What's the current Llano APU vantage wr?

[chew*]

Ok so some more stuff.

Numero Uno Be very very carefull with LLano on phase, just becasue you can run it does not mean your not doing damage.

1.735 just claimed this cpu right after cpu test. The IGP is actually what died, I got it to post again OCed but the screen artifacted at windows load and then it was all over.

This is actually a good thing as there were questions. What if, what if LLano's gpu was killed, would cpu clock better? Answer no it won't post....

Also answered the "what is safe voltage" I would say .200-.100 under what I was running on phase.

Now how does CPU scale benches? Not really much.

I say Dial in GPU, then ram then cpu.

A Proper 05 run now


Barely any Scaling in 06

[BeepBeep2]

What about volts on dice? It is a tad colder, though not much...

~1.625v max?

[chew*]

I tested -40 to -80 range......absolutely no gains......

You have to use one of these chips to understand......they do not scale well at all with voltage. Cold yes voltage really not. It's not linear at all.

Brute force will not work here, I tried it a few times and now have a dead chip for the sake of needing to know the limits.

Most of the scaling I found was only on LN2. Figure 3.7 avg on air they only scale to 4.2 at -40 from -40 to -190 they only scale to 4.9-5.2 range all voltages the same phase to ln2......higher volts = pissed off chip failing benches at known good speeds and possibly degraded.

[BeepBeep2]

I tested -40 to -80 range......absolutely no gains......

You have to use one of these chips to understand......they do not scale well at all with voltage. Cold yes voltage really not. It's not linear at all.

Brute force will not work here, I tried it a few times and now have a dead chip for the sake of needing to know the limits.

Most of the scaling I found was only on LN2. Figure 3.7 avg on air they only scale to 4.2 at -40 from -40 to -190 they only scale to 4.9-5.2 range all voltages the same phase to ln2......higher volts = pissed off chip failing benches at known good speeds and possibly degraded.

So if you could get the chip 5-10c colder by using less voltage, it would be better than forcing more, right?

I'm worried about guys running chips 24/7, it seems due to the high idle voltages people aren't afraid to run ~1.5v in ambient temps

[chew*]

Time will tell.

I think the way this chip blew it was more to do with current chip was pulling at XX temps. Lower temps will lower the power consumption.

Internally cpu's are like fuses, what happens when you pull 20 amps repeatedly on a 15 amp fuse. Poof.

The way this chip died was indicative of that.

[Daveburt714]

chew*: Was the chip you were using an unlocked ES, or did you have to OC using the HTRef?

I just wonder if the thing holding it back is the PCIe bus, since it's linked to the HTref as well.

EDIT: I guess my real question is, do you think unlocked multi's would open things up?

[chew*]

My chips unlocked, I chose/choose not to use higher multis for the sake of being fair. I am however using them ( unlocked chips) to get some things in order for some interesting software for when unlocked parts do become avail.

To answer the question no unlocked parts really won't open doors to higher clocks on air. It will help users with ram that is barely compatible with platform or boards with shoddy bios support.

Certainly more fun and info to come for LLano, stay tuned

[Daveburt714]

Thanks man, I really had plans to buy a Llano rig to play with (and for my HTPC down the road), but decided to wait for a Black Edition...
There's no NDA on the these chips anymore right?

Could you show us what Llano can do with unlocked multis.....

[chew*]

Ok I have to edit some stuff out of the shot though......this was tested at AMD HQ with 100 ref clock.

Llano has a muti range of 16-47 normally ( for unlocked models ), and with some tricks voila 50X in vantage.

I might add I only tested 1 chip, sure with binning/retails better can be done.

From what I have seen this chip i have isn't really that impressive compared to others on phase with retail.

[Daveburt714]

WOW!!!

15694 CPU score.... I knew those bad boys had more in them...
Thanks Bro, You've just made me feel better about waiting for a BE!

Shame you had to edit the SS, but I understand.

[utilizatori]

WOW!!!

