The Tech Repository Forums  

Go Back   The Tech Repository Forums > The Tech Repository > Guides and Articles
FAQ Members List Calendar Search Today's Posts Mark Forums Read

Reply
 
Thread Tools Search this Thread Display Modes
Old April 5th, 2008, 10:37 PM   #1
Tony
Old Jedi master
 
Tony's Avatar
 
Join Date: Sep 2006
Location: Manchester, UK
AMD Phenom - the B3 Stepping finally breaks cover and does not disappoint

Part 1: First look and general guide.

Ever since AMD decided to release Phenom we have been waiting for clock speeds to rise, along with CPU's that had faster memory controller clock speeds.
What follows is my experience with the new 9850BE, a shop bought example I might add and so in no way a cherry picked example.



Phenoms initial release featured Phenom 9500 and 9600 along with a 9600Black Edition unlocked version, clock speeds at 2.2 and 2.3GHZ with most over clocked examples hitting 2.6 thru 2.8GHZ and NorthBridge memory controller clocks limited to 1.8GHZ. BE (Black Edition) did allow for some flexibility but overall the performance was not where most enthusiasts were expecting and issues focused around NB over clocking often left over clockers with a somewhat foul taste, plus the constant talk of the L3 cache issues/TLB errata caused sales really to look to be slow with the enthusiast community...most just bought an Intel Q6600 and had done with it.

Now we have B3 stepping, no TLB errata, NB running at 2GHZ and no bios files seeming to lock this speed down hard, with the BE version any way ' so could this be the CPU the AMD enthusiast has been waiting for?...read on!

I have no preference with motherboards, tests here will be conducted using the Gigabyte 790FX DQ6 or the DFI 790FX M2R, or a combination depending on how flexible I need the motherboard to be. The DQ6 has shown a willingness to over clock the HTT bus a little easier than the DFI but both boards managed to run the 9850 at 3.4GHZ, so really there is nothing to choose between them. DFI do allow more flexible voltage manipulation and the ability for force any CPU multiplier or multi divider as well as any NB multiplier and divider we choose, saying that the DQ6 comes very close and both have all the memory tweaks and setting you will need to push the system memory to a new high.


DFI's excellent LP UT 790FX-M2R Mother board, click the pic for the product page.


Gigabyte GA-MA790FX-DQ6, an equal match to the DFI although lacking a few of the more trick bios settings.

The basic differences between K8 and K10 that matter to the enthusiast

K8 AM2 brought us dual core, either 90nm or 65nm CPU's with 2Mb(2x1024k) or 1MB(2x512k) of L2 cache and no L3 cache. DDR2 was the memory of choice with clock speeds officially limited to 400, 533, 667 or 800MHZ. Due to the way the memory ratios were created along with the CPU multiplier and clock speeds it was often the case DDR800 would not run at 800MHZ with some of the CPU's at stock speeds, often end users would see 750MHZ as AMD had programmed the NB to NEVER over clock the system memory but to under clock it if the algorithm dictated it...this has all changed with Phenom

Also K8 featured a 128bit memory controller, this means with the ram in dual channel mode the memory would be seen by the CPU as 1 huge 128bit wide module, again this has changed on Phenom with either 128bit or 2x64bit modes being selectable. this was something I first saw on old dual channel SiS chipsets for Intels Northwood CPU's where 2x64Bit always seemed faster but created less synthetic bandwidth, the same holds tru with Phenom, 128bit will get the the most synthetic bandwidth but the 2x64bit mode will prove the fastest overall on your system.

Bios either shows ganged or unganged modes:
Ganged =128bit
Unganged = 2x64bit

Many applications still show unganged as single channel, this is incorrect. both are dual channel so please don't be fooled.

So what are the basic specs for Phenom?

AMD lists its official specs for Phenom here, lets break a few key parts down so you can see what this CPU has to offer.

1 Direct Connect Architecture, providing direct and rapid information flow between processor cores, main memory, and graphics and video accelerators.

2 Low latency access to main memory, (same for K8 as the NB is on the CPU die)

3 True quad-core design.K8 was only Dual core max.

4 Integrated memory controller and shared L3 cache(Now we see something different to K8)

5 HyperTransport™ 3.0 technology (K8 officially topped out at 1000MHZ, K10 runs substantially faster than this)

6 Cool and Quiet Technology. most enthusiasts turn this feature off so I won't touch on this during the article.

So How does K10 Phenom work that different to K8 Athlon?

