[chiphell] kepler rumors suggest 15% better than 580.. price and transistors

Page 6 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Grooveriding

Diamond Member
Dec 25, 2008
9,107
1,260
126
True, that could happen, yet does not correlate at all with an otherwise very predictable code name pattern. Just like AMD, nVidia has been cutting down their flagship GPU to get a pair (and in some cases a 3rd) of models from that flagship GPU (this has been going on for every generation dating back to the GeForce 2. The GTX580, 570 and now even the 560Ti 448core are all based off of the GF110 flagship GPU. To assume otherwise for Kepler doesn't get us anywhere, particularly when we're starting from nothing other than what we can best guess at based on their history.

Right now you're just really grasping for straws from information of what is likely a false rumor to come to that conclusion

I'm not grasping at straws here, I am pointing out a possibility of a completely speculative rumour like we have in this thread meaning something other than you think. Taking the $400 price point, and again using historical based logic as you are in your post above, correlating that price point to the second in line GPU.

It is no different than some of your speculation like below:

whos to say the GK104 won't have just as much potential when overclocked from its stock speeds? or that GK100 won't be that much beastlier?

or here:

I actually wouldn't be surprised if we see close to 100% if not better




I interpreted this information as GK104 vs. GF110 (ie GTX760 vs. GTX580)

As a rumour it is open to interpretation. You are seeing it one way based on the given nomenclature, I am seeing it another way based on the price point.


Where you might expect only 40-50% improvement, I actually wouldn't be surprised if we see close to 100% if not better, as we got nearly just that with the GTX480 vs. the GTX285
285 vs 480

This is not accurate. I have ignored the compute benchmarks as I have been discussing strictly gaming results even though I haven't said it explicitly. Just looking at your own link there are cases where the 480 is only 30% faster than a 285. and nowhere do I see 100% gains.

As a whole the 480 was across the board 39% faster than a 285. If you'd like another example prior to that the 280 was 37% faster across the board than a 8800GTX

We could start cherry picking benchmarks to show larger gains than the average for either situation. But, we should make that sort of judgment in an across the board manner, where we see the 7970 40% faster across the board than its predecessor. In line with what we've seen from AMD or nvidia in the past. This is still ignoring the fact that your example is using recent drivers vs launch drivers and the 7970 has just released and seen no driver optimizations.

This is going off on a new tangent though. But there seems to be a lot of wishful thinking that we will see the sort of performance jump with kepler we haven't seen in an across the board improvement since 2005 or 2002. I saw much of the same sentiment from posters about waiting on the 480 instead of buying a 5870 and they wound up getting bit in the ass.

Just curious, why do you think there will be this groundbreaking leap, rather than something more in line with what we usually see, based on what little we know about what nvidia has coming ? If it's just wishful thinking I can understand that, but there is nothing to make us think we will be getting an 8800GTX type leap.
 
Last edited:

VulgarDisplay

Diamond Member
Apr 3, 2009
6,193
2
76
People often remember things in the past in a better light than what actually happened. I mean people still think that the original crysis had these magical and amazing textures, but when you go back and play it you see bad textures all over.
 

JAG87

Diamond Member
Jan 3, 2006
3,921
3
76
That is more telling of the Sli scalling, GTX 480 was around 45% faster then a GTX 285... seems familiar, doesn't it? (6970,7970)

So if the 480 is only 45% faster than the 285, you're implying 285 SLI scales only 45%? You're a genius.

SLI scales 80-100% (minus a few exceptions), and the 480 is obviously not just 45% faster than the 285.

http://www.anandtech.com/bench/Product/167?vs=158

According to these results, it's 50-80% faster, like I've stated previously.

7970 being 40% faster than a 6970 is laughable.
 

Arzachel

Senior member
Apr 7, 2011
903
76
91
So if the 480 is only 45% faster than the 285, you're implying 285 SLI scales only 45%? You're a genius.

SLI scales 80-100% (minus a few exceptions), and the 480 is obviously not just 45% faster than the 285.

I never said that Sli scales only 45%, but thank you for putting words in my mouth and arguing against a strawman. While other posters have already proven, that my 45% estimate was actually optimistic, lets go with a GTX 480 being 145% and a GTX 285 being 100%. (100% + 100%)*magical Sli scaling coefficient which comes out slightly bellow 75%.
 

Crap Daddy

Senior member
May 6, 2011
610
0
0
I had a 285 and I have a GTX570 which is basically a 480. At my res. 1680x1050 I can tell you that the 570 is almost twice as fast as the stock 285, at least this is how I feel it.. Both cards were bought for around 350$, one a half years difference. In 8 months time I expect to be able to buy a card with the equivalent of around 350$ that will perform almost twice as fast as the 570. Considering all that's happening now with both companies I find it hard to believe this will happen.
 

