September 15th, 2012

AMD Radeon HD 8870 and HD 8850 Specifiation Leaked

Well, maybe this is not a leak, but prediction. Hard to say, it’s quite good source though — they leaked Tahiti specs few months before launch.

AMD Radeon HD 8000 series are believed to feature Sea Islands GPUs. This updated Graphics Core Next architecture would still be made with 28nm process. It’s worth saying that HD 8870 and 8850 would not be the fastest Sea Islands GPUs in the pack. The one that we are going to describe should be compared to Pitcairn, which has 2.8 billion transistors. The GPU that is powering Radeon HD 8870 and HD 8850 is called Oland and it would feature 3.4 billion transistors. Die size of this Oland processing unit is estimated at 270-280mm2.

According to the source, Radeon HD 8870 would be as fast as GeForce GTX 680, that’s 30% faster than GTX 660 Ti. Card would be clocked at 1050 MHz for base, and 1100 MHz for boost. So it seems that AMD will also implement their PowerTune feature for next series. There’s a noticeable increase in memory bandwidth (192GB/s). New Radeon HD 8870 would have effective memory clock around 6GHz (assuming it would be 256-bit interface). Leaked data also reveal that new mid-range/high-end GPU would have a TDP of 160W, that’s a 15W lower than Pitcairn-based HD 7870.

The mid-range Radeon HD 8850, which is comparable to GeForce GTX 670 in terms of performance, would also feature the boost technology (clocks are 925/975 MHz). In case of TDP, this model would have a similar power requirements as HD 7850 — 130W.

I’m not sure where did those specs originate, but there are also few interesting sentences like:

High power efficiency;
High cost performance;
Improved texture unit

It’s unclear yet how many stream processors would those cards feature, but it’s said that they may have 1792/1536 cores respectively.

There’s a note about January 2013, I’m guessing that’s the release or paper-launch date. Furthermore, we might expect around 22% lower price in comparison to HD 7800 series.

Comparison	Radeon HD 7870	Radeon HD 8870	Diff	Radeon HD 7850	Radeon HD 8850	Diff
GPU	28nm Pitcairn XT	28nm Oland XT	-	28nm Pitcairn Pro	28nm Oland Pro	-
Transistors	2.8 billion	3.4 billion	+21%	2.8 billion	3.4 billion	+21%
Die Size	212mm2	~270mm2	+27%	212mm2	~270mm2	+27%
Base Clock	1000 MHz	1050 MHz	+5%	860 MHz	925 MHz	+7%
Boost Clock	-	1100 MHz	-	-	975 MHz	-
Primitive Rate	2prim/clk	2prim/clk	-	2prim/clk	2prim/clk	-
Computing Performance SP	2.25 TFLOPS	3.94 TFLOPS	+75%	1.76 TFLOPS	2.99 TFLOPS	+70%
Computing Performance DP	160 GFLOPS	246 GFLOPS	+60%	110 GFLOPS	187.2 GFLOPS	+70%
Texture Fillrate	80 GT/s	123.2 GT/s	+65%	55 GT/s	93.6 GT/s	+70%
Pixel Fillrate	32 GP/s	35.2 GP/s	+10%	27.5 GT/s	31.2 GP/s	+13%
Memory Bandwidth	153.6 GB/s	192 GB/s	+25%	153.6 GB/s	192 GB/s	+25%
TDP	175W	160W	-9%	130W	130W	0%
MSRP	$349	$279	-20%	$249	$199	-25%

UPDATE:
It seems that not only HD 8800 series are being prepared. Radeon HD 8570 and HD 8490, probably some re-branded cards from 7000 series.

