If Barts is 230mm2 and cayman is 50% bigger, wouldn't that mean (maybe aside from clocks) would have a 50% increase in specs?.
Last edited:
It is possible, but going through his post history some of his narrative comes apart, even if some parts sound somewhat close to what we know. For example, his claims:
A. Barts = ~240mm2
B. Barts = 2/3 Cayman
C. Cayman = 15% larger than Cypress
D. Cayman XT = 7gbps
E. Cayman between 333mm2 and 400mm2 (just now)
So, we know that Cayman XT is 6gbps. Possible that it changed last minute, but that definitely weakens his credibilty. Now, lets look at sizes and do some math!
To begin with:
A. Barts is actually 230mm2. It has been measured. This isn't really close or far away enough to confirm or shoot down his original claim of ~240mm2, but we can use that in our calculations.
B. Cypress has been documented as 334mm2
C. Going by Silverforce's claim that Barts is 2/3 Cayman, we can reverse that and call Cayman = 1.5 Barts.
In this scenario both Barts and Cayman share a 256-bit memory controller, and like all GPUs in a family share parts like UVD (although Cayman may take up a few extra transistors for DP as was the case for Cypress vs. Juniper). Seeing that Cypress is just a hair over twice the size of Juniper (and it also doubled memory controller width -- but it had the same tesselation power), we can reasonably assume that Cayman will be *at most* 1.5 times the size of Barts, and quite possibly smaller.
So, to calculate:
230mm2 * 1.5 = 345mm2 (at the *very* most)
and then, according to his earlier claim of cypress + 15%
334mm2 * 1.15 = 384mm2
So, going by Barts = 2/3 Cayman, his previous claim that Cayman is 15% larger than Cypress is out to lunch. At the most this theoretical Cayman would be 3.3% larger than Cypress (and probably less than that thanks to being 256-bit which should save more area than scalable tesselation will take up). That also opens up another can of worms: If Cayman is 1.5x Barts with a 256-bit bus, then the decision not to make Cayman also pin compatible with Cypress when it has basically the same die size eludes me.
There are too many holes in this guy's story. When combined with the memory speed difference he has lost all credibility.
I didn't even know WoW was moving onto DX11 APIs. That explains the new water effect - is that tessellation based? I'm not sure if you'd know, but since you mentioned it figured I'd ask.
I think I'd be one of those people who would side grade from a 5870 to a 6870 if I could do it under say $50 just for the improved tessellation performance.
Tessellation is an obvious handicap on the 5k series moving into the new generation of games.
I'm not arguing anything? I'm asking what the 1+ year old card that Barts is barely beating. Just because you feel the 460 should have been released a year ago isn't relevant. nVidia released it 3mos. ago.
I don't doubt that. Marginally better is still better than slower and hotter. Follow this logic: NV will release marginally better cards in 6 months, and then ATI will release marginally faster cards in 6 months thereafter, etc. - next thing you know it's 5 years and performance has increased only 100% :awe:
hhee I guess AMD faithful have different expectations. They did purchase a $165 HD5770 over the faster $180 HD4890 and a similarly performing and cheaper $125 HD4870 during the last generational launch. I suppose releasing a 10% faster card than the 5850 at $249 13 months after HD5850's $269 launch is acceptable to the same people. I mean we have already seen NV come out with marginally (i.e., 10%) faster cards 6 months late and that did the market real good ....... So I guess getting another launch of cards 6 months after Fermi with marginally faster performance isn't so bad..... :hmm:
Also, your -5% performance expectations seem unrealistic. Based on 3dMark scores at least, HD6870 will barely beat a 5850.
Do you really expect an 800 SP card @ 775 mhz GPU, 48 TMUs, 256-bit to be within 5% of the 1440 SP card @ 725mhz GPU, 72 TMUs, 256-bit???? That's not going to happen unless each SP is almost 44% more efficient (assuming the 5th shader in HD5850 is never used ==> 288x4 vs. 200x4) and your TMUs are 50% more efficient.
