What's the current Llano APU vantage wr?

[chew*]

I don't measure actual cpu temps tbh when cold.

Evap is going -40C idle to -29C at lowest depending on the mood it's in some days -34C is lowest.

My other unit will idle -40C and dip to -38C max, not that it matters really with these chips.

The unit on right is also much beefier and tuned diff, unit on left is the one currently in use.

Buddy was interested in buying it so making sure it's still in tip top shape other than it's typical moodiness that i'm used to.

[Alex-Ro]

Thnks chew,so it seems problem is with asus boards.You wouldn't know where to get some beta bios for them,would you ?

[chew*]

Thnks chew,so it seems problem is with asus boards.You wouldn't know where to get some beta bios for them,would you ?

I don't have that board and am not working on it Alex so I doubt they would send me the bios's for it.

[chew*]

I'm having to back off on the latency with these sticks to get them up to speed which is accounting for the lack of speed in comparison me thinks.

You sure that it's all in the latency?

[Alex-Ro]

Hey chew in gigabut bios you have that option CPU Northbridge frequency,does that work like a multiplier for CPU-NB freq(like old phenom) and does it actually apply the settings or not?I am saying cos there isn't such an option in asus bios and curious if you could actually run CPU-NB higher on gigabut boards.

[chew*]

Hey chew in gigabut bios you have that option CPU Northbridge frequency,does that work like a multiplier for CPU-NB freq(like old phenom) and does it actually apply the settings or not?I am saying cos there isn't such an option in asus bios and curious if you could actually run CPU-NB higher on gigabut boards.

They pulled that option on giga, it's useless ( no measured performance change no matter what multi )

[Alex-Ro]

I see.Such a same that we can't play with northbridge freq like on old phenom II.Maybe with next llano be chip?

[chew*]

Gappo's Latency Test

Because I like him and he's a cool guy.





Really no diff other than Bench Run variation here........

[chew*]

I see.Such a same that we can't play with northbridge freq like on old phenom II.Maybe with next llano be chip?

Doubt it, this chip is completely unlocked and NB didn't do a thing. Dud setting in bios that just needed removing.

[chew*]

Surprised no one else has cracked 7K vantage yet, Even locking GPU at 600 and using crap timings 7K is doable..........

The my board doesn't allow clock speed control excuse is what it is.......an excuse.

[Alex-Ro]

Sorry to sound like a douchebag,but how do you lock your gpu at 600?Another gigabut option that it's not on asus boards ?

[chew*]

Sorry to sound like a douchebag,but how do you lock your gpu at 600?Another gigabut option that it's not on asus boards ?

Well your locking the base GPU clock at 600, or leave it auto which is default 600.

The gpu speed still increases from raising CPU base clock but not from GPU.

For instance, I heard users complain about GPU clocks not applying on UD2H with retail chips.

Also at 600 ( default ) GPU should be on a lower divider. I will see if i can get a chart of where the straps are, they change based on bclk.

So far I have just been testing performance at a given clock to find out where they are and when they are changing.

Let me fire up again and run 3d06 multiple textures fill rate test, that should give an idea of what the actual clock is.

[TsunamiJuan]

The asus board doesn't have an option to change the base GPU speed yet. I Know i have looked. The biostar does have the option, however i am not sure it functions correctly, from the testing i was doing with it yesterday it didn't appear to make any differance in gpu performance at all. So It makes me think that it no properly implemented yet in the beta bios.

[The Stilt]

The bios northbridge frequency control on Gigabyte and Asrock is working just like it should = not working.

The northbridge clock (NCLK) can be indeed adjusted by dividers ranging from 2 to 15.75 (/MainPLL).
However there is a rule which limits the maximum NCLK frequency: NCLK must always be less or equal than the memory frequency (MEMCLK). Also not all of the NCLK dividers are considered desireable. They do work, but there is a minor hit in memory write performance. The "undesireable" dividers are the reason why the "clock to clock" performance is worse on certain dividers. Not because of these dividers are used, but because they are not used. Luckily this can be fixed

I´ve noticed that these APU:s really seem to hate any memory which uses Micron made IC´s.
None of my Micron kits (1GB & 2GB sticks, D9GTR, D9KPT) can even do 1600MHz unless the tCL and tRCD is latched to 10CLK.
The same sticks can easily do 2000MHz+ with tighter timings on K10 platform. Thats on Asus F1A75-M Pro. On Gigabyte A75M-UD2H they did not even post no matter what.

Is everyone using Elpida and PSC at those monstrous clocks?

[chew*]

The bios northbridge frequency control on Gigabyte and Asrock is working just like it should = not working.