Source: read2ch

Author: WhyCry

BoBoBoBo
Awesome
BoBoBoBo
Price even lowered too that awesomeness
stop bloody priced!
D.a.rkKNI._.ght
never mind about so called rumor about specs.mean u cant guesss the performance now days just simply looking at the specss..& i remember when i said “eew” when i saw 256bit interface of gtx680 & thought how the hell it was gonna compete with a card with 384bit….
Pingback: Seria AMD RADEON HD 7000 - Page 71
Merdiso
Sounds too damn good to be true. :)
Syed Ahmed
the price are wrong amd is going to be first to pull out his second generation card so the price are going to high as the performance until nvidia strikes
Raghar
These number doesn’t add. If it would say 22 nm, then possibly. However AMD din’t have that bad performance/watt, and these large FP power would mean something must go, for example integer compute performance.
270/212 = 1.27358490566 Which implies much smaller gain in FP, and rise to 220W in power consumption. Naw it doesn’t look like anything else than a guess.
Erenhardt
That is sad. If they release 8000 series in jan 2013 it is going to be expensive. nV just is launching 600 series (GTX650Ti is coming?). There is no way nV is going to do anything about that. They are not going to release 700 series 2 months after 600. The only thing they can do will be to lower prices – just like AMD did when NV released 600 series cards.
BTW. Any leaks about GTX650Ti?
Pingback: Filtrado el rendimiento de las AMD Radeon HD 8870 y HD 8850 | El Chapuzas Informático
BestJinjo
3.94 Tflops single precision compute @ 1100mhz implies 1790-1800 of so shaders! Going from 80 –> 123 GigaTexels in texture speed implies 112 TMUs
HD7870 has:
1280 SPs –> ~1790-1800+ shaders (+40% increase)
80 TMUs –> 112 TMUs (+40% increase)
That would be extremely impressive for a GCN 2.0 (refresh), especially since they are claiming a drop in power consumption? And a price decrease? Sounds too good to be true!
Pingback: El Thread de AMD Radeon HD 8000 Series
BestJinjo
GTX670/680 launched in March 2012 though. GTX650/GTX660/660Ti are 6-7 months late. You cannot use those to gauge how soon GTX770/780 will launch. NV won’t let AMD have a 6 months head start on the high-end. I expect GTX770/780 to launch in the first half of 2013 and if NV needs to, they can rebadge GTX660Ti as GTX750Ti, GTX670 as GTX760Ti, GTX680 as GTX770 (just guessing, but you get the picture).
Pingback: Ya se cual creo que voy a comprar....
http://videocardz.com/ VideoCardz.com
Well, they must have read your comments and drawn some conclusions :)
Anyway, if GCN2 would be so fast, then GK110 might not be enough to compete.
http://www.facebook.com/FwixGamer Anas Azghari
emm i am not sure about those price if those spec are true !!
and also how the oland xt gonna cunsume less than the Pitcairn XT with +21% Transistors i’ts just impossible with the same 28 process
Erenhardt
Didn’t thought about that. Will rebadging bring price drops? 750ti cant be 300$, it is bloody x50Ti(low end GTX)!
DavidJamesIsTung
Sounds too good to be true. I just hope they won’t end up like the Hd 6xxx series.
BestJinjo
I can’t predict prices. I am saying NV can bring out the flagship GK110 (or some other chip) and make 2 SKUs out of it say GTX770/780. They can then rebadge or release slightly higher clocked GTX670/680 for mid-range at $299-349. Who knows how much of a performance increase we’ll get next generation but the rumors keep pointing to 30-40%. I am guessing GTX680/7970GE will only be mid-range by 2013 :)
BestJinjo
It could be that HD8850/8870 will have higher performance increases from 7850/7870 than 8950/8970 will have over 7950/7970 GE?
It could be the specs are fake?
It could be that NV unleashes a monster GK110 around Crysis 3 launch to show off extreme tessellation? :)
BestJinjo
Pitcairn XT is a 212mm^2 die and GK104 is a 294mm^2 die and both are “pure gaming” chips if you want to call them that (i.e, Tahiti XT has a bunch of added computational transistors for double precision compute for example, etc.).
