Anand has the details about r520,rv530,rv515
- Thread starter tEd
- Start date
Hellbinder
Banned
Um Slappi...
I suppose you better wait to see the actual reviews. A lot of people are making assumptions based on preconceived notions about Clockspeed and Fillrate based on prior designs and even the current GTX.
Dont make foolish and rash jumps about how Ati failed this and failed that based on nothing but a few facts with no context.
- You should end up more than happy with the performance.
- You definitely will appriciate the IQ
- I susspect you will very much appriciate the Features
Remember that PowerVR never needed high clocks or Fill rate to be competative becuase of design. This is a new design not a rehash of the R300.
I suppose you better wait to see the actual reviews. A lot of people are making assumptions based on preconceived notions about Clockspeed and Fillrate based on prior designs and even the current GTX.
Dont make foolish and rash jumps about how Ati failed this and failed that based on nothing but a few facts with no context.
- You should end up more than happy with the performance.
- You definitely will appriciate the IQ
- I susspect you will very much appriciate the Features
Remember that PowerVR never needed high clocks or Fill rate to be competative becuase of design. This is a new design not a rehash of the R300.
What do people make of the 512Mbit memory rumor posted by the Inquirer?
http://www.samsung.com/Products/Sem...tegory_GraphicsMemory_20050621_0000137176.htm
It appears that samsung does have a product.
http://www.samsung.com/Products/Sem...tegory_GraphicsMemory_20050621_0000137176.htm
It appears that samsung does have a product.
I'm not in the position to buy either 7800 or r520 at present so to me whether one is faster than the other is not relevant ( though interesting all the same ), but taking Ati's past record of doing well with DX9 shaders I would hardly say that the r520 can be terrible before it is released, as Alan Partridge would say it is likely to be " at least competitive".
I also agree with John Reynolds a few posts back, apart from a couple of games at present there is more than enough speed, for once the hardware people have outpaced the software games people I think, and this is not really going to change much before G7X/G8X or R580/R6XX comes along .
As I will not be buying this time around I am still very interested in the r520, I want to know how the 90nm process allows overclocking and how much heat it puts out, what this new memory ring bus is all about, how does 512bit v 256bit give you an advantage etc etc.
The G70 is a fine card and very fast, but it wasn't very exciting to me, the only thing that I really liked was the way nvidia were clever with the different clock regimes to save power and cut down heat .. nice work that.
I also agree with John Reynolds a few posts back, apart from a couple of games at present there is more than enough speed, for once the hardware people have outpaced the software games people I think, and this is not really going to change much before G7X/G8X or R580/R6XX comes along .
As I will not be buying this time around I am still very interested in the r520, I want to know how the 90nm process allows overclocking and how much heat it puts out, what this new memory ring bus is all about, how does 512bit v 256bit give you an advantage etc etc.
The G70 is a fine card and very fast, but it wasn't very exciting to me, the only thing that I really liked was the way nvidia were clever with the different clock regimes to save power and cut down heat .. nice work that.
memberSince97
Newcomer
DI and LN2 benchmark whores are gonna be lovin this part from ATI, let alone myself.
rwolf said:
You are confusion bits and megabits.
And theinquirer is confusing internal and external.
the samsung release is about capacity, not bus width.
And thinquirer's hype about the 512-bit bus is like hype saying that the next generation toyota corolla will have 4 cylinders,
or that the next pentium will have integrated floating point unit.
ATI has had internal 512-bit bus since radeon 9700.
Matrox has had internal 512-bit bus in parhelia even before that.
Theinquirer just does not understand what 512-bit internal bus means and draws many stupid /incorrect conclusions from that.
Chip's internal busses are not DDR. So 256-bit DDR bus is "converted" into 512-bit SDR inside the chip.
original geforce was also called "geforce256" due this; it had 256-bit internal bus for it's 128-bit DDR memory.
Last edited by a moderator:
The_Wolf_Who_Cried_Boy
Newcomer
<HALF-BAKED SPECULATION>
Taking Anands figures at face value then the memory controllers are comfortable running at 700MHz. If the 600Mhz figure is an accurate base clock, how much could Overdrive potentially offer, IIRC R480 gained around 50MHz.
</HALF BAKED SPECULATION>
Taking Anands figures at face value then the memory controllers are comfortable running at 700MHz. If the 600Mhz figure is an accurate base clock, how much could Overdrive potentially offer, IIRC R480 gained around 50MHz.
</HALF BAKED SPECULATION>
Jawed
Legend
OK, I have to admit, I'm completely confused what you mean nowAiluros said:Jawed,
It was just a reminder about TMU-arrays which we might see more often in future designs IMHLO.
Pathetically oversimplyfied in my layman's terms I'd also think that "arrays" work better with "arrays" and probably more so in a unified architecture. Have to leave it be right now cause I feel like I'm in the middle of some folks crossfires (no pun intented) LOL
Jawed
Jawed
Legend
One thing I hadn't considered when thinking about R520 sharing its TMUs between vertices and fragments (if indeed it does) is point sampling - which I presume is going to be the predominant form of texturing in vertex shaders. In Xenos it seems that point sampling can run in parallel with texture filtering. 16 pipes of each.
Although I don't think point sampling actually requires pipelines, as such - just a set of coordinates and a wait until the results turn up (it would be great to get clarification on this). Although, having said that, the texture coordinates need to be calculated, which would be pipelined. Erm, I dunno...
Anyway, if R520 has only 16 pipelines to access textures, whereas Xenos has "32", split equally into point-sampled and filtered, then R520's vertex shaders are looking down the barrel of a serious bottleneck...
Jawed
Although I don't think point sampling actually requires pipelines, as such - just a set of coordinates and a wait until the results turn up (it would be great to get clarification on this). Although, having said that, the texture coordinates need to be calculated, which would be pipelined. Erm, I dunno...
Anyway, if R520 has only 16 pipelines to access textures, whereas Xenos has "32", split equally into point-sampled and filtered, then R520's vertex shaders are looking down the barrel of a serious bottleneck...
Jawed
memberSince97
Newcomer
good theory Jawed , it is the cleanest and plausible.
X800 Gto.
