Official AMD Radeon HD 6850/6870 Performance Specs

Architecture, specifications, and design highlights for the AMD Radeon HD 6850 and Radeon HD 6870 – this is what everyone has been waiting for. While many sites are publishing exhaustive dissertations spanning many pages, we’re looking to quickly bring out the essentials of what you need to know. The time for guesswork regarding AMD’s 6000 series graphics cards has come to an end and benchmarks are publishing at many sites as you read this.  In the absence of solid facts alongside a plethora of supposed “leaks”, the buildup to the release of the AMD 6000 series was fed from a consistent diet of rumors, guesswork, and imagination. Sites and communities have lit up with discussion regarding what to expect and with few reliable facts to work with the discussion has a way of turning rabid. As much as we all enjoy getting fired up and trying to figure out whats coming next, everyone welcomes solid details and specifics – its time we take the rumors out behind the shed and give them the “Old Yeller” treatment. Unlike the end of the Disney movie however, I doubt any of us will shed a tear laying these rumors to rest! Fortunately I gained some first hand insight in talking with AMD in LA a week ago, and I’ve been burning to bring this information out to our readers. It’s time to pull the trigger so let’s get started.

6850/6870 Quick Summary

AMD Radeon 6800 Series Architecture Highlights

AMD Radeon 6800 Series Architecture Highlights

Performance per Dollar

Performance per Dollar

The top AMD slide above is a quick-summary confirmation of what we are dealing with in the Radeon 6800 series – generational refinements upon the 5000 series architecture. The second slide is important in how it frames the 6800 series – looking at the X axis we have “Compute Power/Launch Price” against “Performance per $ in GFLOPS”. The positioning of the Radeon HD 6850 and Radeon HD 6870 in the discrete graphics market is a primary factor targeting the sweet spot of the market around $200. The focus these pictures present is on communicating the improvements and work that went into refining the architecture while reducing the cost for the consumer, but let’s break down these details.  For starters, its important to understand that the first statistics presented are comparing AMD’s 5850 against the 6870.  “More performance per mm²” and “over 35% better than previous generation” begs qualification so let’s take a look at the math and ensure we are setting expectations correctly. AMD used 3dmark Vantage Extreme to benchmark performance on the following testbed:

  • AMD Phenom II 1090t (3.2GHz) processor
  • Gigabyte 890FXA-UD5 motherboard
  • 8GB DDR3 memory at 1600MHz (timings: 9-9-9-24)
  • Windows 7 RTM 64bit
  • Driver version 8.782RC1

In this test, the 5850 scored X7403 compared to the 6870 which scored X7730. That doesn’t sound like much of a difference and there is good reason for that – there is only a moderate improvement in performance on this benchmark with the 5850 vs 6870. Underwhelmed? Try not to get ahead of me here! As mentioned previously, this slide is demonstrating architectural refinements so lets look closer at the size of these chips. The 5850 comes in at 336 mm² while the 6870 shrinks things down to 255 mm². So while the performance improvement of the 6870 is moderate in 3dMark Vantage Extreme, the performance per mm² is rather remarkable:

  • Radeon HD 5850 – 7403 (performance)/336 (mm²) = 22.03
  • Radeon HD 6870 - 7730 (performance)/255 (mm²) = 30.31

Crunching those numbers, we see that actually “perf per mm²” is 8.28 points better on the Radeon HD 6870 which comes out to a 37.5% improvement that AMD rounded down to a nice even 35% for the slide.  Similar but better performance from fairly remarkable architectural refinements… Everyone loves a good car analogy right?! You can use a car’s engine compartment as a metaphor – your engine design needs to be more efficient and perhaps a bit more elegant to do more with less. I spent a bit of time focussed on this single statistic and thats because it strikes to the core of the character of these cards – optimizations doing more with less.

What More Are We Getting?

2nd generation DirectX11 design brings faster tessellation and geometry throughput.

AMD reports twice the tessellation performance with the 6870 compared to the 5870 based on “AMD internal synthetic testing”. I don’t know about you but the first thing that makes me think is that I’ll need to see some benchmarks – however its important to note that tessellation performance is very sensitive to the number of pixels per polygon.

