NVIDIA Kepler speculation thread

[UniversalTruth]

650$ is a joke

It is a joke from the standpoint that if they continue with this trend, some people won't be surprised at all if the next big card- perhaps GTX 880 costs over 750-800 $.
This is a very serious problem...

Everything else written in the quoted post is a joke

 

[RedVi]

They're going for $750-850 here.... for comparison, a 7970 GHz is around $420 (very comparable to US). Who would pay that much more? If you're going to be haphazard with money, you should have bought a Titan and been done with it.

 

[UniversalTruth]

7970 GHz is around $420 (very comparable to US). Who would pay that much more?

Everyone who wants a GPU from a higher performance class. The 7970 is a mainstream part and is no match for GTX 780.

If you're going to be haphazard with money, you should have bought a Titan and been done with it.

I feel sorry for everyone who bought the titan.

 

[Blazkowicz]

That's like saying a BMW series 7 is mainstream because you think a Mercedes S is better.

Mainstream part would be something like the 7770 or at best GTX 660 and 7850.

 

[RedVi]

Everyone who wants a GPU from a higher performance class. The 7970 is a mainstream part and is no match for GTX 780.

Uh, right.

It looks bad enough there, add another $100-200 to that and no one should buy it. Why some cards sell for the same price as they do in the US and others go for $100-200 more is beyond me, I think when the supply is limited the retailers just feel like they can get away with gouging. In the US everyone has a proper MSRP to compare to and throw up their arms if it's well above that. Here, most cards now sell for close to US MSRP, but they like to gouge on most new products because they can, for old times sakes. Anyone buying at that price just doesn't care about performance/dollar and should have gone for a Titan.

 

[Blazkowicz]

Even that MRSP is just high. If you really want a 780 but $650 is too much for you then wait till it's $500 (why not, after 770 has dropped from its introductory price)

 

[lanek]

Even that MRSP is just high. If you really want a 780 but $650 is too much for you then wait till it's $500 (why not, after 770 has dropped from its introductory price)

Well, if it happend with some luck, we will finally get them here for less of 700$.. ( average price of the 780 here is 730 to 749$ ( Titan 1050$ ).

But no worry, there's some 680 OC who still sold here for 630$. ( Damn, thoses shops are just crazy for some prices, hopefully for CPU and MB / RAM, prices are in part of USA ). ( the problem is if i order them in Deutshland, France or UK.. i will be charge by the customs. ( ~ +15% of the price + the UPS charge )

 

[UniversalTruth]

That's like saying a BMW series 7 is mainstream because you think a Mercedes S is better.

Mainstream part would be something like the 7770 or at best GTX 660 and 7850.

See what. GTX 680 and 7970 GE share almost the same absolute performance. If the former is the second-tier GPU from NVIDIA. what exactly opinion do you expect from me to have about something with equal capabilities. It is fact that 7970 significantly underperforms, something doesn't go right in AMD as a company and I am negatively surprised they will be able to reach the GTX 780 performance only when using the next process node shrink. May you imagine what will do to it the Maxwell Titan? I think it will humiliate it...

So, no. 7970 (or GTX 680 for that matter) is more like an ordinary BMW 5 series, while 780 (or Titan with almost same performance) would resemble BMW 8 series.

@RedVi:

I don't care so much about the performance per dollar in that segment. It is quite obvious that the ratio worsens with more expensive products.
Of course, titan is more like an exception because the difference between 1 k$, and 650 $ is huuuge
Care more about Absolute performance where the 7970 disappoints from many viewpoints

 

[lanek]

See what. GTX 680 and 7970 GE share almost the same absolute performance. If the former is the second-tier GPU from NVIDIA. what exactly opinion do you expect from me to have about something with equal capabilities. It is fact that 7970 significantly underperforms, something doesn't go right in AMD as a company and I am negatively surprised they will be able to reach the GTX 780 performance only when using the next process node shrink. May you imagine what will do to it the Maxwell Titan? I think it will humiliate it...

So, no. 7970 (or GTX 680 for that matter) is more like an ordinary BMW 5 series, while 780 (or Titan with almost same performance) would resemble BMW 8 series.

@RedVi:

I don't care so much about the performance per dollar in that segment. It is quite obvious that the ratio worsens with more expensive products.
Of course, titan is more like an exception because the difference between 1 k$, and 650 $ is huuuge
Care more about Absolute performance where the 7970 disappoints from many viewpoints

If we follow you, the 580 is now a low budget card. the problem is the card have been released 3years ago and the 7970 / 680 1 1/2 years ago.

