New Llano leaks

[informal]

Can you say limited on purpose? Note how the score in lame superpi is 27 seconds @ 4.7Ghz,which is 2x slower than deneb . Note also the poor 3dmar99 score of 189.This is single core benchmark and Agena at 2.5Ghz scores roughly 330pts,74% higher. On chiphell forum someone mentioned the Turbo core bug,which "seems" to plague BD ES also (early superpi scores of B1/B0 were around 27s for 3.2Ghz Zambezi,very low score).
http://forum.coolaler.com/showthread.php?t=266652

Edit: looking at C10 score and speedup,in order to get to 6.68x speedup the clock speed in single thread benchmark(and probably in all other single threaded benchmarks) has to be locked at around 1400MHz,despite CPUz showing 2.4Ghz.

[cesariuth]

5.4GHz ??



I no understand if this is on LN2 or Air

[informal]

The guy who ran the tests said the software cannot read the real clocks so he is not sure if 5.4Ghz is real or bug. He also said that there is some bug in the ES/bios so that scores don't scale with clock increase.Sometimes the test finishes up in line with clock increase,sometimes it's 2x slower.So the ES is definitely "broken" in some way. This is clearly evident by looking at the superpi and cpumark99 scores which are 1.7x-2x lower than Deneb at same clock.

Should be right ... Because able individual Super-
In fact, the software used to display information is not updated, see do not know is true or false ...


4.7GHz runs a bit confused
5.4GHz result in reasonable points

Little brother has tried, even running times PI or CPUmark99 score and some will fall by half
May is the relationship between ES ... Not final

[charged3800z24]

Anyone know when these hit retail? I am looking to buy one but don't see boards or Chips. I keep seeing the 990FX Sabertooth show up here in there on new egg, but no FM1

[Macadamia]

btw guys, take note of lack of the "Hypertransport" term, with 4x PCIe2 connecting to the southbridge.

Either this makes overclocking easier by not involving the interface... or 32nm is pretty sweet.

[cesariuth]

But the stock clock is 2.4GHz...right?

[Macadamia]

It's a 3800- an SKU that in terms of clocks won't exist.

38X0s (desktops) should have higher clockspeeds, no?

[informal]

Looking at supposed 5.4Ghz AIDA64 cache/memory results,it's safe to say that 5.4Ghz is not real. Actual clock can be easily extarpolated via the cache BW numbers and cache latency numbers (vs Deneb):

Compare with Deneb at 3.4Ghz and its results:


Actual clock speed when he selected the multiplier 54 is around 3-3.1Ghz according to AIDA64 numbers.So real/effective multiplier is either 15x or 15.5x and something in BIOS is responsible for the bug (or "feature") .
This clock speed(5.4Ghz <=> ~3.05Ghz) lines up perfectly with what I said in first post.Actual clock is around 1.7-1.77x lower than what CPUz reports,especially in single thread workloads where Turbo bug is most pronounced.

edit:
Look here. 5.4Ghz supposedly has 4300pts in CPU part of 3dmark06.This lines up perfectly with actual clock speed of 3-3.1Ghz . Another confirmation it's NOT 5.4GHz.And those Chinese guys have no idea ,they still think it's really 5.4GHz.. They just had to compare AIDA64 scores and this simple 3dmark06 cpu test with Deneb at 3Ghz,that's all...

[duploxxx]

It's a 3800- an SKU that in terms of clocks won't exist.

38X0s (desktops) should have higher clockspeeds, no?

valid ES sample, SKU will exist

[mrcmtl]

5.4GHz ??



I no understand if this is on LN2 or Air

The CPUMark score doesn't seem bad as my 4Ghz Thuban scores 553.

[informal]

The CPUMark score doesn't seem bad as my 4Ghz Thuban scores 553.

The score is not correct for the clock speed since the real clock is not 5.4Ghz...

[FlanK3r]

very nice! Seems a bit better in Cinebenchs than Athlon x4 clock to clock. Impressive is 4.7 GHz with stock voltage?!
Commercial version of listening to a locked multiplier of Dong Lin a pity ... you ...
Because ... Llano is a super strong friends overclocking
Only adjust the multiplier to 47 (BIOS to the most strained to 47 only), the rest did not move, did not increase the voltage directly on the 4700MHz ...

[jimbo75]

Informal has it nailed.

It's probably true that AMD has been deliberately gimping clocks on this and BD for the past x months.

Do we know the ES number of this one Coolaler got? Hexus got one as well and it would be nice to know if the samples are the same.