We’ve all seen that NVIDIA favors the Unigine Heaven benchmark in promoting the Fermi architecture, and this is related to the level of tessellation used in the benchmark. This is relevant because the Fermi architecture is particularly strong in floating point performance which is great for tessellation however floating point performance is considerably less important in real world gaming. Bringing this line of thought full circle draws out the fact that you have to be careful when considering synthetic benchmarks – they do not always reflect the level of performance you can expect while gaming.

For this reason, AMD took some extra time explaining that the sweet spot in their perspective is aiming for around 16 pixels per polygon to ensure the best image quality with efficient performance – too many pixels per polygon and performance can suffer, too few and image quality deteriorates. Fortunately for AMD, with being the first to market with DX11 support in the 5000 series a lot of game developers designing for DX11 built using AMD hardware. AMD was the only DX11 game in town for a solid 6 months and its reasonable to think this lended some influence to AMD regarding design decisions throughout the development process while they worked with developers to optimize performance on their hardware.

Why did I take the time to explain that? It’s a point AMD spent some time on, and the message it is looking to demonstrate is that in actual game performance it can be very unlikely for you to see the stark performance contrast you can find in a synthetic benchmark.  In reading on Overclockers Forums, I recently read a forum post which I’ll loosely paraphrase – “There are lies, damn lies, statistics, then benchmarks”. Sometimes that can hold pretty true, so as potential buyers who want to be happy with our purchases, its important we take a well measured evaluation when looking at benchmark results to ensure they are setting our real world performance expectations accurately. If a particular benchmark draws an unexpected outstanding result that contrasts with results in many other real world benchmarks, its likely a particular feature of that benchmark rather than a permeating advantage of the card.

Enhanced Image Quality

New anti-aliasing modes and better anisotropic filtering bring better image quality to the table compared to the 5000 series.

Looking at AA we have something called “Morphological AA” which can be enabled in Catalyst Control Center. This brings full scene AA accelerated with DirectCompute that processes more elements in a way thats faster which improves image quality while offering more performance than other AA levels. It works with any game in DirectX 9, 10, or 11. Anisotropic filtering has also had its algorithm improved reducing visible filtering imperfections.

The CCC has also had user interface improvements made to the way you can control texture filtering and Catalyst AI.

What Less Are We Working With?

Less GPU area at 255 mm², less power draw, and less money out of pocket with both cards coming in below $250.  The following images summarize the exact technical specifications, and highlight how AMD has managed to tighten up the 6800 series yet deliver some solid performance for the price.

6850 Vitals

6850 Vitals

6870 Vitals

6870 Vitals

6800 Architecture Features

6800 Architecture Features

6870 Compared

6870 Compared

6850 vs GTX 460 768MB

6850 vs GTX 460 768MB

6870 vs GTX 460 1GB

6870 vs GTX 460 1GB

Bottom Line: Moderate Performance Improvements, Competitive Pricing

This has been a bit short and sweet with the Overclockers.com review sample still bouncing through the postal system. If you haven’t picked up on the chorus throughout this article this final heading puts it in a nutshell for what you can expect to find once we publish our review, or as you read elsewhere around the net.  This introduction to the 6000 series clearly has a heavy focus on price/performance targetting the moderately priced gaming GPU segment. Keep in mind however that AMD has expressed strongly that they’ve been listening to what gamers want, and they are focused on delivering it. While these cards address the “sweet spot” and break the lid off the AMD Radeon 6000 series, if you find yourself still waiting to have your hair blown back there are more good things on the horizon.

Stay tuned, we’ll have our own hands-on review up soon, and in the meantime we’ve already published a round up of a few other great reviews while you wait!

Leave a Reply

Your email address will not be published. Required fields are marked *

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>

Discussion
  1. I.M.O.G.
    I dunno, 5870 is still worth something now if you were to sell and upgrade in the near future. I've heard people saying in response to the reviews that these new cards are scaling better in crossfire too, but I haven't looked that close at CF testing first hand. May be something you want to consider looking into.

    A lot of people go CF to drag out the longevity of their cards tho, so your plan may be good. But once they aren't that great anymore as they get older, they really won't have any value on the secondary market. Its a game of give and take any way you look at it I guess.