We can argue that Nvidia was have it ready at this time, but looking how they have improve their process, im not even sure the card will have run at this clockspeed.. maybe just at 750mhz .. this will not have leave a big margin on performance over the 7970 or even the 680. ( 10% ? ) ( the 7970Ghz is at what ? 22% under the 780 ( stock reference version ).

Its a good card, and personally i was even expect an higher MSRP ( 700$ ), maybe because i was think they will not cut the DP performance and push more ram with it. ( not important figure, its a just way to justify the price difference with the Titan . ). If i have not my actual system, i will surely take it.

 

[ShaidarHaran]

Uh, right.

It looks bad enough there, add another $100-200 to that and no one should buy it. Why some cards sell for the same price as they do in the US and others go for $100-200 more is beyond me, I think when the supply is limited the retailers just feel like they can get away with gouging. In the US everyone has a proper MSRP to compare to and throw up their arms if it's well above that. Here, most cards now sell for close to US MSRP, but they like to gouge on most new products because they can, for old times sakes. Anyone buying at that price just doesn't care about performance/dollar and should have gone for a Titan.

That's only one metric, and not nearly a complete picture of the performance of either card at that. I think Anandtech states it nicely:

The 22% performance advantage that the GTX 780 enjoys over the Radeon HD 7970GHz Edition cements NVIDIA’s performance lead, while the price difference between the cards means that the 7970GE is still a very strong contender in its current $400 market and a clear budget-saving spoiler like the 6970 before it.

No one is arguing about the value of this card compared to anything below it - it's not a value card. The only value it offers is to people who would have otherwise bought a Titan.

 

[RedVi]

Sure, a $150-200 premium I can understand, like the GTX580. But an MSRP premium of $250 is bad, the (Australian gouge-fest) actual premium of $350-400 is just ridiculous. There's no clear reasoning to buy one. My friend wants to upgrade his 580, he was thinking a 690. I said no, wait for 780 or get a Titan if money is no object. Now I have to put the 780 in the latter category also. Honestly, nothing he can buy is a cost effective, decent upgrade.

EDIT: Just to compare like someone else here already has: The 7970 launched with 30-60% more performance than the 6970 for $200 more. Get's attacked at being too little for too much. 780 launches with 20-40% more performance than the 680 for $200 more. Great card? I don't think so. For $350-400 more (gouge), no way in hell. Especially as the 7970 had the excuse of being a new architecture, which has proven to be a valid reason with performance improving quite a lot over time. Some people just have double standards it seems.

 

[Xalion]

Uh, right.

I am trying to find a way to be diplomatic about this - but this chart is just plain idiotic.

99th percentile means that 99% of the results lay below the line. So for example, Consider two systems.

System A) Renders at 1 fps for 98 seconds, 150 fps for 1 second, and 120 fps for 1 second.
System B) Renders at 60 fps for 100 seconds.

According to this chart, the System B would show up almost 60 fps LOWER than the first - because it's 99th percentile is 60, while the first systems 99th percentile is 120. So which system would you rather have?

I could see a chart like this with average frame rate. I could see a chart like this for average frame rate excluding the top and bottom 5% of results to get rid of noise. But seriously - this chart is just plain bad. It is one of the worst cases of "marketing science" I've ever seen.

 

[xDxD]

Here in UE reference cards are at 560€ too, great price by me.

 

[RedVi]

I am trying to find a way to be diplomatic about this - but this chart is just plain idiotic.

99th percentile means that 99% of the results lay below the line. So for example, Consider two systems.

System A) Renders at 1 fps for 98 seconds, 150 fps for 1 second, and 120 fps for 1 second.
System B) Renders at 60 fps for 100 seconds.

According to this chart, the System B would show up almost 60 fps LOWER than the first - because it's 99th percentile is 60, while the first systems 99th percentile is 120. So which system would you rather have?

I could see a chart like this with average frame rate. I could see a chart like this for average frame rate excluding the top and bottom 5% of results to get rid of noise. But seriously - this chart is just plain bad. It is one of the worst cases of "marketing science" I've ever seen.

Fair enough, I'm not sold on representing 99th percentile in an overall graph either. I take the results on a case by case basic looking at all measurements.

