Hellbinder
Banned
I Also think that it may be more along the lines of the 5800 in certain ways. Like 16 pixels per clock in certain situations up to 32 pixels per clock in others.
Again.. Just an opinion..
Again.. Just an opinion..
eight shader units for R520
- Thread starter kyetech
- Start date
5800 was never capable of outputting more than 4 pixels per clock, but I guess you're just referring to the greater number of ALUs in a rather obtuse way?
Speaking of the 5800 and double stencil ops per clock, I'd be curious to know if 16 "extreme" pipes can compete better with 24 nVidia pipes in Doom 3.
Speaking of the 5800 and double stencil ops per clock, I'd be curious to know if 16 "extreme" pipes can compete better with 24 nVidia pipes in Doom 3.
I still wonder if the statement of a 512bit memory bus would be coming soon during the 9800 Pro Launch web broadcast will ever make daylight. My only thoughts was that it was referenced to the r400. And if so could it of been carried over to the r520? And what ever happen to Micron and there 1GHz+ GDDR3 memory?
Raystream
Raystream
quest55720
Regular
Why is every one assuming 32bfp on the 520? I thought that 24bfp was still the min? I figured that there would be a good chance ATI would skip 32bfp to save trannys. After how consevitive ATI was with the tranny budget on the 420 I figured they would do the same this time around.
quest55720
Regular
tEd said:
Ah ok thanks
Some rumours are often confusing; I'd guess that there might be =/>300M Transistor count chips in the make, yet I'd rather think of embedded ram and anything outside the PC desktop market.
As for R520 my speculations from a layman's perspective are in line with people like Mufu/XMas/Wavey etc. In short a sizeable increase in ALU OPs per cycle in combination with higher core frequencies as up to now. If I think about it a NV40@600MHz could theoretically yield 50% higher shader performance than a usual 6k8U.
As for R520 my speculations from a layman's perspective are in line with people like Mufu/XMas/Wavey etc. In short a sizeable increase in ALU OPs per cycle in combination with higher core frequencies as up to now. If I think about it a NV40@600MHz could theoretically yield 50% higher shader performance than a usual 6k8U.
well im thinking, 24 pixel pipe`s 16 rop`s and 8-10 vs units. i dont think that there`s a need for more then 16 rop`s now-day`s the pixel fill rate is more the enough(@450+mhz).
and Clock-speed-wies the core will probably run at 630-650mhz, and the memory will run at 1200-1300mhz(no much improvement here i guess).
i dont think they`ll go on a 512bit memory cont, its too costly. but they`ll brobably will have cards with 512mb.
to cover the "slow"(
) memory, and if there enough die space left, they could put some on-die cache(512-2048K?)
oh and i dont think they`ll go for SM3.0+ there realy no need for it.
i rather have more PPL and higher clock speed then SM3.0, but the nVIDIA`s propoganda about how SM3.0 is so important that you cant live w/o it is driving the whole market to SM3.0.
and Clock-speed-wies the core will probably run at 630-650mhz, and the memory will run at 1200-1300mhz(no much improvement here i guess).
i dont think they`ll go on a 512bit memory cont, its too costly. but they`ll brobably will have cards with 512mb.
to cover the "slow"(
) memory, and if there enough die space left, they could put some on-die cache(512-2048K?)oh and i dont think they`ll go for SM3.0+ there realy no need for it.
i rather have more PPL and higher clock speed then SM3.0, but the nVIDIA`s propoganda about how SM3.0 is so important that you cant live w/o it is driving the whole market to SM3.0.
radar1200gs
Regular
Last time I checked, it was Microsoft, not nVidia or ATi that sets the specs and provides the forwards direction (which the IHV's then have to impliment in silicon), so I fail to see where you think nVidia is dragging the market towards SM3.0.
Having said that, ATi is late for the SM3.0 party and it's about time they showed up.
Having said that, ATi is late for the SM3.0 party and it's about time they showed up.
radar1200gs said:
SM probably isn't their thing. They sound like they're into swinging though - so many partners.
