AMD: R9xx Speculation
- Thread starter Lukfi
- Start date
Bouncing Zabaglione Bros.
Legend
Anyways, so on with the speculating:
I'm guessing announcement date is approximately 7 days after my 4870 kicks the bucket
No, it'll be seven days after you've replaced it with a 5850.
Well it seems to depend on the day really.
Some days seem to be ok, other days I get visual RAM artifacts all over the place.
Anyways, so on with the speculating:
I'm guessing announcement date is approximately 7 days after my 4870 kicks the bucket
Have you tried using CCC to downclock your memory? It makes more sense to downclock the card 5%-10% in real performance terms, even if you have to go back to what is essentially an HD 4850 with poor idle power.
Will there be a Dual GPU version as well?
If it is slightly faster than 512CC GF100 I wonder why they bother with it. Unless it will be cheaper to make, which frankly anyone doubts as its probably a bigger chip...
Bouncing Zabaglione Bros.
Legend
It's more like an evolution of the current gen I would think.
It would give back the single chip performance crown. It will no doubt be cheaper, cooler, quieter, and use less power than GF100. Nvidia can't make a 512 GF100, so assuming that a new revision comes out at about the same time with fixed yields, it keeps the margin pressure on Nvidia, who will have to match the cheaper price.
Oh, so you mean ATI is starting to behave like nVIDIA? The end is nigh...
Bouncing Zabaglione Bros.
Legend
If you've got to make a refresh product to bridge the gap until the delayed next process node, then of course you are going to try and make it better than your competitor's product.
The alternative is to allow Nvidia to bring out a fixed GF100 under a new name, and then throw away all the good progress you've made over the last year in the hope that Global Foundries or TSMC can get the next node working correctly and on time (yeah right!).
Don't forget, by October/November, ATI will have had Evergreen out in the market for a year, where Nvidia will have had only six months. By sticking to their upgrade schedule, ATI can claim they've obsoleted their competitor product after only two quarters. This is especially true if ATI can do another big bang release of top to bottom products to cover the entire market when Nvidia is still struggling to make, let alone sell their top end products.
If we take Charlies latest GF104/106/108 Article, with their launches slipping, we will probably see a refresh before nV is able to launch top-to-bottom Fermi. With what I know now, Fermi2 promises to be more of the same, a Fermi refresh that is not coming this year, despite what they promised everyone and again big. that all is supposed to launch well into 2011, still on 40nm where we would like to see 28nm products, or we actually will see 28nm products.
good time for a new buy then.
my 5850 is a ye
If my assumptions are correct, the "new" 5850 should perform as a 5870. Probably just Clocks&Optimizations.
Err, easy version:
Cheap Evergreens for OEMs and no reason for sane people to get nVidia in the channel?
You are also presuming that it is going to be substantially bigger/hotter, and the fact that older products do get less interesting with time and all...
Why can't they strive to sell more while trying to get an interim-milestone product out just to make sure their next proper successor chip isn't screwed up? R&D is R&D, people will get paid nevertheless, and a refresh keeps your image positive in people's eyes.
A refresh tackling some future-oriented (cache and threading, tri setup, tessellation performance, even lower idle power, more checkbox features) problems without majorly sacrificing TTM, perf/mm or /W seems to be a way more ideal refresh than guess what... GF100.
A GTX 480 is about 15% faster than a 5870 (having 1920X1200+4XAA in mind). A GTX 512 (lol) would be like another 10% faster, thus reaching 125% of the 5870. Something slightly faster could add another 5%, so the 6870 could be as much as 30% faster than the 5870.
Now, Neliz said "probably" pin compatible. If that is true, then it would mean that the chip is roughly the same size, no? Moreover it could still be 256bit and partners could just add an rv970 on the current boards, slap in a new BIOS and presto, the 6870! 30% faster than the 5870!
Now if they can offer that at decent prices, (lower than what the 5870s are today) it could be a double win situation.
I do find it extremely difficult that ATI can squeeze 30% on the same transistor count though. Unless they overclock it to hell and kiss the Evergreen's awesome performance/power ratio bye bye.
Bouncing Zabaglione Bros.
Legend
I doubt they would do that given the problems Nvidia have had and the benefits that ATI have gained due to their respective power/performance/heat/noise positions.
ATI is likely to leave it to OEMs to do the bonkers boutique factory overclocked models at the bleeding edge, while sticking with their current philosophy mainstream/sweet-spot strategy.
If Fermi has taught us anything, is that the main market considers noise/power/heat to be much more important than a couple of generations ago. A lot of people will simply not buy a product if it's too noisy or hot, especially if it's barely faster than the cooler, quieter, cheaper competition.
Wel... if GF104 is only slightly faster than 1440SPs RV870, you should also wonder why nVidia bothers with it
After reading your posts I feel, like anything ATi did recently was terribly wrong...
Pin compatible does not imply that the chips are the same size - the chip connects to pins of the packaging, and the packaging seems to allow much larger chips than the current. However, if the chip is to be pin-compatible, it implies that it has no alternative to being a 256-bit maximum wide bus, and the point is indeed to allow as rapid deployment as is at all possible.
If they do introduce these parts in october/november, I wonder what we can infer about the date of availability of product on the next lithographic node.
Bouncing Zabaglione Bros.
Legend
I think we can infer that the current foundry roadmaps are not to be trusted, and AMD (at least) has learned not to put all their eggs in the basket of a next-gen node being ready on time.
I would think that ATI are working on 28nm products for 2/3Q 2011, which will give time for the 28 nm process to mature and be more viable for mass production. It will give anything launched in Oct/November 2010 a reasonable six month life.
If we go back to the older leaked info about the taped out "HD6700" etc it still makes sense, even if the first 28nm parts won't come much later than the 40nm refresh:
A refined evergreen design on the well known 40nm, then a mainstream 28nm chip (could be with the existing architecture as it's just fine for the lower-end parts), and THEN the scaled up (HD6800) new design on the new node, when both are better known.
I don't see much reason for a low-mid end 40nm refresh.
A refined evergreen design on the well known 40nm, then a mainstream 28nm chip (could be with the existing architecture as it's just fine for the lower-end parts), and THEN the scaled up (HD6800) new design on the new node, when both are better known.
I don't see much reason for a low-mid end 40nm refresh.
Wel... if GF104 is only slightly faster than 1440SPs RV870, you should also wonder why nVidia bothers with it
After reading your posts I feel, like anything ATi did recently was terribly wrong...
Hmm.. GF104 is not a chip to substitute GF100, afaik
So I dont understand your point.
I didnt say anything ATI have done was bad. But usually it is nVIDIA who is very concerned with the crown above anything else. Thats what made them to the GF100, a monolythical, underperforming chip, after all.
GF104 will replace GTX465 and probably even GTX470. That are 2/3 of current GF100 line-up...Hmm.. GF104 is not a chip to substitute GF100, afaik
So I dont understand your point.
Similar threads