If they build from Pitcairn XT, they can expand the chip to 270-280mm^2 and make it nearly as fast as a GTX670/680. Not saying it will happen for sure, but it’s not out of the realm of possibility. Remember you have 2048 SPs, 32 ROPs and 128 TMUs in 365mm^2 in Tahiti but that’s not 2x more units than HD7870 packs in 212mm^2. So clearly they are some non-gaming related transistor space in Tahiti XT that isn’t needed for mid-range HD8850/8870 parts.
Bringing a solid mid-range HD8800 series is actually not hard. Trying to retain the full compute of Tahiti XT and then expanding die size from 365mm^2 (while keeping the GPU at 250W TDP or less) will be a lot more challenging for HD8900 series. Although AMD may have no choice in the matter if NV is aiming to drop a GK110 7 billion transistor monster!
Pingback: AMD Radeon: 7000 Series Thread(News and Updates ) - Page 27
skr13
HD 8870 for just $280? or $300 custom versions, seems alright for my pocket, and great card for upcoming Crysis 3, BF4.
Don´t know if i should wait for HD8000 or expect see a HD7950 for $250.
If Nvidia will take many time to launch 700 series, AMD will abuse on price again.
Syed Ahmed
i dont thing so that gtx 650 ti is going to come cause nvidia never goes for the main stream cards like amd does so if they are going to release its going to gtx 660 se or they are going to release gtx 650 ti just before launching the 700 series cause the yields of 28 nm are still low
Jerome
Wow AMD is going to have a huge lead this time around
Jerome
Thats the problem they can’t compete with ta 256 bit bus. The 7970Ghz beats it in most games and when the resolution is 2560x1440p and beyond the weakness of the 600 series becomes evident.
Jerome
The GK110 as it is now is a raw compute monster, have no idea what the consumer version of the card will bring. Really doubt it will trickle down to us with 7 billion transistors. GCN 2.0 will probably have more compute power than the GK110 anyways. Rumor was that the GK110 will be around 25% faster than the 680, while rumors of the 8970 put it 45% ahead of the 7970. GCN was a gamble by AMD, now entering its second generation its going to be refined, tuned and optimized. Kepler is just an evolution of Fermi
Sena
Lower price is expected because 28nm matured, but some things sound too good to be true, but we will see.
Those HD 8870s might be superb replacement for my two HD 7870.
BestJinjo
I thought GTX650 Ti is launching in October?
BestJinjo
Right, but I think the point is HD7970 GE has 50% more memory bandwidth than GTX680 does but it’s not 50% faster. We’ve seen before when certain GPUs had too much memory bandwidth that was not utilized fully (HD4890 or GTX285). Right now the memory bandwidth for HD7900 series is mostly marketing. If I overclock from 264GB/sec to 345GB/sec, my benchmark scores barely change. AMD needs to increase pixel fill-rate, geometry processing and texture performance instead. I am guessing 2560 SPs, 48 ROPs on the 8970 with the same 384-bit 288GB/sec bus.
Pingback: Specifiaties AMD Radeon HD 8870 en HD 8850 gelekt? | Tech-nieuws
Erenhardt
I would love to see something between 7770 and 7850 and gtx660. Call it whatever you like. There is nice gap in performance and price.
Syed Ahmed
its just a rumor not conform news so we dont know but there is a huge gap between gtx 650 to gtx 660 i think if they are going to launch it would be like this gtx 650 ti 512 kepler cores and gtx 660se 768 kepler cores plz God make this true i really want to upgrade to gtx 650ti
Jerome
Yeah but double bandwidth doesnt mean double performance in the first place. The bus width on the 7970 is not marketing at all, it really depends on the application. Alien vs Predator is an example of a bandwidth hungry game, as such performs significantly faster on Tahiti 384 bit based hardware.
Raghar
Punctuation?
Think about GTX 660 as about GTX 680. GTX 680 spawned GTX 670, which was followed by 660 Ti when there were enough cards that didn’t pass trough 256-bit tests.
Now we have a GTX 660, some of these would have defect on SMX, others would have defect on the only SMX on cluster, or on cluster’s memory controller. Thus Nvidia would release some lower end variant. The question is how would it be released? If it will be released as OEM only cards… then lets hope for decrease of GTX 660 prices.
