Mobile Radeon HD 8000M GPUs announced, shipping now

By Rick · 17 replies
Dec 18, 2012
Post New Reply
  1. AMD has extended its line-up of mobile GPUs with a few brand-new models filed under its latest class of Radeon: the HD 8000M-series. The new chips deliver better performance, but are expected to do so with reduced power consumption, lower...

    Read more
  2. amstech

    amstech IT Overlord Posts: 1,936   +1,101

    Looks like AMD is really taking ther GPU architecture seriously, and they should. With all the mishaps and layoffs in other parts of thier company they need another strong year or two from thier Radeons/GPU & APU branch. They have really challenged Nvidia in the last year or so, much more then I have seen in the last 5 years.
  3. mevans336

    mevans336 TS Enthusiast Posts: 161   +11

    GCN is serving AMD well so far. It's a strong architecture and once AMD works out the driver oddities, will really give Nvidia something to worry about.
  4. veLa

    veLa TS Evangelist Posts: 781   +235

    The CPU division could really learn from the GPU division
  5. Blue Falcon

    Blue Falcon TS Addict Posts: 161   +51

    All these GPUs look very weak though. When AMD launched them they targeted GTX600 mobile parts but NV will unveil GTX700 mobile parts in 2013 as well. Looking at the specs of 8500-8870M, they are very underpowered parts. AMD is comparing 8870M to 650M as I presume they are comparing them based on price? Perhaps AMD is going to price HD8000M much lower than HD7000 then because there are only 2 models above HD8870 (HD8950/8970M) and I am guessing AMD will put those up against GTX675MX and GTX680MX. Those won't come until Q2 2013 as they are based on Sea Islands. Can't help but think the mid-range 8870M is very underpowered though since it's essentially a desktop HD7770 with 30% lower clocks.
  6. EEatGDL

    EEatGDL TS Evangelist Posts: 568   +241

    veLa They actually did and it was a fail; they used the architecture design methods of the GPU division for the Bulldozer [I'm not trolling, it's true]; they "automated" several parts of the CPU design, not "by hand".
  7. dividebyzero

    dividebyzero trainee n00b Posts: 4,891   +1,264

    Ugly naming conventions even for mobile parts which are known for their "pick a number out of a hat" MO

    HD 7700M....512 core...........HD 8700M....384 core......both GCN......max core 675MHz v 850MHz
    HD 7800M....640 core...........HD 8800M....640 core......both GCN......max core 800MHz v 700MHz

    In fact, they don't look appreciably any better than the models they are ostensibly replacing
    HD 7000M parts are already priced much lower than Nvidia's 600 series (many of which are Fermi based) - Hasn't stopped OEM's and consumers shopping for Nvidia's parts...which comes down to marketing, a lack of high ASP pro parts (last mobile FirePro card was two years ago- the M8900), and in large part, drivers -see AMD's enduring Enduro woes as example.
    AMD is playing catch up wrt Kepler (perf/watt - which has greater weight in the mobile sector than desktop). I can't see that changing with this series- both vendors are locked in to their respective architectures on the same process node. Either way it comes down to sacrifices of power for speed and die size/yield/profit for compute power. What we are seeing now is how the vendors jumped from 3+ years ago.
    My guess is that these parts were accelerated into production mainly because GCN 2.0 is needed for Trinity's replacement (Richland), and of course, hybrid Crossfire with those parts.
  8. Arris

    Arris TS Evangelist Posts: 4,730   +379

    I have to say if it hadn't been for the problems with the 7970M drivers/Enduro MUX graphics switching I'd probably have opted for that over the GTX680, the performance per £/$/? is far in AMDs favor. Not sure how great this latest offering will compare to Nvidia's top mobile GPU.
  9. Blue Falcon

    Blue Falcon TS Addict Posts: 161   +51

    Ya but that's why I said the marketing is misleading/confusing. The parts you listed are not replacements like that.

    HD7670M is replaced by HD8770M at the same price. So it's not 7870M --> HD8870M. The latter is priced lower and is slower in performance. Similarly 7590M is replaced by 8690M, and 8870M is priced at the same price as Nvidia's 650M.