That's true. But most importantly HD5770 lacks double-precision, which makes it useless for distributed computing projects!!! :twisted: BTW, on average HD4890 is a good 20% faster than HD5770, so the SC2 example is one of the few games where HD5770 is better. My point was that we have seen AMD successfully sell a lower performing card based on features. So they can just as easily sell HD6870 at $249 based on UVD3, 6 monitor support etc, without it being any faster than HD5850.
I have no doubts that the majority purchased HD5770 > 4870/4890 because 5770 is DX11 with a higher number in the name
Blastingcap, I am not sure what you mean by saying you'd take a DX11 card over a DX9 card because of DX11 alone. In games like Dirt 2, BF:BC2, you almost can't tell the difference between DX11 and DX9 code paths but there is a massive performance drop. However, in games where you CAN tell the difference between DX11 and DX9 (like Metro 2033), the HD5770 chokes so hard, it might as well be running in DX9 hehe.
Also, in the Dirt 2 example I linked, HD5770 was producing 33 fps at 1920x1200 vs. 52.2 fps of HD4890. Sorry, but almost no one serious about playing a racing game would accept 33 fps with 25 fps minimums as smooth.
And if DX11 performance is so important, then you'd surely not recommend a single HD5000 series card considering how badly Fermi whoops them in DX11 games
I am lagging in Tales of Monkey Island for crying out loud [snip]
When combined with the memory speed difference he has lost all credibility.
I didn't even know WoW was moving onto DX11 APIs. That explains the new water effect - is that tessellation based? I'm not sure if you'd know, but since you mentioned it figured I'd ask.
I think I'd be one of those people who would side grade from a 5870 to a 6870 if I could do it under say $50 just for the improved tessellation performance.
Tessellation is an obvious handicap on the 5k series moving into the new generation of games.
If you bought a Radeon 5870, you're better off skipping the 6xxx series.
If you're already got a current generation card, its never a cost effective update to buy the next card, refresh or not. If you bought a Radeon 5870, you're better off skipping the 6xxx series. If you're like me, and have a 4870 or older, you'll be looking at these cards. Its now on a mature manufacturing process, drives have matured, and the cards should be relatively smooth.
And therein lies the rub. AMD would certainly prefer you NOT to skip a generation. They're competing not only with NV, but with existing products. I didn't upgrade to a 260 or 4870, so I found the 5770 likewise not "worth it." If the 6 series offers the same performance as the 5 series I can skip another generation. That's a potential $500+ lost from one customer alone due to perceived lackluster performance increases.
We may well see g92 cards like my prehistoric overclocked 8800GT soldiering on into 2012!
We may well see g92 cards like my prehistoric overclocked 8800GT soldiering on into 2012!
If Barts is 230mm2 and cayman is 50% bigger, wouldn't that mean (maybe aside from clocks) would have a 50% increase in specs?.
Is it just me or is XFX rubbing it's 6xxx cards in nVidia's face?
Anyone know what PCI-E 2.1 has over 2.0?
I think its just you, lol
They wanted Fermi, but Nvidia said NO (humor)
http://www.vizworld.com/2010/03/xfx-geforce-gtx-480-470-images/
I'm running a G92 8800 GTS still, and quite honestly, it runs everything I play very nicely with details on high. I don't have AA/AF maxed and I do occasionally see stuttering, presumably from lack of texture memory, but I could probably skip the 6000 series if I wanted to and hold out and see what nVidia's next major refresh looks like.
However, I am a geek and I am feeling an upgrade urge, so we'll see if I go for that or some new home theater speakers.
Exactly, it's payback for just that.
I doubt it's payback. There may be some people at XFX that aren't happy wth how their relatonship with Nvida ended, but selling 6xxx parts is just XFX trying to make money. I certainly wouldn't expect them to not sell the cards. I think XFX is enjoying the success of AMD GPU's right now.