Nv40 16 full pipelines- The Inq.

Is it so hard to believe that it might have 16 pipelines? I mean, this *was* done in 1999-2000 with PS2's GS. The GS is in fact a 16x0 configuration. or call it 8x1. how many transistors does GS have not counting the 4 MB eDRAM? (total is 43 or 46 million) - probably less than 20 million is logic in GS.

NV40 is at least 175 million transistors. could be as high as 210 million.

I'm starting to think perhaps it could be a 16x1.

 

Uhm, I think so...unless this is some kind of trick question about how we define "full"....I'm kind of confused right now.

I no longer believe the Inquirer's story that we're discussing in this thread, if that's what you're talking about. I got an interesting PM from someone who talked to some guy who spoke to a person from nVidia today who I don't want to mention and they said "no".
(He-he-he, I just re-read that paragraph and laughed out loud...it sounds like a REALLY bad gossip columnist's line, but it's actually accurate. LOL~~~ )
It may have a 16x1-type functionality, but it ain't gonna be a "full" 16x1 by any means is what I now believe. This is just purely a rumor, IMHO.

 

Final clocks almost certainly won't be set yet - those were probably extrapolated from current performance data.

Seems to be some reputable objection to it being labelled "16x1" now by people under NDA, hehe. Round and round we go...

MuFu.

 

No trick q's from me dig, I was asking if it was a true 16x1. Maybe it can do the 8x2/16x1 trick depending on the situation, as was suggested a while ago. This makes more sense, as I couldn't see how they would fit 16 pipes into 200M transistors, with all the other things that are req'd with said increase in pipes.

BTW, my cousin knows this guy who knows this chick that's dating this guy who knows someone that works as a cleaner at nVidia who overheard a conversation saying it would be 64x1! But he didn't know what "it" was.........

 

I didn't design it, ThermalTake did. Talk to their engineers about thinking about case airflow. The XaserIII SuperTower comes with 7 case fans, I didn't install them, it was designed that way. But you can independently control them with an analog throttle on the front of the case. Do you have 6 10,000 rpm drives in your case, plus CD-RW and DVD+RW?

I bought the case because it looks cool, has uber drive bays to hold all my disks, and I like the way it works internally compared to Antec's designs. The fan noise is minimal, but fan noise, or any white noise, doesn't bother me at all.

Yeah, and how many times have we heard this? SFF's are great for the living room, but there is no "revolution" with people replacing their desktops with SFFs. The closest revolution we've had in years was blade servers, which vastly outnumber the tiny SFF market. SFFs remind me of the diskless revolution. Hardcore users talking about how great these ergonomic, easy to use, devices will be for everyone else, yet they continue to buy ubertowers. Or people who complain about the environment and how everyone should drive hybrids or electric vehicles, but drive sportscars themselves.


The CPU vendors are designing for performance, for desktops and servers, and not for quiet home theater devices. The trend is that CPUs will continue to increase in performance and draw more power, because that's what the market is demanding. SFFs are a niche market, just like iMacs, just like the Apple fan-less cube, just the glamour cases.

If someone makes a chip with 50% more power, but it needs a larger case, more power, and a better cooling solution, I'm there, because performance is my number one concern on my desktop. Yes, GPU vendors will sell cut down versions for the non-performance market, and my bet is, the vast majority of customers who buy SFF designs aren't buying the top performing parts, if anything, they'll be the PC equivalent of iMacs and e-Machines for the general market. I see the majority of SFF users looking more at cards like the Radeon 9600 AIW instead of the Radeon 9800 XT.

 

Seems silly to me. Nvidia hasn't claimed anything official. This in Inq rumormill. Based on the R500/Xbox2 rumormill hype (uber amounts embedded ram), if it doesn't pan out, can we see lots of wise cracks about ATI hype?

I have so far seen atleast 4 different rumors about Nv40 that are all different. Hell, before the NV30, it was supposed to be called "Eclipse" according to rumors.

 

I'm prety sure I read here at B3D that 1 TMU costs 10 - 12 million transistors. How the bleep could they increase NV35 by 4X2 pipelines if thats over 80 million transistors right there? What about all the other necessary support functions for that level of processing? Perhaps the 10 -12 estimate was wrong, or has been improved on? I have many questions? And not many answers?

Caps?

I need help over here?

 

That's kind of an apples to oranges comparison. The "pipelines" in a modern GPU are alot more complex than in GS, which is why the transistor counts are so much higher.

