GF100 evaluation thread

[Silus]

You are free to shop for your card whichever way you like. In case you haven't noticed so far, the people who design gpu's care. The people who make B3D the best online technical forum care about die sizes, perf/mm and perf/W. A lot. I am not expecting you to care about these things. But please excuse me if I (and possibly others) don't share you indifference for die sizes, perf/mm and perf/W.

It's not indifference. I'm just not fixated by it, as you and others seem to be. Please excuse me for having an opinion

I hope Intel makes you lead architect for LRB2.

I hope so too! Thanks

If it heats up like a duck, and it is about as big as a duck, costs almost as much as a duck, then......

At least for me $500, is more than $400, but it's not closer than more than $600...But maybe that's just me...

Haven't seen it either.

So your 58% larger comes from...?

 

[Harison]

So you are stating (like some before) that NVIDIA wanted to be late ? Ok...that makes no sense, but if that's what you want to believe, then go for it...

AMD told everyone and their dog about their strategy for leaving high-end for dual-gpu cards exclusively, nVidia 100% knew what GTX480 will face, and it has nothing to do with being late or not. Fermi getting beaten by 5970 is not so much different than getting beaten by it half a year ago.

Single die ? You mean single chip...and of course it will. Dual GPU cards come later, not first. And I love all those inside sources you seem to have, on what NI will be, its performance against undefined competition and also that everything will go smoothly with it

As we all know, dual-gpu's come for nVidia after refresh, while AMD can launch it almost at the same time, because its their strategy, its not like I'm telling something new and unheard of :smile:. And yes - you can bet on NI high-end being dual (sometime in the future - probably multiple dies a la LRB). No idea if NI launch will be smooth or not, but basing on execution lately it makes sense to assume it probably wont be flop, and maybe on time even

 

[air_ii]

I couldn't care less about what NVIDIA says (though I would like to see where they've said that on record, just so I see it for myself).
It's about common sense. NVIDIA didn't want to be late. They wanted to release these cards at least at around the time Windows 7 launched i.e. before the HD 5970. Just because it's late you are somehow thinking that the chip needs to magically gain performance it was never designed to have.

But of course, you can compare it with whatever you want, even if the GTX 480 is not "more in the 5970 region".

GTX 480 MSRP is $499
HD 5870 is at best $390-400
HD 5970 costs much more than $600

How exactly is the GTX 480 more in the HD 5970 region ?

You misunderstood me. I just stated the obvious. All I said is that what it's going to compete against is going to be determined by prices that will settle after Fermi's become available. I didn't specifically say which card GTX480 is going to compete against. So, just try to read what I posted without overinterpreting it.

 

[trinibwoy]

AMD told everyone and their dog about their strategy for leaving high-end for dual-gpu cards exclusively, nVidia 100% knew what GTX480 will face, and it has nothing to do with being late or not. Fermi getting beaten by 5970 is not so much different than getting beaten by it half a year ago.

Those days are long gone ever since the 4870X2 - it consistently beat up on the GTX280. Nvidia's strategy isn't to do the impossible - which is beat two 2.1billion transistor chips on a combined 512-bit bus with a single 3.1billion transistor chip on 384-bit. If they or anyone else expected that to be the case they are out of their minds.

Their other angles with Tesla and Quadro sort of demand a big chip strategy though. AFR performance isn't relevant there. From a purely academic standpoint I admire what they're trying to do with product differentiation and growth of the business. It's execution that's killing them.

 

[Silus]

AMD told everyone and their dog about their strategy for leaving high-end for dual-gpu cards exclusively, nVidia 100% knew what GTX480 will face, and it has nothing to do with being late or not. Fermi getting beaten by 5970 is not so much different than getting beaten by it half a year ago.

It has nothing to do with it being late ? So they wanted to release their products based on Fermi, at the time of Windows 7 launch (October 2009) and you're arguing that they should already have enough performance in the GTX 480 to beat the dual GPU from ATI, that they knew was going to be released eventually ? Ok

As we all know, dual-gpu's come for nVidia after refresh, while AMD can launch it almost at the same time, because its their strategy, its not like I'm telling something new and unheard of :smile:. And yes - you can bet on NI high-end being dual (sometime in the future - probably multiple dies a la LRB). No idea if NI launch will be smooth or not, but basing on execution lately it makes sense to assume it probably wont be flop, and maybe on time even

Seems you do. You're already predicting performance and everything against an undefined competitor, which you think will be Fermi 2.

As for good execution, yeah NVIDIA was there since the NV40 days up to G92 and that didn't help them much, when they make such radical changes in their architecture. Problems happen. Let's see what happens with NI. Unlike you, I can't predict performance of something I don't even know a single spec of...

