I am optimistic based on history not on die size.Huh? they designed their gpu that way.
Except it pertains to your wildly optimistic outlook.. you believe NV is going to get such a large % boost in die size again on the same node?
Lets see how much bigger they can inflate from GK110 with its >500mm2 die shall we..
I am optimistic based on history not on die size.Huh? they designed their gpu that way.
But you are basing your optimism on a PERFORMANCE delta from gk104 to gk110.. when you clearly failed to see that the performance gain comes at a cost of a massive die, something that isn't going to occur with Maxwell since its already at the limits of manufacturability for TSMC.
So, if you tell me Maxwell brings an awesome new architecture full of optimizations that will enhance performance, IPC (as for CPUs) or performance per mm2, perf per w, perf per transistors.. then sure, your optimism is on solid grounds.
Thought you were on about "the same node".
Do you have faith TSMC can make a 700 or 800 mm2 die on 28nm with any good success rate? Because its what your comparison using gk104 vs gk110 ultimately means, for us to see the same performance jump again over gk110 with Maxwell.
Yeah Maxwell should be very efficient because we have never seen such a large jump in performance in the same node before, 680->780Ti is indeed very impresssive.
We've never seen such a large jump in performance in the SAME NODE before, with this comparison: gk104 -> gk110... ergo, Maxwell should be very efficient.
That's the gist of your statement and the basis of your optimism.
All I am saying is gk104 to gk110 performance delta is largely due to a MASSIVE increase in die size on the same node.
LOL, naive? Maxwell is expected to be here much sooner than whatever island GPU you refer to. And yes, I expect Maxwell to be a significant improvement in power efficiency vs. the competition.
My guess is that maxwell will target at launch a mid range but high compute product. Nvidia's professional line is hurting from kepler's lack of DP. Doing so would allow them to sell a much smaller gpu at a higher price for the professional market, most of which have been waiting on 580's and will upgrade if the product is attractive enough.
Nvidia could achieve a massive 40+% jump in the mid/low end mobile market simply by not selling gimped 384 core 900 mhz+ chips on 64 bit DDR3. They own the high end market (due to enduro issues on the 7970m a lot of OEM's have dropped high end mobile AMD mobile GPUs). That said margins are higher on mobile cards but Nvidia already has so much of the market.
I don't have a R290 to compare your results with, I do however have a 780 and can see some obvious conflicts. I have no reason to believe you deliberately botched the 780 results, nor do I feel your 4.7GHz Haswell was found wanting.. What I am saying is if it's not those, then what? Vram seems to make sense? Especially as you go to highlight 1600, which is more sensitive to bandwidth than 1080 is.
That's two reviews, one of which is from a guy not known for his love of AMD. Keep going. :thumbsup:I'm fine with hinging my idea of performance on professional reviews, so long as they make sense. I don't see why you're so hung up on old reviews with sketchy results, we have newer ones of overclocked cards to consider as well.
My guess is that maxwell will target at launch a mid range but high compute product. Nvidia's professional line is hurting from kepler's lack of DP. Doing so would allow them to sell a much smaller gpu at a higher price for the professional market, most of which have been waiting on 580's and will upgrade if the product is attractive enough.
Nvidia could achieve a massive 40+% jump in the mid/low end mobile market simply by not selling gimped 384 core 900 mhz+ chips on 64 bit DDR3. They own the high end market (due to enduro issues on the 7970m a lot of OEM's have dropped high end mobile AMD mobile GPUs). That said margins are higher on mobile cards but Nvidia already has so much of the market.
I don't think anyone's trying to convince you to buy a reference R9 290/x anymore. The custom cooled cards are mostly quieter than a 780Ti and very well cooled.
current aftermarket 290x prices are on par with 780ti (and aftermarket 780ti still run more cool and quiet and are faster)
Nope, only at gougegg and a couple of other opportunist etailers.
There are a steady stream of cards at or very close to retail (microcenter, amazon itself, not third party).
There's a couple reference 290x's $590 (third party) at amazon right now, amazon itself doesn't have any atm.
R9 290 gaming $470 (not x, but close to retail, or is it at retail)
The custom 290/x's go quick.
current aftermarket 290x prices are on par with 780ti (and aftermarket 780ti still run more cool and quiet and are faster)
But what actually turned me off from the 290/x was how loud the thing can get. I'm not a quiet freak but having a card getting that hot and producing alot of noise for no reason is annoying.
That's two reviews, one of which is from a guy not known for his love of AMD. Keep going. :thumbsup:
He just doesn't care for their social media marketing, for all the talks of bias he uses a lot of easily reproduced game tests.. I haven't seen one person refute his perf/clocks results in any single review, given how cynical people are I'm sure more than a few have tried.
Where is the preponderance of evidence? 20+ reviews in one hand and 4 in the other.I can't make more recent OC reviews, [H] is your third, Linus your forth, how many more are we going to need showing a 780 beating a 290x?
