RV870 texture filtering

Jawed's ixbt link has different results than AMD released which is strange.
http://www.ixbt.com/video3/cypress-part2.shtml

The cloth sim isn't testing ALUs much. It's really a test of dependent streamout and since the peak vertex throughput didn't change from 4870 the improvement is mostly due to efficiency.

Around the G80 launch Nvidia released a cloth simulation demo which is basically identical to the 3dmark Vantage Cloth test. I didn't find it with a quick search, but it's probably out there with source code if someone is interested is digging further.

I'm guessing compute shaders will be used for cloth sim going forward.
 
I'm sure they are working on a Dx11 graphics test, hopefully this time not tied to PhysX but instead using compute shaders.

Regards,
SB

I'm sure they are not, and even if they are they won't publish it anytime soon.
 
Are they in bed with Nvidia then? *SCNR*

Their policy, from what I recall, was to include support for something only when at least 2 IHVs exposed it. I also don't think they have a team that's large enough to handle both their game as well as a new iteration of the benchmark. I may be wrong, however, which wouldn't be the first time. Sorry if I didn't provide the expected entertainment by answering "yes", or something of the sort.
 
G80 and GT200 used separated mini-SPs for this work. They are located in the shader core, but not used for general shading.
This is a bit an unfortunate wording imho. As you mention later those SFUs not only handle interpolation but also special functions - in particular transcendentals. Hence, they are indeed used for general shading, it's just the MUL which isn't available for general shading and hence those SFUs don't really count for the GFLOP number.
 
Thanks! Hardware.fr measured triangle setup (tri/s) rate the same as R700.

I pushed the theoretical limit through the hw (850 MTris/s), FWIW. And from what I've gathered that's accurate since they're still at 1 tri/clock, and they've only doubled scan conversion rate.
 
Texture interpolators have been removed from the design and is done on the shader core. In general we are seeing this as a performance improvement - its also the reason why one of the Vantage feature test gets a disproportionate increase over the previous gen.

Anyone want to fully clarify what "texture interpolators" actual means?

Besides the confusing Anandtech article, HardOCP says "There are now 80 texture units and the HD 5870 can perform 68 billion bilinear filtered texels per second and up to 272 billion 32-bit fetches per sec.", implying that point sampling (no filtering) is 4x faster (L1 bound returning unfiltered results to the shader)?

Seems clear that vertex attribute interpolation is done in the SIMD units.

What is the word on texture filtering "interpolation"?

(a.) Still in completely in texture units?

(b.) Lower precision filtering in texture units with high precision using the SIMD units?

(c.) Or all texture filtering interpolation done in the SIMD units? Meaning texture unit only provides unfiltered results to the SIMD units which have special purpose instructions to do the interpolation?

Anyone profile fetch4 throughput?
 
Hardware.fr had some i think.

Oh cool, I was waiting for their review. They're one of the few sites that still do that sort of analysis. Also, I noticed that not one single person asked why there's no B3D review. How could our expectations have fallen so far in such a short time? :(

Oops, just noticed that Rys promised an article....

Rys said:
We're done analysing our sample here, so it's a matter of checking and double checking and completing the write up. We'll also have an in-depth look at DX11 in the near future, and a close look at HD 4770 too which we finished recently but didn't quite get out before today's HD 5870 announcement.

It better be awesome!!! :D
 
Besides the confusing Anandtech article, HardOCP says "There are now 80 texture units and the HD 5870 can perform 68 billion bilinear filtered texels per second and up to 272 billion 32-bit fetches per sec.", implying that point sampling (no filtering) is 4x faster (L1 bound returning unfiltered results to the shader)?
Gather can return 4 point-samples at once.
 
What is the word on texture filtering "interpolation"?

I'm still a bit confused on this as well. In the context of some of the reviews it almost sounded like all kinds of values are now interpolated on the SPs, courtesy of the new instructions, including both vertex attributes and texture samples. After reading this discussion though, it sounds like only vertex attributes are interpolated by the shader core (which was my initial impression) and texture filtering is still fixed function. Is that correct?
 
Oh cool, I was waiting for their review. They're one of the few sites that still do that sort of analysis. Also, I noticed that not one single person asked why there's no B3D review. How could our expectations have fallen so far in such a short time? :(

Oops, just noticed that Rys promised an article....



It better be awesome!!! :D

Oh boy, when did he get the chance to say that:O
 
Back
Top