 

The 10-12 million estimate is most assuredly wrong, as you can see by simply plugging in some numbers. NV10 had 4 TMUs, and the entire chip was 22 million transistors, far bellow 40-48 millions that otherwise would be required for TMUs alone. 4x4 Parhelia has 80 million transistors, as opposed to 160-192 for TMUs alone, etc.

 

What is a "bleep"?

 

ah, thanx Geeforcer.

Caps

 

Its the sound the Matrix makes when you accidentally get whacked in the bleep.

Caps

 

My, my, I hope B3D has some spare bandwidth tucked away.

My first thought was, "How's nV gonna fit 16 pipes in 200M transistors, when they only shoehorned four into 130M?" Plus, a leap from four to 16 pipes seems, well, large.

My second, when I read Dave's comments, was that this NV40 rumor may be indirect confirmation R420 is 16 pipes.

Then I thought, hey, if both NV40 and R420 will be 16, then the mainstream cards may be eight, which may explain the recent nosedive 5900 and 9800P prices. But will they keep 256-bit buses? If not, then each pipe may be somewhat starved for bandwidth, and given the propensity for newer games to use more shaders, four extreme pipes on a 128-bit bus may make more sense for DX9 games.

But if NV40 is really 16x1, then does NV36 really have "4 rendering pipelines" (along with "2 pixel and 3 vertex" shaders), as this Xbit news item says? Did nV make some further improvements to their core which accounts for the 5700U's surprising performance relative to the 5900 in some benchmarks--improvements that made their way into (or came from) NV40? Or did I just cross a bridge too many in speculation town?

I concluded that I really know too little about this to contribute. Plus, when rumors begin flying that nV is going to FP24, then the whole world's gone mad. I think I may just warm the sidelines until ATi and nV decide to toot their own horns.

 

Re: My, my, I hope B3D has some spare bandwidth tucked away.

My first thought was, "How's nV gonna fit 16 pipes in 200M transistors, when they only shoehorned four working pipelines into 130M?"

 

We have enough trouble determining what "full" pipes are, now you gotta go and bring "working" into this?

Seriously, are you implying that NV35 has some borked pipes vacationing amidst its rolling plains of transistors?

 

I didn't say they always wrong.. just lately they have been getting it(some stories) wrong.

Anyways.. we'll see when the cards get laid on the table.

Just btw.. they also reported the that NV40/45 would get Doom3 vouchers .. which I take is true since there were rumours well before they posted this that Nvidia were after the Doom3 deal. <-- So this I take to be real. So ye not all news they post is crap.. just some of it i.e. DirectX9.1 and most probably this.

US

 

In this eventuality although the NV40 would rock at D3 & DX7/8 games I would question how much actual PS/VS power this thing's going to have and how much of an improvement we're going to see in areas such as AA. Although 210 M Transistors is a big jump over the NV3x chips it's not an unlimited supply if they want 16 rendering pipelines (whether some of them are VS based or not), compromises are going to have to be made somewhere.

 

Currentlly, I don't know if I'm still sleeping when I read all that's been said since last night....

Oh god........

 

Well, ATI fit 8 high performing PS2.0 pipelines in 107million transistors, so it stands to reason that if you can produce a chip with 214 million transistors (ignoring the duplication of vertex shaders, memory controllers, 2D, mpeg decoding, etc), you should be able to produce a 16 pipeline behemoth with high performance PS2.0 pipelines.

Of course, looking at # transistors alone is a perilous way to argue.

NV30 was a 120M transistor behemoth, but it had alot of extraneous stuff: FX12 units, register combiners, FP16 support, etc.

 

That doesn't mean too much either, TT tends to go for the wow-factor quite aggressively. There were a number of extensive tests about various case fan configurations after the blowhole-era, and it was a common agreement that two-three well placed case fans can outcool most uber-fan setups. But it's your box after all, so if you're happy with the fan noise, who am I to argue...?

Not so. YOU may be a power user with your dozen drives and whatnot, but I agree with Ed at overclockers.com on how the average user is getting more and more pissed at fan noise and outworldish power consumption. I don't know when you last saw the specs for an SFF box, but they're already quite powerful - and not just for HTPC but also for gaming, video editing, etc. I would not compare them to the dickless workstations - and don't even get me started on how borked the car analogy is ( oil lobby, anyone...? ).

( Sorry for the OT. )