found this today looks really good spec wise.
http://www.techpowerup.com/172312/AM...-Unveiled.html
http://videocardz.com/34981/amd-rade...fiation-leaked
Printable View
found this today looks really good spec wise.
http://www.techpowerup.com/172312/AM...-Unveiled.html
http://videocardz.com/34981/amd-rade...fiation-leaked
No shader count/SP specs, all this raw data doesnt help much in guessing the performance. And the price there is misleading, 7870 is <240 bucks AFAIK .
If true, the increase in transistor count is small - are they planning to push just about enough to beat 660Ti ?
EDIT: OK 5%+ increase in clock and 75% increase in SP FLOPS => ~60% increase in shader count ,so ~2048 SP ? Thats about as fast as 7970 with a crippled memory bus.
That is a 78xx series replacement... Why is it bad that it will be as fast as 7970?
If those are even remotely true, then 89x0 would be monster sized chip.
Could be a monster,it has to fight with GK110 and it's AMD's last 28nm video card,so process is better and they can increase the transistor count ;)
All of that at 160W will make it a monster of efficiecy... Can't wait to see it.
Also can't wait to see how Nvidia will come late with gk110 and be only slightly faster than the 8970 but people will call it the best thing since sliced bread.
The current specs describe a card that has theoretically equivalent compute performance, in reality it wont be as fast as 7970 because the memory bandwidth will hold it back - the bus is 2/3 in width and lower clocked. I think this will be the case across all mid-high segment unless GDDR5 speeds keep on increasing.
Thoses specs are not official and be leaded from Read2ch: http://translate.google.com/translat...2FID%3AhKcZJDn
Take it as rumor or even not real specs..
For the memory bandwith... in reality taken in account with 1100mhz boost vs 925mhz on the 7970, outside 2560x1600, add some optimisation and the results should be really close of each other. ( if we trust theses specs )
SP performance look close, but for DP (not really important on a middle range card ) , we are speaking about 1/16 vs 1/4 DP rate .. ( 25% performance of the 7970 )
All specs aside the retail pricing of the unit will give the best overall indicator of average performance.
Here we go again with the baseless rumors and people getting their hopes up.....
Without rumors there would be no debate, and the interwebs would be just damn boring...
well if we don't want it to be that good. then why would amd even attempt to make our expectations happen.
we're the customers speaking having shown an interests in that.
everyone else ain't the market. if that appears attractive at that price to us. then we're the market and mass production might be able to make such financial feesable but if no demand for such, they do not produce as much and prices can only be dropped when demand is proven, later in its life.
exactly how it shows us in the market actions and behavior.
if we create a noise. we're heard and its produced more. if the price is set too high and profited beyond our value of it. they price drops due to supply exceeding demand.
i personally have not bought a gpu since 4870x2 was new. and 8850 looks attractive and i will buy at that price. otherwise i'll just wait for 9900 series to adapt to my market.
cause its appearing to me the market has been reviewer driven. accepting just what is offered. not buying the product. and if that is who they're making them for. no wonder they're overloaded with gpu's and price drops. they didn't listen to the market. only waited for the market to be marketed to by reviewers. and focused reviewing to sell product that wasn't meeting our expectations in the market they were trying to sell to.
and it would be like the good old days, of before "economic crisis" and there would be ONE REVIEW of the card, for many months. cause it met expectations in price and performance gains for the market its aimed at. and no cards gonna be justified to give away.
and the exceeding our expectations for the price. results in health recover of R&D on their side and if initial demand exceeds supply. a healthy markup for our our respective economies occurs by retailers and both win.
with some maths i did 8870 flops *7870 clock/8870clock/7870 flops=shader ratio. then shader ratio * 7870 shader count and i got 2038, so assuming logical counts and rounding i would agree on 2048.
but since flops offer no info on actual performance just clock and shader count it looks to be a nice gain but i am not sure on how power consumption will go this gen. we could work all of the numbers back but i am to lazy
that cause unicorns are predatory and all died off. they had a spike on their heads how all horses and deer and moose say hello with a head bunt.
but they would not see it , they not say anything about it. and they had no say in the matter
he can only not say , he wants free toys. nda , and he can be sued in quebec they got civil law to silence dissidence.
