AMD unleashes higher-end, 28-nm Radeon HD 7000M GPUs

A long time ago in a galaxy far, far away—last December, actually—AMD introduced the first members of its next-generation mobile GPU series: the Radeon HD 7600M, 7500M, and 7400M. As it turns out, those weren’t really very next-gen at all, since they were based on the same 40-nm fab process and VLIW5 Terascale 2 architecture as, well, last-gen models.

Today, AMD has injected some actual newness in its mobile GPU lineup with the Radeon HD 7900M, 7800M, and 7700M. Code-named Wimbledon, Heathrow, and Chelsea, the new GPUs are fabbed on a state-of-the-art 28-nm manufacturing process, and they’re based on the same Graphics Core Next architectural framework as the desktop 7000 series. And they have the same other niceties, like PCIe 3.0 support, AMD’s VCE hardware video encoder, and ZeroCore power.

These are, in other words, truly deserving of the leading 7’s in their model numbers.

Judging by the look of the packaged chips and their hardware specifications, I expect Wimbledon has a lot in common with Pitcairn (of Radeon HD 7850 and 7870 fame), while the other two new mobile entrants are probably awfully similar to Cape Verde (which powers the Radeon HD 7770 and 7750). Here’s a detailed look at those specs:

  Radeon HD 7900M Radeon HD 7800M Radeon HD 7700M
Shader ALUs 1280 640 512
Core clock 850 MHz 800 MHz 675 MHz
Texels filtered/clock 80 40 32
ROP pixels/clock 32 16 16
Memory interface width 256-bit 128-bit 128-bit
Memory 2GB GDDR5 2GB GDDR5 2GB GDDR5
Peak memory speed 4.8 GT/s 4.0 GT/s 4.0 GT/S
Peak memory bandwidth 153.6 GB/s 64 GB/s 64 GB/s

The Radeon HD 7900M has the same unit counts and memory configuration as the desktop Radeon HD 7870; the only difference, as far as I can see, is that its core speed is a couple hundred MHz slower. The 7800M is similarly impared versus the Radeon HD 7770, and it’s also had its memory speed curbed somewhat, from 4.5 GT/s to 4 GT/s (or 1125MHz to 1000MHz, if you’re looking at the raw memory clock speed). As for the 7770M, that model has the same number of units enabled as the desktop 7750, but it’s 125MHz slower, and it also runs its memory at 4 GT/s instead of 4.5 GT/s.

Not content with bringing over the finest silicon from its desktop family, AMD has given mobile users a little something of their own to brag about: Enduro, which can best be described as AMD’s answer to Optimus. AMD says Enduro Technology automatically switches between integrated and discrete GPUs based on “application graphical requirements,” and it claims the transition is “smooth” and “automatic”—just like with Optimus on Nvidia’s mobile GeForces. The discrete GPU is also selected when the notebook is plugged in, and users can switch manually via the control panel if they so desire.

Best of all, AMD hasn’t limited Enduro to its own platform. The technology is supposed to work with both Intel processors and AMD’s A- and E-series APUs.

Laptops based on the new mobile Radeons are already here. In fact, AMD’s Radeon HD 7970M GPU is featured as an option in the latest iteration of the Alienware M17x, right alongside Nvidia’s new GeForce GTX 660M and 675M. The machine also packs a 17.3″ display and a choice of Sandy Bridge-based Core i7 processors.

Comments closed
    • h4x0rpenguin
    • 7 years ago

    AMD had the foresight to keep the 7900M/7800M/7700M name slots free for their real 28nm next gen parts. So does Nvidia plan on letting AMD win this round or are they going to cram a complete 28nm lineup into the 68xM/69xM name space (and/or spill over into 7xxM territory)?

      • Washer
      • 7 years ago

      The 640M, 650M and 660M are all based on Kepler. The 675M and 670M are based on 48nm Fermi, though the 630M is a mix of Fermi 28nm and 40nm and the 620M is Fermi 28nm. Notebookcheck.net is showing that the 680M is Kepler based as well. It’s also worth noting that AMD has renamed the previous cards in some cases too, like the 7690M which is a 6770M. The notebook graphics card are mighty confusing sometimes.

    • ULYXX
    • 7 years ago

    Forgive me if i missed something… but wasnt there talk of XDR2 type of memory? I keep seeing GDDR5 in the high ends when there was multiple sites speaking of another type of memory. Was XDR2 a myth for the 7000 series??

