It's funny but it almost seems like some sites were 'persuaded' to only compare against only lower priced reference models. I'm glad a few credible sites got it right and included competitor models that compete based on price and non-reference custom design.
Apparently, one of the benefits of HBM is very low idle power: http://techreport.com/review/28612/asus-strix-radeon-r9-fury-graphics-card-reviewed/11
I guess the water pump in the Fury X prevented us from noticing that.
The problem with many reviews, IMHO, is that they put those customs R9 Fury up against reference blower designs for other GPUs. It's kind of a nice gesture towards AMD hehe but it doesn't paint the whole picture. There can be a huge gap between reference designs and custom ones with the GTX 900s. Of course it's really annoying to have to look at 2 designs per GPU and time is limited. But I don't see how you can provide readers with any insightful analysis of the Sapphire/Asus R9 Fury performances by just looking at the reference GTX 980 benchmark numbers.
Frustrating for AMD to know their Fiji chips are at max clock speed at shipping....?
Doubting unlock voltages would help even the overclock for them.
From a market positioning point of view, it makes some sense to compare the Fury against the 980. From a technical point of view, it's kinda ridiculous to compare a 600mm2 chip against one that's only 400mm2.
yes, you don't know usage of memory during freeze but you can see much of memory is free after freeze, I also posted video comparing 980oc vs fury vs 980ti and only 980ti did't get any freezeWrong thread, but you think this is the 4GB who is a problem ? i see only 3GB used when there's this stutter. And we cant say that Bullet use a lot of vram lol.
alien iso= 20.8%
unity = 22.7%
batman = 20.4%
bf3 = 24.9%
bf4 = 11%
bioshock = 29.8%
cod aw = -3.5%
civ = 30.3%
crysis 3 = 23%
dead rising 32.4%
da:I = 4.6%
far cry 4 = 29.9%
gta v = 16.5%
metro last light = 12.4%
project cars = -15%
ryse = 18.9%
SoM = 25.6%
Witcher 3 = 16.8%
Tomb Raider = 23.8%
Watch Dogs = 11.6%
Wolfenstein = -10.3%
WoW = -2.6%
Let's ignore for a moment your incredible but faulty reading between the lines ability, because I never suggested such a thing, because, yes, I actually absolutely do think that the base BOM difference (GPU silicon + memory) is likely to be larger than $50.You think the manufacturing difference between a Fiji chip and a GM204 is over $50, considering that Fiji Pro is a lower binned part whereas the 980 uses the higher-binned ones?
Even if the difference weren't $50, it'd still be a faulty way of looking at, because jt ignores the markups along the chain.Because that's the price difference between the cards.
Let's ignore for a moment your incredible but faulty reading between the lines ability, because I never suggested such a thing, because, yes, I actually absolutely do think that the base BOM difference (GPU silicon + memory) is likely to be larger than $50.
Even if the difference weren't $50, it'd still be a faulty way of looking at, because jt ignores the markups along the chain.
But let me reiterate my main point: it's stunning that a 400mm2 chip with half the BW manages to perform close to a 600mm2.
Is there any kind of on-chip resource where Fiji (even in cripples form) doesn't blow away gm204? How is it possible that this chip was able to get past the performance modeling stage?
It's not just the Fury review. TR also used OCed cards in the Fury X, 980 Ti, and TITAN X reviews. None of them are marked as OC in the conclusion charts.TechReport is hands down the worst review for Fury. HardOCP at least test without nvidia settings to give a picture, TR are starting off with Project Cars.
If Fury stutters more than 980 that is a legitimate point, however their average numbers seem rather different from other reviews as well.
Fury at par or only a fps faster in games where it demolishing 980 in other reviews. So I go looking at the test notes, they are using OCed models of nvidia cards which behooves them to label them as such in the graphs where it looks as if the vanilla versions are being used.
And many games are showing 20% or more advantage for Fury at 4k, so even if it only were to improve 7%, 980 would have trouble matching it in theory much less in practice. Even TR's review numbers look too close for the advantage that Fury has over 980.
Asus Radeon R9 290X
According to TPU's numbers, the Gigabyte 980 performs 9%-11% better than the regular 980 at 1440p and 2160p, which are the resolutions used in the TR review. (Fury, Gigabyte 980)Gigabyte GeForce GTX 980
ASUS Radeon R9 Fury STRIX review (stock clocks)
1080p 1440p 2160p
GTX 980 100% 93% 87%
R9 Fury 100% 100% 100%
R9 Fury (w/o Proj. Cars) 101% 102% 101%
Gigabyte GTX 980 G1 Gaming review
1080p 1440p 2160p
GTX 980 93% 92% 90%
Gigabyte GTX 980 100% 100% 100%
1080p 1440p 2160p
Gigabyte GTX 980 100% 100% 100%
R9 Fury 93% 99% 103%
R9 Fury (w/o Proj. Cars) 94% 101% 104%
Everyone reads the test setup section to find out what cards and drivers are used. Next time you see a 980 performing on par or better than the Fury just realize the review site's GPU comparison is based on price. Fury's price is just too high for what you get.Fury at par or only a fps faster in games where it demolishing 980 in other reviews. So I go looking at the test notes, they are using OCed models of nvidia cards which behooves them to label them as such in the graphs where it looks as if the vanilla versions are being used.
Really! I would never have imagined from the sound of your earlier posts praising how well they performed in some reviews;-). But lets hope you are right ....And apparently the 15.7 driver doesn't give a boost vs. the older drivers, so AMD have allowed us more hope.
Let's ignore for a moment your incredible but faulty reading between the lines ability, because I never suggested such a thing, because, yes, I actually absolutely do think that the base BOM difference (GPU silicon + memory) is likely to be larger than $50.
Even if the difference weren't $50, it'd still be a faulty way of looking at, because jt ignores the markups along the chain.
But let me reiterate my main point: it's stunning that a 400mm2 chip with half the BW manages to perform close to a 600mm2.
Is there any kind of on-chip resource where Fiji (even in cripples form) doesn't blow away gm204? How is it possible that this chip was able to get past the performance modeling stage?