    The parts below HD7700M are VLIW and are all replaced by GCN HD8500-8800M. So the performance increase will be 20-50% across the same price levels:

    My beef is there are no replacements for HD7870M-HD7970M. Because AMD only has 2 SKUs scheduled above HD8870M, that means HD8950-8970M will be 2-2.5x faster than HD8870M. That's stupid as they will have no mid-range GPU in the mobile sector. They really should have called HD8870M as HD8770M and introduced some mid-range product. Looks like they are going after NV in the lowest price segments. I suppose it makes sense since that's where the volume is but for actual gamers looking for a gaming laptop, they are going to be waiting another 4-6 months.
  10. Blue Falcon

    Blue Falcon TS Addict Posts: 161   +51

    "AMD is playing catch up wrt Kepler (perf/watt - which has greater weight in the mobile sector than desktop). I can't see that changing with this series- both vendors are locked in to their respective architectures on the same process node."

    Ya to clarify this is not true. AMD's sub-HD7700M products were still made on 40nm VLIW. That's why Kepler beats them easily in performance/watt and features. HD8500-8870M are going to play a major role in making AMD's mobile GPUs more competitive since they introduce 28nm, DX11.1, upgrade UVD and ZeroCore power to the same price levels previously occupied by Northern Islands HD7000 parts (this is because low-end HD7000 mobile parts were rebadged HD 6000 series).
  11. dividebyzero

    dividebyzero trainee n00b Posts: 4,891   +1,264

    Don't care what the price point replacement is tbh. I was merely pointing out that that the numerical naming of the GCN parts doesn't engender confidence. Lower core count in the 8700's vs 7700's against a slightly higher freq, and equal core count in the 8800's vs 7800's with lower frequencies. Both these series are the 3rd and 2nd tier SKU's of their generation. Price v Price is generally a moving target. Performance within the product stack is not.
    Sub-8700's I have little interest in when it comes to comparison because you're dealing with two different architectures on two different process nodes. Moreover, the core freq/memory speed/memory type for each 8500/8600 SKU hasn't been published, whereas the 8700/8800 has.
    Like what????? AMD have a mainstream chip (Mars: the Cape Verde/Chelsea replacement) and a performance chip (Neptune: the Pitcairn/Heathrow replacement) that doesn't leave a whole lot of options. If you drop Neptune to 8600/8700 series you have two options, neither of which is realistic:
    1. You overclock the snot out of Neptune to become the 8800 series - Goodbye standard cooling, hello power consumption
    2. You move Sun XT (HD 8970M) down to the HD 8800M series, which means you downclock and neuter what will likely be a 400mm^2/5.1bn tranny Tahiti replacement for the HD 8970M. A quick look at what you're dealing with:
    Tahiti LE : 1536 shaders (67% enabled of full spec)....for an 8.7% increase in performance over Pitcairn for a 52% increase in power consumption(175w vs 115w). The numbers are likely to work out pretty much similar for a 8900 LE (2048 shaders active from 2560) vs Oland XT (8870), and that is disregarding the fact that the 8900 series GPU might not eventuate until well into Q2.

    Whichever way you slice it, you're still faced with the facts that:
    1. AMD's strength lies in the middle of the product stack for performance per watt, and
    2. Nvidia can fit their top tier GPU into an MXM package, where AMD cannot. Unless you believe that AMD can add 25% to the core count whilst simultaneously 30-40% of the die size then that situation isn't likely to change- which is borne out by the fact that there is no mention of any mobile part of higher performance than the HD 8870M excepting one- the HD 8970M (Sun XT based on Oland, the Pitcairn successor)
    To clarify (once again) I was referring to the two series (8700 and 8800) in my post, and secondly, how do you know it is not true? The majority of these low end mobile GPUs are going to 64-bit DDR3 P'sOS, and margins and ASP's are razor thin in general at this price point if its a bare bones MXM package who has the inside running- possibly the vendor churning out the most boards? Cape Verde is 123mm^2 and GK107 is 118mm^2. From the numbers I've seen, AMD are shrinking down to 77mm^2 at the expense of 4 TMU's/16 shaders, and GK107 has a better ROP/TMU/Core balance and a wider (128-bit) memory bus, so don't be too sure that AMD have a slam dunk on their hands. The HD 8600's might be the better proposition -depending upon what GK117 brings to the table.
  12. Blue Falcon

    Blue Falcon TS Addict Posts: 161   +51


    "Don't care what the price point replacement is tbh."

