Valve Half Life 2 Presentation and Benchmarks

What do you mean by ain't gonna be the same NV40??????

 

There were some rumors floating around last week that nVidia was going to rush out a bumped up NV38 and just call it the NV40 or else they were lowering the original NV40 specs so they could get it out the door earlier.

Just rumors, but I wouldn't put it past nVidia right now.

 

Then what is the point for Nvidia, destroying the worthy/successful name NV40. [People then will think NV4x same crap as NV3x]

 

I'm not disagreeing with your asessment at all as I happen to agree with it, but from their marketings point of view I believe the logic would be to be able to have a card out by X-mas that would give ATi at least a run for their money....but I don't see how a hacked NV3x would either.

 

Either NV40 isn't ready or isn't a ATI killer. This means that in lieu of a competative product, the only thing Nvidia has to sell is... more marketing spin.

As has been discussed in other threads, Nvidia has painted themselves into this corner and now that's all they *can* do.

 

What are you saying; story repeats, like ATI R300 vs. Nvidia NV30.

 

Yeah, that's kind of what I think it's going to shape up to be like...for the next couple of years actually.

I really don't think that nVidia has "got it" yet. I have no idea why they wouldn't, but I truly feel like they don't.

It still has a ways to get worse before it'll start getting better for them I fear.

 

Based on the last Shadermark2.0 tests I know of, the NV35 ( FX5900-Ultra ) is only around 41,67% as fast as the R350 when using PS2.0 ( regardless of _a or not when using FP32 ) and around 59% as fast when using PS2.0-pp.

This means an hypothetical NV40 with 8 Pipelines would have to be around 510MHz fast to match the R350 when the PixelShaders would not have been improved.

If the speed of the NV40 PixelShader @ FP32 matches the speed of the NV35-PixelShader @ FP16 then the NV40 would have to be clocked at only 360MHz to match the R350.

But, if it is true that the R420 is an 12-Pipe design at 500MHz then the NV40 would fall back compared to the R420 again cause the R420 would be nearly 2times faster than the R350 even then when the PixelShader of the R420 would not have been improved at all.

So if Nvidia really uses an evolutionary step from NV3x to NV4x they will most likely loose again. Not very nice for them.

 

I believe ATI R420 will be alike R300.

[example R300 325MHz Core & 310MHz memory, 8 pipelines and 1 texture unit.]

So R420 will be 525MHz core 510MHz memory, 12 pipelines and 1 texture unit.

I don't have a good guess about Nvidia NV40.

How can you be sure Nvidia NV40 will follow NV30 footsteps???

 

Can hypothetical NV40 have 12 pipelines instead 8...

 

Okay, where did this "12 pipes" vogue come from?

 

It's a Dead thing, "The Twelve Pipes of Christmas"...similar to the Yuletide Logs tradition but for those who don't like joints.

 

Very unlikely, the NV40 has a transistor budget of 175milion the NV35 allready has 135milion transistors adding 8 pipelines with only 40milion extra transistors is impossible.

I am going to very very suprised if the NV40 is not 8x2.

Edit: I just can't write English

 

Hmm, is pipelines in the traditional way still all that matters in this day and age?

 

Pipelines are going to exist. The only thing is where and in what sense. Of course there will always be the concrete sense for increased clock rate and then there is the abstracted sense which is peddled to the masses because it conveys path and flow.

As for pixel pipelines, they'll exist at some level, leaning more towards the abstract and I think the next offering from the big 2 will change what we mean by pixel pipelines, we'll be talking about units, which will perform close enough to arbitrary operations on each pixel.

Additionally, I'm not convinced that the Nv40 will resemble the Nv3x series. The latter is a hodgepodge of crap. Many claim that sea of math engines garbage is forward thinking, ultimately the statement itself maybe, but what Nvidia interpreted it as was nothing but super old skool. Specialty units every where rather than a large powerful shared execution resource pool. I expect something either in between P10 and R300 or maybe something even more general.

Finally, I believe that ATI is going to be doing something very interesting with the R4xx, I have a feeling that it will not necessarily offer insane performance increase, but rather it'll break down the somewhat arbitrary boundaries put in place by the current organisations of GPUs.

 

Could you expand on that a bit please? You have me most curiously intrigued.

 

Oh screw that!....I want a higher benchmark score!

 

I'm not sure if I should be saying much further. First off, this is the underlying trend I see:

R100 -- lots of interesting features implemented by many fixed function units.
R200 -- more generalised execution resources and super set of features with minimal legacy architecture and less restrictions.
R300 -- further generalised execution resources and a greater set of features with minimal legacy architecture and further reduced restrictions.

Here is something to chew on:

Realise that their competitors were leveraging their legacy architectures/organisations much more heavily. Rather than evolving, ATI makes a departure from the old and carries forward lessons learned. This subtle difference is rather important. Now this is the easy part, what's difficult is how far forward will the next step be. If you follow this to conclusion, how far off does a larger pool of execution FP resources for at least the shader units seem? I think it's a trivial step and thusly things will go much further -- especially, if current restrictions are to be reduced.

It's interesting to think about what EXACTLY did Intel and ATI swap in their agreement, it must be more than simply the bus license. What if they got access to various technologies that one might find in CUs, BP ... -- not a typo.

 

Thats a pretty interesting picture you've drawn there....pretty believable too. We've got some nice things happening coming up pretty soon.

 

In everybody's eyes who will take the crown ATI R420 or Nvidia NV40 in terms of technology advancement/Quality/Performance.

Who have good perspective could/will answer this question wisely?