And so it ends at 4GHz, again.



A 625MHz overclock is nothing to sneeze at, but it certainly does not come near those 6GHz~7GHz clock speeds we see on a regular basis with the Phenom II series. Granted, reaching those high clock speeds requires LN2 cooling but there is another important reason. We have discussed it several times and still do not have an acceptable answer from AMD about the inability of the Phenom II to clock much past 4GHz with a 64-bit operating system. Even with LN2 cooling we have not successfully benched past 4.4GHz with a 64-bit OS.

Once again, we tried XP 64-bit, Vista 64-bit, and Windows 7 64-bit and the results are always the same. As we near 4GHz, the voltage requirements increase dramatically and the clocking ability of the processor decreases in much the same manner. This does not occur in a 32-bit operating system, which happens to be the recommendation for any sort of benchmarking activities with the Phenom II.

That said, running this particular processor in the 3.8GHz~4GHz range offers performance that should satisfy just about any desktop user. Moreover, it is easy to do, in fact, much easier than our collection of 955BE samples scattered about the labs.


Windows 7 Ultimate x64 - Stock Voltage Overclock

The stock Core VDDC (VCore) is 1.4V as delivered from AMD. At this stock voltage and a slight bump in NB Core voltage to 1.250V, we hit 3.8GHz with a 2.6GHz Northbridge speed on our MSI 790FX-GD70 motherboard. The AMD specific 4GB DDR3-1600 kit from OCZ was set at 1.65V with 7-7-7-18 1T timings. We used the retail air cooler and load temperatures stayed around 58C~62C with Windows 7 Ultimate x64 as the operating system of choice.


Windows 7 Ultimate x86 - Stock Voltage Overclock

At the same voltages but utilizing Windows 7 Ultimate x86 we end at a final core speed of 4.066GHz (19x214). Northbridge speed is set to 2.567GHz and memory at DDR3-1426 with 6-7-6-20 timings. This is a similar pattern when testing the Phenom II cpus on air-cooling; we typically can clock the processor about 200MHz higher at similar voltages and settings when using a 32-bit OS in place of a 64-bit OS. Our particular sample made it to 4.2GHz on 1.50V with a NB speed of 2.4GHz using our Vigor Monsoon III LT air cooler. We almost hit 4.3GHz at 1.52V with a 2.7GHz NB speed on water cooling before temperatures spiraled out of control.



We swapped out our retail air cooler for the Corsair Hydro H50 to hit our top two overclocks. We reached 3.89GHz on 1.45V with the NB speed of 2.66GHz and memory running at DDR3-1640 on 7-7-7-18 timings. When running higher NB speeds, it is a requirement to keep the core as cool as possible to guarantee stability under load conditions. This was our preferred setting as temperatures stayed below 54C under load and the system was very responsive in a variety of applications, especially gaming.

Our best core clock reached 4.025GHz on 1.475V (shown at the top of the page) with a 2.76GHz NB speed and memory operating at DDR3-1532 on 6-7-6-20 timings. Temperatures crept back up to the 58C range under full load conditions.



The 4.025GHz setting offered the best results in several video/audio transcoding benchmarks due to pure processor speed, it was actually slower in gaming than the 3.89GHz setting thanks to improved memory bandwidth and internal latencies.

Power Consumption Final Words
POST A COMMENT

56 Comments

View All Comments

  • andrenb91 - Tuesday, August 18, 2009 - link

    amd's only hope to beat the i7s is the istambul core, if it brings istambul to the desktop market, I guess this future cpu can beat some high-end i7 processors, and after some revisions on the deneb core, amd will place it to ''fight'' the i5s leaving the athlon x4 playing against the i3s, but most denebs must be at 95W to be efficient against i5. of course this strategy depends if amd is economically capaple of putting a 300mm squared die in the desktop market...deneb is already too large to compete against the i7!
    Reply
  • Denithor - Friday, August 14, 2009 - link

    You measured performance in video encoding and then power consumption under the same test. Why not take the obvious next step to calculate performance/watt and post those results?