    +1 :thup:

    Always consider the quick trade up to a better single card solution. Especially if you have a decent extra card on the side that can hold you over if you need to sell first before upgrading.
    Loo
    This is why I love my 5870... It still does not fall into middle class with performance. I'll stick with my old idea, get another one 5870 for CF and enjoy that for 2011.


    I dunno, 5870 is still worth something now if you were to sell and upgrade in the near future. I've heard people saying in response to the reviews that these new cards are scaling better in crossfire too, but I haven't looked that close at CF testing first hand. May be something you want to consider looking into.

    A lot of people go CF to drag out the longevity of their cards tho, so your plan may be good. But once they aren't that great anymore as they get older, they really won't have any value on the secondary market. Its a game of give and take any way you look at it I guess.
    This is why I love my 5870... It still does not fall into middle class with performance. I'll stick with my old idea, get another one 5870 for CF and enjoy that for 2011.
    deathman20
    8W? wow would of expected what 28W from a 18% increase in clock not a 5% increase (mind you off the base TDP of 160W)


    I would have thought it would be higher too, but playing a very demanding Crysis mod, I only see ~230 at the wall. It's about 222 watts with stock GPU clocks, and ~130 gaming on the IGP.

    By comparison, my old GTX 280 showed 330 at the wall playing the same mod with everything else in the system the same.

    Anyway :p
    ratbuddy
    Not enough to require more power connectors. My stock 715mhz 460 is running 850 on 1.0v vs 0.987v stock and uses ~8 more watts at the wall than it does with stock clocks and volts.


    8W? wow would of expected what 28W from a 18% increase in clock not a 5% increase (mind you off the base TDP of 160W)
    CharlieCS
    And I expect a 850mhz gtx460 to consume more power as well.


    Not enough to require more power connectors. My stock 715mhz 460 is running 850 on 1.0v vs 0.987v stock and uses ~8 more watts at the wall than it does with stock clocks and volts.
    There are plenty of overclocked reviews.You really haven't looked at many besides guru3d it seems. Issues with some websites is that AMD used different VRM this time Chill instead of Volterra and software like MSI afterburner just doesn't support new voltage controller ,after it gets an update am sure 1GHz will be easy with some Vcore increase and these should reach 1GHz more consistently then say 5870 but 1050mhz is most I see these do 24/7 ,which is impressive considering it's still the same 40nm process.

    Of course both 6850 and GTX460 have more headroom,but none of them will reach 6870 clocks :) And I expect a 850mhz gtx460 to consume more power as well.
    ratbuddy


    For now, I'm wondering why so many reviewers aren't OCing the 6870s. Maybe it was one of the conditions to be sent an early card..


    Not as far as I know. We were supposed to have one pre-release, but there were extenuating circumstances that led to some not getting them on time. Ours is reported as on the way and there is no such restriction applied.
    Nathan54AB
    This isn't making sense. 3870 < 4870 < 5870 right? Ok, so why isn't 5870 < 6870? Also, isn't the 5850 more powerful than the 5770? If that's the case, why would the 5770 be close to a 6950 and 5850 be close to 6870? :sly:


    Check again, I pm'd him and he mentioned he got his numbers mixed up. It's corrected now - just a friday and a long day of work, it tends to do that to people. :beer:

    Depends what you are looking at. For the most part, its going to trade punches.
    EarthDog
    Whats the point of reviewing it like that (460 vs 6870 or even 6850)? From the already published reviews, it spanks it across the board. The only difference would be overclocking headroom (which I believe what you may be getting at) and the 6870 being clocked so high, it may not have much headroom.


    That's exactly what I'm getting at. Last I looked, only one review published results for 6870 OCing and it was pretty dismal, 954 core vs 900 stock IIRC. If they're all clocking that bad, suddenly a $170ish 460 1GB doesn't look so bad considering they pretty much all do at least 850 vs 675 stock. Anand included an EVGA FTW version (850mhz core) in their review and it beats or matches the 6870 and GTX 470 in most (all?) games.

    That's the real story most reviews are missing. Nvidia for whatever reason released the 460 pretty much underclocked by default. Throw in a full 384 shader version (bet the GF104 yields are good enough to do it today, if they wanted) and clock it up to 750 or so by default and the game changes. Of course, they'd completely gut GTX 470 sales, and that wouldn't be good for business. I think we'll see such a card as soon as 470 stock is gone, probably priced to match the 6870.

