Printable View
Here is the NDA information on the HD 4800 series from ATI, straight from AMD:
http://www.hardware-infos.com/img/st...l_document.jpg
Courtesy: Hardware InfosQuote:
Originally Posted by NH
Original http://img294.imageshack.us/img294/6...ocumenthx5.jpg
English: http://www.nordichardware.com/news,7755.html
(Yes this is posted in the ATI thread, but it is lost in the stupidity of people posting old info after new info x( )
Perkam
249 and 349!? more than I tought...
I put in bold the statement about ATI's use of GDDR5, it is intriguing why did that considering the official GTX 280 specs that came out today.
As for the price, I'm predicting:
HD 4870 X2 ?? ?? ??
HD 4870 1GB GDDR5 $349
HD 4870 512 GDDR3 $300
HD 4850 512 GDDR3 $250
HD 4850 256 GDDR3 $225
Perkam
Woot! Finally some almost official specs :D.
I was getting tired of the many possible specs and fake benchs all over the www.
nice job on the info!
might have to get one. I really hope it rapes Nvidia hardcore.
looks good until the 16Rops. that's going to hurt real gaming performance and AA performance big time. why increase shadders and mem amount and bandwidth and not improve the part that is hurting your current gen cards already? anyways time will tell about game performance.
3.92GHz?....sweet mama!
The pieces of the puzzle are coming together nicely. Great job on snagging the advance info....
Damn, almost 4 Ghz memory. I'm still shocked with GDDR5 speeds.Quote:
Originally Posted by perkam
The first image, seems to say that the 4870 will work at 850 Mhz, and your text says 725.... which one is true?
With the ease that ATI can make these chips in high volumes with good yields, it is not too far off to assume that the R6xx architecture has the ability to go 640/32/32, but ATI may not need to bring that out until the R800s.Quote:
Originally Posted by [XC] hipno650
Also, the fact that the HD 4x00 series barely fits the 150W envelope (4870) tells us that the reason for the lack of an ROP increase may have been power draw, which also supports the theory that higher configurations are possible if ATI forgoes the 150W power envelope.
Perkam
its going to be a 4870 for me or 4870X2 if it doesnt cost an insane amount.
that is if nvidia fails to deliver..
Interesting. I hope this is a real letter and not a faked memo, I'm sick of waiting for this launch.
WHOA at the GDDR5! Freeking fast stuff!
It has been verified by both Nordic Hardware and Hardware Infos, both of whom had the information for weeks now but were told not to break NDA.Quote:
Originally Posted by [XC] gomeler
Read the NH article.
Perkam
4850 512 looks extremely nice. If that turns out faster than or equal to the 9800 GTX, I may very well go with that.
wonder if memory is going to be pooled in Crossfire. As in, 512MB + 512MB = 1024MB not 512MB
If so, 4850 crossfire looks very nice
Hmmmm, interesting. The thing is, ATI really needs a halo product, power consumption and cost be damned. This strategy has worked very well for Nvidia. Having the top spot in the graphs is a very powerful thing, even if that top product is <1% of your sales.Quote:
Originally Posted by perkam
They do AA through shaders remember so 16 ROPS isn't as big a deal.Quote:
Originally Posted by [XC] hipno650
ya that would be interesting to see.Quote:
Originally Posted by Miwo
Hm...I'm thinking 4870 in CrossFire, would be nice to see how the 4870X2 go up against 4870 in CrossFire mode.
Don't know how this is "official" and/or "confirmed..."
Lot's of people "in the know" don't exactly agree with it.
Some of them have even said they think it is AMD trying to find leaks.
Check the NH article, this was supposed to be under NDA but leaked by Hardware Infos.Quote:
Originally Posted by LordEC911
The only thing not certain are the final clocks, everything else is as it is in the document.
