Some model that we can found here rightnow :

1. X3-710/X3-720 BE “Heka” Core with Triple core + 6Mb L3 cache and AM3/AM2+ compatible

2. X4-805**/X4-810 “Deneb” Core with Quadcore + 4Mb L3 cache and AM3/AM2+ compatible

3. X4-920/X4-940 BE “Deneb” Core with Quadcore + 6Mb L3 cache and only AM2+ compatible

4. X4-910**/X4-925**/X4-945**/X4-955 BE “Deneb” Core with Quadcore + 6Mb L3 cache and AM3/AM2+ compatible

** –> These series is not retail product and use only for OEM partner

AM3 platform is using and sporting dual channel DDR3 until 1333Mhz and for 1600Mhz, 1866Mhz, and up usually can only achieve with overclocking method. For backward supporting with older socket such AM2 and AM2+ AMD capability to using dual memory controller (IMC) on their processor. So any user with easy can upgrade what their choice for motherboard platform such AM2+ with DDR2 or AM3 with DDR3 support

Specification Test :

AMD Phenom II X3 720 BE cooled by TRUX 120 Extreme + 12cm TT Smart case fan.

MSI 790FX - GD70 AM3 -- > thanks to MSI Indonesia

PDP Patriot DDR3 2×2GB PC8500 with D9-GTR based IC

WD Blue 320 AAKS

Gainward HD 4850 Golden Sample

PC Power & Cooling 750W

AOC 931FWZ

ATI Catalyst Suite 9.5

Windows Vista Ultimate x64 SP2 with latest update

Software Test :

EVEREST-Ultimate-Edition 5.0

3D Mark Vantage v1.1

Cinebench R10 x64

World in Conflict

FarCry 2

Super PI Mod XS 1.5

Intel Linpack Stability 2.1

Hyper PI

In short time, there is 5 main parameter that mostly as major effect for balancing in the “Dragon” platform :

1. BIOS or Basic Input Output System, these is look like a soul for any motherboard. Without good support from motherboard manufacture and then a good motherboard with high quality component is looking just wasting cause the limited tweaking ability. Even now that so many software to help us  like AOD, Asus Tuner, etc. They tweak easy from the windows, actually are directly connecting to BIOS by translating a hex code to change parameter from installed hardware.

Some basic BIOS parameter that you must optimizing at beginning :

a. Disable any motherboard component that not been installed or we are not using it, like : floppy disk, paralel or serial port etc. That mean for saving more power in the circuit and eliminate bug if found.
b. Disable more for feature processor that we do not using such SVM or Secure Virtual Machine (Pasifica), Cool & Quite, CPU Micro update, CIE, etc. These means for maximizing a power processor itself.
c. My recomended use AHCI mode before installing operating system especially for Vista OS. That means is modern harddrive feature “Native Command Quering”  can working for minimize seek time while loading file system later.

CPU VDD or Vcore –> it’s your processor VID voltage control
CPU NB VDD — > allow your control voltage of CPU Northbridge or IMC controller inside processor. Juice with small step cause can make a lot of heat. I found in 1.4v can flying IMC to 2.8Ghz
CPU Voltage & CPU-NB Voltage –> Offset voltage for CPU VDD & CPU-NB VDD if you take for > 1.55v
CPU PLL –> CPU Phase Locked Loop, it’s your CPU clockgen voltage control.
CPU DDR-PHY –> Clockgen for CPU-NB, add in tiny step for test stability in IMC overclocking
DRAM Voltage –> it’s your DDR3 DRAM vdimm, default value at 1.6v
NB Voltage — > 790FX NB voltage. I have found that 1.2 - 1.25v very help in stability at full load overclock test
NB PCI-E and I/O Voltage –> Help system stability when you play with multi GPU environment more than 2 with PCIE clock more than 120mhz.
HT Link Voltage –> Hyper transport voltage control. In small step can make HT Link speed up to 2.6Ghz.