    I just told you that you are comparing the wrong parts and you ignored the entire point why price is a factor. You don't compare HD7870M with HD8870M like you did. The only way to make a proper comparison when the brand names are not logically replaced is to look at price at which AMD is replacing the older HD7000 VLIW parts. You can blame AMD's marketing but your comparison you listed is flat out incorrect. It's like the time when AMD replaced HD5850 with the slower HD6850. HD6850 was not a replacement for HD5850 even though it had a higher #.

    You spent SO much time with some technical details like die sizes and bus width and none of it is relevant to what I said.

    You misinterpreted my comment when I said "Ya to clarify it's not true." You said:

    "I can't see that changing with this series- both vendors are locked in to their respective architectures on the same process node."

    ^ This only applies to higher-end HD7700-7900 series. Did you forget that the volume sales are low-end to lower mid-range parts? The highest volume selling parts are lower end ones and as I said but you ignored, AMD's sub-HD7700M products were still made on 40nm VLIW. So the performance increase will be huge for AMD since they are moving from 40nm to 28nm.

    Here is HD7690M vs. HD8690M. At the same price, AMD is giving roughly 60% more GPU performance from the review and the new 28nm part uses just 2W of power more at load to get there:

    If you are only care about flagship mobile GPU performance, then no one is disagreeing with you that NV will continue to be in the lead with GTX780M(X). These parts they are launching now are all low-end to mid-range and my guess is all of them will be faster than NV's outgoing GTX600 series at the same prices. This is why AMD positions HD8870M against 650M, meaning you'll be able to get a laptop for the same price as a 650M with way more performance.

    Right now the HD7970M is the 3rd fastest single mobile GPU after GTX680M/MX:

    NV actually charges more $ for the much slower GTX675MX than HD7970M:

    The problem is HD7970M is trading blows with GTX680M, not 675MX

    And NV charged $350 more for GTX680M over HD7970M:

    At that point you can almost get HD7970M Cross-fire in a laptop. So sure, if you just want to focus on who has the best/fastest mobile GPU, NV certainly does but they charge an arm and a leg for a 25% performance increase on the 680MX which on price goes against HD7970 Crossfire. That's not even looking at 680M vs. HD7970M where the former is insanely overpriced.
  13. dividebyzero

    dividebyzero trainee n00b Posts: 4,891   +1,264

    Incorrect. AMD moved the whole product stack of the 6000 series (some rebrands aside) up one series from the 3800's - 5800's, so the top tier went from x8xx to x9xx. This was publicized by AMD well before launch. AMD are keeping the same naming convention for the upcoming cards as for the previous bad analogy on your part.
    I don't need to ignore it, the answer is staring everyone, except you it seems, in the face
    Yeah, that low ASP/low profit line per unit is reaaaallly working for AMD- bit of a head scratcher on why it hasn't been adopted by Nvidia, Intel and Apple.
    Nope, but the only full specifications are for the upper tier cards. Do you know what 8500 and 8600 cards are DDR3? which are 64-bit ? which are 64-bit AND DDR3?. No, and nor does anyone else outside of AMD. I keep the speculative analysis to a likely specification, whereas you have a Walter Mitty viewpoint on all things AMD.
    Couple of points...1.ALL of Nvidia's low end mobile GPU's are Fermi based (610-635), and also the high volume mainstream GTX 670M/675M (rebadged 570M/580M). First of all you complain about metrics used that compare VLIW parts with GCN, then go right ahead and compare an UNRELEASED GCN against a two-generation old Nvidia range.
    Awesome! AMD should be golden if they ask Nvidia not to release the GK11x/20x around the same time as the 8000 series. You do realize that Nvidia had 28nm mobile GPUs launched before AMD for the current generation? And now you're betting that Nvidia sits around twiddling its thumbs while AMD supposedly run amok through the mobile landscape. That Sunnyvale Kool Aid must be very refreshing...pity the reviving qualities don't last through a quarterly earnings call.
    Here's some other tidbits about the glorious HD 7970M, since you seem to have a blind spot
    and of course, the all important benchmarks...
  14. Blue Falcon