Perkam
hopefully 4870X2 will winQuote:
Originally Posted by iTravis
i think that having more TMU will be better, and as assassins creed showed, if a game is programed correctly ati is much better at AA since it has dx10.1 (and its easily implemented on 10.0)Quote:
Originally Posted by [XC] hipno650
can do aa in shaders probably if they bothered to try
So one website says a confidential source has told them yadda yadda yadda, another website(Fudzilla) has mentioned final clocks, so NH divulges information they were told not to share? I should take this as fact because? :rolleyes: If these are true then great but anyone can fake a document :down:Quote:
Originally Posted by perkam
why would it hurt aa performance that much? The 9800gtx has 16 rops and still outperforms the 8800ultra in most cases except for ultra high resolutions and extreme detail in which case you won't be able to play the game anyways. Besides AA is handled by the shaders with the r600 design so it shouldn't really matterQuote:
Originally Posted by [XC] hipno650
He speaks sense. But I think we were just hoping for more extreme increases ^^ We won't know until we see the hard numbers i guess how this will pan out :yepp:Quote:
Originally Posted by AliG
ati dont care, tehy make the cards to directX and ogl specs so just becouse devs are instructed not to by some pople dont mean that ati needs to change how they do things, they sell more cards to labs and industrial apps than gamersQuote:
Originally Posted by villa1n
in dx10 the same buffer is used for the pre and post render for AA, so its not needed to have special hardware models for it, unless u use AA from dx9 or ogl 1.5
well im just saying. why upgrade so much of the other parts only to keep one of the weakest points there. why not cut shadders for more rops if they wanted to keep the power envelope. Look at R600 it had way to much bandwidth and lots of shadder power but only 16 rops and of course while it not the only reason it was a major factor in it's "lackluster" performance. and RV670 kept the same number of rops and now RV770??? it's knid of a let down to see. hopefully RV870 (or whatever it will be called) will have at least double. also the reason AMD cards AA performance is not as good if because they don't use the ROP's, if they had more to use then the could very possibly move back and thus the hit for AA would not be as much. because the way it stands is the 3870 can pull playable frames normally but turn on AA and it can be right on the edge or comfortable frame rates. and i don't want to buy a brand new "high end" "upper midrange" and have it not be able to do AA and max out all of my games (not crysis) on a reasonable resolution.Quote:
Originally Posted by AliG
i really hope AMD can pull up and have a good performing and well priced product but by the looks of it GT200 will be a hard competitor to beat. BUT if they price the 4870 well it will be a hit.
If that's why their holding back then they are idiots!! Maybe they like being raped by nvidia though.Quote:
Originally Posted by perkam
They're idiots if they let it hold them back - but unless you're telling us you know their architecture better than they do, I can't say thats for certain that is whats holding them back. In fact, their 16TMU's seem to have been the weaker point, which will be doubled. Also, keep in mind 16 raw ROPS doesn't mean they didn't double their operations per second... there's a lot of optimizations they can do you won't notice until real world tests are done.
So SINGLEDIE and 256BIT bus. Now can the dualcore/MCM etc superhype fanboys rest? Or will it already start again with HD5xxx having blablabla :/
Are you drunk?Quote:
Originally Posted by Shintai
What does HD4870 have to do with R700?
Still in denial?Quote:
Originally Posted by LordEC911
Not at all.Quote:
Originally Posted by Shintai
I was simply asking a few questions, which you have yet to answer.
There is more to RV770 than what meets the eye, or ear.
Proof?Quote:
Originally Posted by LordEC911
last time i asked you for proof, you didnt deliver it :p:Quote:
Originally Posted by Shintai
Not happening; the cards would have to be connected with a super fast interconnect, as fast as memory - Crossfire bridge nor PCI-E don't have that kind of bandwidth.Quote:
Originally Posted by Miwo
If perkam's price predictions are close, then I'm crying, haha, hopefully there'll be a nice price war to push things down a little between them and Nvidia.
HD 4870 1GB GDDR5 - most expensive and most useless. :( Anyway, I already have a HD 3870 since February - I'm would feel crazy and stupid to get an alternative after 3 months... who knows maybe by October they'll release HD 5870... :shakes:
ATi/nVidia stop messing with our brains, one generation per year should be more then enough. :down:
Nor did you. People claim miracle chips each time only to wakeup to a harsh reality on release day.Quote:
Originally Posted by GoThr3k
If you think AMD is havign a technological giant leap in their sleeves instead of basicly doign what they did for the last 15years+. Then its up to you to prove it. But this ends with the rest of the GTX killers and 50% faster clock for clock than Core 2 Barcelonas. In a wishdream that never materialized.
The Rops and Bus look like they are really going to hurt this card badly in terms of actual real world performance, especially when you up the AA at high res.
Its probably going to be an absolute monster of a 3DMark benching card though, just a shame that 99% of people couldnt give a crap about that though.
Sry to disappoint u, but this is fake and not the final specs. If u dont believe me, just wait a few weeks.
FAKE
They'll have a die shrink and slight update in around 6 months. Yes, a 6 month product lifetime. :rolleyes:Quote:
Originally Posted by XSAlliN
Oh well, I guess they feel like they need to make up for the 8800GTX/Ultra being the single-chip king for nearly a year and a half. :p:
If only game developers would follow their steps. Apart from silly games like Crysis every other game is a console port that runs perfectly with a single 9600GT..
i dont think the things your are suggesting...Quote:
Originally Posted by Shintai
Do we think core clock/shader speeds on the RV770 will be BIOS locked like the RV670?
