Nv40 16 full pipelines- The Inq.

This is where most of us draw the line. Selling your product as best you can is one thing - trying to defraud your customers is another. Personally, I don't want to buy products from a company who's business model involves trying to intentionally defraud me.

 

Food for thought: If nv40 is the monster card every one says it will be and ATI will be so surprised, I think a question is in order:

What will be the nv40 price point and can ATI deliver a dual R420 for that cost?

PS. they should have the same power requirements.

 

The original quote I'm replying to is that NV40 just has to be as fast as R300 with better features. I reckon it will have to do a lot better than that, and will have to be a lot better than R420, and that includes it's power/heat/noise requirements.

I disagree that no one cares about noise/power/heat. As has been said elsewhere in this thread, people don't want to have to buy new cases/PSU's to support a card that already costs a lot. People don't want to have their systems overheat and lose stability. People don't want to have to pay $500 and then still have to live with unbearable noise.

 

If the world was really simple, we would evaluate all consumer products to exact technical specifications, price, performance, etc and buy the optimal product without regard to marketing, image, branding, how it makes you feel, etc. Each generation, companies would go head to head on price/performance/features, and you would have an exact formula for which card to buy.

But this isn't a simple world. 100% of sales do not go to "the winner". People act like NVidia should have said "ya know, you got us. We can't beat ATI's chip across the board, so we should hang it up this generation. We will not be releasing any cards for another 12 months. " If you're not #1, you may as well not even bother? Despite the inferiority of the NV3x, it still works, it plays games (trust me, if you put it into your PC, it will play games) and since there aren't a whole lot of PS2.0 games, or even PS1.4 games, someone who buys a 5700 or 5900 isn't losing out that bad. I own an ATI card, but the NVidia cards are not inherently worthless.

 

Noone said they were wortheless, and that since the begining.

People said that Nvidia should have used prices to be competitive instead of cheats. The performance of the cards are not the problem, but he way that company decided to compete is.

 

A new card is a new card is a new card - thus should be judged by its own merit. I believe we'll see pretty soon after launch what it will be capable of, and that's the whole point, is it not? You buy a card for its perfomance, image quality, price point ( and some for longevity )... right? Do you think nVidia can come up with such nefarious cheats ( be it in silicon or driver ) that they won't be discovered shortly after the first few thorough tests? Or do you buy hardware blindly, only judging by the previous generation?

Memory is only useful if you know when and to what extent it is just to go out from previous experiences.

P.S.: just noticed DemoCoder's post - while this is truly not a simple world, we all should at least TRY to approximate the ideal buying decision. Cursing at either of the companies ( even for real injuries ) does NOT help that - especially if it's for the gazillionth time.

PatrickL : you can only play so much with price tags, even if you are as big as nVidia.

 

Are you suggesting that it's okay for companies to cheat and lie to their customers just because their competitors have better products?

ATI were not number one for a long, long time. They competed on price, IQ, other features (such as AIW). They didn't lie and cheat their potential customers.

Nvidia could have competed on price, on their brand name, customer loyalty, the (previous) good reputation of their drivers, the fact that a lot of developers used their cards as baseline, etc. Instead they decided to lie and cheat.

 

I agree that, given equal performance and quality, one would probably choose the quieter card. My point was that people will excuse loud cooling solutions if the card is faster than the quieter competition. Plus, undoubtedly quieter, slower parts will follow.

Power is a separate issue, and I'm not sure ppl who are spending $300+ (likely $400+) will particularly mind tossing in an extra $50 for a new PSU. I'm talking about the non-tech freaks, like us.

 

I don't think a 150W+ card will use power from main PSU. Most likely it will arrive with its own

 

People are more sensitive to sound than ever. In fact, the geeks like us who have lots of high end equipment and thus need more cooling are likely to be generate more noise than "normal" people. Our high end equipment generates more heat, needs more cooling, and can be far more noisy because of it. Noise is an issue for us geeks that it isn't for the low end brigade.

If you've already got a lot of gear and a high end PSU to run all that top of the range stuff, NV40 could double your power requirements. Check out the price of a high end PSU that can manage double your current power requirements stably. I know my 470 watt Vantec Stealth cost £100, which probably adds 20 percent to the price of NV40, and I'm not sure it could supply enough juice on the relevant rails.

There could be significant extra expense and hassle if you need new case/cooling/PSU once you've bought a NV40. Especially if you have to live though all the grief of trying to track down heat and power related failures beforehand. I tell you, it's a customer services disaster waiting to happen.

 

27 pages hm? Hopefully you've reached a verdict yet. If not, I'm afraid that the number of "pipelines" might even increase if it continues like that.

 

I can't see why nobody has proposed 16 full "pipelines" ... total.
Make 16 pipelines, capable of PS/VS 3.0. They will spend most of their times as PS. But once in a while, they'll run a batch of vertices.

Or maybe interleaved PS and VS, using the same functional units and register banks (certain registers allocated to PS, others to VS), but having separate control units. A higher level control selects which control unit (PS or VS) to issue next instruction.

Pretty optimal for load balancing, if they've got rid of the register penalty.

 

Or at least reduced it to a manageable level.

 

It would score great in separate PS and VS tests...

 

If NV40 indeed has interchangeable PS and VS units, how would it weigh the workload to assign any number of them to a certain task?

 

Yes:, the verdict is .... there's no verdict

 

Another benchmark winner ?

nVIDIA : "look, now 3dmark has returned to a good looking product, it evaluates our products in a good manner......"

But that would be really flexible.

 

Seriously? I haven't heard of anything like that since the V6!

 

Like this?

 

There seem to be many possibilities for this along the lines of implementing a vertex output buffer monitoring solution and then dedicating all pipelines with sufficient "backlog" (as others have mentioned in other threads), most speculation revolving around theories of what nVidia would need to be doing to the NV3x basic design to implement this.

Perhaps time to revisit some of the details of the "gatekeeper" patent documentation for analysis?