    Blue Falcon TS Addict Posts: 161   +51

    I seriously don't understand what you are actually trying to explain with all your posts. Time after time, you constantly revert to either attacking AMD products or desperately defending NV, and thus the posters who say positive things about AMD products and god forbid anything negative about NV. You then turn everything anyone says about AMD as "product shilling" or blame someone as AMD biased right away.

    I never defended the comparison of VLIW to GCN. I simply pointed out your ignorance that there will be a 25-50% increase in performance/$ because AMD is moving from VLIW to GCN. This has nothing to do with NV although you want to keep talking about them at any opportunity it seems. I never once said AMD will for sure take market share from NV in 2013. I never once said AMD products are better on the high-end in the mobile sector. My point is you dissed the crap out of HD8000M and you said they are still suck on the same node and I proved you wrong on both of those accounts. AMD is increasing performance by 25-50% and is moving from 40nm to 28nm with these parts. They will be released in January. It's not unusual for AMD to launch product 3-4 weeks before actual launch (HD7000 series was done the same way).

    Regarding high end parts, we have no information on HD8900M series so how can you say they are going to be crap when you haven't even seen their specs? Regarding the new 28nm HD8000M low to low-mid-range GCN parts, AMD is positioning them against NV's GPUs from 650M and below. It doesn't matter if they are Fermi or Kepler parts - actually NV mixes and matches 28nm and 40nm all over the place in that price range. Consumers want good power consumption and performance and HD8000M once it launches in January will be better than competing HD7000M/GTX600M parts at their respective price levels. Again, this doesn't mean AMD will take away market share since they still need to win OEM design wins. This is no different than AMD delivering superior performance/$ and superior performance for single GPUs in the desktop market this generation but they lost market share regardless. Enthusiasts/gamers don't care about market share. Only fanboys and shareholders do to tout some inherent superiority in their choices. No different than people who own a Honda Civic/Accord or Camry/Corolla claiming they are good cars because they are best selling. It's non-sense for people who are enthusiasts in the car of GPU industry.

    No doubt about it that GTX700M could end up much better but we are talking about HD8000M improving performance/watt and price/performance launching next month, not in 5-6 months. Instead, you completely downplayed the move from 40nm to 28nm and instead started focusing on AMD's margins/profits and market share out of nowhere like a desperate shareholder of NV. Are you a gamer or a shareholder? I can't decide. Consumers want faster GPUs for less $.

    Why do you always spin everything I say instead of reading exactly what I state. You have a tendency to focus on useless metrics like financials, market share or high-end mobile GPU performance, why, you don't like gamer and enthusiast specific metrics like overclocking, performance/$, performance/watt and absolute performance? Ironically 3 of those 4 NV lost this round.

    BTW, both AMD and NV use DDR3 for their parts. I never even defended the idea of mixing and matching 64-128bit buses or mixing DDR3 and GDDR5. You automatically assumed I am oblivious to these marketing gimmicks and I am frankly not fond of them at all despite you painting an image of me as I am pro-anything AMD. That's a bunch of bs.

    Funny how you attacked AMD for its horrible strategy of mixing bus width and DDR3/GDDR5 but you completely didn't mention that NV is even worse. Not only do they mix DDR3 / GDDR5 but they mix 40nm and 28nm under the SAME codenames. The consumer has no chance to find out which part they are getting.