^Ummm... RV670 is not bios locked.
No the lack of texture power was the biggest blow imo, this time around it'll have more than enough shader power for aa and hopefully enough TMUs (I personally would have liked to see the 64 g92 has, but that's just me), but only time will tell.Quote:
Originally Posted by [XC] hipno650
looking forward to these.
one of these 48xx and maybe a new gtx thingy for my 2nd peecee. sometime soon mehopes.
Will one of these cards then be what I finally upgrade to from my x1950xt? That x1950xt has just held out too well for me to really bother spending the money so far :p:
these specs are the same posted here ??? >> http://www.xtremesystems.org/forums/...&postcount=818
Quote:
Originally Posted by w0mbat
regards
so repeating a lie makes it right? :ROTF:Quote:
Originally Posted by mascaras
OK so now i say the sky is green, and i post this in different forums, and some newssites and then it must be true.
maybe there wont be any 48XX cards :D; but if there are then the specs are whatever they end up being.
lower gpu speeds than i thought...
4ghz gddr5.... wow...
qimonda also has 4,5ghz gddr5 in the pipeline. ;)Quote:
Originally Posted by saaya
here we go again, sigh.
Yea I also predict prices to end up like that. Which I think would give ATI at least a few months - half year nice sales numbers in performance section until performance segment is filled out with GT200 chips as I bet these 4xxx cards will perform a bit better than G92/G92b while NVIDIA will probably dominate the highend section with GTX 260 and 280 which I think will end up as $450 and $550 pricetag and I bet ATI will get more profit per card than NVIDIA if prices would be around those levels.Quote:
Originally Posted by perkam
at least they got a bump. but not bumping the rops?? another problem with this many shadders is games are not coded to use them well. so yes 3dmark will fly but real games will see less of an increase until games code for the mass of shadders. and yes i do agree that the Number of TMU's will be a problem but at least they got doubled but they are still far behind nividia's count.Quote:
Originally Posted by AliG
that doesn't mean anything. I can tell you that i've been under an NDA and show u some fancy pictures too...Quote:
Originally Posted by perkam
Then what MEANS something really? Let me answer for you: nothing. Am I correct?Quote:
Originally Posted by Dave_Sz
Credibility.Quote:
Originally Posted by Dave_Sz
That is what makes the difference. It is the reason a newbie posting a 6Ghz Intel quad core pic would be shunned for being a photoshopper, but when hipro puts one on XS, you KNOW its real, because, as an extreme legend, you can trust what he posts.
I do not know about hardware infos, but NH is a credible site, but that is not ruling the possiblity that it could be a plant by AMD, since Kinc from NH hasn't (or can't) post in this thread to confirm.
We'll just have to wait and see, but a document as original as that is hard to call a fake seeing as both websites are staking their credibility on the line to confirm the specs, which means SOMETHING, not nothing.
Perkam
Ok that's it, now I KNOW Marcus (kinc) has them :rofl::rofl::rofl: :ROTF::ROTF::ROTF::ROTF: :clap::clap::clap:Quote:
Originally Posted by w0mbat
Another note, 480SPs on single slot cooler for 4850 FTW :up:
Perkam
maybe this will finally be worthy of a crosfire-x setup (4870x2 + 4870)
I'm just hope having a x-fire board pays off eventually.
AFAIK they dont have them right now but will get them soon.Quote:
Originally Posted by perkam
Actually, when Fuad posted the RV770Pro clocks, I got permission from the source to post the rest.Quote:
Originally Posted by [XC] gomeler
Indeed ;)Quote:
Originally Posted by w0mbat
//Andreas
When was the launch date again? I can't wait for the cards...errr I mean WR scores to be posted. :p:
Either 16th or 18th june.
I agree, they should have made the rv770xt have something like 32 ROPs and the rv770pro have half of them disabled, that would allow them to bring both a nice power envalope and also it would further seperate the 4870 and 4850, which I personally think needs to be done for them to sell more of both cards.Quote:
Originally Posted by 003
why not somethink like a 4830 with the disabled functionallity you describe?Quote:
Originally Posted by AliG
4870 RV770 gpu, high clocks
4850 RV770 gpu, medium clocks
4830 RV770gpu, low clocks and some tiny bits disabled (like for exemple 16 rops)
i think this would able them to get higher yields, get an enthusast part, get an high end card and a high midrange card
If the 4ghz ddr5 w/ 256mbit bus = ddr3 with 512mb bus, why pay more for the ddr5? Am I missing something?
causeQuote:
Originally Posted by TheKarmakazi
gddr5 with 256bit -> cheaper pcb, cheaper chip, more expensive ram
gddr3 with 512bit -> expensive pcb, expensive chip, cheap ram.