    And I was quite disappointed to see that you posted only about half of each CPU list on each of those charts - a few chips overlap but many do not so we cannot even do the calculation for ourselves except in less than half the cases.
    Reply
  • - Monday, August 17, 2009 - link


    Two things : Intel's SSE extentions are used by everyone, and should be the difference in some of these tests.



    http://www.anandtech.com/cpuchipsets/showdoc.aspx?...">http://www.anandtech.com/cpuchipsets/showdoc.aspx?...

    http://blogs.zdnet.com/Ou/?p=719">http://blogs.zdnet.com/Ou/?p=719

    The other disturbing thing is the FarCry benchmark, the writer:

    FarCry 2 is another example of a title well optimized for Intel's architectures and thus we see that the 965BE can't even win against its Q9550 competition. Thankfully for AMD, I do not believe FarCry 2 is representative of the majority of titles on the market.

    I believe this is an example of how SSE extentions deliver; but looking at the game benchmark data closer, we see that all cpu's are comparatively the same even the i7's vs Intel Core ll. Most, if not all vendors optimize in Intels favor

    asH
    Reply
  • - Tuesday, August 18, 2009 - link

    Note the extention differences between the two designs-

    Phenom ll X4 945
    Processor core Deneb
    Core stepping C2
    Manufacturing process 0.045 micron SOI
    758 million transistors
    Die size 243 mm2
    Data width 64 bit
    Number of cores 4
    Floating Point Unit Integrated
    Level 1 cache size ? 4 x 64 KB 2-way associative instruction caches
    4 x 64 KB 2-way associative data caches
    Level 2 cache size ? 4 x 512 KB 16-way associative caches
    Level 3 cache size 6 MB shared 48-way associative cache
    Virtual memory (TB) 256
    Features MMX
    3DNow!
    SSE
    SSE2
    SSE3
    SSE4a ?
    Advanced Bit Manipulation ?
    AMD64 technology ?
    AMD-V (virtualization) technology
    Enhanced Virus Protection ?

    Low power features Cool'n'Quiet 3.0
    CoolCore Technology ?
    Dual Dynamic Power Management ?
    Core C1 and C1E states
    Package S0, S1, S3, S4 and S5 states

    On-chip peripherals Integrated 144-bit DDR2 Memory Controller
    HyperTransport 3 technology

    -----------------------------------------

    Type CPU / Microprocessor
    Family Intel Core i7
    Model number ? I7-920
    CPU part number AT80601000741AA (Q1CM, Q1H7, SLBCH, SLBEJ)
    Box part numbers BX80601920 (SLBCH, SLBEJ)
    BXC80601920 (SLBCH, SLBEJ)
    Frequency (MHz) ? 2667
    Bus speed (MHz) ? 2400 MHz QPI
    Package 1366-land Flip-Chip Land Grid Array (FC-LGA8)
    Socket Socket 1366 (LGA1366)
    Introduction date Nov 17, 2008
    Price at introduction $284

    Architecture / Microarchitecture
    Processor core Bloomfield
    Core steppings C0 (SLBCH)
    D0 (Q1H7, SLBEJ)
    Manufacturing process 0.045 micron Hi-k metal gate technology
    731 million transistors
    Die size 263 mm2
    Data width 64 bit
    Number of cores 4
    Floating Point Unit Integrated
    Level 1 cache size ? 4 x 32 KB instruction caches
    4 x 32 KB data caches
    Level 2 cache size ? 4 x 256 KB
    Level 3 cache size Inclusive shared 8 MB cache
    Features MMX instruction set
    SSE
    SSE2
    SSE3
    Supplemental SSE3
    SSE4.1 ?
    SSE4.2 ?
    EM64T technology ?
    Hyper-Threading technology
    Turbo Boost technology ?
    Virtualization technology
    Execute Disable bit ?

    Low power features Thread C1, C3 and C6 states
    Core C1, C3 and C6 states
    Package C3 and C6 states
    SpeedStep technology ?

    On-chip peripherals Integrated triple-channel DDR3 SDRAM Memory controller
    Quick Path Interconnect



    Reply
  • - Tuesday, August 18, 2009 - link


    Type CPU / Microprocessor
    Family Intel Core 2 Quad
    Model number ? Q9650
    CPU part number AT80569PJ080N (QHGF, SLB8W)
    Box part numbers BX80569Q9650 (SLB8W)
    BXC80569Q9650 (SLB8W)
    Frequency (MHz) ? 3000
    Bus speed (MHz) ? 1333
    Clock multiplier ? 9
    Package 775-land Flip-Chip Land Grid Array (FC-LGA8)
    1.48" x 1.48" (3.75 cm x 3.75 cm)
    Socket Socket 775 (LGA775)
    Introduction date Aug 10, 2008
    Price at introduction $530

