r520 performance speculation

[hovz]

all marketing features aside, what do you guys think the r520 will bring to the end user.

curious to hear your opinions

 

[Unknown Soldier]

SM3.0+
700 Mhz Core
24/32 Pipes
eDram?
700+Mhz DDR3(GDDR3?) 512mb memory
PCI-E & AGP?
VIVO
MPEG-2 Decode/Encode
MPEG-4 Decode/Encode
DivX Decode/Encode?
AMR
Tile Based Rendering

US

 

[hovz]

wow..

 

[K.I.L.E.R]

TBR? I wish.

People have constantly said this for upcoming cards since several years ago. We still don't have it.

 

[Tim]

SM3.0+

Standard SM3, no big extraes.

700 Mhz Core

More like 600MHz

24/32 Pipes

Most likely 24

eDram?

No

700+Mhz DDR3(GDDR3?) 512mb memory

GDDR3 (later GDDR4)

PCI-E & AGP?

Only bridged AGP (native PCIe)

VIVO

MPEG-2 Decode/Encode
MPEG-4 Decode/Encode
DivX Decode/Encode?

Only using shaders.

AMR

Maybe

Tile Based Rendering

No

 

[nobody]

SM3.0
550MHz core
600MHz ram
16 pipes

 

[PowerK]

Is R520 supposed to be a desktop version of R500 of XBOX 2 ?? or are they supposed to be based on different base architectures ?

 

[CJ]

Is R520 supposed to be a desktop version of R500 of XBOX 2 ?? or are they supposed to be based on different base architectures ?

R520 is based on the R3x0/R4x0 architecture.

R600 is the first desktoppart that is based on the R500 aka X-Box2 chip.

I'm expecting R520 to be SM3.0+. They'll probably add some extra stuff apart from the current SM3.0 specs so they've got something to boast about compared to nVidia's SM3.0 implementation.

 

[nobody]

I'm expecting R520 to be SM3.0+. They'll probably add some extra stuff apart from the current SM3.0 specs so they've got something to boast about compared to nVidia's SM3.0 implementation.

There is nothing like SM3.0+.
And you cannot use things that are not written in the D3D specs.

 

[PowerK]

Is R520 supposed to be a desktop version of R500 of XBOX 2 ?? or are they supposed to be based on different base architectures ?

R520 is based on the R3x0/R4x0 architecture.

R600 is the first desktoppart that is based on the R500 aka X-Box2 chip.

I'm expecting R520 to be SM3.0+. They'll probably add some extra stuff apart from the current SM3.0 specs so they've got something to boast about compared to nVidia's SM3.0 implementation.

Hmm.. so, does that mean graphic/visual processing power of XBOX 2 will be quite a bit ahead of PC until R600 is released for desktop ??

 

[bdmosky]

Well... my guess is that the next generation desktop hardware will be out before the Next Xbox.

 

[PowerK]

Well... my guess is that the next generation desktop hardware will be out before the Next Xbox.

That's what I think as well. XBOX2 will be released near the end of 2005, I think. R520 will be released sometime in Q2/Q3 2005.

Hmm.. so, does that mean graphic/visual processing power of XBOX 2 will be quite a bit ahead of PC until R600 is released for desktop ??

Stupid me. Nvidia won't be sitting around doing nothing. Which will make ATI to push their tech as hard as they can. Competition is indeed a good thing.

 

[CJ]

There is nothing like SM3.0+.
And you cannot use things that are not written in the D3D specs.

There wasn't anything like SM2.0+ until nVidia launched the FX series and dubbed their version SM2.0+... and MS added stuff like PS_2_0_a to the specs to prevent major disaster for nV. And as we all know it contained features that were more SM3.0-like than SM2.0 (and lacked some SM2.0 features but I won't go into that).

So I'm suspecting ATi might add some new stuff that might be headed more into the SM4.0 direction and call it SM3.0+. I know the specs aren't finalised for SM4.0, but that does not mean that there is absolutely nothing known about the direction it's headed in...

 

[PeterAce]

I wonder what the ratio of ALU ops to Texture ops will be on R520?

R3X0, R4X0 and NV4X all are 1 to 1 (ALU to Tex).

R500/Xenon is 4 to 1 (ALU ops to Tex ops).

So what will R520 be?

 

[Dave Baumann]

R3X0, R4X0 and NV4X all are 1 to 1 (ALU to Tex).

Best case, all of these are about 4:1 (possibly a little more for NV40).

 

[Luminescent]

In NV40, the ALU to TMU ratio is 2:1 or, given the mini-alus, 2.5:1.

In R420, it is closer to 1:1, but more like 1.5:1.

It should be noted that Ati and Nvidia define/structure their ALUs differently.

 

[Sage]

650MHz
16-pipe
fancy new memory architecture (with virtual memory)
SM3

 

[Xmas]

I wonder what the ratio of ALU ops to Texture ops will be on R520?

R3X0, R4X0 and NV4X all are 1 to 1 (ALU to Tex).

R500/Xenon is 4 to 1 (ALU ops to Tex ops).

So what will R520 be?

R500 is 4:1 only if you count the vec3/scalar split of the ALUs.

Using the same method, R4x0 is usually 2:1, but sometimes 4:1. You can't really give a ratio for NV40, because most of the time (not always) you can use either TMU or SU1, but not both.


I'm expecting about double the shader performance of R420 for R520. Branching and vertex texture fetch might be much faster than on NV40. No shader features beyond SM3, and no multisampled FP16 rendertargets yet, but FP16 texture filtering and blending. And, just a dream, some hybrid dynamic super-/multisampling.
Possibly 550/650 core/mem clock, maybe less.

 

[Dave Baumann]

What will the lowest quality of pixel output be?

 

[Richard]

no multisampled FP16 rendertargets yet

I'd be happy to get fp blending, though not aa not being orthogonal is a thorn, any particularly difficult hurdles to overcome on the hardware side to make this happen?