      • TREE
      • 7 years ago

      I reckon AMD will be looking for a different type of memory/bus interface in future architectures. I guess the tell tale sign is the increase to a 384-bit bus used in Tahiti, clearly AMD wants more memory bandwidth and given the cost of wafers from TSMC, it makes sense for AMD to decrease that bus width in order to increase yields.

      • khands
      • 7 years ago

      Yeah, XDR2 hasn’t shown up this gen, maybe next gen (or possibly in a console?)

    • shank15217
    • 7 years ago

    The GTX 675M doesn’t stand a chance against 7970M

      • pogsnet
      • 7 years ago
    • rrr
    • 7 years ago

    And some rebrands for lower end parts.

    • krazyredboy
    • 7 years ago

    Why is it “Enduro,” sounds dirty, to me?

      • shank15217
      • 7 years ago

      lol nice one

      • crabjokeman
      • 7 years ago

      I think of this (yes, I’m really dating myself with this link): [url<]https://en.wikipedia.org/wiki/Enduro_%28video_game%29[/url<]

        • UberGerbil
        • 7 years ago

        Oh, man, they gave away a Datsun 280ZX when they introduced that game. [url=http://www.youtube.com/watch?v=kWF-hH1nloo<]Black Gold![/url<]

      • UberGerbil
      • 7 years ago

      Maybe because [url=http://www.google.com/search?q=enduro+race&hl=en&safe=off&prmd=imvns&source=lnms&tbm=isch&ei=vs6WT6mIB-GciQKh2O2uCg&sa=X&oi=mode_link&ct=mode&cd=2&ved=0CA4Q_AUoAQ&biw=2048&bih=1051&sei=eM-WT__3HMfaiQLD-qCOCg<]enduro racing[/url<] tends to get kind of muddy?

      • tfp
      • 7 years ago

      [url<]http://www.youtube.com/watch?v=a-RGhAe6Rrg[/url<]

    • DancinJack
    • 7 years ago

    [quote<] AMD says Enduro Technology automatically switches between integrated and discrete GPUs based on "application graphical requirements," and it claims the transition is "smooth" and "automatic"—just like with Optimus on Nvidia's mobile GeForces. [/quote<] Finnnnnallllyyyy We'll have to see how well it works.

      • Goty
      • 7 years ago

      Their automatic switching technology already in use isn’t that bad, really. It is smooth and seamless, but it is a bit dumb and doesn’t always pick the best card for the job (though you can assign programs to particular cards in CCC). Any sort of improvement on this setup would make it great.

        • codedivine
        • 7 years ago

        Agreed. I am more interested in knowing if they have resolved the driver distribution problems yet. Currently it is difficult to just download and install regular Catalyst releases on machines that do automatic switching. One has to wait for the laptop vendor to update the driver, which either never happens or happens too slowly.

        • DancinJack
        • 7 years ago

        After reading the short Anandtech article, it may not matter a whole ton. With ZCP, the GPU might consume a very small amount of power at idle which would make the Enduro stuff not extremely helpful. We shall see I guess.

          • xeridea
          • 7 years ago

          ZCP is only when the display is off.

            • DancinJack
            • 7 years ago

            Thanks for clearing that up for me.

    • Left_SHifted
    • 7 years ago

    How about pairing these with the new ivy bridge mobile parts, now that would be a serious next gen combo

      • derFunkenstein
      • 7 years ago

      because they haven’t been released yet?

        • Left_SHifted
        • 7 years ago

        [url<]http://vr-zone.com/articles/hp-gives-the-pavilion-dm4-an-ivy-bridge-makeover-with-a-splash-of-beats/15649.html[/url<] read this recently, thought some more juicy stuff might be in the offing now that the iron curtain of press lockout has been lifted :S

    • dpaus
    • 7 years ago

    Any indication of the TDPs for these? I wonder if AMD can produce a version of Trinity with the 7900M as the integrated graphics, or even added to a much lower-end GPU with the Enduro (but as an all-in-one ‘gamers APU’) in a 125W package?

      • SomeOtherGeek
      • 7 years ago

      Man, that would be awesome. If AMD were smart they would do exactly that.

        • dpaus
        • 7 years ago

        Especially if AMD would implement some of the same on-die GPU-CPU integrations Intel has, and especially-especially if they can fix their cache latency issues, given the claims of 20-25% performance improvements in Trinty over Bulldozer, AMD could become a serious competitor again.

        Ah, early morning pipe dreams 🙂

          • Deanjo
          • 7 years ago

          [quote<]20-25% performance improvements[/quote<] You should know better by now that when it comes to AMD cpus to take any claims with a heavy pinch of salt.

