AMD Zambezi news, info, fans !

[Particle]

Historically, node shrinks haven't worked magic for AMD. I know you haven't seen many shrinks yet, but that's just how it has been for them to date. The most lackluster shrink in my memory was when Thoroughbred cores came out on 130nm. They had trouble matching what 180nm had been doing at first.

HKMG may change things or maybe it will be a good node for AMD regardless. You just can't assume that a new node will mean a magic 20-25% frequency increase based on history, and that's my only real bone to pick here. People are expecting 5 GHz when even 4 GHz is a mixed bag? They're setting themselves up for disappointment.

[BeepBeep2]

Historically, node shrinks haven't worked magic for AMD. I know you haven't seen many shrinks yet, but that's just how it has been for them to date. The most lackluster shrink in my memory was when Thoroughbred cores came out on 130nm. They had trouble matching what 180nm had been doing at first.

HKMG may change things or maybe it will be a good node for AMD regardless. You just can't assume that a new node will mean a magic 20-25% frequency increase based on history, and that's my only real bone to pick here. People are expecting 5 GHz when even 4 GHz is a mixed bag? They're setting themselves up for disappointment.

Hasn't Llano already hit 4.6 Ghz? Thats why I made the post I made. ...and theres a GPU on that die so

[drfedja]

Bulldozer is high frequency design. Only limit for clocking 8 core BD@5 GHz is TDP. 3.8 GHz version has turbo core for half modules up to 4.8 GHz! That is stock frequency. Here is my calculation:


This is for only 20% of IPC improvement over Stars core.

some prediction for overclocking:

[XRL8]

Historically, node shrinks haven't worked magic for AMD. I know you haven't seen many shrinks yet, but that's just how it has been for them to date. The most lackluster shrink in my memory was when Thoroughbred cores came out on 130nm. They had trouble matching what 180nm had been doing at first.

HKMG may change things or maybe it will be a good node for AMD regardless. You just can't assume that a new node will mean a magic 20-25% frequency increase based on history, and that's my only real bone to pick here. People are expecting 5 GHz when even 4 GHz is a mixed bag? They're setting themselves up for disappointment.

Uhm.
Im old.
Palominos were pretty lackluster in overclocking department, 1.7-1.8ghz was really max what they do.
Thoroughbred-A had a problem, yes, it overclocked 1.9-2.1 and was hot.It was a failed design, AMD quickly had done Tbred-B, which overclocked like crazy.2.3ghz up to 2.5 on water.Had one DLT3C if i remember correctly, done 2330 on air.Than came the barton, which on the same node pushed up to 2.7ghz on water.My barton mobile did post on winter air at 2.95ghz ,My palomino 1700+ on the same board did 1.85ghz max POST.Dont tell me thats no difference.

However, if you want to poimt out failed shrink.
Theres the brisbane core.
65nm brisbane clocked worse than 90nm windsor, at the same time being slower clock for clock.

[crazydiamond]

did brisbane have any 125w cpus?

[BeepBeep2]

did brisbane have any 125w cpus?

I dont think it was able to clock high enough to fit in a 125w TDP.

They hit about 3100...that was, the 5000 BE, some hit 3.2, but they werent close to as good as Windsor F3s. 6400+ non BE was 125w....

[chew*]

I've got an idea. Stop speculating and just wait for the real deal.

[duron]

I've got an idea. Stop speculating and just wait for the real deal.

Speculation is in out Blood BUT if you have something to chew on pls. pls. pls. give us a peak

even they are speculating

http://www.overclock.net/hardware-ne...x-8130-es.html

[BeepBeep2]

They are always speculating.

[chew*]

Speculation is in out Blood BUT if you have something to chew on pls. pls. pls. give us a peak

Hmm ok I will leak something. Unlike Intel's P67 launch. SATA works.

[KiSUAN]

Hmm ok I will leak something. Unlike Intel's P67 launch. SATA works.

[duron]

wohoooooo we have a leak
SATA works!!!
now is that SATA2 or SATA3?

[Barr3l Rid3r]

990fxa-ud3 = best bet

[Vladutz20]

@Rider Please belive me ... it is inconceivable that Gigabyte doesn't have a digital vrm,or buttons to start your system or to make an overclock or maybe a EFI bios or maybe a debug led.
By OBr Crosshair 5 is no1 right now. Belive me i had to be a gigabyte fan but right now I hate them. We live in 2011 not in 1990 )

[FlanK3r]

I've got an idea. Stop speculating and just wait for the real deal.