2. Processor clock speed, that is the main point here for optimizing. More speed means  making your data process run faster but the limit depend on your  processor overclock ability. Some good CPU can do 3.9 - 4.1ghz in moderate vcore but they commonly average  do 3.6Ghz until 3.9Ghz  in  good air cooling. You can increase that with two way, FSB BUS speed or your multi processor if your using Black Edition type. Increasing CPU VDD from default AMD specification can make help stability but must be anticipate with good 3th party cooling. In this case Black Edition processor is more easy cause you only need increase multiplier in 0.5 step or 100mhz in real speed if my assumtion your FSB still 200Mhz. Test your system right away with Hyper PI in 4M or 8M calculation or you can using 10 loops full stress Intel Linpack stability for final test. If passed and your CPU temperature still below AMD specification max case temp please overclock more…..

For non BE processor (lockup multi) the only one to know your max CPU are make isolation on that processor. Lowering your HT Link speed,CPU-NB clock, your RAM divider and latency.

Ex. with X3 710 + DDR2 800 :
BUS : 200
Multi : x 13
HT Link & CPU NB Clock : x 8 (default = x10)
Divider ram : 1:1.33 (DDR 533) with CL 5-5-5-x or 6-6-6-x

With above example we can eliminate your motherboard and memory from holding up your processor speed. Increase BUS with 3 or 5 Mhz step while your testing it. ACC or Advance Clock Calibration is an exclusive feature that have in SB 750 motherboard only. AMD itself  did not give a brief explanation about them, but from several source that i read ACC can can make stability with VID and frequency on the weaker core. I self not in deep investigating about that, look like ACC will working optimum in extreme cooling or subzero environment. AMD famous overclocker “Machi” form XS says that you must set from “zero” for any core and add positive value at weaker core.

I prefer to like using Intel Linpack stability for testing cause only wait for about 20 minute (10 loops continues) to know stability my system to passed 64bit Vista environment

3. Integrated Memory Controller (IMC Clock) or CPU-NB clock. AMD had introduce since release first Phenom which they seperating between cpu core with memory controller itself with independent multiplier for solving speed problem on first generation Phenom. Like AMD Phenom AM3 X3-710 have CPU-NB clock 2ghz with 10x multi or X4-9950BE and X4 940BE AM2+ have 1.8Ghz. When you increase memory divider to 1066Mhz you will gain only small performance but if you increase CPU-NB clock let says 2.2Ghz with 800Mhz DDR yuu will gain more better performance than 1066Mhz DDR only.  Of course if you increase CPU-NB to 2.2Ghz and popping up DDR to 1066Mhz will gain a optimal performance with your bandwidth and latency. The main point here is memory efficiency directly connected with CPU-NB cause the default CPU-NB is not efficient. With increasing 400mhz CPU-NB and  keep your DDR at 800Mhz you will gain extra performance like you have running 1200Mhz DDR

The more speed you can achieve DDR clock you need CPU-NB higher too. So you don’t need to try to increasing DDR clock up 2Ghz with your CPU-NB keep at 2Ghz… you only wasting your time. This is can applied on CPU with DDR2 or DDR3, the only different are DDR2 CPU only achieve about 2.6Ghz max and DDR3 CPU is easy to achieve 3Ghz with good air or water cooling system.

Allright, rightnow i will give diferent performance effect between 2Ghz CPU-NB and 2.6Ghz on some sensitive memory application. On 2.8Ghz CPU-NB you need more juice higher while the performance itself almost didn’t give advantage than 2.6Ghz, maybe i must crank up my DDR3 speed around 1.800Mhz too. My analysis with 3.6Ghz to 4Ghz CPU are enough with 2.6Ghz CPU-NB and the most importannt is safety for daily used. My CPU clock at 3.72Ghz (200×18.5) and my DDR3 DRAM running at almost every DDR3 ram can achieve is 1600Mhz CL8

2.6Ghz CPU-NB clock faster in synthetic benchmark but how are they perform in real application ?