            • dpaus
            • 7 years ago

            I do, seriously, but, well, I guess I’m also an eternal optimist 🙂

            Actually, I think they’re claiming 15-20% and I’m just excited…

            • wierdo
            • 7 years ago

            I thought 10-15% was the claimed gain, is this a new estimate or are we talking about integrated GPU performance?

            • khands
            • 7 years ago

            IGP gain is supposed to be 50% I thought. I recall aiming for a 10% IPC increase from the Piledriver core and then who knows what other optimizations make up that last little bit, all bcs of course.

            • dpaus
            • 7 years ago

            It was originally 10-15%, and then fairly recently I heard 15-20% being tossed around (by AMD). I always take those with a grain of salt (the big blue kind the cows lick…)

        • helboy
        • 7 years ago

        sadly there lies the problem ” If AMD were smart ……” 🙁 .As much as I admire their technology and innovative products….i wouldn’t bet on them being anywhere near “smart”.

          • dpaus
          • 7 years ago

          AMD: the [url=http://www.youtube.com/watch?v=-WmGvYDLsj4<]Pakleds[/url<] of the CPU universe....

      • ImSpartacus
      • 7 years ago

      I’m not up to date with Trinity rumors, are they seriously rumored to be throwing a glorified Pitcairn GPU next to four Piledriver cores? That is going to be massive.

        • Left_SHifted
        • 7 years ago

        Hell, i’ll even settle for a Cape Verde, as long as it is paired with a proper Piledriver refresh.

          • ImSpartacus
          • 7 years ago

          Yeah, I would be worried about battery life and bandwidth if they tried to fit a Pitcairn on the CPU die.

          Something smaller like Cape Verde seems much more likely.

          • Alchemist07
          • 7 years ago

          Actually, an APU with Steamroller (upgrade to Piledriver) + Cape Verde is already planned for release in Q1 2013…I cant wait 🙂

        • jensend
        • 7 years ago

        No, dpaus is either engaging in wishful thinking or trying to give unsubtle hints to invisible watchers from AMD.

        The wildest the Trinity rumors get is the claim that its 384 cores will be GCN cores rather than the VLIW4 (i.e. 6900-style) cores everybody has been pretty sure it would have. While that would be neato- 384 GCN cores would be 3/4 of a 7750- it’s quite a stretch, esp. since Trinity is 32nm and all of AMD’s GCN stuff has been 28nm.

          • dpaus
          • 7 years ago

          [quote<]dpaus is either engaging in wishful thinking or trying to give unsubtle hints to invisible watchers from AMD[/quote<] Wouldn't the latter also imply the former? Anyway, yeah, it's a common trait of OO programmers who have moved on to management: if you see two Lego pieces, why not try putting them together? EDIT: a colleague read that over my shoulder and said "Because somehow, you manage to find Lego pieces made from refined plutonium"

            • ImSpartacus
            • 7 years ago

            Reading TR at the office? Are we setting a good example, Mr. Paus?

            • dpaus
            • 7 years ago

            I’d like to think I’m doing marginally better than [url=http://ca.news.yahoo.com/blogs/daily-buzz/aviva-investors-mistakenly-sends-email-firing-1-300-204417481.html<] this outfit[/url<]. But then, 'reading at the office' played a major, negative role in that fiasco, too 🙂

            • ImSpartacus
            • 7 years ago

            Their HR guys were reading TR, so they accidentally fired everyone?

            We have to get US politicians on TR…

      • MadManOriginal
      • 7 years ago

      Unlikely, because it would result in severe lack of GPU memory bandwidth. Unless perhaps they added some ‘sideport’ memory like their older northbridge integrated graphics had.

      • ish718
      • 7 years ago

      Well, HD 7900m is close to a HD7870 in spec and the HD7870 has a TDP of 175W…

        • BobbinThreadbare
        • 7 years ago

        Where did you get that TDP from, as under load with an i5 TR measured system load at 182 watts. I think the i5 and motherboard probably use more than 7 watts.

        [url<]https://techreport.com/articles.x/22705/7[/url<]

          • ish718
          • 7 years ago

          [url<]http://www.hwcompare.com/12345/geforce-gtx-680-vs-radeon-hd-7870/[/url<] That is the max TDP, maybe TR wasn't loading the HD7870 enough.

            • BobbinThreadbare
            • 7 years ago

            I think it’s more likely that is an extremely conservative TDP.

Pin It on Pinterest

Share This