Thuban performance scaling

[LightSpeed]

Sorry if this has been posted already, but AT has a great article about the CPU/NB scaling of Thuban

Link here

Edit: just finished reading and really, those are WTF gains from the NB/Mem overclocks, especially in games. 3.2/2.8 Ghz CPU/NB beating 4/2 Ghz CPU/NB? Have a look. Seems like a lot of performance can still be extracted from K10.5

[Alex-Ro]

Yea looks interesting for sure,wonder if on deneb things are the same...

[mrcmtl]

I wonder why AMD doesn't release thubans with 2.5Ghz on the NB by default. Maybe TDP limited?

[TheBlueChanell]

This is exactly what every knowledgeable/die hard AMD clocker has been saying for the past two years. I'm glad somebody finally took notice.

[crazydiamond]

this is what bugs us when "reviewers" oc the chip to 4ghz & dont touch cpu-nb

[richierich]

this is what bugs us when "reviewers" oc the chip to 4ghz & dont touch cpu-nb

seriously.

[I.nfraR.ed]

Yep, even when testing some high frequency memory (let's say 2000MHz) and compare it with lower-speed RAM sticks. Well, 2GHz CPU-NB is really not sufficient for those speeds.

[Leeghoofd]

Edit: just finished reading and really, those are WTF gains from the NB/Mem overclocks, especially in games. 3.2/2.8 Ghz CPU/NB beating 4/2 Ghz CPU/NB? Have a look. Seems like a lot of performance can still be extracted from K10.5

Read it again, where does the 3.2 config beat the 4ghz testbed ? HawX ? the most unrealible benchmark of them all, run it 4 times and get 4 times different scores, sometimes miles apart...

Why didn't he test with the same ram speeds, why do you think the scaling is so high ? coz he goes from 1066 to 1600mhz, or from 1333 to 2000mhz, not alone NB speed gains...

He always refers to NB OC, not adding the memory OC, this till the last phrase where he mentions both jield the gains...

2000mhz on the ASrock, darn man I need proof for that, text is nice, screenshots are better.... warning Asus ASrock inbound !!

(I'm just awake though , could have misread something) I'm just sceptical with everything I read and some things would have been done better via another approach...

[LightSpeed]

Read it again, where does the 3.2 config beat the 4ghz testbed ? HawX ? the most unrealible benchmark of them all, run it 4 times and get 4 times different scores, sometimes miles apart...

Why didn't he test with the same ram speeds, why do you think the scaling is so high ? coz he goes from 1066 to 1600mhz, or from 1333 to 2000mhz, not alone NB speed gains...

He always refers to NB OC, not adding the memory OC, this till the last phrase where he mentions both jield the gains...

2000mhz on the ASrock, darn man I need proof for that, text is nice, screenshots are better.... warning Asus ASrock inbound !!

(I'm just awake though , could have misread something) I'm just sceptical with everything I read and some things would have been done better via another approach...

I never said 3.2/3 CPU/NB beats 4/2 in all or even most benchmarks. Yes, it was HawX and even in SC II it was very close to beating it. Also, they have done the bench not once but four or so times.

Yes, it wasnt NB alone, but even then I thought memory gains alone dont impact performance much. Atleast thats what I recall from my original Phenom, but times may have changed. Dunno

Still, impressive gains and can only mean good things about Llano.

[Leeghoofd]

Cripple your system and run your ram at 1066 and then at 1600Mhz, do your fav tests , think you will see a gain... once above 1600 the gain is less... 1800 is still a sweetspot...otherwise it would be pointless to buy DDR3 is everything went super fast at only 1066Mhz

Secondly I still want to see them 2000mhz ASrock speeds... now that would really impress me... gonna ask me a sample... the PSC work like crap on the 890GX , would surprise me if they flew on the FX... but that aside, maybe I can find it out myself...

[demonkevy666]

this go's all the way back to original phenom when someone noticed it was scaling to 106% from fsb/HTT Over clock.

I haven't noticed any gains from going from ddr3 1600mhz from ddr2 800mhz in my fps.
I noticed gains going from a 1gbs 5870 to 2gb 5870 in dirt I went from 60fps with 8X to 80 Fps now.

[lopri]

Running DDR3-1066/CL5 doesn't seem to cripple 2000 MHz NB.

http://www.xbitlabs.com/articles/mem...x6-memory.html

Actually the above review might be to Leeghoofd's likings. Running 1066Mhz to 1600Mhz with lots of timings and finding there is no difference in performance. Well, obviously no gain there since NB is the bottleneck. He just wasted lots of time for finding out nothing..

