Athlon II (Regor) In the house!

[Smartidiot89]

Max stable on Air? :O

[Lightman]

WOW you're brave mAJORD!
Popping lid in such a new CPU
Has it died on you or you were sure it will survive this operation?

When we can expect first results from you?
BTW in Athlons 64 mem controller was in sync with CPU clock, so X2 6000+ ran at 3GHz

Independent IMC clock is required for independent core clocking. Also enabled K10 to maintain same memory frequency across a range of CPU multipliers (K8 memory speed varied depending on CPU multi.). Same as in Griffin, Regor will have ability to clock cores at different speeds. I only hope AMD added option to completely shut down a core when not needed! That would be cool on a laptop!

[G.Foyle]

Regor is 117 mm^2

117.5, to be exact

[Glow9]

Maybe I missed it but does this chip have unlockable cores?

[cbjaust]

Maybe I missed it but does this chip have unlockable cores?

It's a new mask, so no, it's dual core only and a such has no additional cores to unlock.

Sempron II anyone? Single -> dual core !!

Cheers

[mAJORD]

WOW you're brave mAJORD!
Popping lid in such a new CPU
Has it died on you or you were sure it will survive this operation?

When we can expect first results from you?
BTW in Athlons 64 mem controller was in sync with CPU clock, so X2 6000+ ran at 3GHz

Independent IMC clock is required for independent core clocking. Also enabled K10 to maintain same memory frequency across a range of CPU multipliers (K8 memory speed varied depending on CPU multi.). Same as in Griffin, Regor will have ability to clock cores at different speeds. I only hope AMD added option to completely shut down a core when not needed! That would be cool on a laptop!

thanks Lightman..

The only CPU I'm not sure if it still works is the 9850BE, due to the heating process to flow solder, I bent the package, and failed to straighten it before it cooled (it was ****** hot!) and it's stayed bent. Have to try reheat+straighten and see if it hasn't fractured any of the bumps under the die!

I'll post results tonight, as an idea though, Agena/Kuma like perf. clk/clk Looking at around..
2.5-10% on basic integer varies wildly
10-15% for gaming/3dmark, (With a 2000NB)
POV ray in the 20%+
Encoding 10-15%

and an interesting one.. SSE2 Mandrelbot bench.. ~75%

What about power consumption?

Coming tonight also, need to swap a low power vid card in, or maybe throw them in a MINIX

Quick and dirty check:

45nm Regor: 8.2mm x 13.0mm = 107 mm^2
65nm Griffin: 11.1mm x 16.8mm = 186 mm^2


Regards, Hans

wow, Griffin was larger than I thought

117.5, to be exact

Is that an official figure? My measurment may be a bit off, i'll borrow some decent verniers


CBJ: Re-run at 3ghz

[Glow9]

It's a new mask, so no, it's dual core only and a such has no additional cores to unlock.

Sempron II anyone? Single -> dual core !!

Cheers

Was curious due to the thread in the news section

[Trunks]

t



Is that an official figure? My measurment may be a bit off, i'll borrow some decent verniers

yes

@xoqolatl darn you!

[G.Foyle]

I cannot confirm nor deny if it's official or not you will have to believe me
Based on die picture and this number I have estimated die width and length to be ~ 13.8 mm x 8.5 mm.

[Hans de Vries]

wow, Griffin was larger than I thought

No, it's not that big, something got messed up here

This should be a better measurement from the photos,
assuming Socket S for Griffin (35x35mm) and Socket AM2
for Regor (40x40mm):

45nm Regor: 8.45 mm x 13.9 mm = 117 mm^2
65nm Griffin: 10.0 mm x 15.6 mm = 156 mm^2

Regards, Hans

[qcmadness]

( Edit: Jun, 1st ) This should be better a better measurement:

45nm Regor: 8.45 mm x 13.9 mm = 117 mm^2
65nm Griffin: 10.0 mm x 15.6 mm = 156 mm^2

Regards, Hans

117mm^2

A new breed of possibly mobile processors.

[mAJORD]

Sounds better hans.. Still seems a good shrink considering the larger cores

Anyway I've updated the OP with out-of-box stock vs stock 6000+ comparisons, brisbane and windsor.

NOTE that's NOT my uARCH comparison tests.. I'm Re running them Vs Windsor at 2.8ghz to get equal RAM clocks (even multi), and at 2800NB (sync)

[qcmadness]

Sounds better hans.. Still seems a good shrink considering the larger cores

Anyway I've updated the OP with out-of-box stock vs stock 6000+ comparisons, brisbane and windsor.

NOTE that's NOT my uARCH comparison tests.. I'm Re running them Vs Windsor at 2.8ghz to get equal RAM clocks (even multi), and at 2800NB (sync)

most of the ~750M transistors go to the cache but not core.

Each core (with L1 + L2 cache) only consists less than 100M transistor.

Barcelona / Agena (4 core, ~4.5MB total cache): ~450M
Shanghai / Deneb (4 core, ~8.5MB total cache): ~750M

[informal]

mAJORD great work man,you have ballz of steel .
The core looks sexy naked .
I'm looking forward for your head on duel with K8 @ 2.8Ghz and synced NB for Regor at the same clock . Keep it coming!

[Hans de Vries]

Sounds better hans.. Still seems a good shrink considering the larger cores

Anyway I've updated the OP with out-of-box stock vs stock 6000+ comparisons, brisbane and windsor.

NOTE that's NOT my uARCH comparison tests.. I'm Re running them Vs Windsor at 2.8ghz to get equal RAM clocks (even multi), and at 2800NB (sync)

Nice! AMD's much needed μP for the mobile segment.

Regards, Hans

[mAJORD]

Any requests for added benchmarks?

I'd like to re-run tonight, so pls post here

[qcmadness]

Any requests for added benchmarks?

I'd like to re-run tonight, so pls post here

I would like to know if the L2 cache is slower or faster than Phenom equivalent

[mAJORD]

hi QC, I did test that with cpu-z latancy tool, but forgot to post results.. it's the same as Phenom..

L1: 3 Cycles
L2 15cycles..

so slower than windsor (12 cycles)

[qcmadness]

hi QC, I did test that with cpu-z latancy tool, but forgot to post results.. it's the same as Phenom..

L1: 3 Cycles
L2 15cycles..

so slower than windsor (12 cycles)

Hello mAJORD

15-cycle L2 cache latency... but now the caching is more complex than before...

[imamage]

Can you please try to run some Sandra / Everest Memory Bandwidth Test?
It should be interesting to see the difference with K8 and K10.5
Thanks !

[informal]

hi QC, I did test that with cpu-z latancy tool, but forgot to post results.. it's the same as Phenom..

L1: 3 Cycles
L2 15cycles..

so slower than windsor (12 cycles)

Slower but has double width iirc(256bit wide).

[high5]

mAJORD,

Could you please check whether kink's PhenomMsrTweaker works with Regor.

Thanks.

[mAJORD]

no worries high5, if someone else hasn't by tommorow. Still getting k8 results.

imamage, sure. am including all the benchmark suite too.

[hstuehmeyer2000]

Hey mAJORD, could you give me some undervolting numbers? My 7750BE is kind of toasty

[mAJORD]

K8vs K10 uARCH comparison results posted!

Awaiting feedback and more benchmarks (and prepared to run new tests on both CPU's again!)

Updated OP with results, and descriptions.

Hey mAJORD, could you give me some undervolting numbers? My 7750BE is kind of toasty

will try to fit that in this w/e. i am interested also i have to say.

mAJORD,

Could you please check whether kink's PhenomMsrTweaker works with Regor.

Thanks.

a no go. CPU type not supported on install