Thuban & Zosma info/code name from anandtech "Juicy"

[ajaidev]

I didn't say it is hw based... It's called C-state performance boost,in other words using low power state of idle cores to boost clocks on fewer non-idling ones. Wheter it's required to have AOD or it will be BIOS' responsibility ,I do not know.

Ya thats what i was forgetting the C state AM did talk about it somewhere i think

I hope its handled by the mobo and not AOD tough...

[accord99]

Dude, HT on 920/860 is around 15-20% extra performance in multithreaded apps.

But then each core on the i7 is also more powerful.

a 6 core Thuban at 3ghz will beat the 920/860 (they revolve around 16-17k in cinebench 10)

I suspect around 20k in Cinebench 10, which is around what a I7 975 gets out.

It'll be more like 17K.

[Nedjo]

its just seems like its taking forever to get out something that for most of us was suppose to be as simple as deneb with 2 more cores.

well probably isn't that simple 'cos it's not just Deneb+2 more cores

[Oliverda]

the delay to simply make it, not saying their target dates were off. its just seems like its taking forever to get out something that for most of us was suppose to be as simple as deneb with 2 more cores.

It's not Deneb with 2 more cores. It's Istanbul with 0 more core.

BTW I've just found some evidence about the Istanbul's DDR3 support. It has combo DDR2+DDR3 compatible IMC like Deneb and all other AM3 CPU.

(Istanbul is a HY-D0 stepping CPU.)

[Mats]

well probably isn't that simple 'cos it's not just Deneb+2 more cores

I think you're right.
A better question is how Thuban could take so long time when Istanbul has been around since last summer?

I mean, the biggest step must have been to go from 4 to 6 cores, with Istanbul, so how come it took twice as much time making the desktop counterpart (counting from the first 45 nm)?

Shanghai - nov 2008
Istanbul - july 2009
Thuban - april 2010

[Nedjo]

It's not Deneb with 2 more cores. It's Istanbul with 0 more core.

I admire to your persistence

[noee]

Is it feasible to assume, being conservative, that AMD would make some design changes for Thuban in preparation for 32nm? Just in case.

[Lokinhow]

in every thread you see the same reply, delayed, why not released already....etc..

AN UPDATE AGAIN:
THUBAN + ZOSMA are based on LISBON core architecture, which is due to release soon on server platfrom
ALL AM3 released CPU's are AM2+ backward compatible if it is supported by BIOS also Thuban and Zosma



informal, don't give people to much hope - ideas on the turbo boost feature, there is no such thing build into the core, it is probably SW based, overdrive ......

Can you tell me what Lisbon is different in comparsion to instabul? Differences that may be relevant in desktop.

[Mechromancer]

I'm stoked that I have a great upgrade path for my AM2+ platform! This is truly great news. My Folding@Home/main rig will become a BADASS.

[doompc]

Zosma must be Thuban with two disabled cores (probably for some ACC unlocking action)

The C-State thing is the hardware C-State switching present on C3 rev Phenoms. (faster than software based)

The Leo Platform is DDR3, so we won't see any 890GX/FX with DDR2 memory, but Thuban is probably DDR2 compatible.

Zambezi (Bulldozer) will come in AM3r2 form and probably all AM3 boards will support it with an updated BIOS, but it won't support DDR2 memory anymore.

[Oliverda]

Can you tell me what Lisbon is different in comparsion to instabul? Differences that may be relevant in desktop.

Faster HyperTransport (from max. 5.2GT/s to 6.4GT/s). It is important for a platform with two or more CPUs but doesn't really matter on desktop (1 CPU). Anyway the HT can't run at higher freq than NB so if they would like to use this new capability then the NB must run at a >2.6GHz freq.

Cool Speed:

"The processor provides protection to processor integrity by reducing the power states when it reaches a temperature limit. This will help platform providers in reducing system fan speeds which eventually could result in better efficiency."

...and + C1E.

Zosma must be Thuban with two disabled cores (probably for some ACC unlocking action)

The C-State thing is the hardware C-State switching present on C3 rev Phenoms. (faster than software based)

The Leo Platform is DDR3, so we won't see any 890GX/FX with DDR2 memory, but Thuban is probably DDR2 compatible.