JAG87

Diamond Member
Jan 3, 2006
3,921
3
76
I had a 285 and I have a GTX570 which is basically a 480. At my res. 1680x1050 I can tell you that the 570 is almost twice as fast as the stock 285, at least this is how I feel it.. Both cards were bought for around 350$, one a half years difference. In 8 months time I expect to be able to buy a card with the equivalent of around 350$ that will perform almost twice as fast as the 570. Considering all that's happening now with both companies I find it hard to believe this will happen.

Exactly, but trying to convey this concept to the pro-AMD people here who have to defend their honor seems very hard. Regardless of what Kepler turns out like, Tahiti is a pretty big disappointment. I have nothing left but hope (and some track history) that Kepler will bring more tangible increases that are actually worth spending $600 for, but besides that it's all FUD.



I never said that Sli scales only 45%, but thank you for putting words in my mouth and arguing against a strawman. While other posters have already proven, that my 45% estimate was actually optimistic, lets go with a GTX 480 being 145% and a GTX 285 being 100%. (100% + 100%)*magical Sli scaling coefficient which comes out slightly bellow 75%.

What? 200 * 75? What is this magic math you are trying to do?
 
Last edited:
May 13, 2009
12,333
612
126
Exactly, but trying to convey this concept to the pro-AMD people here who have to defend their honor seems very hard. Regardless of what Kepler turns out like, Tahiti is a pretty big disappointment. I have nothing left but hope (and some track history) that Kepler will bring more tangible increases that are actually worth spending $600 for, but besides that it's all FUD.

I go back and forth on this. Maybe the days of those kind of performance increases are over? Heck look at intel. I went from a i7 930 to a 2600k and there's really not much difference. Ivy Bridge is coming out and it's not even much faster than my 2600k. That's not a whole lot of difference and this is using intel fabs.
 

Vesku

Diamond Member
Aug 25, 2005
3,743
28
86
Not sure if I want to drag all this up if the GTX flagship is roughly 20% faster with launch day drivers (~40% faster than GTX 580) than the 7970 and priced $50-100 more...

It is an illogical argument if you:

1. Compare raw 7970 pricing to 6970 ("It's priced out of line of the 6970 launch MSRP")

But

2. Compare performance pricing of the Radeon 7970 3GB to the GTX 580 1.5GB. ("It's a new node, it shouldn't be more expensive than the GTX 580 that it only barely beats.")

And

3. Ignore that there hasn't been much if anyone saying the 7970 is priced for value.

Choose either 1 or 2, do not cross the streams.
 

JAG87

Diamond Member
Jan 3, 2006
3,921
3
76
I go back and forth on this. Maybe the days of those kind of performance increases are over? Heck look at intel. I went from a i7 930 to a 2600k and there's really not much difference. Ivy Bridge is coming out and it's not even much faster than my 2600k. That's not a whole lot of difference and this is using intel fabs.

Well, not everything in the industry is a product worth your attention. Why did you do it?
 

Panopticon

Member
Dec 27, 2011
125
0
71
I dont understand the confusion... Nv had the fastest single card for around $500 and no one really complained. Amd releases a card for $50 more that performs better and clearly has potential for a quick refresh as shown with the great overclocks and the how well the clock speed translates to real performance. All said and done it appears the 7970 will wind up about 30% better than a 580 and about 45-50% faster the a 6970 and I bet you the price settles to around 500 within months. I really feel like the 7970 will end up slightly behind the Nv flagship in the same way that the 6970 was behind the 580. I also feel that by the time you can actually buy a 780 the 7970 will be a better value for whoever is buying then because if its slower then the 780 amd will price the card competitively. So whats the main difference between tahiti and kepler? One is in stores and one has NO official information on the internet... This is not an Nvidia flame post I have gone back and forth between the two companies for years I have zero brand loyalty. It all boils down to when you need a new card and how much you want to spend. I just built a new rig and money isnt a huge issue at the moment so why would I buy a 580 now when tahiti is here today.

sorry for txt block and punctuation I typed said txt block on a tablet (lame I know)
 

bunnyfubbles

Lifer
Sep 3, 2001
12,248
3
0
Do you play at 2560x1440/1600 as well?

I could see that point. At 2560x resolution even SLI 580s struggle at times with stuff like Witcher 2 and ubersampling, metro 2033 with ultra dx11 settings.

I've just become comfortable running 1 notch below full detail, both of those games run 60+ fps fluid with slightly lower settings :p

I have a Dell U2711 and a BenQ XL2410T, if I'm not pushing 1440p I'm trying to push 1080p @ 120fps. I hate it when my frame rate drops much blow 90 and really try and strive to have my minimum frame rate above 60fps.




