The K10.5 Mystery - Llano's potential

[deadjon]

We all know about Llano - Its AMDs answer to all things integrated - Decent performing Dual/Tri/Quad Cores with embedded Graphics that make Intels HD3000 look like my Grandma in terms of pixel pushing power.

The recent release of the Llano based Athlon IIs (631 + 651) and their alleged Overclocking potential (apparently largley fake, an MSI Marketing Fad) has made me wonder - what Could AMD have done with the K10.5 Arch if they combined the tweaks from the Llano architecture with L3 Cache. The K10.5 revised architecture could be Made for higher clocks, lower power consumption and lower heat - Overclocking potential through the roof and Performance per clock higher than Deneb.

This could have helped them keep the Price/Performance crown.

Any thoughts on this - I would like some expert conjecture on the possibilities this could have brought.

Regards,

Deadjon

 

[deadjon]

Bump for intelligent discussion.

 

[Soul_keeper]

I think the majority of the improvement compared to the propus incarnation of stars come from the increased L2 size
AMD's L3 is relatively high latency and low bandwidth compared to the competition

It's hard not to draw a comparison to the yorkfield/wolfdale
where they had large L2 up to 12MB in size, only now are we seeing AMD break that with L2+L3 sizes

speaking from a core:core ipc point of view:

There are still ways in which the wolfdale is better than stars
ie: 3-issue vs. 4-issue for the integer core
AMD would need something beyond just increasing the cache sizes to make stars definitively beat wolfdale/yorkfield
avx would be a start, but not enough on it's own

ultimately imo stars was/is inferior to core2 (not a landslide tho) let alone the newer i-series
The use of 6 cores in thuban, large L3, and relatively high clockspeeds helped mask most of these shortcomings.

I think the hope should be placed on trinity/piledriver

 

[amdfangirl]

Definately.

AMD took a gamble with a new forward looking arch.

It lost.

See, problem is... less risk would mean that it would need to take the plune eventually.

 

[xaira]

i thought the 631 and 651 didnt have onboard graffix??? anyway, what messes with me is that the 631 and 651 have 100w tdp, i thought when the process got smaller the tdp was supposed to go down??? the 45nm athlon II series were 95w, so the 32nm is supposed to be 65...

and to answer your question, if amd had kust done k10.5 at 32nm and released it as phenomII, bulldozer would just sit on shelves, because the new phenom II quads owould pretty much have to start at something like 4.6GHZ stock, and everyone would buy these and leave the bulldozer octocores on the shelves, it would pretty much be netburst, but without the heat...

the phenom 980 at 3.7 would have to be srsly challenged, so it stands to reason that they might have even released a stock 5.0ghz quad, although i must admit i would like to see the performance with 8 real K10.5 cores on a single die

 

[nforce4max]

The problem comes mainly from poor yields early on coupled with already high volts at stock. The rest is the gpu being slightly a power hog as far as the whole apu goes during overclocking. The unlocked versions have proved to be amazing overclockers. It is a shame that the mobile versions turned out to be more popular and are unlocked making them popular for those who want the most on the cheap for mobile gaming. LoLs when I got my A8 3530mx at 3ghz on pretty much stock voltage and rock solid. A small bump on the volts and still stable at 3.6ghz

 

[BaronMatrix]

We all know about Llano - Its AMDs answer to all things integrated - Decent performing Dual/Tri/Quad Cores with embedded Graphics that make Intels HD3000 look like my Grandma in terms of pixel pushing power.

The recent release of the Llano based Athlon IIs (631 + 651) and their alleged Overclocking potential (apparently largley fake, an MSI Marketing Fad) has made me wonder - what Could AMD have done with the K10.5 Arch if they combined the tweaks from the Llano architecture with L3 Cache. The K10.5 revised architecture could be Made for higher clocks, lower power consumption and lower heat - Overclocking potential through the roof and Performance per clock higher than Deneb.

This could have helped them keep the Price/Performance crown.

Any thoughts on this - I would like some expert conjecture on the possibilities this could have brought.

Regards,

Deadjon

Going with BullDozer is future proof. K10.5 wasn't designed to go higher than 4GHz. When Intel HyperThreading came out it lowered perf in MANY cases because it needed a Windows patch. Bulldozer uses the HT scheduling which doesn't work. I would think that they are planning the release of the patch for the A10 Trinity. The proper scheduling in Win8 got up to 10% on a Pre Beta while most Core i5 tests (here at Tomz I believe) showed an overall decrease BECAUSE it's a PreBeta.

AMD is going away from the monolithic cores and looking more towards ISAs like XOP, AVX, AES, and FMA. I don't think SuperPi uses those.

At most Llano would get to 3.4GHz. The 2.9GHz quad is at 100W already. They probably don't want a 125W - though I see little problem because it has a GOOD GPU. The same GPU with Trinity would probably get a good 500MHz initially at 100W. If we extrapolate the Win 8 numbers and assume 15-20% (25%-30% total) more speed from the release version, Win7 can improve by at 15%, perhaps 20% for EVERY BD and Trinity is PD which has additional tweaks and everything that may end up in the B3 rev of FX. We can almost be certain that power will go down over time.

All in all, Deneb is a dead end. It can't be tweaked anymore without making a new arch.