But,

Isn't much better. That's a 5-6 FPS (10-12%) delta for $200. At $550, I'd say great card. $600, okay. $650, meh, if you have the money. $750-850 (local prices), HORRIBLE.

 

[silent_guy]

I am trying to find a way to be diplomatic about this - but this chart is just plain idiotic.

99th percentile means that 99% of the results lay below the line. So for example, Consider two systems.

Your explanation would make sense if he did the 99% percentile of FPS. He doesn't: he uses frame time, which doesn't suffer from the FPS 1/x distortions when you get to ultra short frame times.

Please redo your scientific analysis on that and come back with the result. (Hint: you'll end up with exactly the opposite of your initial findings.)

There sure is a lot of crap on the web, but TechReport has been fine tuning their stuff for almost 2 years. Give them just a little bit credit, even if you don't agree with everything they do.

 

[A1xLLcqAgt0qc2RyMz0y]

Pricing for the upcoming GTX 770

With the GTX 780 at $650 and the current prices of the GTX 680 hovering around the $450 area and the HD7970GE also around the $450 area my guess is that the GTX 770 gets priced at the same $450.

 

[Xalion]

Your explanation would make sense if he did the 99% percentile of FPS. He doesn't: he uses frame time, which doesn't suffer from the FPS 1/x distortions when you get to ultra short frame times.

Please redo your scientific analysis on that and come back with the result. (Hint: you'll end up with exactly the opposite of your initial findings.)

There sure is a lot of crap on the web, but TechReport has been fine tuning their stuff for almost 2 years. Give them just a little bit credit, even if you don't agree with everything they do.

You failed to understand the basic problem. It wouldn't matter if he used frame time or any other metric. The problem still exists. It is also a very obvious problem - which is why it annoys me.

Let me state it plainly without an example. Percentiles do not include information about "typical" behavior for the included range. They only indicate outliers for the excluded range.

For frame times, you could have a scenario where one card has 98 frames at 1/10th of a second, 1 frame at 1/5th of a second, and one at 1/2 of a second while the other has 98 frames at 1/60th of a second, 1 at 1/5th of a second, and one at 1/4th of a second. They both have the same 99th percentile, even though one is clearly better than the other.

This is a problem that students in the US start running into in the 9th grade with standardized testing, so is something I expect people to be familiar with even if they are not scientists. I thing it is pretty obvious that percentiles should not be used for that type of comparison.

 

[lanek]

With the GTX 780 at $650 and the current prices of the GTX 680 hovering around the $450 area and the HD7970GE also around the $450 area my guess is that the GTX 770 gets priced at the same $450.

effectively somewhere between 450 and 500$.. it will depend by how much it is faster.. ( with 230W, they should have some margin on clockspeed ).

 

[silent_guy]

You failed to understand the basic problem. It wouldn't matter if he used frame time or any other metric. The problem still exists. It is also a very obvious problem - which is why it annoys me.

In your initial contrived example, you'd end up with a 99% percentile of 120fps. If you had used their methodology with the same contrived data, you'd end up with 1fps. Which very well reflects what a user would be experiencing. It's fair to say that your initial takedown was just a little bit flawed and a somewhat weak foundation to call their methods, and I quote, "one of the worst cases of marketing science."

Nobody is saying that 99% percentile is the one and only number to rule them all. News flash: there doesn't exist such a number. Which is why they post a crap load of graphs, including a histogram with frame times from which the 99% number is derived.

However, if you need to select one and only one number that has the best chance of representing what a user will experience, I think it's a pretty good proxy.

Your second example illustrates that it can't distinguish between 2 cases that behave differently for 98%. No single number metric is perfect. But remember that 99% percentile at 60Hz means a stutter once every 1.5 seconds. You can probably make some case that, perceptually, our brain will be more disturbed by this 1% than what you'd intuitively think. Of course, that again is based on the assumption that those 1% frames are spread uniformly across the frame time recording, which is unlikely. Etc. Etc.

Every reductionist metric can be refuted with a contrived counter-example. On your second try, you've managed to come up with such an example. Good, here's a cookie.

Now be constructive and come up with a better single number performance metric that can be used to calculate !/$.

 

[swaaye]

$650 is certainly not a new tier. It's also not unprecedented that these companies play pricing mind games. And not surprising that there's a pointless uprising about it here.