Althon bas a reasonably basic set of rules. These are:

1 Hyper transport HT operates at a max frequency of 1000MHZ(officially) and has multipliers of 1 2 3 4 5 multiplied by the base HTT bus speed (200MHZ)

2 CPU clock speed is derived from the base HTT bus speed (200MHZ) multiplied by the CPU multiplier (partially locked for none FX or some Black Edition CPU's) Example...4000+ has a base CPU Multiplier of 10 so gave an overall clock speed of 2000MHZ (10x200MHZ base HTT bus)
There is no FSB on AMD CPU's, this is an Intel bus architecture. AMD often call the base bus frequency HTT and overall bus frequency Hyper Transport or HT.

3 The NB clock speed = CPU Clock speed, so if you have a 6400+ BE Athlon which is rated at 3200MHZ the NB is also running at this frequency.

Regarding rule 1 HT speed, it was well known most good K8 CPU's would run 1500MHZ on the HT bus. Many overclockers would either lower the HT multiplier to 4 but then push up the HTT bus speed, keeping the HT overall speed between 1000MHZ and 1500MHZ.

Regarding rule 2 CPU speed, unless you have an FX or BE cpu the ONLY way to push up the overall CPU clock speed is with HTT Bus over clocking, many will set 250MHZ+ so pushing up the CPU overall clock speed by 25%+.

So a break down of a typical over clocked Athlon 6000+ would look like this in regard to bus speeds. Forget the 3dmark scores, lets break down what CPU-Z is telling us.



Under the clocks section I have highlighted the important info you need, this being:

Overall CPU core speed is 3300MHZ
CPU multiplier applied is 12
HTT Bus speed is 275MHZ
HT Link speed is 1375MHZ

So how does this break down

CPU speed is easy,CPU Multiplier X HTT base Bus speed..so 12x275=3300MHZ
HT Link speed = HTT X HT multiplier, but CPU-Z does not give us the HT multi so lets rework the info we do have so show what HT multiplier I had applied.

HT multiplier = HT Link speed / HTT Bus:
so 1375MHZ / 275 = 5

HT Multiplier was 5

The last part of the equation with regard to Athlon over clocking is the memory, as mentioned often due to Athlon architecture memory speeds would be lower than expected at stock CPU speed. This is wholly dependent on CPU multi applied, with some CPU's having DDR750 applied instead of DDR800 at stock speed, or 520MHZ instead of 533MHZ for lower clocked modules. As ever as you push the HTT bus higher the memory speed would rise also as it also is based off this HTT base bus of 200MHZ.

Remember, NB core clock frequency on Athlon is = core frequency the CPU is running at.


So when over clocking Athlon you had to remember the following

1 Do not push overall HT speed to high

2 Do not push memory speed to high and watch for slight down clocks due to CPU multiplier used.

So overall life was quite easy with Athlon...but now for Phenom.

Phenom is much like Athlon is that is uses the same 200MHZ HTT bus frequency, it also has an overall HT frequency although this can run a lot faster than the "official max of 1000MHZ" on Athlon.

So lets break the HT section down on Phenom:

Phenom uses the same base HTT bus clock of 200MHZ BUT its base HT speed is 2000MHZ for the B3CPU's such as 9850 or 1800MHZ for the lower spec B2 Phenoms or near 2X what we saw on Athlon. It also has multipliers available to push this HT bus to 2600MHZ.

Multipliers available are 1 2 3 4 5 6 7 8 9 10 11 12 13

Most bios files show this as actual bus speeds, so you may see the following reported. 200 400 600...>...1400 1600 1800...>..2400 2600

Try to keep in mind that they are just multipliers, if you set 1800MHZ it is NOT FIXED at this speed if you are adjusting the HTT bus for over clocking. The overall speed will be 9 (for the 1800MHZ setting) X HTT bus speed applied.

Next under the microscope is the Memory Controller or NorthBridge which the same as Athlon is integrated onto the CPU die.

On Athlon the NB multiplier and bus speed were directly linked to the CPU multiplier and HTT bus speed, hence NB frequency = CPU frequency

THIS IS NOT THE SAME ON PHENOM

We now have total control over the NB multiplier, bus speed is still the same HTT bus so as you increase the CPU clock speed you are also increasing the NB core clock speed UNLESS you lower the multiplier applied to it.

There is also a very specific rule you must abide by:

HT speed can be equal or less than NB core clock speed but must NEVER exceed it!

This is totally independent to the CPU core speed so this is something new we have to be aware of.

How is NB core clock applied?


With Phenom we have the ability so set both NB multiplier and the Divider used on this multiplier. Confusing i know...let me explain.

