUs Talk about graphics, cards, chips and technologies


Reply
 
Thread Tools Display Modes
  #2261  
Old Today, 06:26 AM
Stuckey Stuckey is online now
>intel 4004
 
Join Date: May 2012
Location: 173000G
Posts: 4,539
Stuckey will become famous soon enough
Default

Given that Vega should have at least twice the geometry performance of Fiji, and not have to waste time rendering millions of polygons that are hidden/smaller than a pixel, I doubt it will have many problems.

Same with the fill rate. 90 Gpix/s is actually 90 billion pixels per second. Which if you do the math equates to enough pixels to fill a 4K screen at ~1000 fps..So basically, I don't think the difference between 90 and 130 Gpix/s (on the 1080ti) is going to make any difference, as I'm pretty sure both cards will be bottlenecked by other stuff long before they get to use even 1/10th of that.
__________________
To find the right answers you must ask the right questions.
Reply With Quote
  #2262  
Old Today, 06:40 AM
Bigos Bigos is offline
itanic
 
Join Date: May 2011
Posts: 100
Bigos is on a distinguished road
Default

Quote:
Originally Posted by Stuckey View Post
Same with the fill rate. 90 Gpix/s is actually 90 billion pixels per second. Which if you do the math equates to enough pixels to fill a 4K screen at ~1000 fps..So basically, I don't think the difference between 90 and 130 Gpix/s (on the 1080ti) is going to make any difference, as I'm pretty sure both cards will be bottlenecked by other stuff long before they get to use even 1/10th of that.
You cannot only count the final frame pixels when calculating the fill rate required. There are many intermediate textures that need to be filled before you can even think about rendering the final frame. There is also a lot of overdraw involved (though new techniques try to minimize this effect), making the same final pixel be filled many times for a single frame. Multi-pass techniques basically draw to the same texture many times (not sure if they are used today, though). Also, what pixels are they counting for the fill rate in the graph? RGBA8888? 16? FP32? What is the number of samples? Was the output easily compressible or not?
Reply With Quote
  #2263  
Old Today, 07:04 AM
DCO DCO is offline
>intel 4004
 
Join Date: Jun 2010
Posts: 4,203
DCO will become famous soon enough
Default

Quote:
Originally Posted by del42sa View Post
Oh yes we know, it was in presentation slides , footnote

It doesn't say Vega 10 has 4 geometry engines, only that Vega is designed to handle up 11 to polygons per clock with 4 geometry engines. It can have 4 or not, we have to wait the reviews.
Reply With Quote
  #2264  
Old Today, 07:09 AM
Toettoetdaan Toettoetdaan is offline
8-bit overflow
 
Join Date: Dec 2011
Posts: 364
Toettoetdaan is on a distinguished road
Default

Quote:
Originally Posted by DCO View Post
It doesn't change what I've said, it is exactly what I've said, Frontier edition as the pro duos don't come with certified and optimized drivers for CAD/CAM applications. For that you will pay 2x-3x more.
AND still it is a pro card for another market segment with their own pro drivers, not sure what is so difficult to grasp.
Well then I agree with you, my original point was just that we should not compare a card where you pay for certified and optimized drivers versus a card where you don't pay for it on price.

Quote:
Originally Posted by del42sa View Post
In reality they will be glad for matching 980/980ti...
That is all they need, Fury X is on pair with 980TI and 1070 with just 50% of the throughput.
Also, note that Nvidia performs above it specs here because of the usage of some form of tiled rendering. AMD is probably going to do the same with the 'draw-stream binning rasterizer'.

But there is also good news:
Quote:
VEGA will have a terrific price/performance ratio, according to my sources.
(https://twitter.com/BitsAndChipsEng/...72194755837952)
__________________
Back in 2015....
Quote:
Originally Posted by Toettoetdaan View Post
Zen will be equal to Intel's 14nm offering.
Reply With Quote
  #2265  
Old Today, 07:22 AM
del42sa del42sa is offline
2^10
 
Join Date: Oct 2011
Posts: 1,262
del42sa is on a distinguished road
Default

Quote:
Originally Posted by DCO View Post
It doesn't say Vega 10 has 4 geometry engines, only that Vega is designed to handle up 11 to polygons per clock with 4 geometry engines. It can have 4 or not, we have to wait the reviews.
what ? It states clearly, that Fiji = 4 polygons/clock with 4 GE

Vega up to 11 primitive/clock with the same amount of GE. It means it can handle two primitives per clock per /GE. The rest is with primitive discard accelerator/primitive shaders.

if it had more GE, then I guess it probably can handle more polygons than 11 don´t you think ?