I guess at least one of these cards, GK106-250, would get released. Current low end segment is dominated by GTX 650 “lets not talk about it” Nvidia card, which says Nvidia can place an additional card to 170 – 200 $ segment. Lets hope yields would be nice enough and price would drop sufficiently that releasing a new card at 200 $ price would be out of question. There is bucket that was filled by GTX 560 SE, and GTX 560. However retailers are getting rid of these cards already, and it’s unlikely theirs stocks would be here for long.
The GTX 660 would have grace period for about 2 weeks, but after that…
Jerome
A program like Unigine uses probably 200X more tessellation than any game would likely to use. That being said both 7970 and 680 can handle unigine at max settings no problems. I doubt crysis 3 would be any better on a 780 as far as tessellation in concerned.
hey
GK110 has 15 SMX and 2880 CC. So, GK110 is %87,5 more powerfull than GK104, as same clock speeds. This is theoretical difference between GK110 and GK104.
It must be generally %50-60 faster than GK104 in the games.
Pingback: AMD Strikes Back?
Pingback: Primeros datos de las Radeon HD 8870 y Radeon HD 8850 (Oland) | MadBoxpc.com
Pingback: AMD HD 8800 series detailed
skr13
According to 3Dcenter.org: 8870/8800CHIP »
2 Raster-Engines, 1792 Shader-Einheiten, 112 TMUs, 32 ROPs, 256 Bit DDR
For the price $280 it can be a winner card if performance reach 7000 flagship and NVIDIA 600.
Pingback: Spesifikasi VGA AMD Radeon seri HD 8800 Terkuak? | Jagat Review
Pingback: Primeros datos sobre las próximas AMD Radeon HD 8800
Pingback: AMD Radeon HD 8870 ve HD 8850 modellerinin özellikleri sızdırıldı – Cizgim – Oyun ve Teknoloji
Pingback: Radeon HD 8800 Oland XT et Pro : Nvidia a du souci à se faire ?
Pingback: AMD Radeon High definition 8000 йѓЁе€†и¦Џж јжµЃе‡є | indianakirwan
http://www.facebook.com/TheHeartOfAsiaa Pratik Rawankar
woww
everyday i read many comments on many sites but sometimes I didn’t read all of those comments
but I never forgot to read ur comment @BestJinjo:disqus
its nice HD 8970 will definetly breaks all performance boundaries and no chance for NV no matter they launch it late with Crysis 3 release
BestJinjo
Sorry, you have not taken into consideration that GTX680 uses about 180-190W of power under Peak Load at 1058mhz GPU boosted clocks:
http://www.techpowerup.com/reviews/Point_Of_View/GeForce_GTX_680_TGT_Ultra_4_GB/26.html
You cannot have a 15 SMX 2880 SP, 240 TMU, 48 ROPs 384-bit or 512-bit GK110 @ 1ghz clocks and stay under 250W of power consumption.
You are either looking at a severly underclocked 15 SMX part (like 850mhz GPU clocks), or a chip that has lower # of SMX clusters (say from die harvested failed GK110 parts that couldn’t be sold as K20 Tesla parts). Alternatively, NV may produce a custom-based GK1xx part for GTX780 series that we haven’t heard about (for example, since Kepler architecture is scalable, you can always add more SMX clusters and memory controllers but this may require a brand new chip such as GK1xx).
You cannot just increase the die size to nearly 520-600mm^2 on 28nm (and maintain the full double precision compute of the K20 chip) and not go back to Fermi GTX480 power consumption levels without some compromise. NVidia does not have access to alien 28nm technology at TSMC.
That tells you right there your estimates are way off since you are not assigning any penalty for GPU clock speeds or the dramatic increase in power consumption that would result from a full-blown 28nm 500mm^2 die chip, with the added double-precision compute (which has negatively affected the performance/watt of Tahiti XT as well). NV is not immune to a double-precision transistor power consumption penalty, which is why they carved out this feature from GK104 when they moved away from GF100/110 consumer designs.
Pingback: First AMD Radeon 8000 Series Details Emerge « News « FlyingSuicide.Net
Pingback: Primeros datos sobre AMD Radeon HD 8870 y 8850
Jerome
Agreed but it is very interesting I mean bandwidth is what feeds the GPU internally, Nvidia and AMD are always finding way to increase bandwidth with every new architecture, whether it be widening (expensive) or stacking possiblity (future). As far as gaming goes Kepler is incredibly efficient, in this regard.
Pingback: Rumores sobre las nuevas gráficas 8800 de AMD
BestJinjo