 

[Silus]

You misunderstood me. I just stated the obvious. All I said is that what it's going to compete against is going to be determined by prices that will settle after Fermi's become available. I didn't specifically say which card GTX480 is going to compete against. So, just try to read what I posted without overinterpreting it.

Well, your post was somehow assuming I was saying what I did, because NVIDIA said so ? When that's obviously not the case and you over-interpreted me as well. The case is in the prices we know about and they don't put the GTX 480 against the HD 5970 in any way, even if some want it to be

 

[XMAN26]

We can look on it from the die-size perspective => 60% larher than RV870
...or from price-segment perspective => 50% more expensive than HD5870
...or from price/performance perspective => same league as HD5970
only the performance perspective situates it at the 5870's level

overclockers perspective is also interesting - GTX260-216 at the end of fall was nearly three-times cheaper than GTX470 now. Good OC makes it only 10-15% slower than GTX470 in many games...

%50 more than 5870, exaggerate much? Try 20-25% more as I dont see the MSRP being in the 6-700 range in any review. Current average cost for 5870 is in the 390 range, 480 is about 530. Yeah, 50%.

In 6 months after several driver updates, price/performance for the 470 and 480 should be revisited, but right now, they prices dont make any sense.

 

[air_ii]

Let's take the ever popular PhysX. By switching I eliminate that option and the decision is made for all titles - upcoming Mafia II included for example. What each person should do is discount their future perceived value of all PhysX effects in all games they will play during the lifetime of the card. For some that will be zero. For others it will be higher, maybe significantly so. I for one own Batman and though I haven't played it yet I have seen what the additional PhysX effects bring and whenever I get around to it I plan to have those effects enabled.

Same goes for CUDA. Thus far its only use to me has been running benchmarks, demos, playing with the SDK, contributing to OpenCL/CUDA comparisons, messing around with Just Cause 2 settings etc. Nothing practically useful but still something I have access to and have used in the past and therefore is of value to me as someone interested in the technology. I don't have to just read about it on the internet.

Now the argument on the other side of the coin is that all of that is useless and I should give it up for the quieter fan. Not very convincing....

That's fair enough. I was more excited about the fact that I can run some DX11 titles, demos, benchmarks and DC5 fun. Sure, they were available at even lower quantities (4, 5 maybe) than PhysX titles, but it was (or I was hoping it to be) something refreshing. I did keep the 285 for a while (mainly for PhysX), but sold it shortly after.

BTW, you should check this out. It looks like nVidia were holding back on CPUs while developing PhysX.

 

[rpg.314]

So your 58% larger comes from...?

Stuff's gonna come out even if nv doesn't acknowledge it. I meant, no 'blessed' numbers were given out.

 

[XMAN26]

You do realise that just because nVidia says that it's not competing against 5970, it doesn't make it necessarily true? It's all about price and performance. If price is more in the 5970 region, it will compete against that card.

You misunderstood me. I just stated the obvious. All I said is that what it's going to compete against is going to be determined by prices that will settle after Fermi's become available. I didn't specifically say which card GTX480 is going to compete against. So, just try to read what I posted without overinterpreting it.

You didn;t, sure as hell reads like you did?

 

[air_ii]

Well, your post was somehow assuming I was saying what I did, because NVIDIA said so ? When that's obviously not the case and you over-interpreted me as well. The case is in the prices we know about and they don't put the GTX 480 against the HD 5970 in any way, even if some want it to be

Not really. You said that GTX480 was never meant to compete against 5970, as in it was not designated by nVidia to do so. I don't know how else you can read that?

You didn;t, sure as hell reads like you did?

No, see that if (as opposed to since or because)?

 

[Harison]

Those days are long gone ever since the 4870X2 - it consistently beat up on the GTX280. Nvidia's strategy isn't to do the impossible - which is beat two 2.1billion transistor chips on a combined 512-bit bus with a single 3.1billion transistor chip on 384-bit. If they or anyone else expected that to be the case they are out of their minds.

As one Nvidia fan is claiming, die size isnt important, performance is If you are assuming GTX480 wasnt the competition for 5970, is it a sign nVidia is abandoning the fastest "halo" card? I dont think so, therefore IMO nVidia hoped GTX480 will be good enough to be THE fastest card. X2 after refresh wont really matter, AMD will dominate highest-end till then, and at X2 launch they might have refresh of their own, and thats assuming Fermi wasnt late, now NV refresh will have to face NI.

 

[XMAN26]

Situation is strikingly similar (R600 and Fermi launch), and yes - nVidia knew AMD's fastest new gen card will be dual, while they insisted of making massive single die high-end card to compete with... and it didnt went so well.

