Bulldozer's Power Management

AMD confirmed that the power management of the Bulldozer core is an improved version of the power management improvements that are part of the “Llano” CPU. Just like Llano, Bulldozer has a Digital APM Module. The APM modules samples a number of performance counter signals and these samples are used to estimate dynamic power with 98% accuracy. Now combine this power estimate with Bulldozer's power gating at the module level and vastly improved clock gating and you can start to understand what is possible. 


Bulldozer reduces the number of active and power consuming circuits by vastly improved clock gating

If your application runs only one or a few threads on your 8-module, 16-core Interlagos CPU, several of those modules might be power gated. Or if you run integer-only threads, the fact that quite a few unused parts (i.e. the FPU) of the module will be clockgated might be enough to stay under the configured TDP. So in those cases, it won't be necessary to limit the clock speed. And that is really great, especially in the real world.

In the real world, only a few HPC application behave like the SPEC CPU rate benchmarks, which spawns threads accross all cores.  Most server applications do not fully utilize all available cores all the time. Sometimes, only one thread will be really critical and the perceived application performance will depend on it. A little bit later several threads might demand CPU power (but not all cores will be busy). Only a certain percentage of the time are all the cores used. That is exactly the reason why the cheaper Magny-Cours make so much sense for HPC applications, yet it struggles to keep up with the higher clocked, higher IPC Xeon Westmere cores when running OLTP and ERP applications. Putting a power cap on a Magny-Cours means even lower frequencies, and as a result even higher response times (as we have measured here). 

By adding power consumption measurements to the CPU, Bulldozer will run most server applications at full speed unless you lower the TDP too far. (Obviously, if the TDP is lowered enough, the CPU will not be able to operate at higher frequencies, thus degrading the response time performance too.) The maximum throughput will be a little bit lower, but most server applications almost never run at maximum throughput. In fact, maximum throughput only matters for HPC applications and benchmarks. For real human users, response times are the only thing that matter.

The beauty of this new power cap system is that in normal circumstances (e.g. the server is running at 40-70% load), the response times will hardly be any longer. At the same time, the adminstrator can make sure that the server cluster does not exceed the capacity of the cooling equipment and the power lines.

This TDP Power Cap technology could be very interesting to small and medium businesses too, and not only to owners of large server clusters. TDP Power Cap could be a way to make sure that your collocated servers never exceeds the maximum amount of amps allocated to you, and as result you will not have to pay unexpected high electricity bills. However, whether or not this ideal world of low response times and low electricity bills will become a reality for the Bulldozer server owners will also depend on the availability of a good and decently priced management software tool that allows the administrator to configure the TDP on all servers simultaneously.

On a standard server, you will get a section in BIOS that allows you to tweak the TDP in 1W increments (or a maximum of 64 power settings), a good step forward compared to the current p-state setting. But to control a server cluster in an efficient way, good management software is needed. Currently, you either have to buy all your servers from the same vendor (HP for example) and then pay for management software such as HP's Insight Control software. To really unlock this technology, AMD or one of their partners needs to make sure this kind of software is widely available--some open source code perhaps?

TDP Power Cap Final Thoughts and AMD's Future Plans
POST A COMMENT

59 Comments

View All Comments

  • stmok - Friday, July 15, 2011 - link

    No it doesn't. It expressly says CPU for Komodo. Not APU. Look carefully!

    Even the latest leaked slide (Bit-tech.com source in my thread) refers Komodo as a CPU with "Next-Gen Discrete Graphics" as part of the Corona enthusiast/performance desktop platform for 2012.
    Reply
  • Kristian Vättö - Friday, July 15, 2011 - link

    I have added a (?) to suggest that it's not certain whether Komodo features an IGP or not. Reply
  • jjj - Friday, July 15, 2011 - link

    I don't see any GPU on the slide TimCh provided (and do read the fine print).Anyway there is only 1 botched slide with GPU and 8 cores and if you think about it, it makes no sense.
    Why have Sepang with 10 and Komodo with 8,it would be a waste of time and money and there would be no reason to have a different name for it,instead of Trinity.Also how big would the chip be with 8 cores and GPU,even if they drop the L2 cache and why add a GPU to a chip adressing the high end where most don't need it (and make the same mistake as Intel using the HD 3000 for K chips).
    Reply
  • jjj - Friday, July 15, 2011 - link

    edit to prev post: obviously i ment L3 Reply
  • rnssr71 - Saturday, July 16, 2011 - link

    'Also how big would the chip be with 8 cores and GPU,even if they drop the L2 cache and why add a GPU to a chip adressing the high end where most don't need it (and make the same mistake as Intel using the HD 3000 for K chips).'

    well, 8 core bulldozer(4 module) is going to be over 300sq. mm. larger than llano(on 32nm) but smaller than thuban(45nm).
    i would guess that 8 meg of L3 cache that the 8 core bulldozer will have would take up 2/3 as much room as the current gpu in llano. so, quite a lot of space.
    so you're right, it would be a mistake to have a full sized gpu......maybe ANY gpu until gpu computing really takes off.
    Reply
  • Casper42 - Friday, July 15, 2011 - link

    Just an FYI, while it is true that Insight Control is required for power capping a cluster of Rackmount (DL) servers, you get Dynamic Power Capping withing a single c7000 chassis today for free.

    So for Clusters of 16 servers in the same chassis, you don't need IC licenses to see a pretty big benefit. When one server needs more power, the other servers can all be throttled back slightly to let the one burst and still get the workload done, like you mentioned.

    DPC in blade chassis also has 3 different settings.
    One is Average Load for Thermals
    Two is Average Power Draw (80% rule on individual circuits)
    Three is Maximum Power Draw (Do Not Exceed - Circuit protection)
    So you can not only unlock excess capacity with the Max Draw setting being lower than faceplate value, but you can also tune the other values to hit the appropriate thermal values for the DC.

    I've heard there will be an Insight Control powered Multi-chassis DPC coming next year as well. And for those familiar with HP gear, Insight Control licenses, when purchased in 8/16 packs along side a new chassis are only like $50 more per server than iLO Advanced which alot of people already purchase. With normal Enterprise discounts the gap can be even smaller.

    Johan/Kristian - where are you guys located? (roughly)
    Reply
  • Kristian Vättö - Friday, July 15, 2011 - link

    I'm from Finland (GMT +3 now) but I don't know were Johan lives. FYI, Johan covered the power capping section so you have to wait for him if you think there are any changes necessary.

    I would suggest shooting him an email (click his name on the top) if you want to contact him. Not all of us read the comments, even though we are supposed to :)
    Reply
  • JohanAnandtech - Friday, July 15, 2011 - link

    I am located in Belgium. You know the land without a government but with the best beer in the world. Paradise thus ;-).

    Thanks for the valuable feedback. Are those 3 settings the only choices you have to tune your power draw?
    Reply
  • StevoLincolnite - Friday, July 15, 2011 - link

    Spelling error, Johan!
    "It will aslo be compatible with AMD's current San Marino and Adeleide platforms (Opteron 4000 Series) for socket C32."

    It should be Adelaide, not Adeleide. :)

    Other than that, good job.
    Hanging out badly for Zambezi's release, my rig is ready to drop a new 8-core chip into it. :)
    Reply
  • Kristian Vättö - Friday, July 15, 2011 - link

    Actually, I wrote that part. I have fixed it now along with a few other minor edits. I added that the latest word appears to be October release (just saw it in my RSS today) and I also added a (?) to the Komodo's IGP as it seems to be uncertain.

    Thanks for the feedback :)
    Reply

Log in

Don't have an account? Sign up now