Personally, I can believe the specs but I am reluctant to believe price drops as well. If AMD manages a head-start launch for the next generation, I see no reason they won’t continue with the $299/349/449/549 price line-up. It’s hard to imagine if HD8870 actually does deliver near GTX680 level of performance that AMD would sell such a card for just $279 by January 2013.
Pingback: 2chに書き込まれたRadeon HD 8800のスペックがリーク情報として海外で拡散！ | 2chまとめのまとめ
Pingback: AMD Radeon HD 8800 series leaked | t-break: Tech @ Its Fastest
Pingback: ข่าวรั่ว!! รายละเอียด Radeon HD 8800 series รหัสชื่อ “Oland” ถูกเปิดเผย ? « HardwareData : ข่าวสาร Hardware,Hardware review และอื่นๆ
http://www.facebook.com/FreshPrinceOfLC Tyler Roemhildt
With OpenCL and HSA on the horizon, it makes sense to focus on graphics. NVidia will no doubt have some amazing cards as well, but probably more focused on the highest end and Tesla. AMD knows that if they can get their GPUs improved enough, the APUs will benefit as well.
SpycyMitaball
8870 is probably a better choice there. low costs, performance is probably better or at least around the 7950, and with the lower power consumption most systems could probably handle two of these :)
skr13
Yes, but that 8870 will take some months to be in the market, and right now i don´t have a decent graphics card, and really need a upgrade. I don´t like multi-gpu solutions, but 8850 for $200 can be a lower cost option CF.
Pingback: 8850/8870 rumors on the net ... would you buy 7870/7950 now? - Hardware Canucks
http://www.facebook.com/claudiu.peagu Claudiu Peagu
…one can dream…
Here’s hoping they’re EOL-ing HD7900s and can empty stocks soon… so it actually makes sense financially too, in this forsaken world we live our humbles lifes on.
http://www.facebook.com/claudiu.peagu Claudiu Peagu
Really? I don’t understand you people… It might make some sense if it’s that cheap and performs even above expectations… but really? Upgrade every gen? Especially now since I don’t see anything above PCI-E 3.0 and DX11.1/OGL 4.3 coming out anytime soon… not to mention the applications…
http://www.facebook.com/claudiu.peagu Claudiu Peagu
A 768 shader/192 bit 7790/7830 would have been nice…
http://www.facebook.com/claudiu.peagu Claudiu Peagu
We don’t know anything about GCN2 or just how good 28nm has got in the mean time, jeez… calm down…
skr13
If somebody is thinking about upgrade from 5000/6000 series, better wait for 8000, because is expected to bring good performance/watt ratio compared to 7000, ya some people upgrade every geneneration, but is good for the used market, where you can buy a cheap 7970/7950, maybe for half price.
mic08n64
If those specs are accurate, I believe there are are a few things Nvidia can do to stay competitive and not fall behind. Since Gk104 is in reality their midrange chip, they could rebrand the GTX 680 as the GTX 760Ti and a GTX 670 as a GTX760. Or, use a GK114 with higher clocks, same cuda counts, and increased energy efficiency to compete against the 7870 and 7850
BestJinjo
Thanks Pratik!
BestJinjo
AMD doens’t have proprietary tech. DirectCompute and OpenCL are industry open standards. Nothing is stopping NV from performing dynamic lighting, contact hardening shadows, post-processed AA, HDAO/ambient occlusion passes via directcompute shaders on the GK104. The problem is NV’s GK104 chip has static scheduler which makes it much slower for computational work
k2
Hey videocardz! The source is just a guess in a thread on 2ch.net (similar to 4chan), so it’s absolutely not a leak!
Many users of the thread is laughing at this article.
Pingback: Filtradas las posibles especificaciones de las AMD Radeon HD 8870 y HD 8850 | Team Hardware Venezuela
Jerome
All that means is that AMD drivers and crossfire profiles will be ready on the launch day for FarCry3
BestJinjo
I am hoping at least one of the new AMD Gaming Evolved titles (Far Cry 3, Tomb Raider, Bioshock Infinite) uses global illumination via Direct Compute. I think the industry is moving towards using more and more direct compute to enhance the lighting model in games.