    Also, you mentioned that GTX670/675M are volume parts? Do you have evidence for that? Not that this matters since right now AMD is improving the performance of low and low-mid-range product lineup. HD7970M already creams GTX670/675M beause it beats both for the same price. Only a fanboy would buy a GTX670/675M instead of the 7970M.
  15. Blue Falcon

    Blue Falcon TS Addict Posts: 161   +51

    Regarding benchmarks, you seem to continue to ignore price. How convenient.

    Alienware 17x with HD7970M costs $50 less than the same laptop with GTX675M:

    Problem is 675M is destroyed by HD7970M in gaming performance:

    Like I said, if you want the best GTX680M/680MX are the ticket but they'll cost $350+ more for 20-25% more performance on average than 7970M. Like I said only a fanboy would buy a GTX670/675M GPU as it's often 50-60% slower than HD7970M in modern games on Ultra settings.

    Regarding Enduro vs. Optimus, Enduro is much worse but who is lugging a 9-12 lbs gaming laptop around and cares about 15 min more of battery life? Most people use it as a desktop replacement at home plugged in or when they travel / in hotels. In those cases, you plug the laptop into the outlet and get full performance.

    And honestly next time you want to discuss GPU market share, at least come off as knowledgeable. A very large reason AMD lost so much market share suddenly is not because their products was much worse but because they couldn't afford to spend a lot of $ on winning OEM design wins with HD7000. This is contrasted with NV who spent millions more to get 300 design wins with Kepler because they set this as a goal after Fermi.

    "The policy of cutting implementation and other costs has reduced the company?s operating expenses from circa $610 million to about $450 million per quarter this year. For a struggling company, $160 million in cash is a significant amount of money.

    But reduction manufacturing costs in many ways causes market share decrease. Many criticized Nvidia Corp. for pumped up OpEx due to implementation costs and other manufacturing-related charges that the company faced during the Kepler GPU family ramp up. As the time has shown, Nvidia is now the No. 1 supplier of notebook GPUs (based on data from Mercury Research provided by Nvidia) because of AMD?s reluctance to help integrate its Radeon Mobility products based on the recent architecture."

    You don't just lose 20-25% mobile dGPU market share on technical metrics alone. Other factors are at play. If hypothetically AMD priced their GPUs at 50% less and they are 2x faster, it still doesn't mean they'll get OEM design wins. To get those design wins you need to be willing to spend $ on custom designs and be willing to work together with OEMs. It's not as simple as delivering a good performance/$ and performance/watt part only.

    I hope you are saving up $ for GTX700 series by buying NV stock while it's under $13. The "biased" AMD owners are going to have bitcoin mining $ from HD7900 generation and will once again be upgrading to free GPUs in 2013. Happy Holidays!!
  16. Blue Falcon

    Blue Falcon TS Addict Posts: 161   +51

    Forgot to say, those AnandTech benchmarks you posted are with Enduro. Enduro is worse than Optimus but any form of switchable graphics will cause a loss in performance. That's what separates enthusiasts from people who fall into the marketing traps. People who spend nearly $2000 on a gaming laptop want to use the full potential of the their GPU, because that is what they paid paid for. That means the best gaming performance is plugged into an outlet with Enduro disabled to begin with. I already showed you what happens when you have the laptop plugged in without Enduro but you didn't read the review carefully.

    HD7970M keeps up with GTX680M for less cost than a 384 shader GTX675M. But eh if you think 15 min of battery life saved is worth it to get a GPU 50% slower in games, or spending $350+ more for a 680M is worth it for that, by all means.
  17. dividebyzero