R520 should be back from the fab by now, given the reported tape out timeframe.
It's pretty much always the IHV's who provide the forward direction, and Microsoft who then writes specifications based upon new hardware. It's only been very recently, with DirectX 9, that Microsoft has attempted to set the spec before any hardware was available (with SM 3.0). Do you think it's a coincidence that the R3xx is essentially the bare minimum for PS/VS 2.0?radar1200gs said:
And even with PS/VS 3.0 on the GeForce 6800, the NV40 still managed to provide forward direction with its support of floating-point blending and filtering.
that`s my point, i don`t think that there is much or any improvement with SM3.0. i`d rather like 2.0+(2.0 with higher shader limit, and some form of instancing/branching maybe) and more "horse power" then a full 3.0 support. i think SM3.0 need to be called 2.0X because it`s realy not much of an improvment over the 2.0. and the DXnext spec should be called SM3.0 because there`s should be "some" improvment there...carpediem said:
and i realy dislike the MS "all or nothing" aproach to DX.
i dont care that they take a bunch of feats and call it SMX.X. what i do care is about card`s with partial support, i dont think that a test for GI support sould be "IF VS=3.0 then..." it should be "IF GeomtryInstancing then".
I'll abuse this thread and make a question about R520.
It seems to be common knowledge (?) that R500 is using unified shader model marchitecture etc. while R520 is based on R300. Why R520 is based on old generation after all this time between R300 and R520 where as R500 is really a new generation? And has this "R520 is based on R300 marchitecture" been stated somewhere?
It's also quite big risk to develop completely new marchitecture for console which is millions - i.e. you can make only minor mistakes. In PC world we customers are used to beta testing new hardware so it would be not so big risk to test something so new in PC environment first, IMO.
It seems to be common knowledge (?) that R500 is using unified shader model marchitecture etc. while R520 is based on R300. Why R520 is based on old generation after all this time between R300 and R520 where as R500 is really a new generation? And has this "R520 is based on R300 marchitecture" been stated somewhere?
It's also quite big risk to develop completely new marchitecture for console which is millions - i.e. you can make only minor mistakes. In PC world we customers are used to beta testing new hardware so it would be not so big risk to test something so new in PC environment first, IMO.
Paju, read earlier for the reasons for why one not the other.
As for the risks - its thought that the unified architecture isn't new for ATI as this is what the, defunct, R400 was going to be, but has since been extended on for the XBox and will likely form the basis for their Longhorn graphics architecture.
As for the risks - its thought that the unified architecture isn't new for ATI as this is what the, defunct, R400 was going to be, but has since been extended on for the XBox and will likely form the basis for their Longhorn graphics architecture.
I think I read a mumur from someone here (could it have been you Dave? or maybe from a CC?) that hinted that the R400 did actually tape out. Any idea if they had one up and running?
That would be as good a test bed as any to prove that ATI could handle a unified shader part, albeit far more basic.
That would be as good a test bed as any to prove that ATI could handle a unified shader part, albeit far more basic.
tEd said:
That's what you normally get when you go from one process node (120 nm) to the next (90 nm) ;-)
Though I hate always to think in terms of transistors count. Normally (at least in my company) we tend to impress people with gate counts and RAM/ROM sizes. If you have lots of embedded memories, you get lots of transistors, but memory is something easy to do.
DaveBaumann said:
Does this mean the XBOX2 GPU is again a design from the former ArtX people ? Makes sense, then the other design centers can just further hack & extend the R300 (m)architecture.
Or in different words, mid next year the R500 design is handed over to the other team to make PC derivatives... (R620 ?).
R300, R420, R520 --> West Coast.
R400, R500 --> East Coast.
R400, R500 --> East Coast.
I find it funny that people think the 420 was conservative. When I first heard the specs I thought it was very agressive, still do. It's just that Nvidia was even a little more agressive. I'd bet R420 and NV40 are both two of the biggest chips ever sold in the consumer market.quest55720 said:
