news six core phenom ii to feature turbo core tech high frequencies

Six-core Phenom II to feature Turbo Core tech, high frequencies

Looks like the rumor mill was right to be buzzing about AMD incorporating a dynamic clock speed boosting mechanism similar to Intel’s Turbo Boost into its upcoming six-core processors. AMD has confirmed to us that the chip code-named Thuban does indeed include such a feature, dubbed Turbo Core technology.

At its heart, Turbo Core is essentially similar to Intel’s Turbo Boost. When not all of the processor’s cores are fully occupied, this feature raises the clock speeds on active cores above the default frequency while staying within the CPU’s defined thermal limits. However, Turbo Core is a little bit different from Intel’s technology in its operation—and generally simpler.

The processor becomes eligible to enter the "boost" state when three or fewer cores are heavily occupied with work. This boost state is a simple, binary condition. When the CPU enters this state, the peak clock speed for the three affected cores will jump by as much as 500MHz above the baseline, improving performance. To give a possible example, a future Phenom II X6 CPU might have a base clock speed of 3.4GHz but range up to 3.9GHz when only one to three cores are busy. Meanwhile, the three inactive cores will be in an idle state, with greatly reduced power draw. Three cores is the magic number here because that’s the split on six-core parts; three cores go into boost, and three do not. Quad-cores based on Thuban may have a 2/2 split.

Below is a quick diagram from AMD that will either clear up exactly how it all works or leave you in a state of deep confusion.


Unlike Intel’s Turbo Boost, the Turbo Core mechanism doesn’t establish multiple frequency thresholds based on the number of cores occupied or the present thermal conditions. For instance, there’s no separate peak speed allowed when only a single core is busy, nor does Turbo Core rely on sophisticated real-time thermal monitoring. AMD will define the peak Turbo Core frequency for each CPU model, and the CPU’s model will determine its behavior. All chips that bear a given model number will behave identically with regard to frequency. The particulars of the individual chip or the thermal environment won’t influence how much time the CPU spends in a boost state.

This may be a less sophisticated approach than Intel’s Turbo Boost—and perhaps not as likely to wring every possible bit of headroom out of a given thermal envelope—but AMD cites the deterministic behavior of its Turbo Core-endowed processors as a positive trait.

Turbo Core will act in concert with a host of other existing mechanisms to regulate clock speed and voltage, and that symphony is becoming increasingly complex. AMD tells us Cool’n’Quiet will still reduce clock speeds dynamically on cores that aren’t busy, although we’re unclear on whether AMD’s decision to lock the P-states of all cores together on the Phenom II will remain in force here. The dynamic voltage scaling portion of Cool’n’Quiet still applies the same voltage across all cores, and Thuban does not include the ability to gate off power entirely to an idle core like Intel’s Nehalem-derived CPUs. The upcoming Llano APU is slated to be the first AMD processor to employ power gates for individual cores.

Regardless of the finer points, Turbo Core should give AMD a credible answer to Intel’s Turbo boost. That’s true in part because of Thuban’s other virtues, which aren’t limited to dynamic clock speed increases. Thuban ought to have more thermal headroom available to it thanks to a process technology change. For this chip, GlobalFoundries has added a low-k dielectric to its high-performance 45-nm SOI fabrication process, which should reduce leakage power.

As a result, AMD expects Thuban’s six cores to fit into established power envelopes while featuring base clock speeds similar to current quad-core Phenom IIs. In fact, the firm plans to introduce a Black Edition six-core Thuban processor at speeds "over 3GHz, substantially," so the example we cited above may not be far from the truth. That part will gain 500MHz from going into a boost state, which means effective clock speeds with lightly multithreaded applications could approach or exceed 4GHz. Lower-end Thuban derivatives will get smaller frequency bumps out of Turbo Core, and AMD intends to offer a range of Thuban-based products in its current suite of power envelopes.

Those processors will see their formal introduction in the coming weeks, almost surely under the Phenom II X6 brand name, and they should be drop-in compatible—Turbo Core and all—with existing Socket AM2+ and AM3 motherboards. Although the move to six cores and the addition of Turbo Core may not be enough to wrest the performance crown from Intel’s 32nm Gulftown-based Core i7-980X monster, AMD anticipates that the new Phenoms will offer an attractive price-performance proposition, in keeping with its recent tradition.

0 responses to “Six-core Phenom II to feature Turbo Core tech, high frequencies

  1. Indeed, AMD is finally executing well, they have been over the past year or so and the results are starting to show up…. so far nothing appears to be nearly as bad as the Barcelona gaff.

  2. Maybe it has something to do with compatibility across AM2+ to the latest AM3. A simpler implementation is easier to validate and gets basically the same results.

  3. These will be a serious challenge to Lynnfields; it’ll come down to MONEY
    Can’t wait to see what the X6 price is.

    AMD is impressing me more and more every day. Someone on these forums pointed out to me that a lot of things AMD is doing now is to keep the ship floating as well as they can, and betting on Bulldozer. Magny-Cours and these X6’s are great moves in that direction.

    Kudos to AMD; these product optimizations are hitting the targets so well that it’s starting to sound like the company is being run by engineers…

    I might even say that purchasing ATi was a smarter move than I ever gave them credit for.

  4. My 785G motherboard (Gigabyte 785 mATX) had a recent BIOS release that claims support for the new X6 Phenoms + Turbo, so I’m pretty sure that this will work in AM2+ 🙂

  5. Does that mean this actually works with AM2+??? I’m too attached to my 780G mobo, plus if I did a new board…might as well go Intel.

  6. GF killed their 32nm bulk before TSMC did IIRC.
    In any case, there isn’t a single 32nm bulk mentioned on their website.

  7. That’s a bit of a stretch. Their only real hiccup was the original Phenom, and that’s all it was. They’re past that now.

  8. Perhaps, but games that are multithreaded enough to actually take advantage of a 6-core processor won’t benefit much from adding a few hundred megahertz to three cores.

  9. No mobile Celeron processors supported Speedstep through to the 45nm (Penryn-3M) T3300. Thankfully, the Arrandale-based P4500 does support it.

  10. Yeah, let’s just wait to see the actual PII X6 turbo frequencies before making such sweeping predictions, m’kay?

  11. Looks like Intel Turbo Boost put them into damage control mode.

    Hopefully this actually works, unlike their hyped per-core CnQ clock downscaling that caused considerable performance degradation, for example. And that little quirk with their software-based C1E mode…..

  12. If you are playing Xcom, in DOSBOX, turn down its cycles and watch Cool’n’Quiet drop clocks due to low CPU utilization. Or, wander your little way into your OS’s power options and set it to power saving.

    I can’t believe I’m seeing people complain about this smart way to increase per-core performance. This is what we need in this world of >2 core CPUs and the not-happening multithreaded future.

  13. There are plenty of examples, but a good one would be if in a game you get 70FPS at 1.5Ghz but the game keeps pushing more onto the CPU to get 150FPS and runs it up to 3.2Ghz or whatever. The extra frames don’t make any difference in gameplay and rendering them just uses extra power. Why not lock the CPU at 1.5?

    I agree with #44 100% and would like more control over CPU speeds and such. If I am playing x-com for example I could turn my CPU down to like 100mhz and play just fine.

  14. If true, that should make a 1090T running three of its cores at 3.6 GHz a highly credible alternatives to i7s

  15. I think that was only rumor; the latest official word from GF is that 32nm bulk is still on. It’s possible that it was killed but given these conjectures:

    TSMC killed their bulk because most of their cuting edge customers work on half-nodes (nVidia, ATI); so why spend billions developing a process that no one is going to use.

    GF was developing 32nm as their first market commercial process, and as their first bulk process (everything at Saxony is built SOI since AMD built Fab30).

    I think it’s still on, even if it’s only a proof-of-concept. It’ll be the first bulk process GF has built (Chartered excepted)

  16. Their last (leaked) roadmap shows the 3.0GHz “960T” above the 3.4GHz 965. A small IPC boost would account for that.

  17. I see that as likely too, considering socket compatibility and that we didn’t really even know they were going to be 6 core till early this year/late last.

  18. Touche; I should have phrased that as “someone calculated that the losses from deferral of top-end Phenom II X4 sales would be more than made up for by new sales gained from customers who deferred their i7 purchases”

  19. Yes, they killed their bulk 32 nm node, but not their SOI 32 nm node and that latter one is what AMD will be using for their CPUs.