If nVidia would have planned to launch with X2, their chip strategy would have been different. We can expect X2 after refresh, but Fermi2 will be single-die, again, unless they learned from their mistake. And by the way, Fermi2 will face dual NI card, and probably will get beaten again. As you can see, AMD strategy pays off not only for mid-level cards margins, but eventually they took over Top card as well, and it seems for a long time.

Nvidia's stratagy is to have the FASTEST SINGLE GPU in single GPU comparisons. They will eventually release a dual based card designed to take on the 5970. This insistence that the 480 was ment to take on the 5970 is just completely insane.

 

[rpg.314]

Those days are long gone ever since the 4870X2 - it consistently beat up on the GTX280. Nvidia's strategy isn't to do the impossible - which is beat two 2.1billion transistor chips on a combined 512-bit bus with a single 3.1billion transistor chip on 384-bit. If they or anyone else expected that to be the case they are out of their minds.

I wonder where you stood during the days of Cg?

Their other angles with Tegra and Quadro sort of demand a big chip strategy though.

/scratches head
Why does Tegra need a big geforce chip?

 

[XMAN26]

No, see that if (as opposed to since or because)?

Here, let go over your statment

You do realise that just because nVidia says that it's not competing against 5970, it doesn't make it necessarily true? It's all about price and performance. If price is more in the 5970 region, it will compete against that card.

The first 2 sentances you are insinuating that the 480 is ment to compete with the 5970. Then in your last sentance you TRY and justify you claim with an IF. only the price is nowhere near 5970 range, so to try and make that claim is simply nuts.

 

[air_ii]

Nvidia's stratagy is to have the FASTEST SINGLE GPU in single GPU comparisons.

There, there.

 

[trinibwoy]

If you are assuming GTX480 wasnt the competition for 5970, is it a sign nVidia is abandoning the fastest "halo" card?

There's certainly a dual-chip Fermi something or the other coming eventually. Just because they're late it doesn't mean they aren't trying.

That's fair enough. I was more excited about the fact that I can run some DX11 titles, demos, benchmarks and DC5 fun.

Which you knew was only temporarily unique to AMD hardware. The difference is that you know CUDA and PhysX won't be running on AMD's stuff anytime soon.

I wonder where you stood during the days of Cg?

Not seeing the relevance...

/scratches head
Why does Tegra need a big geforce chip?

Oops, that should have been Tesla. Darn marketing gobbledygook.

 

[Silus]

Not really. You said that GTX480 was never meant to compete against 5970, as in it was not designated by nVidia to do so. I don't know how else you can read that?

From this:

The GTX 480 was never meant to compete with the HD 5970. Never did NVIDIA counter a dual GPU with a single GPU and this time, it's not different. Even more so, because of the delays

How you conclude what you did is beyond me. I even mentioned the delays. They didn't want to be late, so they clearly wanted to release their cards with Windows 7. I really don't know how they were supposed to magically have the performance up to beat the HD 5970, 5 months later, given that the design was done and they were bringing up prototypes during GTC 2009...
NVIDIA wanted to pitch their best against ATI's best, but they certainly didn't want to be late.

 

[John Reynolds]

This insistence that the 480 was ment to take on the 5970 is just completely insane.

Wow, deja vu and all that. I do believe you've been thoroughly punked on this specific topic for your posts when the 5870 launched. There are things called search engines, you know, so you might want to work on your consistency when debating relative performance expectations.

 

[air_ii]

Here, let go over your statment



The first 2 sentances you are insinuating that the 480 is ment to compete with the 5970. Then in your last sentance you TRY and justify you claim with an IF. only the price is nowhere near 5970 range, so to try and make that claim is simply nuts.

Fine, if you insist...

"You do realise that just because nVidia says that it's not competing against 5970, it doesn't make it necessarily true".

This sentence means no more, no less than what it says above, which is that nVidia might target or say that the card is to compete against a certain card, but it will be verified by final prices (on both sides - remember that ATI can act as well on that front to effectively shift competition). Why do you insist on reading between the lines where there's nothing there?

I guess you have trouble with both understanding and counting (it's 2 sentences in total, not 3).

How you conclude what you did is beyond me. I even mentioned the delays. They didn't want to be late, so they clearly wanted to release their cards with Windows 7. I really don't know how they were supposed to magically have the performance up to beat the HD 5970, 5 months later, given that the design was done and they were bringing up prototypes during GTC 2009...
NVIDIA wanted to pitch their best against ATI's best, but they certainly didn't want to be late.

I'm glad that we finaly undestand each other. It's all that matters, isn't it?