Zambezi (Bulldozer) will come in AM3r2 form and probably all AM3 boards will support it with an updated BIOS, but it won't support DDR2 memory anymore.

I agree.

[Karolis]

Anyway the HT can't run at higher freq than NB so if they would like to use this new capability then the NB must run at a >2.6GHz freq.

Yet current Istanbul Opterons have HT@2400 MHz, and NB@2000/2200 MHz.

[Oliverda]

Yet current Istanbul Opterons have HT@2400 MHz, and NB@2000/2200 MHz.

AFAIK the HT and the NB are in sync.

[ridney]

i guess we'll see samples of buldozer this year



more here:
http://vr-zone.com/articles/amd-2010...2010/8348.html

[initialised]

AMD Phenom II X6 1075T 6 ?? ?? 6MB 125W ??
AMD Phenom II X6 1055T 6 ?? ?? 6MB 125W/95W ??
AMD Phenom II X6 1035T 6 ?? ?? 6MB 95W ??
AMD Phenom II X4 960T 4 ?? ?? 6MB 95W ??
[B]

Drool, wants ES now

[Karolis]

AFAIK the HT and the NB are in sync.

Well in the AMD cpu comparison 'thingy' you can see the specs anyway. All hexacores have NB clock lower than HT clock.
http://products.amd.com/en-us/Optero...teron%e2%84%a2

[Oliverda]

Well in the AMD cpu comparison 'thingy' you can see the specs anyway. All hexacores have NB clock lower than HT clock.
http://products.amd.com/en-us/Optero...teron%e2%84%a2

You're right. I think that this asynchronous NB and HT clocks ability is an Istanbul specific feature.

[Nedjo]

You're right. I think that this asynchronous NB and HT clocks ability is an Istanbul specific feature.

well you can do the same on any K10(.5) CPU via BIOS settings!

[zalbard]

i guess we'll see samples of buldozer this year

Yes please!
Looking at that slide... Mmmm, delicious!

[ajaidev]

i guess we'll see samples of buldozer this year

[iig]http://resources.vr-zone.com//uploads/yantronic/amd2010/9.jpg[/img]

more here:
http://vr-zone.com/articles/amd-2010...2010/8348.html

Thanks

Nice slides muuumm, 4-8 cores means just 2-4 modules i wanted more modules....

[[XC] Synthetickiller]

One last upgrade path for me. I'll have to figure out what to do with my PII 945 and q6600 since this would be a good, cheap upgrade path for me to finally have a 4ghz cpu.

On a serious note, will this cpu be limited with ddr2 and sb600?

[Oliverda]

well you can do the same on any K10(.5) CPU via BIOS settings!

Yes but the HT freq cannot be higher than the NB freq. Looks like that the Istanbul's HT freq can be higher than the NB freq.

[SocketMan]

Yes but the HT freq cannot be higher than the NB freq. Looks like that the Istanbul's HT freq can be higher than the NB freq.

I think we can do it now as well with C3 revision,not 100% since I don't have any (C3).

[[XC] gomeler]

Dude, HT on 920/860 is around 15-20% extra performance in multithreaded apps.

a 6 core Thuban at 3ghz will beat the 920/860 (they revolve around 16-17k in cinebench 10)

I suspect around 20k in Cinebench 10, which is around what a I7 975 gets out.

a 6 core thuban at 3ghz will be a serious competitor in multithreaded apps.

Only the westmere cpus will be able to beat it, but at a much higher price tag.

Are you forgetting the fact that the 920/860 will do 4GHz easy? C3 965s aren't as hot as C2s but they are still hot enough that adding 2 more cores makes me wonder if you'll be able to not be thermally limited in overclocking. I guess we'll see what happens when the chips become available

[Lokinhow]

Are you forgetting the fact that the 920/860 will do 4GHz easy? C3 965s aren't as hot as C2s but they are still hot enough that adding 2 more cores makes me wonder if you'll be able to not be thermally limited in overclocking. I guess we'll see what happens when the chips become available

but thuban isn't C3...