  20. When your processor market share is like <20%, I think that’s almost guaranteed.

  21. I’d heard they killed it completely, that wasn’t directly from GF/AMD though so it could’ve been wrong.

  22. You are partly mistaken.
    GloFo has canceled it’s 32nm bulk process and moved straight to 28nm bulk on that line. That’s the stuff that most of their customers will probably be using. However, their next “SHP” process is still 32nm SOI, which is what AMD’s CPUs will use.

  23. Yeah, the landscape will have changed enough once these release, or at least I think so.

  24. Global Foundries 32nm doesn’t exist anymore, they’re going straight to 28nm in early 2011.

  25. I think the issue there is that most games that can use more than two cores can use at least 4, and that will keep Turbo boost from coming into play. I’m not sure the way they implemented it will be useful at all, except on 2-thread or less limited applications.

  26. This is looking good, but any word on whether there’ll be an option to turn off the turbo clocking? For me personally, there’s nothing out there right now that would need the extra turbo clocks. Even if running a game, it won’t need more than the default speeds if they’re over 3 Ghz already. There might be times when you’d rather just have all the cores stay at their default speed (or less with cool n’ quiet) regardless of the type of workload being done.

  27. Noooooo! Don’t cut me off! Seriously, though, couldn’t they revamp the DDR3 controller and just leave the DDR2 controller alone. None the less, times change and I’m at peace with moving on from DDR2. It has served me long and well.

  28. they won’t do that until they decide it’s time to cut off AM2+ users, and that probably won’t be until Bulldozer.

  29. 3/6 sounds like an optimal split for normal computer enthusiasts. Three cores for gaming, six for encoding or folding.

  30. Thanks for this to-the-point, non-trollistic post.
    I see you have more insights than AMD when it comes to the amount of shared L3 cache needed depending on workloads and number of cores.
    Please by all means send your CV across to AMD to sort this out ASAP, wouldn’t be very nice of you to withhold crucial information like this..

  31. The cache hierarchy is poor. Down to 1MB per core L3 from the 1.5MB per core on Phenom II X4.

    This thing has no chance against Lynnfield let alone Bloomfield or Gulftown. Stick to GPU’s AMD!

  32. I like that AMD’s Turbo is guaranteed. It might not be as flexible as Intel’s, but the overclock for three cores is not trivial – +500MHz on a 3.4GHz CPU is a 15% boost and it still allows a reasonable level of multi-threading to occur.

    The main point is that a 6 core CPU will operate in the same TDP as the previous 4 core CPU at the same clocks – showing that GlobalFoundries have a good 45nm process (and hopefully boding well for 32nm). The Turbo Boost is a nice add-on feature that will improve performance in certain classes of application that AMD has been behind in, including games – 15% is a lot when it comes to graphs in reviews!

    Obviously the major enhancements are coming first with Llano, then with Bulldozer.

    AnandTech has more details:

    1090T: 3.2GHz -> 3.6GHz (+12.5%) (125W)
    1075T: 3.0GHz -> 3.5GHz (+16.7%) (125W, Q3)
    1055T: 2.8GHz -> 3.3GHz (+18%) (125W and 95W version)
    1035T: 2.6GHz -> 3.1GHz (+19%) (95W)
    960T: 3.0GHz -> 3.4GHz (+13%) (95W, Quad-core)

    a 95W 1055T looks like a good buy to me.

  33. Somebody has clearly done the calculation that shows that slowly leaking juicy tidbits about high-frequency Turbans with dynamic overclocking will cannibalize more sales from i7s than from quad-core Phenom IIs.

  34. I was just referring to Turbo Core, not to Thuban (which has lots of other improvements in many other areas).

    But yes, the way AMD is implementing its Turbo functionality is a lot less advanced than what intel’s been doing for the past year and a half.

  35. AMD sometimes pulls some amazing stuff out of their available manufacturing technology late in the game. 90nm was quite a show and looks like 45nm will be as well.

  36. And what was it they said about an UPDATED memory controller ?….. AHEM!

  37. Right, so adding 2 cores with this functionality in the same thermal envelope without a process shrink is a small step?

  38. We do know that there is no power gating on idle cores, no per-core voltage stepping, no progressive boost with # of active cores, and no TDP controls.

    While Turbo Core may be a positive step for AMD, it’s a very small step.