Leave my pot of gold alone.
$1 in our current fractional reserving banking system is worth over a million in credit. a piece of gold only gets you back 1700 in credit. its a worthless trinket , but has a ever decreasing industrial value in computer processors. devaluing their own currencies. exporting it to the lands with the most gold. which is rothchild banking of europe. and their vaults filled from the french english wars. to bring on another age of sodomy.
you all worked on credits. and really made at best $2 in your life.
china has a fixed market, buying up most the gold. will fix it at an all time low to provide us our processor needs at dirt cheap costs as always. and intel and amd will have to do the same to survive.
so that we can play games. that desensatize us of our humanity. let the brain worm epidemic effecting most on earth. with toys to play their games. to forget the fact, we're the terrorists in the world. only economy is made off of war. and eventually they win, the invaders always lost in war.
so that we never actually get rights in the homes and have to obey our parents commands in violation of our birth rights constantly. and all change has to start in the home
while nations with social security through the UN gave every since not worker, our poverty level in money. of $24,000 year. stolen by the province and states. to pay the workers for war criminals money. and predators upon our health and lifes. enforcing corperate laws on the citizens who only had birth rights, signed seals and delivered.
both our grandparents and ourselves are the slaves of our parent we got the worst quality of life in america , in all recorded history. we just go, less toys and more wise about money then our parents. cause they weren't gonna let us have any of it.
and its us the 55% of the population now without opportunity at work, amd need to think about. cause we don't go much to spare of our charity as its stolen by criminal mafia governments who cops don't do :banana::banana::banana::banana: about being paid by us.
cause those workers, that actually can build a PC are the 1%
HD8850 - 28nm Oland Pro
3.4 billion transistors
~ 270mm-280mm^2 die
925mhz GPU clock (975mhz Boost)
1536 Shaders
96 TMUs (93.6 GTexels/sec texture fill-rate)
32 ROPs (31.2 GPixels/sec pixel fill-rate)
6 Ghz GDDR5 @ 256-bit (192 Gb/sec memory bandwidth)
TDP 130W
(Single-Precision Compute: 2.99 Tflops, 187 Gflops Double Prec.)
HD8870 - 28nm Oland XT
3.4 billion transistors
~ 270mm-280mm^2 die
1050mhz GPU clock (1100mhz Boost)
1792 Shaders
112 TMUs (123.2 GTexels/sec texture fill-rate)
32 ROPs (35.2 GPixels/sec pixel fill-rate)
6 Ghz GDDR5 @ 256-bit (192 Gb/sec memory bandwidth)
TDP 160W
(Single-Precision Compute: 3.94 Tflops, 246 Gflops Double Prec.)
So a 27% increase in die size with a higher clock speed on the same process with a lower tdp, yeah sure. :rolleyes:
LoL. They just took the change in transistor count and applied it to the die size...
If transistor count and specs are true, you would be looking at ~250mm2.
it is supposed to be on goflo and not TSMC, so the drop in TDP could happen. it could also be similar to the change from 45nm to 45gt when changing from the 4870 to 90 and that made a huge difference with voltage needs and leakage.
if it is spec based they can not really destroy the PR like promising better single threaded performance and overall faster but then making something that is slower per clock with the same number of threads, but they gave us some thing the same single thread at the same clock and slower when you use both threads on a module.
I have to agree that these rumors are most likely inaccurate and something fabricated to create page hits. Any early rumors have been way off about AMD lately and performance has basically under delivered since 5870 according to early rumor hype.
