Phenom 9500 w/ MSI K9A2 Platinum

[KTE]

Warning: Verbose post ahead.

I wanted to quickly explain AM2+ - 790FX overclocking.

NOTE: The below method worked for BIOS v1.1, P0C, P0D and 113 BETA. Putting an NB multiplier higher than 9x made the system start downclocking in 0.5x multiplers the CPU and HT frequency.

I told a while ago of the HT ref. limit of ~265-275MHz max on Phenom 9500 so far which limits us to a maximum of 2.9-3.0GHz screenable and still very very unstable. I've been messing about with AM2+ P-States for 2-3 days now. I haven't seen others do it yet here and I got the feeling very few actually understand it, but I did post a 3GHz NB with 2.1GHz HT link speed 2 dys ago and no one really caught on to how or what was happening.

I wanted to explain it for those that are unsure but I needed some more testing time and different BIOS to be able to experiment more and see if it makes a difference. Then the last of my queries was answered my Macci in a thread yesterday (CPU VID): make sure you read it!

Here is how AM2+ P-States work.

790FX BIOS give you extra options to control:

CPU FID
CPU VID
CPU DID

NB FID
NB DID
NB VID

=Frequency ID
=Voltage ID
=Divider ID

Plus you can control:

HT speed multi
HT reference speed

What FID does is decide your clock frequency multiplier.

So for instance, CPU FID 02 equals 9x CPU multiplier (03 is 9.5x, 04 is 10x and so on).

BUT changing FID alone will not be equal to changing the CPU multiplier. There is still part of the equation missing.

There is CPU DID which is the Divider for the FID value.

So, you have options of 1, 2, 4, 8 and so on for CPU DID. What they do is divide the end CPU FID value, and then the HT reference speed is multiplied by it to gain the CPU speed.

Hence, if we input CPU FID value 08 which stands for CPU multiplier 12x, and we choose CPU DID value 1 (which is what its says), then:

FID / DID = CPU Multiplier
12 / 1 = 12x multiplier

If we chose CPU DID 2 then the end CPU multiplier would've been 6x.

So the HT reference clock speed you set (say 200MHz as an example) will multiply x12 to get 2400MHz CPU speed now.

HOWEVER the 9500 I have is multipler locked obviously. So whatever FID/VID value you choose above what gets you 11x multiplier will just result in no change from stock 11x multipler.

CPU VID is the Voltage identity which I did't fully understand till Macci replied and said to leave it at default which is value 28 for 9500 and stands for 1.250V. If you increase it to value 29 you get 1.1875V with my testing.

BUT if you increase the voltage (i.e. decrease the value to 27), you will fail POST.

Now we know how CPU speed is worked out.

Additionally, NB FID, DID, VID are of the same task but for the NB instead and NB VID can be changed to higher. Value 36 is stock which for me was 1.1V.

NB FID values are as so: 00 is 5x NB speed multiplier, 01 is 6x, 02 is 7x and so on.

For NB DID you get two options of divider 1 and divider 2.

What that NB FID does is, get the HT reference clock you set (say 200MHz) and multiply that by the NB multiplier you choose through NB FID / NB DID.

So say you choose NB FID of value 00 which is 5x multi and choose NB DID of value 1 (which is again what it says).

NB FID / NB DID = NB Multiplier
In the above case we have 5 / 1 = 5x multiplier

With a 200MHz HT ref. speed, that would give us 200x5 = 1000MHz NB speed.

BUT=> NB clock speed can only be equal or higher than HT speed. Keep that in mind.

If it isn't, then either you'll have no boot or the system will downclock itself lower than stock. If you set a high NB clock, your system may not boot at all or downclock the NB/CPU/HT to make it bootable.

Now we know how the NB speed is worked out.

Easy as that! Any qs, please ask. I've tried to make it simple but you'll have to give it a go before understanding properly.

Here was my older run using this method I posted within those images

[informal]

Nice work man!IMO,this should be stickied ASAP

[JumpingJack]

Nice work man!IMO,this should be stickied ASAP

Agreed. This is just too much good info.

[justapost]

Warning: Verbose post ahead.

I wanted to quickly explain AM2+ - 790FX overclocking.

NOTE: The below method worked for BIOS v1.1, P0C, P0D and 113 BETA. Putting an NB multiplier higher than 9x made the system start downclocking in 0.5x multiplers the CPU and HT frequency.

I told a while ago of the HT ref. limit of ~265-275MHz max on Phenom 9500 so far which limits us to a maximum of 2.9-3.0GHz screenable and still very very unstable. I've been messing about with AM2+ P-States for 2-3 days now. I haven't seen others do it yet here and I got the feeling very few actually understand it, but I did post a 3GHz NB with 2.1GHz HT link speed 2 dys ago and no one really caught on to how or what was happening.

I wanted to explain it for those that are unsure but I needed some more testing time and different BIOS to be able to experiment more and see if it makes a difference. Then the last of my queries was answered my Macci in a thread yesterday (CPU VID): make sure you read it!

Here is how AM2+ P-States work.

790FX BIOS give you extra options to control:

CPU FID
CPU VID
CPU DID

NB FID
NB DID
NB VID

=Frequency ID
=Voltage ID
=Divider ID

Plus you can control:

HT speed multi
HT reference speed

What FID does is decide your clock frequency multiplier.

So for instance, CPU FID 02 equals 9x CPU multiplier (03 is 9.5x, 04 is 10x and so on).

BUT changing FID alone will not be equal to changing the CPU multiplier. There is still part of the equation missing.

There is CPU DID which is the Divider for the FID value.