Then he compares 2000 NB and 3000 NB with same memory DDR3-1333/9-9-9? It's almost as if he is too stupid or he didn't want to see any gain from NB overclocking. AT review makes more sense in that they are running the memory as tight as possible with given NB Mhz.

[Leeghoofd]

That’s what the next diagrams show. The results were obtained at an Uncore frequency of 3.0 GHz.

The OC tests are nice for ram tests, not NB limited there. But no idea why he only did 1333Mhz C9 for the NB tests...

My point is it's hard to judge the scaling when you alter all sorts of variables like the AT article... Which factor jielded the biggest gains ? The increased NB speed or the ram speed combined with the timings or both ? We know the answers, but an AMD nOOB that wants to know, can't derive it from there...

If you take Xbit's tests, look at the gains from 1000Mhz C7 to 1666Mhz C7 in eg FC2 : that's 8FPS... If he tested the same speed at eg 2000NB then we would be able to tell which setting was more or as efficient yes or no...

If I compare something it's apples vs apples, not apples vs apples mixed with oranges... both articles have their strengths and flaws... sometimes you can't keep all variables in line, but keep them then as close as possible...

[Mechanical Man]

^^ for you to find out that, one would need to run many different nbclock + memclock combinations to find which ever is the bottleneck at given time.

[Leeghoofd]

Yes and no... If they only took the fastest setup of the stock 3.2CPU speed and 1600CL7 --> then vary eg 2000-2400-2800NB speed and do the same for the fastest OC setting... That would have made that article far better to my liking... as you get more variables to look at... It's notmy intention to thrash both articles, it's just if you put up a title like that, make sure you have loads of data to show what happens under what condition... I know how time consuming such subjects are. Heck why I avoid them in the first place lol

[generics_user]

great test!

[feniksgod]

Let's write what they want, but the truth remains one, a exactly high-clocked core speed is a better than overclocked CPU-NB.

[demonkevy666]

Let's write what they want, but the truth remains one, a exactly high-clocked core speed is a better than overclocked CPU-NB.

I don't think so.
slow Nb usually bottlenecks 4.0ghz cores.

[feniksgod]

I don't think so.
slow Nb usually bottlenecks 4.0ghz cores.

Yes, it is true, but I speak for example about 2600mHZ NB Over this is usually decreased core speed or require more voltage to stable corespeed.
P.S Absolute truth is this. Everything depends on specific CPU Unit.

[chew*]

From an effeciency stand point some of you are wrong.

3400 cpu/3300NB is more effecient than 4000cpu 2700 NB and this is what this person probably stumbled on. Both = a combined total of 6700mhz but both are not equal in terms of effeciency.


As nb approaches CPU 1/1 ratio effeciency goes through the roof. Problem is at exactly 1/1 there is an anomally where effeciency goes in the toilet.

I'm sure Raja knows and or would have picked up on this but he's not the one that did the testing.

[Alex-Ro]

Do Thuban's reach common higher CPU-NB frequency compared with Deneb?Revision wise?I saw lots of persons running air with CPU-NB on 3.000 MHZ versus classic Deneb where 2600-2800 top up...

[chew*]

Do Thuban's reach common higher CPU-NB frequency compared with Deneb?Revision wise?I saw lots of persons running air with CPU-NB on 3.000 MHZ versus classic Deneb where 2600-2800 top up...

I think the NB on thuban is just more consistant and not so all over the place, deneb chip to chip NB varied alot up to 400 mhz.

I'm seeing maybe on air a 200 mhz variation at most with thuban.

[FlanK3r]

Chew: ufff, really is better example 3700 MHz + 3000 MHz than 4000MHz + 2600 MHz?! In practice? I thought, in practice are not so big jumps..yes, NB is cool for performance, example setings 3850/2800 than 4000MHz/2000MHz, but so low CPU clock with high NB clock vs high CPU clock and middle NB clock?

[chew*]

Chew: ufff, really is better example 3700 MHz + 3000 MHz than 4000MHz + 2600 MHz?! In practice? I thought, in practice are not so big jumps..yes, NB is cool for performance, example setings 3850/2800 than 4000MHz/2000MHz, but so low CPU clock with high NB clock vs high CPU clock and middle NB clock?

I'm not making mention of overall speed. Only making mention of effeciency.

To make that compare you need to take into consideration the ultimate max effeciency and that is NB 100 mhz less than cpu.

[pcnazz]

It all depends on the application where running . For example wprime does not use the cpu/nb frequency as well as lets say super pi does .