4. Hyper Transport/HT Link speed. Built and developing by Hyper Transport Consortium and have been using by several big microchip company such as AMD, x86 Transmeta, Broadcom, etc. Today they is up to 3.1 version with data speed transfer 3.2Ghz or data transfer speed by twice calculation 6.4MT/s. The Main function is act like communication data path between CPU with Nortbridge to connecting with all connected I/O peripheral, storage device, PCIE express line, etc.

HyperTransport packets enter the interconnect in segments known as bit times. The number of bit times required depends on the link width. HyperTransport also supports system management messaging, signaling interrupts, issuing probes to adjacent devices or processors, I/O transactions, and general data transactions. There are two kinds of write commands supported - posted and non-posted. Posted writes do not require a response from the target. This is usually used for high bandwidth devices such as Uniform Memory Access traffic or Direct memory access transfers. Non-posted writes require a response from the receiver in the form of a “target done”. Reads also cause the receiver to generate a read response. HyperTransport supports the PCI consumer-producer ordering model

By default X3-720 HT Link speed is 2.0Ghz or 4.0MT/s. For these setting i found that 2.6Ghz HT link speed only need  small juice HT voltage to make them stabilize. 2.4Ghz - 2.6Ghz is optimal setting for me.

5. DDR/RAM, Like i said above that they have compatibility which AM3 procie can running 2 DDR SDRAM generation module. DDR2 with AM2/AM2+ mobo or DDR3 with AM3 mobo. Their native frecuency are DDR2 400,533,667,800,1066 for AM2+ and DDR3 800,1066,1333 for AM3. They price begin dive and more cheap than last year for supporting upcoming platform from both side like Intel Nehalem or AMD AM3 of course. You can buy 4Gb kit Patriot DDR3 PC8500 value chip only $70 (in my country). And if you luck you can get a Micron D9 based IC that they are capable to running PC14000 with easy. For advance user, MSI give us advance memory cell menu setting to give advance tweak opportunity like data drive strenght, alpha timing for adjustment stability. For beginner or basic ram user we can begin with filling main ram option like tCL - tRCD - tRP - tRAS, Command rate (tCR) and tRFC with SPD Dram spesification.

My Patriot have SPD rate timing like inside AMD overdrive information. At these point we can know default value for our ram timing to fill in BIOS later. We can ignoring or set AUTO for the other timing than tCL – tRCD – tRP- tRAS, Command Rate and tRFC. In these section i will give different performance between DDR 3 1066 CL6, 1333 CL7 and 1600 CL8. For testing i will clock up my X3-720 to 3.7Ghz, 2.6Ghz CPU-NB and 2.4Ghz HT Link speed. AMD give us 2 option for memory controller choice, they are Ganged and Unganged. On Ganged mode they are combining two channel into 128 bit single one big fat channel. On Unganged mode they are separating two channel controller into two 64bit mode that can be independent accessing. Unganged mode is faster than ganged but more difficult to achieve high clock speed. The reason maybe a limit factor in IMC itself.

Welldone, from the resulting test we can know that DDR3 1600 CL8 only give small advantage on synthetic benchmark only. In real application i doesn’t see differential performance with DDR3 1333 CL7. Cause until i write these articles AMD didn’t give us limit for using DDR3 vDimm on their AM3 “Dragon” platform, i can say that these Phenom II generation with 45nm process production is optimize enough with DDR3 1333 moreover if your ram capable running in CL 6.

Last Word…

Phenom II is more phenomenal that first generation Phenom. They can do about same performance clock by clock with Intel Core 2 Duo/Quad in some benchmark. I only running triple core X3 720 @3.7Ghz, 2.6Ghz CPU-NB, 2.4Ghz HT Link and DDR3 1333 CL6 but the system running blazing fast, more faster than my last overclockable Intel C2D E8400 @4.2Ghz on multithread application of course. Thanks for reading.....

Artikel berbahasa Indonesia bisa anda download di amd-phenom-ii-oc-guide.pdf [1.5 Mb] (Count: 394)