I have some idea ...But u right, better is waiting, I only like not some bull*ts here :-/. Example post "1053", its impossible (4.8 GHz stock turbo clock).

[Particle]

Uhm.
Im old.

I was primarily addressing beep. You do seem to agree with me in spirit though, noting the 65nm shrink's less than steller performance jump. The string of these shrink events are the history I base my generalization on. It's more historically accurate than someone expecting a major increase.

Bulldozer is high frequency design. Only limit for clocking 8 core BD@5 GHz is TDP.

My point isn't in regards to BD. I've been referring to beep's quote, "Considering a simple Stars shrink to 32nm should do 4.6-4.8 Ghz I would guess." It's a bad assumption based on history.

As for BD itself, we don't have any idea what frequency target the BD design had in mind during design, just that it should be higher than Phenom II. Hopefully there aren't any major problems with that goal, but if there are by the time the chips hit retail we'll likely see a later stepping fix them. I'm not worried about it. Also, 4.8 GHz turbo is old info. The latest rumors are that the 3.8 turbos to 4.2. Also keep in mind that these are rumors. We don't know any launch models with certainty.

[FlanK3r]

4.8 Ghz turbo was fake from 1.4., unfortunately. Example...SB likes hight clocks too, but we have stock models with 3.4 GHz and up to 3.8 GHz Turbo (95W TDP), potentially could be SB about 3.9GHz and up to 4.3 GHz Turbo with 125W TDP. So...4.8 GHz BD really not

[marten_larsson]

I have some idea ...But u right, better is waiting, I only like not some bull*ts here :-/. Example post "1053", its impossible (4.8 GHz stock turbo clock).

If you read carefully you'll see that he's projecting OC performance of 4.8GHz. He also states 1GHz turbo for half modules engaged and 400MHz for all cores activated which seems reasonable.

[Sn0wm@n]

ive got a question ... how can turbo be only 400mhz when its 500mhz on the server platform on all 16 cores ????

[undone]

As far as we know the Zambezi ES reach 3.2GHz+, retail version often higher than ES about 20%-30% in freq, so perhaps those spec on the wikipedia(8150,C0,4GHz) may be reliable however as a clue.

[marten_larsson]

Well, his 400MHz isn't unreasonably high, that's what I was getting to. But I agree, some 6-800MHz for all cores if TDP allows is more in line with what I'm projecting.

[BeepBeep2]

I was primarily addressing beep. You do seem to agree with me in spirit though, noting the 65nm shrink's less than steller performance jump. The string of these shrink events are the history I base my generalization on. It's more historically accurate than someone expecting a major increase.
My point isn't in regards to BD. I've been referring to beep's quote, "Considering a simple Stars shrink to 32nm should do 4.6-4.8 Ghz I would guess." It's a bad assumption based on history.

Bad assumption based on history?

...they went from ~2.6 Ghz to ~3.4-3.7 from 65nm to 45nm, also added major power efficiency improvements, keep in mind Phenom II is the same core as Phenom it was a shrink with optimizations and extra cache...and look how much faster the same core is CPC.

Not sure why you would focus on a process two generations back instead of focus on the most recent. If we get gains at 32nm like we did at 45nm, then Stars would easily be doing 5 Ghz and have 5-10% better IPC.

[undone]

...they went from ~2.6 Ghz to ~3.4-3.7 from 65nm to 45nm

Initial Shanghai has only 2.6-3.0GHz, later amd release 3.2-3.4Ghz Deneb, so if Zambezi higher than 3.5Ghz even with turbo it may not be a disappointment. Not to mention ES has 3.2Ghz clock, only ES.

[BeepBeep2]

Fastest 65nm was 2.6...

Thats what I'm saying.

Their maximum on 65nm was about 2.6-2.7 Ghz if they were to push frequencies and on 45nm the same core made it all the way to 3.7 Ghz.

To have the same freq. scaling they would need to start out with a 3.7 Ghz chip. Not that it matters with a different architecture...

[nex_73]

As they managed a 2.5 multiplier turbo on half the cores for the 1055T, i'm thinking in the lines of:
Attachment 116863