[informal]

Read above FLanker,those are wrong readings in CPUz.Real clock is ~2x lower.

[FlanK3r]

I tried wrote him at coolalerforum about hwinfo...we will see, if he use it.

[informal]

New results posted @ 2.7Ghz?(shows as 5.4Ghz in his CPUz):
http://forum.coolaler.com/showpost.p...4&postcount=77
Since one of our fellow users notified me that the real clock is actually 2x lower than what CPUz reports,I will do rough calculation based on this.

Also,I see the OP posted the hwinfo screen showing 3.6Ghz as max T clock and actual clock of 3.1Ghz while multi is set to 47x (maybe showing as 4.7Ghz in CPUz). I think this is wrong since if the scores are done at 3.6Ghz then llano is actually up to 20% slower than Deneb at the same clock,which is not realistic IMO.

Latest scores recap,I still use 2.7Ghz as a real clock of Llano(5.4Ghz showing in CPUz)- note that clock may be higher than 2.7Ghz so below may be not correct :
3dmark11 physics score - Llano is 11% faster at the same clock vs Deneb (X4 Deneb @ 3.5 scores 14fps,LLano @ 2.7Ghz scores 12fps).

C10 64bit- Llano is 14.5% faster at the same clock vs Deneb (X4 Deneb @ 3.2Ghz scores 13419,LLano @ 2.7Ghz scores 12793).

CINEBENCH R11.5 - Llano is around 16% faster at the same clock vs Deneb (X4 Deneb @ 3.5Ghz scores 4.11,LLano @ 2.7Ghz scores 3.68).

edit:
now he posts hwinfo while running Cinebench MT test . CPU "throttles back" down to 2.4Ghz real?? clock while test runs ! Priceless info. Looking at the latest info,the original MT scores may have been correct.So Llano is around 2-5% faster than Deneb with L3 and above calculation is not correct since I don't think the sample ran at 2.7Ghz(5.4Ghz shown).It probably ran at around 3Ghz so 10% has to be taken off .

[Manicdan]

your deneb compare, is that with or without L3? since i though llano will only have L1/L2 cache

[jimbo75]

20% slower than deneb at the same clock is what that german site tweaktown.net got while testing that msi notebook with superpi. That doesn't look so good.

[FlanK3r]

good to know, hwinfo is great software...

Again, I wrote him about possibility problem. Think, this 3600 MHz is CPU turbo ratio (manualy changed from default turobboost) and 2400 Mhz stock clocks

[informal]

your deneb compare, is that with or without L3? since i though llano will only have L1/L2 cache

I said Deneb and Deneb has L3. Propus has no L3,but I didn't use its numbers since it also has 2x less L2.

20% slower than deneb at the same clock is what that german site tweaktown.net got while testing that msi notebook with superpi. That doesn't look so good.

Yep I saw that.But as you can read in my last post,scores are not 20% lower than Deneb,they are actually a bit better at the same clock.Pretty good in my book .

[jimbo75]

I said Deneb and Deneb has L3. Propus has no L3,but I didn't use its numbers since it also has 2x less L2.



Yep I saw that.But as you can read in my last post,scores are not 20% lower than Deneb,they are actually a bit better at the same clock.Pretty good in my book .

Yep I just saw that as well.

Hmm so 10-15% faster clock for clock than Deneb? If Llano comes out the blocks near 4 GHz that surely must be a real threat to the 2500K.

Edit ok so 5%, still not bad.

[Manicdan]

the IPC increase of llano over deneb is kinda unexpected, if its really pushing 10-20% better without L3, then our IPC guesses at BD might be a little off, or these number are way off. i cannot expect much more than 5% plus or minus deneb

[informal]

i cannot expect much more than 5% plus or minus deneb

That should be correct.

edit:
btw I posted at the original Chinese forum and explained how the results are borked.I hope they understood,or at least the OP understood.He seems to be using hwinfo32 now,it supports family 15h and seems to correctly detect the clocks.

[jimbo75]

Yep I'd take even with Deneb in a heartbeat. Now it's just down to GF providing the clocks.

[Nintendork]

the IPC increase of llano over deneb is kinda unexpected, if its really pushing 10-20% better without L3, then our IPC guesses at BD might be a little off, or these number are way off. i cannot expect much more than 5% plus or minus deneb

Llano also has twice the cache L2 (4MB vs 2MB)

What I want know is a quality mobo (Gigabyte FXA-UD3/UD5 cousin). All the boards are "cheapo", specially in the number of sata ports.