As for the performance of the next 8870 having gtx 680 performance. This is completely possible. However I still think these rumors are false. Not so much about being a pessimist about possible performance figures but all these rumors from AMD lately have been completely off. 6970 and 7970 History threads show this. The 6970 performance was supposed to deliver the type of performance you expect from a new node and new architecture and the 7970 was supposed to beat the gtx 580 by 60%. The reason why these leaks aren't likely true is AMD wouldn't let this information be know this early. Price and performance months out? Neither tech company has let this much information leak 5 years ago even and today, information secrecy is even more tight. AMD is one of the best at this right now. I.e keeping bulldozer underwraps and 6970 and 7970 performance until close to launch.
Showing how much performance and your target price this early is a huge mistake. This information advantage lead to the collapse of all 7xxx pricing and for the gtx 670 and gtx 680 looking better than they were.
thread should just be locked, the source is 2ch which is well known for being as reliable as reading the future out of the shape of some random dog turd found on the street.
The specs are quite realistic and very much what we can expect from AMD. AMD knows the GTX 680 is going to be the next gen GTX 760 Ti when the GK110 chip launches as the flagship GTX 780. The HD 8870 chip has identical resources as HD 7950 - Shader count (1792), 112 TMUs, 32 ROPs and other resources like 2 ACE, 2 rasterizers, 2 tesselators. The 256 bit memory controller of Pitcairn will be updated to run at 6 Ghz. The HD 8870 will be gaming focussed and perf/watt optimized like GTX 680. HD 8870 has a good chance of matching the HD 7950 at the same clocks. As we have seen in the past AMD also makes improvements to the tesselators and other performance tweaks will also be done. And with improvements in the 28nm process you can expect lower voltages compared to HD 7000 series to hit the same clocks. Other areas like reducing chip redundancy because of a more mature 28nm process would mean transistor savings. So overall a 3.4 billion transistor chip which matches a GTX 680 is quite expected.
http://www.techpowerup.com/reviews/S...apor-X/31.html
HD 7950 (1100 Mhz) matches a HD 7970 Ghz (1050) . I would expect HD 8870 (1.1 Ghz) to atleast match HD 7950 (1.1 Ghz) at 1080p and 1200p and fall behind slightly at 1440p and 1600p .
What GPGPU resources does Pitcairn have, that can be cut?
I'm referring to Tahiti here as (I would assume) that iteration of GCN would be the basis of any evolutionary changes. Many could even cascade down to encompass the lower end cores as well.
A good example of optimization would be the Compute Unit itself. As NVIDIA did with Kepler, streamlining core functionality with a focus upon gaming starts with the "building blocks" of a given architecture. I wouldn't presume to say how AMD can cut items since I'm not an engineer but any large scale move towards prioritizing certain functions starts here.
The Tahiti refresh will most certainly not be cut down regarding GPGPU. I though we're discussing Oland?
Because there is always a trade-off of some sort. AMD needs Tahitis strong GPGPU part to stand their ground against Tesla. Optimizations in general yes, but not at the cost of GPGPU efficiency (at least for Venus Pro/XT).
A 200% blown up Pitcairn as a possible solution for Venus is out of the question, imo. That is what I essentially meant. Their largest chip won't be purely about gaming, same as Tahiti. That said, Pitcairn is. So I'm unsure what is there to optimize especially.
since high end gpu sales should be focusing on gaming perf. as laptop perf for gpgpu.
cause if a game comes out that sucks on it, cause of finally making awesome use of gpgpu they just ramp up the laptop gpu's perf and call it the desktop one.
