When the GTX 285 hit, if it was ~35% faster then the 2900XT and nothing had launched between(the 8xxx/9xxx generation) for nV, how good of a part do people think it would have been? The term 'catasrophic failure' comes to mind. That is a full process node drop. That is what the 7970 is compared to the GTX 580.
A comparison is only valid and useful when the circumstances are similar and any possible difference is taken in account.
The GTX580 was the fastest single GPU in the market until the 7970 hit the market.
The 2900XT wasn't the fastest single GPU card in the market in your hypothetical scenario, the 8800Ultra was.
Interestingly, the GTX 280 was fabricated on the 65 nm node while the 8800GTX/Ultra were fabricated on the 90nm node. So a full node shrink (or is that 2?).
And look
Is that 37% over the 8800GTX? So 30% or less over the 8800Ultra?
Now that is a similar comparison.
I think what Ben means (could be wrong here) is that it is a catastrophic failure when a competitors top new gen flagship is only at maximum 35% faster than the competitors last gen flagship. Also, he was comparing GTX285 performance over 2900XT (new gen flagship over last gen competitors flagship) to AMD7970 performance over GTX580 (new gen flagship over last gen competitors flagship).
It doesn't matter that the 2900XT wasn't the fastest single GPU card in the market in this example. Not sure how you figured it does?
And 90nm to 65nm is a full single node shrink. As opposed to a half step or two steps. Wasn't there some AMD/ATI GPU done on 80nm at some point? 1/2 step? Unsure.
About nodes:
90/80nm, 65/55. 45/40 (45 was cancelled and probably the 40nm is the 45nm renamed), 32/23 (and again the 32 was cancelled and 28 is probably 32 renamed).
The 2900 was a failure. Comparing anything to it is pointless.
If 35% over last competition flagship is a failure what shall we call a flagship that is only 35% faster than your own last flagship?
A success? Give me a break.
If we want to compare node jumps we either compare at first card in a node or at last card on a node. Reaching conclusion about a new node based on a new card on a new process to a card on a mature process is silly - imagine, that just like NVIDIA and AMD were able to extract around 20% more performance with their refreshes, AMD is able to extract 20% more performance from GCN? Doesn't seem that silly, considering 7970 that are able to OC to 1100+ on core already get that.
2900 was an architectural problem, 7970 seems to be an yield problem.
Have you noticed that TSMC keeps cancelling new nodes? What does that tell you?
Have you noticed that new games keep looking (IQ wise) the same for the last few years? What does that tell you?
I dont expect GK104 to overcome HD7970 in performance but i could see it be close/next to HD7950.
GK104 could have almost the same performance as HD7950 but at a much lower price and that's what i believe Charlie is implying.
You misunderstand the comment. BenSkywalker's comment was talking about a full node jump, not cards, so Gaia is on target here and brings up several valid points. How can you hold a company's progress based solely on a comparison of performance to its competitor's, especially when parts are compared in a multitude of metrics? It'd be like saying the GTX 280 and GTX 260 were complete failures because they were brought out on a 65nm process when AMD's 4800 series came out a few days later on a 55nm process. Each company has its own roadmap and you have to look at the progress from release to release and compare whatever is available on the market. Whenever Kepler arrives, we'll be able to see what NVIDIA can do on 28nm process, until then, it's all speculation about what a "full process node drop" this generation should entail performance-wise.Gaia, forgive me but I still have little idea why you're going on this node tangent.
2900XT was a failure in that it was soooo hyped up and failed to deliver. It was just about as fast as the 8800GTS640, Nvidia's second fastest at the time. Didn't do to badly, just nowhere near as good as hyped. Kind of like Barcelona.
So, unless you wish to get into a node discussion, that probably wasn't the focus of Ben's post which you quoted. Maybe he could chime in here.
I'll repost here what I believe the focus was:
I think what Ben means (could be wrong here) is that it is a catastrophic failure when a competitors top new gen flagship is only at maximum 35% faster than the competitors last gen flagship. Also, he was comparing GTX285 performance over 2900XT (new gen flagship over last gen competitors flagship) to AMD7970 performance over GTX580 (new gen flagship over last gen competitors flagship).
"If 35% over last competition flagship is a failure what shall we call a flagship that is only 35% faster than your own last flagship?"
Is that what we're talking about? A single company's new gen improvement over it's own last gen? Or are we discussing competitors products? I though it was the latter. In fact I'm pretty sure of it. ::shrugs::
Gaia, forgive me but I still have little idea why you're going on this node tangent.
