When Tuesday does the G70 NDA expire?
- Thread starter digitalwanderer
- Start date
A pretty useless set of benchmarks at this point really.
Wake me when we get SC:CT, Chronicles of Riddick etc benchmarks. I'm very curious to see HDR performance, as that's where even the Ultra falls on its face right now.
As for everything being CPU limited - well, not everything. I see that big ol' bandwidth limitation rearing its head as well.
AA with alpha textures - didn't expect that though. Very nice.
Wake me when we get SC:CT, Chronicles of Riddick etc benchmarks. I'm very curious to see HDR performance, as that's where even the Ultra falls on its face right now.
As for everything being CPU limited - well, not everything. I see that big ol' bandwidth limitation rearing its head as well.
AA with alpha textures - didn't expect that though. Very nice.
geo said:dizietsma said:
Nice to see the top oc'ers get some sampling love.Yeah, I understand this is as much a symptom of "king of the hill" (or "capture the flag" as Wavey prefers) marketing as anything else. . .but I always like to see excellence in any field rewarded.
True. Also, Sampsa seems to have his up to 525Mhz as well so that is now 4 that have gone to or past 500MHz ( out of 4 ) . I think this means that nvidia can easily do a faster Ultra at say 500Mhz if they wish.
Also, the 1200MHz effective ram seems to be hitting well over 1350, maybe they have improved the RAM, improved the PCB design, improved the power distribution or maybe relaxed the timings a bit ?
http://www.xtremesystems.org/forums/attachment.php?attachmentid=32835
Anyone brave the waters over at nVidia's forums or nVnews lately? I'm curious to hear about the reaction of the faithful to the news so far, but I can't bring meself to actually visit those places much anymore.
I never said there was one, I said there were two, but each was akin to "half" an ALU since only in combination did they have the full set of SIMD operations. We are all aware of the terminology "shader top/alu top" and "shader bottom/alu bottom" used to refer to this split. The NV4x thus had about sort of a "1.5 ALU" per pipe. Now, the "shader top" part of the NV4x has been upgraded so that it is now capable of pretty much everything the shader bottom is, so one can say that the G70 has 2 full ALUs per pipe, whereas before, it did not.
Now matter how you slice the semantics, the result is the same. 2x the dot-products.
Now matter how you slice the semantics, the result is the same. 2x the dot-products.
Very wise plan, I'm trying to get Hanners drunk myself....Mordenkainen said:They're awaiting B3D's review before passing judgement.
christoph said:
Splitting hairs. I clearly say there are 2 half-ALUs. DaveBaumann uses the "split" terminology too:
DaveB said:
The NV4x does not have 2 full ALUs. In totality, it has something more than 1 ALU per pipe if you compare functionality with the R300 or NV3x, but it is not 2. The G70 top/bottom ALUs now have almost identical SIMD functionality (scalar/special ops will still be different probably)
Demo, I'm not trying to split hairs and would have been okay with the 1.5 ALU statement. I just didn't get how ALU2 remains unchaged, but only counted as .5 before and 1 now, but I do understand your point. Sorry.
While it obviously has more pipes, it may, like the NV40, only be capable of 1 DP per clock while texturing. So maybe best to say 2x under certain circumstances.
While it obviously has more pipes, it may, like the NV40, only be capable of 1 DP per clock while texturing. So maybe best to say 2x under certain circumstances.
digitalwanderer said:
Not raving and not sobbing, people with 6800's seem to be thinking of passing. A lot of people are thinking that cpu limitations are much in evidence and so it is not worthwhile
Is the FX-57 out tomorrow ? If so then the FX-57 will surely help the cpu side out. An FX-57 at 3GHz and a 7800GTX at 500/1300 would be a nice match.
No, I agree. But to that same point, long shaders typically don't need a mad every cycle either. I just think it's more complex than just saying 2x. Certainly you could write a shader that is, but for most code it's an incremental improvement. The compiler should certainly be able to schedule instructions easier now!
Similar threads