If those specs are real, it means the 28nm was incredibly bad and now is finally ok with good yelds, how else would you be able to sell a faster gpu cheaper?
higher demand due to the cheaper price, covering r&d costs over a higher number of cards instead of less.
as they proved with their last round of gpu's. when the price dropped it was an upgrade worth paying for. covering the r&d costs finally that the higher price couldn't do , due to the demand for the product.
so through the higher efficiency of longer runs of the lines. less hydro costs due to less switching products on the line, less programmer payments for the higher paid employees. less reprogramming the line switching between products. using the card that don't meet speed requires as re branding in the lower end market. loss prevention and costs of recycling the material. not forking over voluntary carbon emissions fees. less environmentally friend logo certification stickers. accepting gaming is not environmentally friendly and lowering power consumption reduces the dependence on nuclear power was proven by fukashema to be the least environmentally friendly. so just lowering power consumption proved more environmentally friendly and they don't need a logo of someones to show that. less artists making boxes, selling the cards not marketing to sell cards. moving all the stupidity costs of marketing to the re sellers who buy in bulk of 1000's of units. 1 year warranty. letting re sellers eat the costs for more in their prices. letting the guy selling cards in a oem brown box compete. moving all accessory to the re sellers to cover at their discretion. no partnerships with game developers leaving that to the re sellers. giving up on producing tech demos when games do that fine. helping game developers optimize for their code paths instead.
a new guy making the decisions.
I dont say thoses specs are real ... but you know GCN first "chips" have tapout in march April 2011. production have been start to be ramped around october 2011 ( maybe even september )..
We approach slowly the end of 2012.. the way of product 28nm have certainly evoluate since.. ( the hunt of energy lost, the capacity to put more transistor on an equivalent surface without increase much the TDP etc etc. ).
Even the yield ( and so cost/revenue for AMD or Nvidia ), the risk is now better controlled...
(Remember Nvidia Fermi, who have do a lot of progess on TDP between the 480 and 580 ( and the core is nearly exactly the same.. 90% of the tdp removed was by change some transistor size where they was too much leak )
HD8000 will sport GCN2.0, as allways during a developpement, there's a moment where you let the baby to the "engineer production " team ( the guys who take the project and decide of the final specifications ), letting down many thing the initial team will surely like got more time of include it, we shoud surely see some slight optimisation and correction of the initial GCN .
Dude I done even know why I just read your whole post but seriously what the hell do u mean man? :confused:
Anyway, the current crop of 7xxx cards show that there is certainly room for faster clock speeds as well as lower operating voltages. I feel that the next gen cards will benefit greatly from that along with process improvements (+ what Lanek said). Of course, GCN 2.0 will increase shader efficiency so while these specs might not be real, its not too far fetched to believe either
man sea islands is soooo last week
I want Pirate Islands!!! Yaaaaarggg
http://vr-zone.com/articles/amd-next...mes/17154.html
Quote:
2015: Pirates of… Graphics?
Perhaps one of most interesting codenames is coming down the pipe in 2015: Pirates Islands. Yes, you've read it correctly. Bringing a tribute to legendary pirates such as the Blackbeard, Captain Hook or well, Captain Jack Sparrow, AMD's imaginative engineers are targeting the 20nm process with 14nm APUs in mind. The real work on this GPU architecture only started recently (GPUs are being designed 2-4 years in advance) and guessing what lies on cards is a bit too unpredictable. We know feature set, but performance… your guess is as good as ours.
Read more: http://vr-zone.com/articles/amd-next...#ixzz26ycbZT1D
Next-gen flagship Tenerife HD 8970:
http://www.pcper.com/news/Graphics-C...-Graphics-Card
nice. :D now prices. 3W long idle, loving that idea. most my gpu's time is in idle.
Omg lol, that slide surfaced like 6 months ago and consensus was that it is fake.
or just nda. if it comes out like that. will anyone admit it was real? or accept majority rule and just say , what ever the majority didn't know, didn't happen. :ROTF:
and until someone sells a prototype of the card or breaks amd nda working for R&D at amd, there is no way to know if that was the intentions of the next gen card 6 months ago. but was not feasible financially or in production.
maybe some news about 8970 .( thanks iMacmatician )
Ofc dont knowif thoses spec are true or not... take them as they are.
http://wccftech.com/amd-radeon-hd-89...ations-leaked/
http://www.pcper.com/news/Graphics-C...Graphics-Cards
not sure if this was posted as well
edit lol it was :rofl:
Salt. You need a lot for this one.