2900XT was a failure in that it was soooo hyped up and failed to deliver. It was just about as fast as the 8800GTS640, Nvidia's second fastest at the time. Didn't do to badly, just nowhere near as good as hyped. Kind of like Barcelona.
So, unless you wish to get into a node discussion, that probably wasn't the focus of Ben's post which you quoted. Maybe he could chime in here.
I'll repost here what I believe the focus was:
I think what Ben means (could be wrong here) is that it is a catastrophic failure when a competitors top new gen flagship is only at maximum 35% faster than the competitors last gen flagship. Also, he was comparing GTX285 performance over 2900XT (new gen flagship over last gen competitors flagship) to AMD7970 performance over GTX580 (new gen flagship over last gen competitors flagship).
"If 35% over last competition flagship is a failure what shall we call a flagship that is only 35% faster than your own last flagship?"
Is that what we're talking about? A single company's new gen improvement over it's own last gen? Or are we discussing competitors products? I though it was the latter. In fact I'm pretty sure of it. ::shrugs::
IonusX said:hey dude ill stop comparing it to fermi when kepler launches so how about you tell jason to whip his team harder so he can race around the track too.
i only compare the racers on the field man. if you expect anything more than its YOU that needs help. you want to argue different okay
get me a gtx 680 review right now on the double and hesitation will mean you cant.
and thats how we do things back on the spot..
i dont get why nvidia fans are screaming for kepler. tesla wont be for 2 years if not more giving amd plenty of time to surpass it and reign supreme. 2013 the year without nvidia..
you make justin beiber look good..
You misunderstand the comment. BenSkywalker's comment was talking about a full node jump, not cards, so Gaia is on target here and brings up several valid points. How can you hold a company's progress based solely on a comparison of performance to its competitor's, especially when parts are compared in a multitude of metrics? It'd be like saying the GTX 280 and GTX 260 were complete failures because they were brought out on a 65nm process when AMD's 4800 series came out a few days later on a 55nm process. Each company has its own roadmap and you have to look at the progress from release to release and compare whatever is available on the market. Whenever Kepler arrives, we'll be able to see what NVIDIA can do on 28nm process, until then, it's all speculation about what a "full process node drop" this generation should entail performance-wise.
Exactly. :thumbsup:So if the NVIDIA flagship card is 30% faster than the 7970 is it a failure? Or is that new card competition the 6970?
So the 7970 competes with the GTX580 (assuming that is the name) and the GTX680 competes with the 6970?
So by that logic if the GTX680 is 40% faster than the 6970, being as fast as the 7970, it is a win?
And did the 5870 compete with the GTX285 but the GTX480 competed with the 4890?
In my opinion products compete with existing products.
At the moment the 7970 seems to be a solid product, being a better product and value (at least in my opinion) than the GTX580, which seems to be its competitor on that price bracket.
When the GTX680 arrives we compare it against the 7970.
There's no comparison made to current cards in that second quotation. If you're trying to argue an inconsistency, you didn't find one.So what is it, you can compare future GPUs, you can't compare future GPUs? Make up your mind kid...
I agree, and suggest you argue based on logic as to keep people honest :thumbsup:.I remember already stating this position (since I'm coming from an AMD card anyways) and was basically called a shill.
Looking at the previous AMD cards, the performance increase is consistant across generations. It's just amusing that people argue when compared to nVidia it isn't consistent (which it still is when you factor in nvidia always had the crown at the end of a generation.)
But everyone spins it so they're right. Change the conditions until you win, right?
I remember already stating this position (since I'm coming from an AMD card anyways) and was basically called a shill.
Looking at the previous AMD cards, the performance increase is consistant across generations. It's just amusing that people argue when compared to nVidia it isn't consistent (which it still is when you factor in nvidia always had the crown at the end of a generation.)
But everyone spins it so they're right. Change the conditions until you win, right?
I think the difference this time is the price. It just appears to be much more expensive for the performance gain per dollar.
There's no comparison made to current cards in that second quotation. If you're trying to argue an inconsistency, you didn't find one.
The GTX580 was the fastest single GPU in the market until the 7970 hit the market.
Interestingly, the GTX 280 was fabricated on the 65 nm node while the 8800GTX/Ultra were fabricated on the 90nm node. So a full node shrink (or is that 2?).
If we want to compare node jumps we either compare at first card in a node or at last card on a node.
I think what Ben means (could be wrong here) is that it is a catastrophic failure when a competitors top new gen flagship is only at maximum 35% faster than the competitors last gen flagship.
I think the difference this time is the price. It just appears to be much more expensive for the performance gain per dollar.
A comparison is only valid and useful when the circumstances are similar and any possible difference is taken in account.
The GTX580 was the fastest single GPU in the market until the 7970 hit the market.
