Easy: fist they only had the hd3870 x2 (as the fastest hd3000 series gpu) in this slide. then they decided to add also the hd3870 and they insert a new row which automatically had the same color as the row above.Originally Posted by annihilat0r
@conzymaher: in the first slide the biggest difference is one single number, in the new one its 2, so its clearly that theres a difference there. and u made the first line wrong.
AMD Phenom II X6 1055T@3.5GHz@Scythe Mugen 2 <-> ASRock 970 Extreme4 <-> 8GB DDR3-1333 <-> Sapphire HD7870@1100/1300 <-> Samsung F3 <-> Win8.1 x64 <-> Acer Slim Line S243HL <-> BQT E9-CM 480W
Well as far as the slant goes to be fair, that's due to them going from a 4 digit number to a 2 digit number at the beginning of the gigaflops, vs. the old one where it only went 3 digits to 2 digits.
Not possible at all or NV would end up with selling GTX280 below production price, when AMD would still earn money.I'm obviously speaking from a purely technological standpoint... anyone can re-price things and it happens day to day. So prices drop, and soon you can get two of them for more than the r700 but also with much higher performance..
With R700 beeing 2x 55nm RV770 (which are much less complex than GTX280 so it's much easier to make them) and with GTX280 beeing 65nm giant.
So if these Vantage noumbers are true, AMD is in better situation than NV.
Do you mean they originally had a chart that only had the 3870 X2 in it but not the 3870? Why would they have had the 3870 X2 "saved" then re-opened and edited again to add the 3870 when they didn't even have the X2 until 2 months after the 3870?
Why wouldn't they have cleaned it up before it went public?
Are we there yet?
Look at that real slide of 2900XT and see the last GigaFLOPS column
Which says "Giga FLOPS" whereas all the others have "GigaFLOPS"
So surely AMD, who can't make a mistake regarding white/grey alternating lines or aligning words, couldn't have made a mistake that big? Then that is %100 fake as well??
Lmao as if you know how much it costs to make the boards for both companies? You really think 2 R770's and all that comes with it is cheaper than 1 bigger chip?
I love how your generalizations of "less complex" automatically implies "ridiculously cheap production cost" somehow.
A proper AMD slide would be made by a "designer" or someone who would leave the gigaflops consistent down the right hand side and let the numbers go out of line... It looks ridiculous
I didn't say they can't make a mistake, but missing an extra space is a bit different than a different colored row which stands out in my opinion. I think we'll wait and see who's right here. Will be interesting either way.
Last edited by Sr7; 06-07-2008 at 04:10 AM.
Im just guessing. They made this slide after r680 launch, lets say 2 weeks ago. the insert a quadcore cpu and the fastest ati gpu for comparison, and of cause the rv770 pro & xt numbers. Then they saved it. After that they decided to also add hd3870 ond so they just added a new row which now has the same color as the row above.
and it didnt went public just because it was leaked. maybe they would have changed that before an official presentation.
in a nutshell: i dont know if this slide is fake or not, but u cant argue here w/ things like color of symetrics, cause they are all ok on this one.
AMD Phenom II X6 1055T@3.5GHz@Scythe Mugen 2 <-> ASRock 970 Extreme4 <-> 8GB DDR3-1333 <-> Sapphire HD7870@1100/1300 <-> Samsung F3 <-> Win8.1 x64 <-> Acer Slim Line S243HL <-> BQT E9-CM 480W
What about the "Giga FLOPS" vs "GigaFLOPS" mistake made in the 2900XT slide?
AMD's "designer" can't think of that?
n1
AMD Phenom II X6 1055T@3.5GHz@Scythe Mugen 2 <-> ASRock 970 Extreme4 <-> 8GB DDR3-1333 <-> Sapphire HD7870@1100/1300 <-> Samsung F3 <-> Win8.1 x64 <-> Acer Slim Line S243HL <-> BQT E9-CM 480W
I'm saying they didn't have 4 digit numbers in the real/older slide, so it makes sense that the slant is more noticeable when they're going from a 4 digit gigaflop numbers down to a 2 digit gigaflop. If you have:
569 blah
34 blah
24 blah
you get less slant than:
2322 blah
569 blah
34 blah
24 blah
The new/extra 4th digit pushes the text over to the right more when centered.
Remember the CH roadmap: http://bbs.chiphell.com/attachments/...1g3GQzTQkx.jpg
According to them rv770 is 55nm, 256bit & 800sp.