15694 CPU score.... I knew those bad boys had more in them...
Thanks Bro, You've just made me feel better about waiting for a BE!

Shame you had to edit the SS, but I understand.

ye, but that's 4 cores and 5000 MHz
I would wish better scores

[SF3D]

Very nice runs!

Unfortunately there is no magic efficiency in these runs. You set IGP to 755MHz and set bclk to 150. At those settings you are running IGP around 1100HMz, which is explaining the good results
We could really talk about scaling, if 25% gain in IGP clockspeed would bring 80% gain in end result.

On my UD2H + retail CPU setup I can not change the IGP clockspeed value. It should work with retail, but it don't for some reason. It might be just bios issue.
So other users should try those settings and see if it works on retail setups.

Nice to see scores with ES sample, which is completely unlocked. I have not been able to test those myself unfortunately.

E:
Here is explanation for IGP clocks.
MainPLL (3600MHz) / GPU divider (default 6) = GPU clocks (default 600)
BCLK 150 = MainPLL (5400) / GPU divider 5 = GPU clocks 1080MHz

[BeepBeep2]

WOW!!!

15694 CPU score.... I knew those bad boys had more in them...
Thanks Bro, You've just made me feel better about waiting for a BE!

Shame you had to edit the SS, but I understand.

Will you be going LN2? Otherwise it doesn't matter much to you

[chew*]

Nice to see scores with ES sample, which is completely unlocked. I have not been able to test those myself unfortunately.

Completely unlocked to 47X not 50X higher than 47X is acchieved via a register change.

[cesariuth]

Chew: This chip is a 3850 ES unlocked???

[chew*]

Chew: This chip is a 3850 ES unlocked???

technically it can't be 3850, since 3850 is not nor intended to be unlocked, it's just an ES or we can call it a A-3870 for now....

[Alex-Ro]

What board are you testing with?I have no GPU multiplier in my asus A75-M-Pro

[chew*]

What board are you testing with?I have no GPU multiplier in my asus A75-M-Pro

there is no GPU multiplier, what people seem to not realize is that there is a hidden divider based on dialed in gpu frequency, 750 has a different divider than 800, 850, 900 and so on. Like I said earlier.......Use what is fast not what you think is fast

Btw I was not priveledged to know this info, I discovered it by testing I was preety much handed Llano like all other cpu's, here's the chip learn it's secrets

Apparently my chip is not dead, it's back to life at stock at least for now. I might have usefull info for those benching cold though. Llano's socket has a sort of flaw for cold benching. The hole in center is an air pocket, even with the socket sealed with eraser condensation can occur.

Here's a shot I took of chip. As you can see there is a tiny drop of water, might not hurt to nail polish or conformal this area on LLano cpu's if benching cold.

[Manicdan]

so the gpu slows down based on the bus speed? but nothing is able to detect it?

[chew*]

so the gpu slows down based on the bus speed? but nothing is able to detect it?

The GPU can slow down when ref GPU base clock is adjusted due to a hidden divider changing at certain frequencies.

The GPU clock increases as CPU base clock is raised as well as PLL which in turn has an effect on gpu dividers.

Benches can detect it, so "nothing" would not be a choice word to use. Currently software can not detect correct values however.

[Manicdan]

i hope we soon get control over that, it would be nice to see a real 900mhz+ run of the Llano gpu

[chew*]

i hope we soon get control over that, it would be nice to see a real 900mhz+ run of the Llano gpu

You already did. As SF3D pointed out my clocks were 1083.

[Manicdan]

hmm, some things make more sense, then some things dont

he mentioned IGP clock speed, but then his calculations dont even include the IGP speed

[chew*]

hmm, some things make more sense, then some things dont

he mentioned IGP clock speed, but then his calculations dont even include the IGP speed

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

[JumpingJack]

<snip>
Here's a shot I took of chip. As you can see there is a tiny drop of water, might not hurt to nail polish or conformal this area on LLano cpu's if benching cold.

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

Here is a pix of my llano:
Image003.jpg

Looks clean on this one.