    I do expect that it will continue to work with OC’ing, though, since there is talk of a BE. If Turbo core doesn’t work with BE multiplier OC’ing, that would be epic fail.

  39. Is that some kind of 1337 terminology or are you using handwriting recognition to write comments?

  40. We don’t know the exact details of AMD’s turbo function yet but Intel’s is pretty straightforward – it operates by changing the multiplier to get various speeds. Motherboards vary but they generally fall in to two camps – when overclocked some are still able to ‘Turboboost’ correctly and some aren’t. For the former, the processor will simply use whatever higher overclocked QPI speed is set with the same multipliers it would at default. So if the ‘Turbo’ multiplier is 22x, then a higher QPIx22 would yield the new ‘overclocked’ Turbo clockspeed. Some motherboards are not able to operate Turbo properly when QPI is overclocked in which case you end up with something the same as non-Turbo overclocking. I think most allow you to turn Turboboost on or off entirely as well.

  41. I’ve overclocked with all power saving features ON for the last 4 years. Never had a problem really, I don’t push things with really high voltage or to the ragged edge either. It is not necessary to turn off power saving functions aside from more ‘suicide run’ type situations…and if an overclock doesn’t work with power saving features enabled I’d say it’s not truly a stable overclock, after all those are standard processor features that are designed to work and should work at any speed. Leaving them enabled eliminates the negatives in your last paragraph.

  42. I always leave Speedstep/CnQ on when OCing. Some motherboards let you specify different voltages for different power states. Other times, I’ve had to rely on software solutions like RMClock or CrystalCPUID.

    These days I’m more interested in undervolting and silence than extreme OCs, though.

  43. For overclocking, you turn it off. Anything that changes the clock and voltage settings behind your back – turbo, SpeedStep/Cool’n’Quiet, et cetera, will be harmful.

    Turbo’s bad because the point of overclocking is to run the chip up to the very limits of stability. If it’s stable at, say, a maximum of 4 GHz, and you leave turbo on, it’s going to try to push it past that stable point. That means you’ll end up with an unstable system.

    Power-saving stuff is bad when it comes to overclocking because it generally also changes the voltage. If the power-saving features lower the voltage and frequency, but the frequency remains higher than it expects (because you’ve overclocked it), you can again get instability: there’s not enough voltage to run reliably at that given non-stock, untested frequency. Leaving power-saving features on can cause some really strange behavior, where the system can run forever in stress tests with zero errors, but crashes randomly while it’s idling.

    Overclocking these days can come with some real drawbacks. Your system will run much hotter (and noisier) at idle, and can really run up your power bill compared to one that’s running at stock speed with all the power-saving options on. But hey, you get the increased performance, and sometimes that’s all that matters 🙂

  44. I would refer to them as my Marturb!ng Group. (that looks a lot like another word, but is coincidentally close to what a real Marketing Group does all day.)

  45. Damn, hopefully AMD incorporated tweaks to allow for better performance at the same clock? Wishful thinking.

  46. I’ve often wondered that myself. Not only whether or not it works after a predefined overclock, but if so, how well it works after an overclock. If it works properly, that could create even bigger increases in performance for gaming and such, but if it goes wrong, in some way or another, then it’ll either do nothing or perhaps overheat the processor, either way rendering it useless.

    Anyone have any insight into this dilemma?

  47. You should apply for a job in Apple’s marketing department. Maybe they will issue a new family of smart phones called !phone.

  48. yeah but you (the engineer) dont get to decide the names of the chips, the marketing people decide. and nothing against you, but if I were a marketing person i wouldn’t name my chips that 😛

  49. the way I see all this ‘turbo’ feature is that basically if you run the processor at stock speeds you get a nice little on the fly overclock. but I am more concerned with when you run the processor overclocked from the start does that totally make the turbo boost feature useless?

  50. Grrrr, “the coming weeks” Why can’t we get more hard dates in the hardware industry? Because of corporate secrecy?

  51. If I design a chip, I’m gonna call it the Turb!on [that’s Turbion with an exclamation point in place of the ‘i’ just to show how turbo it is!!!], it’s clock-scaling feature Turboost, and it’s power saving feature TruTurb.

  52. Hm. I wonder how something like this might stack up against the Lynnfield series.