11/4= 2,7x more geometry



and we already know Vega has 4 SE, pretty much like Fiji.

Last edited by del42sa; Today at 07:36 AM.
Reply With Quote
  #2266  
Old Today, 10:16 AM
DCO DCO is offline
>intel 4004
 
Join Date: Jun 2010
Posts: 4,203
DCO will become famous soon enough
Default

Quote:
Originally Posted by del42sa View Post
what ? It states clearly, that Fiji = 4 polygons/clock with 4 GE

Vega up to 11 primitive/clock with the same amount of GE. It means it can handle two primitives per clock per /GE. The rest is with primitive discard accelerator/primitive shaders.

if it had more GE, then I guess it probably can handle more polygons than 11 don´t you think ?

11/4= 2,7x more geometry



and we already know Vega has 4 SE, pretty much like Fiji.
Yes.
Yes. It means what it means, how much comes from primitive shaders is your assumption.
Yes, but it says with 4 GEs, 1000 geometry engines will not change what's the output of 4.
I haven't seen where it says 4SE.
Reply With Quote
  #2267  
Old Today, 10:54 AM
del42sa del42sa is offline
2^10
 
Join Date: Oct 2011
Posts: 1,262
del42sa is on a distinguished road
Default

Kód:
+ case CHIP_VEGA10:
+ adev->gfx.config.max_shader_engines = 4;
+ adev->gfx.config.max_tile_pipes = 8; //??
+ adev->gfx.config.max_cu_per_sh = 16;
+ adev->gfx.config.max_sh_per_se = 1;
+ adev->gfx.config.max_backends_per_se = 4;
+ adev->gfx.config.max_texture_channel_caches = 16;
+ adev->gfx.config.max_gprs = 256;
+ adev->gfx.config.max_gs_threads = 32;
+ adev->gfx.config.max_hw_contexts = 8;
+
+ adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
+ adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
+ adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
+ adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
+ gb_addr_config = VEGA10_GB_ADDR_CONFIG_GOLDEN;
+ break;
+ default:
+ BUG();
+ break;
+ }
Reply With Quote
  #2268  
Old Today, 12:02 PM
Stuckey Stuckey is online now
>intel 4004
 
Join Date: May 2012
Location: 173000G
Posts: 4,539
Stuckey will become famous soon enough
Default

Quote:
Originally Posted by Bigos View Post
You cannot only count the final frame pixels when calculating the fill rate required. There are many intermediate textures that need to be filled before you can even think about rendering the final frame. There is also a lot of overdraw involved (though new techniques try to minimize this effect), making the same final pixel be filled many times for a single frame. Multi-pass techniques basically draw to the same texture many times (not sure if they are used today, though). Also, what pixels are they counting for the fill rate in the graph? RGBA8888? 16? FP32? What is the number of samples? Was the output easily compressible or not?
Thanks, I get that. However I'm pretty sure the point stands-

Quote:
most GPUs will crumple under the load of rendering a high-fidelity graphics at 4K before reaching texture fill-rate pipe limitations.
http://www.gamersnexus.net/guides/17...l-rate-defined
__________________
To find the right answers you must ask the right questions.
Reply With Quote
Reply

Tags
amd, vega


Thread Tools
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off


Forum Jump


All times are GMT -5. The time now is 12:06 PM.


Powered by vBulletin® Version 3.8.7
Copyright ©2000 - 2017, vBulletin Solutions, Inc.
SemiAccurate is a division of Stone Arch Networking Services, Inc. Copyright © 2009 Stone Arch Networking Services, Inc, all rights reserved.