Nv40 16 full pipelines- The Inq.

nv40 info

 

I have to call this bs. Although am a regular reader of the inquirer, I dont know why their latest news seem to be against Ati, I think they love underdogs, if nVidia can be called that right now.

 

You know 8x1 isnt as simple as ATi made it look, Nvidia better walk before they can run.

 

I thought I was exagerating, but that grapefruit DID crack me up!

 

So let's see... going from NV35 to NV40, 130M -> 210M transistors = 60% increase. And they expect us to believe that they quadrupled the number of pipelines? That would certainly be unprecedented. I'm also curious to see how much they will charge for this thing if its die size is actually 60% larger than a NV35. Oh, and the cooling should be fun.

 

Sorry, it is in french, but people tend to forget....

 

Well, given the info and what is now widely "known" of R420, you would expect just that, no less.

 

Well, first Yamhill info were from Mercury 2 years ago

 

It's Inq. when one has found anything non-suspicious from them?

 

If you wanna believe the 16 pipelines talk, what do you think the Z/Stencil rate will be like?

On a mildly releate not to this thread, I've recently been speaking with an analyst about the transistor count thing and here's something that he wrote:

 

Perhaps nvidia adds up all the transistors from ALL the dies (good and bad) on one wafer, and then divids that by the number of good dies?

This way, we can monitor how their yields are going simply by asking nvidia to give us the "NV40 transistor count of the day" at any given moment...

 

that is interesting dave...

so the tranny count and relative performance would be explained then (but the real-estate occupied? even on a smaller process )

 

Does nVidia think by redefining their terminology/definitions they're going to get a performance increase?

 

Silly dig....it's all about a marketing increase.

Though personally, all else being equal, fewer transistors is more impressive IMO.

 

It the 16 pipes are real, the likely it's 8 full and 8 Z/stencil only. Double that of GFFX in other words.

 

don't know if this is off topic, but has anybody established what are
the precise situations in which a NV31 and/or NV34
can act as a 4X1 and which are those in which can act as a 2X2?

 

It's another one of those "money-things", isn't it. I suck with the money-things.

Yeah, but isn't bigger better? Heck, more heat means they can throw on a really noisy cooling solution...and everyone KNOWS that's a "kewl leet status" thing amongst geeks.

 

That's quite suprising. Even a small amount of cache can mean a hell of a lot more transistors. Interesting, nevertheless.

NV30 supposedly had busted logic that would allow 4x2/8x1 operation (so 8 textured or 8 z-only max). I just wonder whether they've gotten it working in NV40. What's odd is that, if true, NV35 and NV38 carry this redundant functionality as baggage.

 

Besides those additional transistors from such places as cache, they could be adding some few million virtual ones no?

 

Hardcore gamers play with headphones 8)