The 2900XT wasn't the fastest single GPU card in the market in your hypothetical scenario, the 8800Ultra was.
Interestingly, the GTX 280 was fabricated on the 65 nm node while the 8800GTX/Ultra were fabricated on the 90nm node. So a full node shrink (or is that 2?).
And look
Is that 37% over the 8800GTX? So 30% or less over the 8800Ultra?
Now that is a similar comparison.
On the other hand the 4800 series undercut prices for quite a good margin, for market share and mind share reasons. i just don't know if it is valid to hold a company to the 4000 series standard, although it would be nice if this time NVIDIA does some fanning the flames.
Why would a consumer defend premium pricing strongly? I understand why an AIB would -- or an IHV -- or an AMD loyalist would at this time because that's the direction the company has taken now.
You see nVidia spending a lot to compete with themselves? How about AMD? The comparison to the competition is the one that is relevant in any market sense. If a company is pushing themselves around they aren't likely to last too long(not that it doesn't happen on occasion, but it isn't good business).
http://www.anandtech.com/show/2549/11
~50%-100% in almost every GPU strained bench(QW:ET being a notable exception). The page I linked isn't the biggest gap, I just try to be reasonable and link Crysis to use as an example of GPU limitations.
The 285 *was* nV's first high end 55nm part- I was careful to make sure I took that into consideration.
I think you might want to look up Intel...You see nVidia spending a lot to compete with themselves? How about AMD? The comparison to the competition is the one that is relevant in any market sense. If a company is pushing themselves around they aren't likely to last too long(not that it doesn't happen on occasion, but it isn't good business).
This line of thinking is incorrect if you're assuming that a full node shrink automatically guarantees a similar architecture improvement as well. You're lumping too many variables into "full node shrink," which alone didn't produce the performance increases we've seen from generation to generation. If you want to make the argument that AMD's architecture isn't as refined as possible on 28nm, OK, but more proof than an assumption is needed. NVIDIA's Kepler performance would be a good comparison to validate your claim. It could also be that we're starting to hit a wall in maximizing the performance efficiency of architectures, in that as GPU designs become more advanced, it's more difficult to improve them further. Again, whenever NVIDIA actually displays Kepler, we'll have a better idea.http://en.wikipedia.org/wiki/Die_shrink#Half-shrink
That is a full node shrink, and an extremely interesting chart you linked to-
http://www.anandtech.com/show/2549/11
~50%-100% in almost every GPU strained bench(QW:ET being a notable exception). The page I linked isn't the biggest gap, I just try to be reasonable and link Crysis to use as an example of GPU limitations.
The 285 *was* nV's first high end 55nm part- I was careful to make sure I took that into consideration.
Thanks for having my back Keys, but they have it right this time
In this industry the limiting factor is process availability overwhelmingly. This generation, for the first time I can ever recall, we missed a half node move and the only option was for a full node drop. Both AMD and nVidia had an oppurtunity to make enormous progress with such an oppurtunity. It looks like AMD didn't redesign when the half node got canned, and just went with their original plans. That is smart in terms of being first to market for obvious reasons. The fact that the cards are showing enormous OC potential backs this up. The problem is going to come when nV launches their parts, barring a major stumble by them, are going to handily best AMD's parts in the same price segment. I think AMD will likely have a refresh ready to roll by then, or perhaps the part they were planning to release on this build process in the first place, either way I was shocked when I saw how small the performance difference was given the huge die advantage they had.
People can try and act like what I am saying is absurd, go look at what these parts are hitting when they are OCd and then come back and talk
Neither Kepler nor Tesla has been released, and in that second quotation he compared unreleased GPU's against other unreleased GPU's. What's so difficult about that? Again, if you have a quotation that actually proves what you're accusing IonusX of, post it. Otherwise stop derailing the thread and wasting forum space. End of discussion.Really? Try reading it again, it will come to you...
That is exactly why NVIDIA didn't improve performance on the 80nm shrink, while it certainly did from the 65->65nm shrink.
Unfortunately techpowerup didn't have a breakdown by resolutions then. Sure, if you overload the 8800GTX/9800GTX memory limits, the GTX280 crushes it.
Likewise, if you use the super resolutions of today, 3xscreens the GTX580 loses by default (aside from custom cards).
Sure, but you could also use the example of 65nm->40nm as well, after all it is more recent.
Take into consideration these are watered down results including 1024 x 768, and the lack of enhancements. Using these makes them all much closer than they really are.
Why would a consumer defend premium pricing strongly? I understand why an AIB would -- or an IHV -- or an AMD loyalist would at this time because that's the direction the company has taken now.