Dave Baumann of AMD has recently commented at Hardware Canucks about the industry’s interest in the new Forward Lighting technique (as was already used in Dirt Showdown, Sniper Elite V2 and Sleeping Dogs). The main reason we have to use the inferior FXAA/MLAA (I hate how they blur textures/details in games) is because right now the lighting model is done using a Deferred Rendering game engine for lighting. With DirectCompute and Forward Lighting model, we can use MSAA without the massive 30-40% performance hit MSAA incurs in BF3 on modern GPUs today.
“DirectCompute is already being used by lots of titles and devs will adopt it more and more as new algorithms are developed using it. Likewise, when we initially demo’ed forward plus with Leo it garnered a lot of developer interested and experimentation because it is a rendering technique that has the efficiencies of Defferred Shading without some of the limitations.”
http://www.hardwarecanucks.com/forum/video-cards/56126-gtx-660-ti-review-comment-thread-15.html
Pingback: AMD-Gerüchte: Spezifikationen der Radeon HD 8850 und HD 8870
Pingback: AMD-Gerüchte: Spezifikationen der Radeon HD 8850 und HD 8870
Sergei
Dirt Showdown is a AMD title, and you know that Nvidia can´t use GPGPU well, so you expect that new games hit very hard Nvidia cards performance.
Pingback: Rumor: As especificações das próximas Radeon HD 8870 e 8850 » PCManias.com
Pingback: AMD Radeon HD 8870 and Radeon HD 8850 Specs Detailed
Pingback: Traccie di Amd HD 8000 nei driver Catalyst | Tecnologia Facile
Pingback: ข่าวรั่ว!! รายละเอียด Radeon HD 8800 series รหัสชื่อ "Oland" ถูกเปิดเผย ? - HardwareData
jon
this makes no sense. the 7970 ~= 680 and the 7950 ~= 670 depending on
the game and benchmark. so if your saying the 8870 is equal to the 680
then how is that a big performance increase? if thats true that would be
a HUGE fail on amds part. maybe if you mean the 8870 will equal the 690
then you have something.
http://www.anandtech.com/bench/Product/645?vs=598
http://www.anandtech.com/bench/Product/618?vs=555
mrhello
Looking forward to 8850 I can run it with my 450 watt psu.
Pingback: AMD Radeon HD 8000 Series Specifications
DnaAngel
Huh? Theres only 2, I repeat 2 AAA titles on market where the 7970 out performs the gtx 680 and in some cases even the 660ti/670, and those games are optimized for AMD’s arch.
DnaAngel
Your getting yourself confused with Architecture and Instruction sets. Modern games have yet to saturate a 256 bus, therefore having the extra depth (width, bit) makes little to no differences Much like the PCI-express x16 slots have yet to be saturated. Even with PCI-E gen 3 boards like my ASUS ROG Maximus V Formula the only time me or other end users will see a difference is when running 3-4cards in Sli/Crossfire and the 3rd/4th cards are getting the benefit of the bandwidth expansion of gen 3 pci-e, instead of running in stock PCI-e 4x mode (gen 2)
As I posted earlier the only thing that is heavily dependent on bus depth is when you are cranking AF/AA to the max, which for most people with normal sized 1080p monitors will see no real visual difference over lower AA settings like 2x/4x. This is even more so on the new UHD monitors with 2500k res.
You will benefit from the higher bus width when running multiple monitors IF you are only using a single card. Most people that do run 2-4 monitors will be using 2-4 gfx cards anyway thus making the added bus width relevant at that point.
DnaAngel
Bandwidth is what feeds the internal Vram yes. much like the memory controller on your CPU or Motherboard, That controller is not what controls the overall performance of the card/cpu/motherboard. There are other factors. The problem is things get over-hyped in media and community’s and people start thinking that things work a certain way for a certain reason. You need to brush up on your knowledge on what the Bandwidth does and what it controls. Other then power Users bandwidth is nothing more then another marketing ploy, a simple google search will tell you all you need to know, im sure you can do that much ;)
DnaAngel
Lol? you need to go read up on some benchmarks dude. your info is way off
Pingback: Rumor: Xbox 720 Silicon Production Points To 2013 Release Date?

Emtek Launches Five XENON GTX 650 and GTX 660 Graphics Cards PowerColor Launches Three Radeon HD 7750s With GDDR3 Memory

VideoCardz.com

AMD Radeon HD 8870 and HD 8850 Specifiation Leaked

Follow us

Tag Cloud