    dividebyzero trainee n00b Posts: 4,891   +1,264

    Maybe you should just read the substance of my posting rather than infer. But like most trolls you are only in posting to talk at people. You have no interest in a dialogue except that it allows you to segue into your own diatribe.
    My post commented upon the naming convention and facts - pricing (see below) , market share, fabrication process, and strategy re: Richland. Personally, it doesn't bother me what you say regarding AMD, Intel or Nvidia- so long as it is factually correct and is not an isolated/cherry picked facet. You want to tell me how great the 7970M's performance is (and you did-see below), fine, but maybe you should also take into account the whole ecosystem (and you did, then promptly immediately contradicted yourself- see below) You're the one using my posting to fly off on a tangent to promote AMD's strengths along with screeds of benchmarks for a series of SKU's that this article- and my posting- had no interest in. All you posting ever seems about is an excuse to jump on a soapbox and write a wall of cherry picked circumstances that show AMD in a glowing light while presenting straw man arguments and then sending up walls of text when someone brings reality into the discussion....refusing to acknowledge any facets that run counter to your junior league AMD no, I don't count "Nvidia/Intel product X is OK for Y, but here's a qualifier to refute the OK part" or "AMD product X has its faults but that's [insert excuse]"
    If you like I can check our respective postings and note verified facts versus the wrong assumptions, outright lies and excuses. I have a fairly reasonable article on TS regarding AMD based entirely on fact from a personal knowledge base...I don't see any remotely unbiased content from you regarding Intel or Nvidia.
    And where did I say that "[AMD's] cards were much worse?" (Hint: Nowhere) Just another straw man argument and a lot of hyperbole...
    Nope. Don't see any mention of performance. Just a mention of Nvidia's better OEM relations, consistant pro graphics ecosystem, and less problematic mobile drivers at the current time.
    If you want to believe that Apple dumped AMD in favour of Nvidia because the latter spent more on PR, then be my guest...plenty of people thought the world was ending last week as well. No accounting for the thought process.
    The reason that OEMs were switching to Nvidia was primarily because of the longevity of their relationships, a reasonable dialogue with the IHV regarding mobile drivers, and the fact that the vast percentage of notebooks use an Intel CPU, and Nvidia's Optimus gives both a solid product as well as cutting AMD out of the picture.

    But enough of the truth...lets revisit you paraphrasing a statement I didn't make...
    I didn't say they were crap. Please point out where I said anything of a sort. Yet more butthurt hyperbole.
    AMD have two choices for the 8900M. A mainstream part already used for the 8800/8700's or a low-power version of the high-end desktop part. Doesn't take an Einstein to work out the probable performance requirements for either...especially since the same dynamic has existed since the first mobile graphics based on desktop GPUs.
    Do you even read about the tech? The reason the Enduro issue is an issue is that for more than a few mobile platforms, Enduro cant be disabled since their machines are muxless
    So, you’d agree that performance and performance/$ aren’t the be-all-and-end-all ! But wait…in the your next post
    If you're going to argue, at least be consistent.

    As for the rest of your rambling, maybe if you read what was writen rather than misreading and hyperbolizing you'd sound more coherent.
  18. dividebyzero

    dividebyzero trainee n00b Posts: 4,891   +1,264

    Because profits mean a healthy company?
    Because profits mean long term viability?
    Because profits mean sustained R&D ?
    Because profits mean keeping staff ?
    Because profits mean maintaining software and app support ?

    Nah, you're right. Nosediving market share and lack of profit line are obviously the winning strategy. It's plainly obvious that OEM's prefer to ally themselves with companies on life support that cant sustain projects due to staffing cuts and employees leaving (that's actually two for the week- Bulldozer's chief architect Mike Butler is now as Samsung too).
    Who said you did? You seem to have a persecution complex.
    I noted bus width and memory config because these are unknown quantitys with the 8500/8600 parts. If you don't know what the bus width and bandwidth are, how can you state what the performance will be? I certainly couldn't, but you think you are able to.
    I said no such thing. You're trolling....and what's worse is you aren't even halfway decent at it.
    Nope. You got that wrong as well :'(
    What I said was:
    Nvidia's mainstream parts are:
    GTX670M (rebadged GTX570M) -in production for 19 months
    GTX675M (rebadged GTX580M) -in production for 19 months
    GTX660M - in production for 9 months
    GTX670MX - in production for 2 months
    GTX675MX - in production for 2 months

    Of those parts, which are the likeliest to have had the highest production volume ?

    Comprehension is the key to understanding. You have very little of either.

Similar Topics

Add your comment to this article

You need to be a member to leave a comment. Join thousands of tech enthusiasts and participate.
TechSpot Account You may also...