AMD Phenom II X6 1055T@3.5GHz@Scythe Mugen 2 <-> ASRock 970 Extreme4 <-> 8GB DDR3-1333 <-> Sapphire HD7870@1100/1300 <-> Samsung F3 <-> Win8.1 x64 <-> Acer Slim Line S243HL <-> BQT E9-CM 480W
Well thats why first Intel dual cores were two separate cores.Lmao as if you know how much it costs to make the boards for both companies? You really think 2 R770's and all that comes with it is cheaper than 1 bigger chip?
I love how your generalizations of "less complex" automatically implies "ridiculously cheap production cost" somehow.
Thats why AMD needed Phenom X3.
GTX280 chip is huge and NV will have problems to get them, thats how it works.
AMD already showed with 38xx agresive pricing what they can do.
So that's why i generelize that less complex and made in 55nm is easier and more flexible in pricing....
So tell me.. if they now use separate shader clocks (which they do), why would they only list core clocks when showing gigaflops?
*sigh* I guess some people just want to believe (you included)
@Sr7: U know so much, its hitting me really hard here. So, can u tell me how do u know that rv770 has a shaderclk? And no, a higher clockdelta isnt a shaderclk.
So u basically say that rv770 has 480sp and a higher shaderclk, like in here:
If so, i can assure u that this is a fake.
AMD Phenom II X6 1055T@3.5GHz@Scythe Mugen 2 <-> ASRock 970 Extreme4 <-> 8GB DDR3-1333 <-> Sapphire HD7870@1100/1300 <-> Samsung F3 <-> Win8.1 x64 <-> Acer Slim Line S243HL <-> BQT E9-CM 480W
They don't.So tell me.. if they now use separate shader clocks (which they do), why would they only list core clocks when showing gigaflops?
You're missing very obvious things, I'm afraid
The shader units don't have a seperate domain, it's all 750 MHz (or 600)
FLOPS calculation: 750 x 2 x 800 = 1.200.000 for 4870; 625 x 2 x 800 = 1.000.000 for the 4850.
LMFAO - you're saying this is fake based on a PR powerpoint and how its edited?
For all you guys know, the highlights were intentional: they grayed the CPU in BOTH cases to show that the second best card in the lineup (in this case the 4850 this time around, the 2600XT last time) perform many many GFlops better than a CPU at 3GHz to show the power of the GPU in process power
Oh and congrats, they centered the Processing Rate hence the slant from oh I don't know, 1xxx GigaFlops to 96 GigaFlops..
Nvidia is doing the same thing with their Tesla & CUDA and ATI/AMD are as well
And unlike all the other rumors, these clocks were confirmed by more than a few members within this own thread who had been harping on that the 850/1000 rumored clocks for the 4870 were fake long before any leaks on these #'s were released. Add in the fact that the math is actually coherent and correct and its harder to argue against it
Sorry, but you are the one missing things. My point is, the mere existence of a shader clock in a separate domain would throw a wrench in the works of the calculation you just made, the number of SP's and also the slide itself being valid.
I'm telling you there is a separate clock domain
Figure the rest out for yourself.
I have a hard time believing they'd randomly always gray the second best card... where is the logic there?
Sorry but I don't think so.
And the cost of a GPU has to go by yield numbers, which we don't know what % of dice on a wafer are used by AMD for the RV770 series
Same goes for Nvidia wafers
But, absolutely DO NOT underestimate how much a few % in yields can change how profitable a chip is. Trust me, I actually studied microfabrication of wafers and a few % can be the difference between 1 million in losses and 5 million in profits for cheap chips. Now if you consider that AMD has 256mm^2 chips versus Nvidia 576mm^2 chips, and if the chips were perfectly square (as an example, cause they aren't in practice necessarily), AMD would have a 16 x 16 mm chip versus Nvidia's 24mm x 24mm chip... meaning in a square area, AMD can make 9 chips vs. Nvidia's 4. If given the same yield %, then AMD can make 9 to every 4 of Nvidia. Obviously, once you factor in yields that ratio changes, not to mention that in reality they aren't perfectly squares and not every bit of area of a wafer is necessarily used up (such as the edges). But that's an easy example of how a smaller chip can be profitable (and why CPU's are constantly pushed to a new process)
Actually the start of the 800 shaders rumour apparently is rooted much earlier than I believed (I believed it surfaced around May and was just based on guesses).
Look at that link provided by a member of B3D:
http://forum.beyond3d.com/showpost.p...&postcount=756
Here it says 160x5 with 32 TMUs, and clocks 100-125MHz higher than what we have here now, and no mention of a separate shader domain clock.
Although if it's indeed 800 that doesn't make sense to me. That would mean a 2.5x increase in math processing power for a less than 2x increase in texture fill rate. We all know t-fillrate was a big bottleneck for any R6xx gpu, so in this case it makes no sense at all.
Bookmarks