They will have to do something big on 20nm end of the year for their pro line though. Intel is dropping Knight's Landing on 14nm in 2015 with 3+ Tflops DP and 14-16 Gflops/w. Performance wise it's going to kick the ass right off of anything nvidia can deliver in the pro market.
I don't know, he was almost foaming at the mouth with some of his tirades. He doesn't strike me as the most unbiased person. Kind of like quoting a review from AMDZone.
Where is the preponderance of evidence? 20+ reviews in one hand and 4 in the other.
Where was the 2nd chart from that you posted above anyway? I didn't recognize it.
We're really rehashing this again? If you can find evidence as conclusive as 20+ reviews showing an oced 780 is faster than an oced 290X, maybe your statements will have more credibility. As is the average oced 780 is 8-9% faster than a Titan and the average oced 290X is 15-23% faster, so...
What I'd really like to see is the 780 vs. 780 ti vs. 290x vs. 290 face off with the average oc statistics.
1187/2151MHz
http://hwbot.org/hardware/videocard/geforce_gtx_780_ti/
1171/2052MHz
http://hwbot.org/hardware/videocard/geforce_gtx_780/
1153/2091MHz
http://hwbot.org/hardware/videocard/radeon_r9_290x/
1150/1759MHz
http://hwbot.org/hardware/videocard/radeon_r9_290/
People here like to claim 1300+ is normal for the 780/ti, yet pretend like the 290/x don't overclock over the average too. Regardless, since everything is down to a silicon lottery I'd like to see these exact clocks to see what the average user may be able to get.
While I'm very interested in max OC's, I only take them in the interest of data, not to support the brand allegiance.
Again, this review was fairly close to the avg OC and I'd like to see the 780 and 290 tossed into the mix. [h], do your thing!
We're really re-hashing this again, I just want to point out that the term average is a discrete value so you have some funky stuff going on here with the "average" being 15-23%, just to remind you, average is not a range - it is a discrete absolute value.
I know you love AMD, and that's cool, but you're continuing to put them in the best light possible. I don't understand WHY you're doing this, but at the same time, there is nothing wrong with that. Hey, we all have our preferences. I like nvidia is far better because of intangible factors and features. You like AMD. Whatever man, that's how some roll, nothing wrong with it. But I don't agree with your assessment at all. I mean, i'm looking at overclocked 290X reviews at Guru3d, hardwarecanucks, pcper, techpowerup, and hardocp and they are most CERTAINLY not averaging 23% faster for overclocked 290X cards. In fact, like I said, the definition of the word "average" is not a range. It is a discrete absolute value. So that brings up problem number one with your claim.
Furthermore, there are numerous reviews at the websites that were aforementioned and their scaling ranged from 6-10% above stock. 6-10% faster than Titan and 6-10% above stock. 6-10 % is not 23%. Now we have a pickle here, huh? The stock 290X is even with Titan or within 1%. 6-10% above stock is certainly not 23%. An average is a discrete value, I should remind you.
How about we do this. Which websites do you look at? I look at techpowerup, hardwarecanucks, hardocp, pcper, and guru3d. Like I said. Aftermarket 780s overclocked basically blow away the overclocked 290X at the above websites. I don't know what kind of funky stuff your'e doing to think that a "range" of values is an average, that violates the definition of the word average itself, or what websites you're looking at.
I'll tell you what we CAN do. With all due respect, I think your assessment is wrong. I do not think Hawaii overclocks that well. But, if you want to debate this further, you suggest which of the above review sites *I* mentioned that you like. Look at my list. And you can suggest your own. Maybe we'll do a sample of 10 or something. Because I think your assessment, no offense, is wack. Then we can look at overclock values and do some absolute math without an "average" suddenly being a range of values. Like I said, a range violates the definition of "average".
So you suggest the review websites that you like, not including kitguru (AMD sponsored and only tests AMD GE games). I like the above websites because they test a mix of both games sponsored by both nvidia and AMD. I already named mine. And we'll look into this further. My list again is HardOCP, PCPer, hardwarecanucks, guru3d, techpowerup. All of these websites test a wide variety of games that aren't necessarily all AMD aligned. Unlike kitguru. Kitguru is AMD sponsored. Oddly enough, so is anandtech. And AT hasn't extensively tested GTX 780 overclocks that I've seen. But I will look again. Anyway, you name which review sites out of above you like. And suggest your own. Then we'll agree to a testing methodology.
And clearly you love NV.
Why "pick" 5 sites out of 20+?
I love numbers, well the numbers above in my earlier post are what I would consider the biggest dataset we have on OCing, including LN2, Water, air, and custom (several thousand on the 780). According to those numbers, not the golden OC numbers some prefer to throw around, let's get some benchmarks. Sure golden clockers are awesome to see, but you can't make blanket statements by them.
We can easily speculate on the 780 based on it's performance hit compared to the 780 ti. The clocks in the [h] review are close to avg, but the [h] review only had a few games.
GM110 (or Big Maxwell or whatever it is called) is expected to be in the same 14-16 DP GFLOPS/w throughput range.