So, you have options of 1, 2, 4, 8 and so on for CPU DID. What they do is divide the end CPU FID value, and then the HT reference speed is multiplied by it to gain the CPU speed.

Hence, if we input CPU FID value 08 which stands for CPU multiplier 12x, and we choose CPU DID value 1 (which is what its says), then:

FID / DID = CPU Multiplier
12 / 1 = 12x multiplier

If we chose CPU DID 2 then the end CPU multiplier would've been 6x.

So the HT reference clock speed you set (say 200MHz as an example) will multiply x12 to get 2400MHz CPU speed now.

HOWEVER the 9500 I have is multipler locked obviously. So whatever FID/VID value you choose above what gets you 11x multiplier will just result in no change from stock 11x multipler.

CPU VID is the Voltage identity which I did't fully understand till Macci replied and said to leave it at default which is value 28 for 9500 and stands for 1.250V. If you increase it to value 29 you get 1.1875V with my testing.

BUT if you increase the voltage (i.e. decrease the value to 27), you will fail POST.

Now we know how CPU speed is worked out.

Additionally, NB FID, DID, VID are of the same task but for the NB instead and NB VID can be changed to higher. Value 36 is stock which for me was 1.1V.

NB FID values are as so: 00 is 5x NB speed multiplier, 01 is 6x, 02 is 7x and so on.

For NB DID you get two options of divider 1 and divider 2.

What that NB FID does is, get the HT reference clock you set (say 200MHz) and multiply that by the NB multiplier you choose through NB FID / NB DID.

So say you choose NB FID of value 00 which is 5x multi and choose NB DID of value 1 (which is again what it says).

NB FID / NB DID = NB Multiplier
In the above case we have 5 / 1 = 5x multiplier

With a 200MHz HT ref. speed, that would give us 200x5 = 1000MHz NB speed.

BUT=> NB clock speed can only be equal or higher than HT speed. Keep that in mind.

If it isn't, then either you'll have no boot or the system will downclock itself lower than stock. If you set a high NB clock, your system may not boot at all or downclock the NB/CPU/HT to make it bootable.

Now we know how the NB speed is worked out.

Easy as that! Any qs, please ask. I've tried to make it simple but you'll have to give it a go before understanding properly.

Here was my older run using this method I posted within those images

Not sure if it was posted later but the formula for the cpu frequency is

CPUCOF = 100 MHz * (CPUFID + 10h) / ( 2 ^ CPUDID) (s.h. K10 BKDB p 325.)

EDIT: Ahh you got your BE. That cinebench results do not show >100% scaling. Have you compared memory subtimings?

[Blacklash]

I can tell you right now Q6600 G-0 = effortless 3.0GHz with the stock cooler on stock volts. Mine needs 1.36v for 3.2, 1.46v for 3.6 and 1.52v for 3.8GHz.

BTW good luck with that BE.

[Titan7171]

would it be ok if I posted my results with the K9A2 here?? Im using 6400X2 though?

[KTE]

Not sure if it was posted later but the formula for the cpu frequency is

CPUCOF = 100 MHz * (CPUFID + 10h) / ( 2 ^ CPUDID) (s.h. K10 BKDB p 325.)

I've posted it around but not posted it in this thread AFAIR. Thanks. There's a few ways to work it out, this is also easy:
CPU Speed = (1/2*HT ref.) * (FID+16) / DID

EDIT: Ahh you got your BE. That cinebench results do not show >100% scaling.

Yep, but it's extremely close to linear. Better tests coming up.

Have you compared memory subtimings?

They're kept constant throughout in that in the above.
Single channel 2T (1 stick) at the same settings as the above 2640MHz gets 8456, which is as bad as you can expect with memory.

I'll test everyrthing better now because you can keep NB/HT/RAM constant.

I can tell you right now Q6600 G-0 = effortless 3.0GHz with the stock cooler on stock volts. Mine needs 1.36v for 3.2, 1.46v for 3.6 and 1.52v for 3.8GHz.

Yep, I've owned a B3 and a G0 before to know.

would it be ok if I posted my results with the K9A2 here?? Im using 6400X2 though?

Well, I'd like to stick with Phenom in general but X2 results with this board are also welcome which highlight or deal with the boards issues/performance/oc.

[justapost]

I've posted it around but not posted it in this thread AFAIR. Thanks. There's a few ways to work it out, this is also easy:
CPU Speed = (1/2*HT ref.) * (FID+16) / DID

DID can be zero

Yep, but it's extremely close to linear. Better tests coming up.
They're kept constant throughout in that in the above.
Single channel 2T (1 stick) at the same settings as the above 2640MHz gets 8456, which is as bad as you can expect with memory.

I found that the max async latency differs ~10ns. That has an impact on benchmarks. Using the same latency leaded to much more linear results. So i keep an eye on that subtiming now. Might be common sense for ocers.

[KTE]

DID can be zero

With this board the options are either 1 or 2.

I found that the max async latency differs ~10ns. That has an impact on benchmarks. Using the same latency leaded to much more linear results. So i keep an eye on that subtiming now. Might be common sense for ocers.

I haven't seen that option on this board.

10ns EVEREST RAM latency? Are you sure about this?

There' nothing I've seen giving you that much gain on this board so far (no patch BIOS). I mean the major timings on K10 are ~seven and 5-5-5-15 2T tRC 35 tRFC 195ns wouldn't get 10ns higher latency than 3-3-3-11 1T tRC 24 tRFC 75ns on my board. Usually 4-5ns is max.
This might help you although it concerns K8/DDR, it is still simple enough to apply in many areas.

gareth170: Using the exact same method you used to flash 1.2, just with the correct v1.1 BIOS file.