View Full Version : NV31 & NV34 - Announced on the 6th, so let's try to do last-minute speculation!
Hey everyone,
Since the NV31 & NV34 are now officially launched on the 6th, let's try to have the correct specs speculated ( although much comes from rumors ) before the official announcement!
NV31 high-end model:
- 350Mhz core
275Mhz DDR-I memory with a 32x4-bit memory bus
85M Transistors
- "True"' 4x1 ( unlike NV30's "false" 8x1 ) *but* with 6 Pixel Shading units ( the NV30 got 8 ) , 75% the per-clock Vertex Shader efficiency ( same pool of units system )
- CineFX & DX9 Compliant.
Uses minor software hacks in the VS, nothing major ( real VS still in hardware, though )
- 3DMark 2003: 2500, runs all the tests
NV34 high-end model:
300Mhz core ( could be 250Mhz )
250Mhz DDR-I memory with a 64x2 memory bus ( could be 225Mhz or 200Mhz )
45M Transistors
- "True" 4x1 ( unlike NV30's "false" 8x1 ) and with 4 Pixel Shading units ( the NV30 got 8, the NV31 got 6 )
5% the per-clock VS efficiency ( no true VS, slow TnL unit )
- CineFX Compliant, DX9 Compatible ( Can't be DX9 Compliant without a VS, ya know! And probably some other stuff too )
- 3DMark 2003: 900, runs all the tests
And you know why nVidia insisted so much that doing skinning for every light is ridiculous?
Because without a VS, that's super slow on the NV34.
Why does the NV34 lose so many marks?
Because in the first test, everything is done in the VS, and thus it's emulated on the CPU.
And in Game 2 & 3, skinning for every light pretty much *triples* the CPU work it would do with software skinning!
You know what? In Doom 3, the NV34 should look a lot better than in 3DMark 2003. And it should also be the case of multiple other games.
So, anyone else got some NV31/NV34 speculation? Or maybe, err, facts? You know we all love "speculation" ;)
Uttar
Nebuchadnezzar
03-02-03, 03:08 PM
nv34 5% vertex shader per-clock effciency?? :p
must be a typo;)
Okay, so let's call it 0% Vertex Shader efficiency. The NV34 *has no vertex shaders* , remember that! It's all CPU work, just like on the GF4 MX ( but the Pixel Shaders *are* in hardware, however )
I said "5%" because in some cases, it would still be faster than nothing, because there's probably a TnL unit.
But yeah, 0% would make more sense :P
Uttar
Originally posted by Uttar
Okay, so let's call it 0% Vertex Shader efficiency. The NV34 *has no vertex shaders* , remember that! It's all CPU work, just like on the GF4 MX ( but the Pixel Shaders *are* in hardware, however )
I said "5%" because in some cases, it would still be faster than nothing, because there's probably a TnL unit.
But yeah, 0% would make more sense :P
Uttar SOunds like a Xabre!!!
GlowStick
03-02-03, 05:33 PM
So if you guess it correctly do u get a prize :D
sebazve
03-02-03, 06:41 PM
Originally posted by Dazz
SOunds like a Xabre!!!
:bs: :barf: :barf: :barf:
Lezmaka
03-04-03, 02:15 AM
Seems like NV34 will be .15 micron
per that little excerpt I heard from jen hsung...
nvidia will have a 10 fold advantage over its competition in a 'very short period of time'
:rolleyes:
now I dunno what nvidia's plans are... I doubt many people are in the 'know' anymore... but for petes sake...
ok enough of my little rant...
I think I am with uttar pretty much because those numbers correspond nicely with other numbers I have seen floating round... :) mainly put forth by uttar lol...
Clockspeeds have been bumped because performance per clock is not particularly stellar and both are still slower than the Ti4200 in most, if not all current apps. Looks as if they are probably having to use more expensive PCBs than they would like as well (due to termination requirements).
:eek: :eek: :eek:
NV31 is SLOWER than 4200? :rolleyes: :confused: :eek:
jbirney
03-04-03, 08:58 AM
Originally posted by MuFu
Clockspeeds have been bumped because performance per clock is not particularly stellar and both are still slower than the Ti4200 in most, if not all current apps.
:eek: I hope your kidding or I am reading that wrong. Both are slower than the GF4 Ti4200?
If thats the case (and I hope note), then ouch baby :afro2: (bad austin powers joke)
digitalwanderer
03-04-03, 09:14 AM
IT'S A BLEEDING MX CARD WITHOUT THE "MX"!
Nope, not kidding. NV34 definitely slower, NV31 - well we'll see. Probably still slower 80% of the time, at least initially.
MuFu.
StealthHawk
03-04-03, 09:19 AM
Originally posted by MuFu
Nope, not kidding. NV34 definitely slower, NV31 - well we'll see. Probably still slower 80% of the time, at least initially.