The northbridge clock (NCLK) can be indeed adjusted by dividers ranging from 2 to 15.75 (/MainPLL).
However there is a rule which limits the maximum NCLK frequency: NCLK must always be less or equal than the memory frequency (MEMCLK). Also not all of the NCLK dividers are considered desireable. They do work, but there is a minor hit in memory write performance. The "undesireable" dividers are the reason why the "clock to clock" performance is worse on certain dividers. Not because of these dividers are used, but because they are not used. Luckily this can be fixed

I´ve noticed that these APU:s really seem to hate any memory which uses Micron made IC´s.
None of my Micron kits (1GB & 2GB sticks, D9GTR, D9KPT) can even do 1600MHz unless the tCL and tRCD is latched to 10CLK.
The same sticks can easily do 2000MHz+ with tighter timings on K10 platform. Thats on Asus F1A75-M Pro. On Gigabyte A75M-UD2H they did not even post no matter what.

Is everyone using Elpida and PSC at those monstrous clocks?

Give me a list of micron you would like tested stilt and I will see what i can come up with, I have a preety wide variety here.

As far as I am concerned elipda ( at least hypers ) is a fail for the platform as far as 3d although i will say i didn't put much time in, lets say "needs further testing"

Yes I'm on PSC atm.

Alex

Ok here's the formula.

default score for Llano= 11856
result in fillrate multi texturing/11856 = % boost in the GPU clock

600MHz (default GPU clock) * result above = actual core

Here's how it works.



19167/11856 = 61.66% boost in the gpu clock ( value is 1.6166 )

600MHz (default GPU clock)* 1.6166 = 969 core

[El Gappo]

600MHz (default GPU clock)* 1.6166 = 969 core

I'm just thinking of 600 as a multi, 161 * 6 = 966. Close enough

trfc had quite an impact. :/

[The Stilt]

Give me a list of micron you would like tested stilt and I will see what i can come up with, I have a preety wide variety here.

As far as I am concerned elipda ( at least hypers ) is a fail for the platform as far as 3d although i will say i didn't put much time in, lets say "needs further testing"

Yes I'm on PSC atm.

Please test with D (D9JNx) & F-die (D9KPx) chips if you have any.
They are supposed to do the highest frequencies at low voltage.

Regarding the GPU clock control in bios:

On retail chips (i.e locked) the adjustment is not even supposed to be working (upwards) as there is a frequency limit.
On A6-3650 chips it is around 457MHz and on A8-3850 620MHz or so. The only way around it is raising the BCLK frequency.
Downwards it should and it is working (on Gigabyte atleast).

Early Asus A75 motherboards advertised "GPU Boost" function.
Even the pre-production board I have has "GPU Boost" silk screened on it.
This was before Asus realised it is not working on retail chips and Photoshopped their product pictures

It is still implemented in the bios code, but it is not visible or working even on unlocked chip.

Attachment 118712

[chew*]

Please test with D (D9JNx) & F-die (D9KPx) chips if you have any.
They are supposed to do the highest frequencies at low voltage.

It is still implemented in the bios code, but it is not visible or working even on unlocked chip.

Attachment 118712

I will toss in the D9KPT now.

Is manual GPU clock speed working on any boards for you with unlocked models?

[The Stilt]

I will toss in the D9KPT now.

Is manual GPU clock speed working on any boards for you with unlocked models?

Great.

On Gigabyte UD2H it did IIRC.

[chew*]

Stilt other than actual posting KPT above 1600 what issues did you have? Timings needed i guess?

Side note, It looks just like the Gskill ECO 1600 SPD bug they had when they released them.

Stilt try this for me, stock volts ( 1.65v ) in my case, 7-7-7-21-28-110trfc set 1333 divider, up base clock to 130 and try that

[The Stilt]

Stilt other than actual posting KPT above 1600 what issues did you have? Timings needed i guess?

Side note, It looks just like the Gskill ECO 1600 SPD bug they had when they released them.

Stilt try this for me, stock volts ( 1.65v ) in my case, 7-7-7-21-28-110trfc set 1333 divider, up base clock to 130 and try that

Well, I can surely get the system post at 1600MHz and even much higher but making it 3D stable anything above that is just impossible.
3DMark06 GT1 especially is a certain killer.

[chew*]

Well, I can surely get the system post at 1600MHz and even much higher but making it 3D stable anything above that is just impossible.
3DMark06 GT1 especially is a certain killer.

Running 32m now with 1333 strap stock volts and 1700+ 7-7-7-21-28-110. Will run 06 as soon as this is done.

[TsunamiJuan]

Interesting stilt, I am running a retail release chip sent out as press review. So i can confirm that GPU boost is disabled for this chip in the bios from around 7-25-2011.

downclocking does work on the Gigabyte ud4h with release f4 of the bios. I did not test pushing it further up yet.

[chew*]

Ok stilt

The naked KPT passed 32m and epic fails 7-7-7, 8-8-8, 9-9-9 in 06 gt1, didn't bother with higher.



This other Corsair kit set uses some form of micron, other than that i have no clue.

Passes 06 at rated timings albeit 1.55v versus the 1.50 rated. 1.50 failed GT1.

[The Stilt]

Ok, thanks for testing.

I made a "calculator" for Llano a while ago.
This version is a bit censored, but I can send the "unlimited" version for you chew if you like to have it.

This is for A8-A3850 only.