What's the current Llano APU vantage wr?

[chew*]

Chew -- I think that is a defect/anomaly from the packaging underflow.

Here is a pix of my llano:
Attachment 118358

Looks clean on this one.

I dabbed it with a paper towel, it was def water.

[JumpingJack]

I dabbed it with a paper towel, it was def water.

My mistake -- looked like a hole or pit in the underflow.

[Lightman]

Excellent thread!

Thanks to all

[Manicdan]

Here's a way to calculate your gpu speed courtesy of Macci.

Run 3D06 multi texturing fill rate test.

Then use this formula

default = 11856

Brian’s OC = 21408

21408/11856 = 80.56% boost in the GPU clock

600MHz (default GPU clock) * 1.8056 = 1083MHz

I'm in the 22400 range now, safe to say 1100 core is broken

ok so your getting almost double the gpu performance, hows important is ram timings and speed at that level?

[chew*]

Good question I am looking into that now. Seems I need 756 and 157 ref clock with loose ram timings but higher clocks to catch the scores that petri said weren't effec........

I would say speed is important no matter what, timings are questionable atm......it's more of a issue where tight is holding back your overall clocks.

[BeepBeep2]

To stop that water from forming, maybe a little peice of paper towel there, or a small bit of shop towel?

When I went DICE for the first time I layered about 5 paper towels under the socket backplate and behind VRM area to stop water from condensing, seemed to have worked alright. I used rice on the top though, you can spill some liquid on it and the rice is so thick it never reaches the board.

[chew*]

Round 2. Macci said it would be tough to push these scores further and right he was but it was possible.

1. Benching this APU can be infuriating at times I like it

2. To push further at least on this board it's going to require some mod research. With the combined high gpu and cpu it's tripping OCP.

3. I got 03 working finally with a different version as well as system info on everything but 05 ( I need to try diff versions for that )

4. This APU is very sensitive to DX9 benches compared to DX10 and up and thus clocks are actually less in older benches, also as the gpu went up I had to tone down cpu or OCP would trip. Some results were slightly impacted by this.

03, ran it once to get a baseline of where I was at score wise.


05 minimal improvement, GT1 is lacking here compared to my other run, sacrificing ram timings + cpu clocks in this bench has it's drawbacks.


06, the gains are ok I guess, Ram timings don't seem to have to big an impact in this bench. CPU being dropped probably cost a little compared to previous run.


Heaven, heaven is just brutal, so much power needed for so little gains.


3D 11 is scaling decent.


Vantage, I just really found a sweet spot here, finally cracked 21 FPS in GT1 and 16 FPS in GT 2.


Gonna take a stab at some mods and see if we can't get Round 3 rolling

[sin0822]

Chew -- I think that is a defect/anomaly from the packaging underflow.

Here is a pix of my llano:
Attachment 118358

Looks clean on this one.

And you were testing with phase?


Thanks for the thread chew, very interesting on the cold scaling.

[JumpingJack]

And you were testing with phase?
.

No, looking at Chew's original pix I mistakenly thought it was a defect in the pcb package and mine did not have it. Chew clarified that it was a droplet of condensation from the dead space due to the socket.

[chew*]

Ok what I am going to show today does not apply to current cpu's. This will only work on future unlocked models.

There will be a simple program that will be public most likely in time for or prior to unlocked models being released so don't get your panties in a bunch.........It will be shared with the public.

How it works.

Default multi range is 16x-47x for unlocked parts. If we change something we can shift the multi range, now i'm not 100% sure of the impacts this can have on total platform performance however cpu intensive multithreaded shows no gains or losses.

So to sum up lets call this Mode 0 = 16X-47x.

Mode 1 = 32x-63x (lower than 32X will not be available)

Applying Mode 1 with an already high clock say 3700mhz will result it 16X higher so 3700+ 16X will = 5300 and most certainly a crash. In order to counter this multi should be dropped in K10 to lowest setting so when applied frequency will be 32X = 3200 which just about any LLano cpu can do on air.

Will this be usefull for air users? Not likely unless the process improves drastically.

Will it be usefull for ln2? Yes, it will help find the max limits of cpu without board or ram as part of the equasion.

I have tested k10 stat K10153 version and it works fine. Unfortunately at this time cpu-z does not recognize the shift so I will have to contact frank to see if he can get it working.

I used Wprime to first measure performance at a given clock and also show the speeds are "REAL" not bugged.

First a Base line run at lowest multi 16x Mode 0



Second Base line run at 42X Mode 0



Mode 1 = cpuz reported + 1600mhz As we see the time is cut in half compared to the above 1600mhz run.



Mode 1 = cpuz reported + 1600 mhz As we can see the time is identical to the above run at 4200mhz.

[Farmer Boe]

Very interesting stuff here. I'm looking forward to tweaking my incoming A8-3850 just for fun. It's supposed to be for my media/download computer but they never stay that way

[Alex-Ro]

Couldn't get any more on air

[El Gappo]

Ok what I am going to show today does not apply to current cpu's. This will only work on future unlocked models.

Can you not use it for down-clocking on locked samples? Custom P-states would be nice

[chew*]

Couldn't get any more on air

That score is nothing to be ashamed of alex.

Can you not use it for down-clocking on locked samples? Custom P-states would be nice

The problem with that would be the entire multi range is shifted elgappo from 16-47 to 32-63 and models that don't support upwards multi adjust won't apply. I tried this at ASUS on another ES sample and the old claim ( ES is all unlocked is bullocks ), it didn't work on their chip.

I would imagine with K10 stat you can already do custom pstates however.

[Alex-Ro]

k10stat supports voltage adjustements on giga board?i mean if i modify it on the fly,it won't change in cpu-z nor ai-suite 2 and i'm too lazy to get my multimeter.Board is asus a75-M-Pro.

[chew*]

k10stat supports voltage adjustements on giga board?i mean if i modify it on the fly,it won't change in cpu-z nor ai-suite 2 and i'm too lazy to get my multimeter.Board is asus a75-M-Pro.

I will check later on tonight Alex and let you know, I haven't tried voltage in K10 yet.

[Manicdan]

PhenomMsrTweaker might too

[Dolk]

Anyone strap a 6670 to the A8 and see how it performans?

[chew*]

Anyone strap a 6670 to the A8 and see how it performans?

Kinda defeats the purpose. We are trying to find out how far the APU can go, not an actual VGA or X fire.

[Dolk]

Meh I was just asking to see if anyone has done it any see how it went. I would like to see this chip do some damage alone, but the Xfire is a bit interesting as well.

[chew*]

Google is your friend.

Here's 6570 crossfired with APU......You can probably add 3-5% increase with 6670.

http://www.techwarelabs.com/amd-llano-a8-3850-apu/5/

And what appears to be X fired laptop with 6620D.

http://forum.notebookreview.com/hp-p...t7738781result

[Pt1t]

On the biostar FM1 Board.

http://www.pt1t.eu/?p=430

[Alex-Ro]

Damn,that cpu score looks great.Also board clocks memory good.Very nice little board

[G.Foyle]

Very nice memory

[El Gappo]

On the biostar FM1 Board.

http://www.pt1t.eu/?p=430

Nice The last bios on the M+ was a pain in the ass, might actually be able to get something done now.