    For now, I'm wondering why so many reviewers aren't OCing the 6870s. Maybe it was one of the conditions to be sent an early card..
    doz
    Because the 6850/6870 are the successors to the 5750/5770. There is going to be a 6950/6970 that will replace the current 5850/5870.

    In other words, if you are looking to buy a 5850, you should just get the 6870 as they are about the same price but better performance/newer card. Same w/ 5770, just buy a 6950 (which will edge out the 5770 as the 6950 is more onpar w/ a 5830).


    This isn't making sense. 3870 < 4870 < 5870 right? Ok, so why isn't 5870 < 6870? Also, isn't the 5850 more powerful than the 5770? If that's the case, why would the 5770 be close to a 6950 and 5850 be close to 6870? :sly:

    That makes more sense. So a 6870 will probably outclass a 5870?
    And to wait and see what the full size cores will do, I can't wait. If the shaders are really powering it that much we'd see roughly a 42% increase in oomph. Might just have to wait and see how the new ones fair before jumping on this... safer then sorry ;) besides its not like I said I was going to get a 5850 earlier this year and still sitting on my GTX285.
    Nathan54AB
    Am I the only one whose noticed that specs on the 6850/6870 are actually lower than that of the 5850/5870?


    Because the 6850/6870 are the successors to the 5750/5770. There is going to be a 6950/6970 that will replace the current 5850/5870.

    In other words, if you are looking to buy a 5850, you should just get the 6870 as they are about the same price but better performance/newer card. Same w/ 5770, just buy a 6850 (which will edge out the 5770 as the 6850 is more onpar w/ a 5830).
    Well, thats the optimizations and part of the whole shebang with "doing more with less".

    Some specs are higher, but ya, a lot of the performance comes from some design optimizations of the architecture.
    It looks like the 6870 and the GTX460 1GB have almost the EXACT same specs, except for the stock clocks. so, I would bet you have to get the GTX460 1GB to the same clocks as the 6870 for them to be roughly equal in performance. Which I doubt could happen, especially once the 6870 is OCed. This assumption is based on the specs, not actual benchmarks though...

    HD6870

    Max Power Draw: 151 W

    Noise Level: Moderate

    Framebuffer: 1024 MB

    Memory Type: GDDR5

    Memory Bus Type: 64x4 (256 bit)

    DirectX Compliance: 11.0

    OpenGL Compliance: 3.2

    PS/VS Version: 5.0/5.0

    Process: 40 nm

    Fragment Pipelines: 1120

    Vertex Pipelines: 0

    Texture Units: 56

    Raster Operators 32

    GTX460 1GB

    Max Power Draw: 160 W

    Noise Level: Quiet

    Framebuffer: 1024 MB

    Memory Type: GDDR5

    Memory Bus Type: 64x4 (256 bit)

    DirectX Compliance: 11.0

    OpenGL Compliance: 3.2

    PS/VS Version: 5.0/5.0

    Process: 40 nm

    Fragment Pipelines: 336

    Vertex Pipelines: 336

    Texture Units: 56

    Raster Operators 32

    "At the same clocks" Comparison

    HD6870 GTX460 1GB

    Core Clock: 900 MHz 900 MHz

    Memory Clock: 2100 MHz (4200 DDR) 2100 MHz (4200 DDR)

    Memory Bandwidth: 134.4 GB/sec 134.4 GB/sec

    Shader Operations: 1008000 MOperations/sec 302400 MOperations/sec

    Pixel Fill Rate: 28800 MPixels/sec 28800 MPixels/sec

    Texture Fill Rate: 50400 MTexels/sec 50400 MTexels/sec

    Vertex Operations: 0 MVertices/sec 75600 MVertices/sec
    Whats the point of reviewing it like that (460 vs 6870 or even 6850)? From the already published reviews, it spanks it across the board. The only difference would be overclocking headroom (which I believe what you may be getting at) and the 6870 being clocked so high, it may not have much headroom. A better comparison, b/c of performance and pricing (now) would be the 470 and overclock those?