    Architecture / Microarchitecture
    Processor core Yorkfield
    Core stepping E0 (QHGF, SLB8W)
    Manufacturing process 0.045 micron
    Data width 64 bit
    Number of cores 4
    Floating Point Unit Integrated
    Level 1 cache size ? 4 x 32 KB instruction caches
    4 x 32 KB data caches
    Level 2 cache size ? 2 x 6 MB 12-way set associative caches (each L2 cache is shared between 2 cores)
    Features MMX instruction set
    SSE
    SSE2
    SSE3
    Supplemental SSE3
    EM64T technology ?
    Virtualization Technology
    Execute Disable Bit technology ?
    SSE4.1 ?
    Trusted Execution technology

    Low power features Enhanced SpeedStep technology ?
    Stop Grant state ?
    Halt state
    Extended Halt state
    Extended Stop Grant State
    Sleep state ?
    Deep Sleep state ?
    Deeper Sleep state ?


    http://www.cpu-world.com/CPUs/Core_2/Intel-Core%20...">http://www.cpu-world.com/CPUs/Core_2/In...BX80569Q...
    Reply
  • - Thursday, August 13, 2009 - link

    Intel's biggest (only?) advantage is hyperthreading; realize Windows 7 had to be optimized (how much more code?)for hyperthreading..how will Intel's i7's react in an openCL, CPUGPU environment (WARP) compared to Phenoms II's and an ATI graphics card, is it cost efficient(less code) and more efficient (faster) to go with CPUGPU over hyper..Will multicores do away with hyperthreading? These current comparisons on vista or XP do not necessary reflect comparisons on Windows 7 or DirectX 11. staytuned Reply
  • - Thursday, August 13, 2009 - link

    I would also love my excel spreadsheets to have the advantage of CPUGPU...Photoshop too Reply
  • - Thursday, August 13, 2009 - link

    oh yeah, Intels game plan to fight AMD's CPUGPU concept-

    a)license the SLI technology from Nvidia for Nehelam
    b)get Microsoft to optimise Windows 7 for hyperthreading (sidebar-Intel pushes Windows 7 for corporate upgrades- can you say payoff))

    innovative genius

    but in reality they will probably make sure this great technological concept dies, thereby assuring comp's remain in the dark ages for another 10 years
    Reply
  • ash9 - Thursday, August 13, 2009 - link

    "Now once you start throwing in background tasks and look at future titles being more threaded then the picture becomes a little more muddy"

    I dont understand where the writer is going with these conclusions. As CPUGPU or OpenCL begins to take hold, the old comparative model of simply looking at raw speed becomes obsolete, now, overall power can be reduced while concurrent events run parallel in multicores and GPU, thats is where AMD is heading. These comparisons with Vista may not be as eye opening as compared on Windows 7 or DirectX 11, this is where AMD planed to rock and roll from the start.
    Reply
  • GourdFreeMan - Thursday, August 13, 2009 - link

    You are looking too far forward into the future for a product that is going to be used by consumers for the next 2-3 years. Yes, there is a general move among all vendors (AMD, nVIDIA and Intel) towards moving largely parallel computation onto GPUs, however you must keep in mind CUDA has been around for 2.5 years and Brook even longer than that, yet GPGPU has only found commercial consumer application in media encoding and Adobe Photoshop. To expect a sudden shift when DX11 is released as a commercial product as part of Windows 7 later this year is wishful thinking at best. Programmers require time to learn new skills, adopt new methodologies and experiment to determine what works more efficiently and what doesn't. That time will be measured in years, not months.

    If we restrict ourselves to the domain of PC gaming, then Anand's comments are accurate, if a little dated. Programmers have finally adopted multithreaded (CPU) development with a vengeance. Most new games are multithreaded, regardless of whether they are a console port or the rare instance of the dying breed of PC exclusives. The first dual core consumer CPUs were released more than 4 years ago. That should give you some idea of how agile PC games development actually is. We still don't have many titles with 64-bit executables despite how long ago 64-bit CPUs and operating systems were released.

    Finally, there remain open questions of where tasks belong and what to do with unused processing capability on the CPU and GPU. You can put physics on the GPU, but what about AI? Game logic? Branchy scripts written by content creators as opposed to programmers? You can use your unused GPU cycles to do more graphically or at a higher frame rate, but can't you also find more tasks for your CPU that will contribute to gameplay? World simulation and more complicated AI immediately spring to mind.
    Reply

Log in

Don't have an account? Sign up now