Nv40 16 full pipelines- The Inq.

My first thoughts, on seeing:

followed by:

were very much along the lines of Toasty's post...

 

Pagh! Get a 5-button mouse and assign copy/paste/enter to the extry keys, it makes it easy. That's what I do.

LOL @ the Jay & Silent Bob Strike Back references. Not Kevin's best effort, but the funnest for long-time fans and totally self-indulgent.

Dave-What is the big deal if it's 16 fake pipelines or not, isn't that kind of what we were all expecting?

 

I just don't see what's fake about them. Uttar said they were full pipelines, and could process 16 pixels in the same way NV30 can do 4. What exactly is fake about them?

 

We don't know, yet..... :?

(This post is dedicated to Vince with hugs & love.)

 

All this talk about pipes. 16 pipes for the Nv40? Are you sure they aren't talking about how many pipes filled with Hallucigenics ATI would need to smoke to think they could hold the performance crown for long? The one they took a year and a half ago?

 

16 pixel pipes, 32 z-only/clk, 400MHz+, <51.2GB/sec raw bandwidth. Niiiiiiiice.

MuFu.

 

wonder if this thread will make 16 pages

 

As someone else asked earlier, isn't that more bandwidth than the AGP port can handle?

 

The bandwidth has little to do with AGP port bandwidth. It's bandwidth between onboard video memory and the GPU. Even if PCI-Express could even come *close* to that bandwidth (it can't, it's not much faster than AGP8x), a typical PC's system memory could not supply data fast enough.


AGP bandwidth is only relevant to texture upload, GPU commands, geometry streams/buffer uploads, and in a minority of cases, AGP texturing. PCI-Express won't change this, except to make some algorithms that work better on unified memory arcitectures possible (e.g. using CPU to post-process framebuffers)

 

That makes sense. Thanks for taking the time to explain it DC, I appreciate it.

A buddy of mine (and boss) over at EB posted up:

Is that right? The 125w figure? (Or should I say, "Has anyone else heard the 125w figure bandied about?")

 

Nothing that low!

 

125W That's more than Prescott, eep. 2 slot cooling wasn't good enough for them this time, apparently - they had to go 10 slots better.

 

125W (3-rail peak) does sound low for a 200 million transistor part @400MHz+.

I wonder what yields are like?!

 

Yeeeeeeeeeeeeeek



Damnit.

 

125W... h0ly cr@p, that's the combined power dissipation of my GPU *and* my CPU... will the card ship with a water cooling setup or what? If so, how could Gainward up the ante - slap a cryogenic unit on their Golden Sample card?

 

--> http://www.kino.orc.ru/js/review/stills/rocketeer.jpg <--
Please, stay on topic after that Dav will burn me with 10 nV40 otherwise !

 

A lot of people would say ATI sucks. The top spot is psychologically important, even if most people don't actually buy it themselves.

 

Sounds like they had some very ambitious plans for NV45 which they've somehow pulled off, and now it's a case of NV40 being an "AGP version" of NV45.

Might be in ATi's best interest now to get R420 out of the gates first - they could buy themselves a couple of months at the top. This could all be an elaborate trick to fool them into doing so though.

 

125W? Mmmmm. What was the 5800 Ultra?