MuFu.
NV31 will be slower even with AF/FSAA? looks like i'll be buying an ATI card for sure then.
I don't know, sorry. I imagine RV350 is faster for FSAA/AF than NV31 anyway.
MuFu.
I just want two more pieces of info now...
MIPS ( real numbers you can get with a synthetic benchmark, no theorical numbers please! )
MSRP
Well, the performance hit in % for AA & AF of the NV31 should be *smaller* than the one of the NV30!
Why? Because there's less transistors, but the memory/clock ratio is the same.
And all the AF/AA technologies in the NV30 should still be in the NV31 ( although they could be slightly less efficient to save bandwidth... )
So, comparing Balanced 4x AF & 2x AA with a GF4 *Ti4600* 4x AF & 2x AA, you can be sure that it's going to be faster. Playing with more, however, would give too low performance ( but it would also be faster than on a Ti4600 )
Now, the problem with the NV31 is its low MultiTexturing power. But ah well!
The NV34, however, might do just fine in its targetted market. Which is, basically, morons who want a "DX9 card" but got no idea what's the difference between two DX9 card. Oh, and they also try to buy exclusively GeForces :D
The NV31 is supposed to have 6 Pixel Shading units, compared to the NV30's 8. But, err, are they "fixed"? You know what I mean... The NV30 is operating, at high instruction counts, as if it had *1* Pixel Shading unit!
Uttar
Typedef Enum
03-04-03, 04:52 PM
Honestly...
Who is going to really care about these lower-end parts when...well, you know...ATI is about to launch even _better_ low/mid products??
I mean, I really fail to see how anybody could/would get excited by any of these low-end cards.
Not excited, just intrigued. This is a period of revolution for nVidia believe it or not. What they are doing now in terms of product design is the way forward, IMO. ATi has a *very* different approach when it comes to designing mainstream products. In playing it safe they've won this round already it seems. I'm sure they will follow in nV's footsteps later on - no doubt without tripping up so much along the way.
Had to delete my post at the request of the powers that be, sorry.
MuFu.
digitalwanderer
03-04-03, 05:01 PM
Originally posted by Typedef Enum
Honestly...
Who is going to really care about these lower-end parts when...well, you know...ATI is about to launch even _better_ low/mid products??
I mean, I really fail to see how anybody could/would get excited by any of these low-end cards.
I think you've failed to take into consideration the considerable amount of pleasure that certain people are going to have running around and bashing the lackluster performance of these new offerings or how much certain people are looking forward to that... ;)
I really don't see how ATI "won" this round.
Sure, they won the NV31 vs RV350 fight.
But what proofs you they won the NV34 vs RV280 fight?
We don't even know the RV280 clockspeeds!
Nor do we know how things like caches have been tuned on the NV34, which could reduce performance.
IMO, the NV34 vs RV280 will be an interesting fight. The RV280 got better VS power - it pretty much can do the same thing as a mid-end CPU, but without using CPU time.
But the NV34 might have better fillrate performance.
It's hard to say, really.
Uttar
Matthyahuw
03-04-03, 05:37 PM
Uttar, stop...
you need help, seriously...
They are no doubt going to sell a lot of crappy low-end NV34's (~250/200MHz).
The entire PCB BOM costs under $8 (4 layers, 8xTSOP chips on one side). :eek:
MuFu.
Originally posted by Matthyahuw
Uttar, stop...
you need help, seriously...
Why? :confused:
Because I lose too much time?
MuFu: I'm going to say this for the billionth time... MSRP? :D
Wanna bet the NV34 might actually cost *less* than $99, but rather about $89 or $79?
Woah... That would be interesting!
Uttar
digitalwanderer
03-04-03, 06:14 PM
Originally posted by Uttar
I really don't see how ATI "won" this round.
Sure, they won the NV31 vs RV350 fight.
But what proofs you they won the NV34 vs RV280 fight?
We don't even know the RV280 clockspeeds!
Nor do we know how things like caches have been tuned on the NV34, which could reduce performance.
IMO, the NV34 vs RV280 will be an interesting fight. The RV280 got better VS power - it pretty much can do the same thing as a mid-end CPU, but without using CPU time.
But the NV34 might have better fillrate performance.
It's hard to say, really.
Uttar
Keep the faith brother, keep the faith. ;)
Originally posted by Uttar
MuFu: I'm going to say this for the billionth time... MSRP?
Never cared about prices, names or anything above EE level. No idea and I don't intend to find out, sorry. :)
MuFu.
Originally posted by MuFu
Never cared about prices, names or anything above EE level. No idea and I don't intend to find out, sorry. :)
MuFu.
Seems logical to me! :)
But then, ignore that request, and check your PM.
Uttar
vBulletin® v3.7.1, Copyright ©2000-2012, Jelsoft Enterprises Ltd.