AMD: R8xx Speculation

[digitalwanderer]

What does triangle rate have to do with rasterisation rate or the count of rasterisation units?

I'm not so sure I understand this fully, but don't you need the triangle setup to rasterize? :|

 

[Jawed]

Do you know of any way to test this case properly?

No. It might only be revealed by some tiling-aligned shading tests, which then run the risk of falling foul of other aspects of the architecture/data-storage (i.e. textures) that are also tiled.

Jawed

 

[Jawed]

I'm not so sure I understand this fully, but don't you need the triangle setup to rasterize? :|

Yes. One triangle could generate 1 or millions of rasterised fragments, which is why a triangle throughput test doesn't say anything meaningful about the organisation of rasterisation (or what other effects might be seen in the way work is scheduled).

Jawed

 

[trinibwoy]

But if it's true that there's two of them then that is technically interesting. See, you keep saying it's nothing but a marketing detail.

You can twist my words however you like but the fact is I'm just asking wth they're referring to in those diagrams. If it's just an increase in scan conversion throughput it's stupid.

What does triangle rate have to do with rasterisation rate or the count of rasterisation units?

Cause that's what everyone interpreted those diagrams to mean and that's what everyone keeps talking about - the need for faster triangle setup. What do you think they mean, and why hasn't it been further fleshed out by AMD or anyone else?

Yes. One triangle could generate 1 or millions of rasterised fragments, which is why a triangle throughput test doesn't say anything meaningful about the organisation of rasterisation (or what other effects might be seen in the way work is scheduled).

Sure if you assume the author of the test is retarded

 

[AlexV]

Sure if you assume the author of the test is retarded

I think that's typically the implied assumption. At least that's how it sounds most of the time.

 

[Lightman]

At least one author had balls to ask AMD the right questions about Dual-Rasterizer

Keeping with the double-up theme, AMD has added a second rasteriser and hierarchical-Z processor to ensure the rest of the GPU can keep up with the kind of workloads it is given. The last thing AMD wanted was for the GPU to not be able to convert complex wireframe meshes into pixels fast enough, and have portions of that massive bank of stream processors left idle.

DirectX 11 introduces multi-threaded rendering for the first time, but it wasn't clear whether the dual rasterisers required Microsoft's latest API to function correctly. When I enquired about this, AMD said that both rasterisers are accessible all of the time. AMD achieved this by modifying the set-up engine, geometry dispatch unit and ultra-threaded dispatch processor so that the two rasterisers work on two 16-pixel scan converters that each focus on non-overlapping pixels.

In other words, it sounds as if the GPU can use both rasterisers no matter the game or DX version it uses – we’ll have to wait until DX11 is released and we have some DX11 games to verify this though.

The whole article is worth reading!

Bit-Tech - Cypress Architecture Analysis

 

[fellix]

Poor Jawed would be pissed off.

 

[Scali]

What's all this confusion about DirectX 11 and its multithreaded rendering?
It's just the addition of being able to build lists of commands from various threads, to be executed by the 'main' thread. It's little more than a driver/API trick, and works on any DX9/DX10 hardware supported by DX11.

 

[AlexV]

Poor Jawed would be pissed off.

He shouldn't be. Dave already explained on these very forums what the story is. As for the bit-tech quip, no comment.

 

[3dilettante]

He shouldn't be. Dave already explained on these very forums what the story is. As for the bit-tech quip, no comment.

I must have missed that one.

 

[trinibwoy]

I must have missed that one.

Me too. Linky please?

 

[AlexV]

Bah, bloody hell, seems I can't get anything right today The rasterizer discussion wasn't on the forums...oh, well, second fail for the day executed.

 

[smw]

You can make amends by posting it

 

[neliz]

soon to appear here: HD5770 with 1120SP's (not 800)

 

[fellix]

Native 14 SIMD design?

 

[-The_Mask-]

Native 14 SIMD design?

Yes and 2 disabled for the HD5750 I think.

 

[MarkoIt]

soon to appear here: HD5770 with 1120SP's (not 800)

Well, I guess that's confirming RV870 has more than 20 SIMDs, and probably more than a 256bit MC.

 

[-The_Mask-]

Well, I guess that's confirming RV870 has more than 20 SIMDs, and probably more than a 256bit MC.

No it confirms nothing, but yes it would be nice.

 

[Arty]

Well, I guess that's confirming RV870 has more than 20 SIMDs, and probably more than a 256bit MC.

How so?

AMD has in the past had RV570 which was 3/4 of the R580. If Juniper is 14SIMDs then it is 2.8/4 of Cypress.

edit: Juniper having 14SIMDs is a bigger surprise for me than the 20SIMDs in Cypress.

 

[-The_Mask-]

I think neliz is sleeping.

Juniper

HD5700 will be with 1120 stream processors and 56 texture units TMU, the memory is GDDR5. Far as we know there will be HD5770 (199 dollars) and HD5750 (149 U.S. Dollar), HD5770 will replace the current HD4890, the HD5750 is to replace the HD4870. HD5700 opponent is to the NVIDIA GTX275, GTX260, GTS250 and the future D10P GT200b series. Others like DX11, ATI Steam, UVD2, Eyefinity, CrossFire technologies are also readily available.