DDR5 also has some sort of timing compensation built in (someone clarify this) so you don't have to worry so much about trace lengths.
Exactly, the problem is with the start of the r600 cards, ati hasn't really had a true high end gpu. So that's why nvidia could disable parts of its high end and still sell them very well. Where as the 2900xtx failed to perform better than the 2900xt and that could barely compete with the g80 8800gts, so ati would have got no where selling a partially disabled r600 as a second high end and the die was too big to be a performance chip.Quote:
Originally Posted by GoThr3k
And since ati isn't planning on ever again creating a monster monolythic die, chances of them having the option they used to with the x1950xtx to do what nvidia will with gt 200 is very slim. But if nvidia screws up like intel did with netburst, then they well have that window open again to broaden their line up and make more profits off the same chips
GDDR5 will also save power versus GDDR3 also
Quote:
Originally Posted by Hornet331
Quote:
Originally Posted by eleeter
Thanks for the answers, they all make sense :up:Quote:
Originally Posted by zerazax
Another thing to keep in mind is that the size of the core will also be a part of if it's possible to implement a 512-bit interface to the memory.
Small core --> not enough space for all the pins on a 512-bit interface.
I'm confused as to why everyone thinks more than 16ROPs is needed...
G92 "only" had 16ROPs.
If RV770 has 16ROPs but tweaked Z, then it will be perfectly fine.
Remember the high-end, meant for high resolutions +20", will be dual GPUs, R700/CF.
could be architecture differences too maybe, like how AMD has always had less L2 cache than intel but smacked the larger cache P4s around :shrug: Don't really know, just saying it could be different implementation perhaps and could be an apples to oranges comparison.Quote:
Originally Posted by LordEC911
as far as i understand, ati does aa in their shaders, which is also why i think it was pusing dx10.1 so much, and why nvidia has not been since it uses the rops. what crippled r600 was the tmus, not rop.
You're missing the fact that there is no such HD48_0 card that has 512bit GDDR3. Only 256bit GDDR3 for HD4850 and 256bit GDDR5 for HD4870.Quote:
Originally Posted by TheKarmakazi
Btw, bus width is measured in "bits" (128bit/256bit/512bit), not "mbit" or "mb".
the couch i'm sitting on is reeeeeal comfy - lets sit back together & watch these clowns turn this thread to the crappers :D :up:Quote:
Originally Posted by biohead
I think without posts like this, most are allready aware of you enjoying seeing a thread "turned to the crappers", your contribution in doing just that has been unsurpassed.Quote:
Originally Posted by tiro_uspsss
On topic, unless NH is plain lying, the specs are known at this point.
Not confirmed though, there still is the possibility that NH is not telling the truth.
Big die, small margins, dead ATI.Quote:
Originally Posted by GoThr3k
People who keep asking for a fast ATI chip (probably so that they can get their GT200s cheap, nothing else :p:), but apparently forgot the R580 situation. The chip was fast, in shader stuff blown the G71 out of the water, but ATI still netted loss on the GPU/line itself. The GX2 with horrible SLI support barely kept up too, especially when you used the card for more than 6 months.
ATI might be aiming for a shorter upgrade cycle with more cards, a strong midrange and hopefully a competent highend. A very nice upside is that RV770/870/etc will be hard/virtually impossible to screw up in execution/ stuff after tape-out. :)
actually - i cant wait for R770 launch, i hope it kicks the crap out of whatever nv has, & even if it doesnt, & at least comes close, i will probably buy, it'll make a nice replacement to my x1950pro :) .. btw, i was 'talking' to biohead - seeing as ur trolling, i'm reporting u :up:Quote:
Originally Posted by Jakko
well, looks like the 4870 will have the GDDR5 memory clocked at just 900MHz. but it seems to be more like qdr instead of ddr memory
ASUS published the specs of it's 4870 with a memclock of 3.6GHz (900MHz DDR5). so 900x4 it is.
Specifications Model EAH4870/G/HTDI/512M
PowerColor Introduces HD 4870/4850 boards
Asus jumps the gun, announces HD 4870
Sapphire and PowerColor announce HD 4870
thats becaus gddr5 has 2 clock domains, 900mhz command clock and 1800mhz write clock. If you double the write clock (ddr) you have your 3600mhz. :p:Quote:
Originally Posted by Der_KHAN
i see. thx for clearing that up :)