In AMD's documentation they quote NB FID and NB DID

FID = Multiplier
DID = Multiplier divider

Some bios files list the ratio's (FID) in Hex, confusing i know so here are some of the more popular options.

Hex DEC
00 = 4
01 = 5
02 = 6
03 = 7
04 = 8
05 = 9
06 = 10
07 = 11
08 = 12
09 = 13
0A = 14
0B = 15
0C = 16
0D = 17

Now we are also able to set the NB DID or Multiplier divider. For the NB this is either 1 or 2....1 is a little useless I know but is the "By AUTO" setting, 2 allows us to set half multipliers in some combinations.

So an example would be:

07 HEX for a FID Multiplier of 11
DID = 2

Would give an overall NB multiplier of 5.5 as 11/2=5.5

Now remember B2 stepping Phenom has a default NB core speed of 1800MHZ using 05 Hex for a multiplier of 9 with DID set to AUTO=1
B3 has a default NB speed of 2000MHZ so we have 06 HEX for a multiplier of 10 with DID set to AUTO=1

Moving on we need to look at CPU multipliers, again these are a little different to Athlon.

Phenom has NB FID and DID but it also has CPU FID and DID acting in exactly the same way. The options are different but the rules remain the same. We have either direct options for FID from bios IE 8, 8.5, 9, 9.5 etc or we have the HEX settings. We also have DID which allows dividers of 1 2 4 8 16, which again has 1 as the by AUTO setting but you can also set 2 4 8 16 to get some quite different multipliers over what was possible with Athlon.

Now...not all manufacturers have all the DID options programmed in their bios and working, some have chosen to give you fixed option such as full and half multi's, DFI allow you to program what you like although some look to not be working.

An example of how different these option are over Athlon is the ability to set quarter multipliers.

To start lets look at what multipliers are available, I mapped out the more popular options from HEX to DEC, this is far from the available options on an unlocked BE edition Phenom and is just a guide.

HEX Multi
00 = 08
01 = 08.5
02 = 09
03 = 09.5
04 = 10
05 = 10.5
06 = 11
07 = 11.5
08 = 12
09 = 12.5
0A = 13
0B = 13.5
0C = 14 ETC

Things look to get good from 11 onwards using the 2 4 8 or 16 dividers and ODD ratio's

Setting 07 FID and 2 DID will give 5.75 as 07HEX =11.5 and 11.5/2=5.75 so it seems you could have 1/4 and 3/4 multipliers as well as 1/2.

Just remember some board manufacturers have limited bios to full and 1/2 multi's, but the option could be there if they wished to add it.


So a brief recap on Phenom clock options are as following:

HT speeds from 200 to 2600 in 200MHZ steps using 1 2 3 5 6 7 8 9 10 11 12 13 multipliers from bios.

NB core speeds completely unlocked starting from 4X200 for 800MHZ rising in full and half multiplier options multiplied by the HTT bus.

CPU core frequency adjustable by using both HTT bus AND FID (multipliers) in full, full+1/4, full+1/2 and full+3/4 options as long as bios has the hooks enabled to alter CPU settings.


A brief breakdown of some Phenom settings, this was a run By Macci who works for AMD...I am sure he won't mind me using his pic



I highlighted the important info in red.

NB core speed 2500MHZ, NB Multi 10X, HTT speed 250MHZ

Bios values.

NB FID HEX 06 =10
NB DID 01

So HTT speed =250x10 for NB core clock of 2500MHZ

Moving to CPU core.

CPU core speed 3000MHZ, multi 12, HTT 250MHZ

CPU FID HEX 08 = 12
CPU DID 01 = 1 so NO change to CPU FID as 12/1=12

CPU core speed 250HTT x 12 for 3000MHZ

Moving to HT bus speed.Approx 2500MHZ

HT multi X HTT= HT speed
10 x 250HTT=2500MHZ

Options for HT multi:

200=1
400=2
600=3
800=4
1000=5
1200=6
1400=7
1600=8
1800=9
2000=10
2200=11
2400=12
2600=13


As you see Phenom, is actually very flexible with regard to how you set it up. Last on our list is the memory option and how they differ to Athlon.

Phenom memory options

Athlon as mentioned in some circumstances would down clock the memory dependent on what CPU multiplier was used, it also had options from 400MHZ to 800MHZ only, this is where Phenom changes the rules.

Phenom sets exactly the memory speed you have chosen, IE if you choose 800 it sets 800MHZ with any CPU multiplier. Also we have an extra option for memory speed directly from bios, this being 1066MHZ.