I guess there's just no getting through to you, that you can't admit you overlooked a crucial point to the entire basis of this rumor, that the GK104 is not the GK100 and thus not indicative at all of what we'll see from upper-end Kepler, or that GK104 wouldn't be any worse than what we see from the 7970 even if this rumor was 100% true

I don't know how many times I have to emphasize that.
 

Despoiler

Golden Member
Nov 10, 2007
1,966
770
136
Slides clearly indicate this is not Nvidia's flagship chip. Still complete FUD. They might aswell be giving us graphs of the increase in wood screws compared to the previous generation at this point.
 

VulgarDisplay

Diamond Member
Apr 3, 2009
6,193
2
76
Not sure if I want to drag all this up if the GTX flagship is roughly 20% faster with launch day drivers (~40% faster than GTX 580) than the 7970 and priced $50-100 more...

It is an illogical argument if you:

1. Compare raw 7970 pricing to 6970 ("It's priced out of line of the 6970 launch MSRP")

But

2. Compare performance pricing of the Radeon 7970 3GB to the GTX 580 1.5GB. ("It's a new node, it shouldn't be more expensive than the GTX 580 that it only barely beats.")

And

3. Ignore that there hasn't been much if anyone saying the 7970 is priced for value.

Choose either 1 or 2, do not cross the streams.

So by your reasoning the the slower gtx580 should still cost $500 because it's an older node, and it barely gets beat (24% stock 40% oc'ed is not barely by the way) and that the 7970 should cost less?

Does not compute.
 

Vesku

Diamond Member
Aug 25, 2005
3,743
28
86
Uh, I was just summarizing the illogical crossroads espoused by some of the people who are very upset with the 7970s pricing. Hence the quotation marks as in a generic quotation of the argument they make.

To your point, if NVIDIA lowers the GTX580 3GB price to something like $449 then as a consumer I would expect a bit of downward movement on 7970 pricing. Not necessarily at a 1:1 ratio but at least some change even if it's just a rebate offered.
 

maddie

Diamond Member
Jul 18, 2010
4,723
4,628
136
I had a 285 and I have a GTX570 which is basically a 480. At my res. 1680x1050 I can tell you that the 570 is almost twice as fast as the stock 285, at least this is how I feel it.. Both cards were bought for around 350$, one a half years difference. In 8 months time I expect to be able to buy a card with the equivalent of around 350$ that will perform almost twice as fast as the 570. Considering all that's happening now with both companies I find it hard to believe this will happen.

]Exactly, but trying to convey this concept to the pro-AMD people here who have to defend their honor seems very hard[/B]. Regardless of what Kepler turns out like, Tahiti is a pretty big disappointment. I have nothing left but hope (and some track history) that Kepler will bring more tangible increases that are actually worth spending $600 for, but besides that it's all FUD.
You can tel what percentage faster by how it feels? This is amazing.
 

railven

Diamond Member
Mar 25, 2010
6,604
561
126
You can tel what percentage faster by how it feels? This is amazing.

Just had to LOL at that. I can tell when something feels faster (or slower), but to flat out claim accurately detect the percentage on the fly, well clearly I'm lacking something.

With FRAPs on, I can see how options affect me games in certain scenarios, and I can read the 100% improvements in some games, but to be frank - I can't FEEL the difference after a certain point (ie, going from 60 FPS to 120FPS on a 60hz monitor.) Thus I ran out and bought a 120hz monitor (and to try 3D - still hate it, nVidia or AMD+IZ3D) and personally losing the IQ to get the extra frames in certain games isn't worth the trade off. Desktop 2D feels more responsive, oddly enough - or that's just me trying to justify a $560 monitor :eek:
 

Crap Daddy

Senior member
May 6, 2011
610
0
0
You can tel what percentage faster by how it feels? This is amazing.

Isn't it? Back to your doubts I can recall only one benchmark which is Crysis 1 where the 285 scored around 30FPS and the 570 around 50FPS. If you still have doubts check benchmarks from reputable sites, if you don't find there the 285 then figure out how is the 460 compared to the 570 because the 460 is around a 285.
 

SirPauly

Diamond Member
Apr 28, 2009
5,187
1
0
So by your reasoning the the slower gtx580 should still cost $500 because it's an older node, and it barely gets beat (24% stock 40% oc'ed is not barely by the way) and that the 7970 should cost less?

Does not compute.

Imho,

Usually new generations reduce the pricing of older generations and yet this isn't the case because some are using a year old 40nm premium priced sku as some sort of objective pricing barometer.

Considering AMD is still on the small die strategy -- the answer is AMD desires to be more of an aggressor and predator when it comes to revenue and margins based on their execution and engineering prowess.