Now there are rules associated to the 1066 setting, I will try to explain.

AMD have built onto Phenoms memory controller the ability to set 1066MHZ directly. It can do this 2 ways.

A memory modules SPD has 3 speed tables, these are linked to CAS latency also and dictate overall what setting and what speed the module will run at. Usually this table stopped at 800MHZ...with Phenom it can stop at 1066MHZ.

So...for those who like to test SPD settings the Byte you are looking for is Byte No9, if this is changed to HEX17 the module will force an initial boot on Phenom with the memory at 1066 MHZ; be careful though other timings need to be changed in order to allow a clean boot and operation, so just setting 1066 may NOT be enough
You can of course just boot with a standard SPD and force 1066 mode thru bios, this is the method most end users will choose.

Now either if you boot with 1066 SPD modules or you choose 1066 mode the board upon a reboot will enter a special 1066 mode. This alters many of the memory timings and the options for those timings in bios.
Cas latency timings are usually unaffected by this, but TRP, TRCD and TRAS all have higher minimum values allowed and higher max values allowed. TRC also has its upper limit greatly increased.

For non 1066 mode, IE 800 mode, TRP has a low limit of 3, with 1066 mode enabled TRP has a low limit of 5, as does TRCD, so be sure to study the bios on the board you use and make notes of the low limits for timings and the available max in 800 and 1066 modes. It may prove better for you to push HTT bus speed and leave the ram in 800 mode to allow tighter timings to be chosen over going straight to 1066 mode where the timings have a loser minimum setting.

So while Athlon had 2 basic rules you had to address, Phenom has 4.

This is the MOST IMPORTANT RULE

1 HT overall speed can be equal to or less than NB core clock speed but must NOT be greater.


2 If you choose 1066 mode the boards memory timing defaults do change, if you have 1066 SPD's modules you will enter this mode by default from a cmos clear.

3 Both CPU and NB control are unlinked with regards to multipliers but linked with regard to base HTT bus speed.

4 Voltage options to the CPU core and NB core are independent on many boards, you have to make sure both are stable so tweaking BOTH voltages may be necessary.

The one common item for both Phenom and Athlon is base HTT bus speed, near everything is derived from it and it is the same base speed for both IE 200MHZ.

CPU speed Athlon= HTT Bus speed x CPU Multi
CPU Speed Phenom = HTT Bus X (CPU FID/CPU DID) or HTT X (CPU multiplier/CPU Multi Divider)

NB core clock for Athlon = HTT Bus speed X CPU Multi
NB core clock for Phenom = HTT Bus speed X (NB FID/NB DID) or HTT X (NB multiplier/NB Multi divider)

HT bus (overall) for Athlon = HTT Bus X HT multiplier ( 1 thru 5)
HT bus (overall) For Phenom= HTT bus x HT multiplier (1 thru 13)

Memory option for Athlon = 400, 533, 667 and 800
memory option for Phenom = 400, 533, 667,800 and 1066

memory speed for both CPU's = HTT Bus X (Applied memory ratio)

I have highlighted HTT Bus, as you see it controls every clock on both Athlon and Phenom.

This concludes part 1 of our look at Phemon, part 2 will involve a little further indepth look at memory setup, 1066 mode and some performance testing.

feel free to discuss anything Phenom here rules on TTR strictly prohibit bashing and fanboy'ism...
Attached Thumbnails
Click image for larger version

Name:	motherboard_productimage_ga-ma790fx-dq6_big.jpg
Views:	14025
Size:	43.5 KB
ID:	623  Click image for larger version

Name:	Phenom breakdown.PNG
Views:	14582
Size:	85.8 KB
ID:	624  
Attached Images
 
Tony is offline   Reply With Quote
Old December 9th, 2008, 07:27 AM   #2
Tony
Old Jedi master
 
Tony's Avatar
 
Join Date: Sep 2006
Location: Manchester, UK
Phenom 2 just round the corner

Seems AMD has a new weapon for us to play with...a mk2 which looks to be near a total redesign.

Stay tuned for more Phenom2 news and hopefully some hands on testing.

Tony
Tony is offline   Reply With Quote
Reply


Currently Active Users Viewing This Thread: 1 (0 members and 1 guests)
 
Thread Tools Search this Thread
Search this Thread:

Advanced Search
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

vB code is On
Smilies are On
[IMG] code is On
HTML code is On
Forum Jump


All times are GMT -4. The time now is 08:50 PM.


Powered by vBulletin® Version 3.6.8
Copyright ©2000 - 2014, Jelsoft Enterprises Ltd.
The Tech Repository