AMD "Piledriver" refresh of Zambezi - info, speculations, test, fans

[Daveburt714]

Some A10 Trinity info (including Vids)...

http://tbreak.com/tech/2012/04/amd-a...st-impression/

Not alot of details, but the DIRT3 demo is pretty impressive for a laptop!
If the desktop version comes out around the same time (as suggested) I may be updating my HTPC in the next few months...

The "Steady Video" thing looks promising too!

[Smartidiot89]

And the speed of the cache or the memory access it's useless for you.
Different families, what you can't compare core i3 with i5 or i7? Scoket 1155 with 2011?
It's a funny thing that Intel Cpu's during last years just improved over and over in Super Pi, with generations and arhitectures, and IPC and overall performance also just over and over improved.

And yet comes BD, with a big hit in Super Pi compared with Phenom.
And what a surprize, single thread is down to earth and multithread almost a fail.
Don't writte of x87, do you think that it's used only in Super Pi.

No modern code runs on x87. SuperPi is useless (but fun) for any performance comparisons. Much of the performance increases we see comes from the new/improved instruction sets, so SuperPi is not worth one thing when comparing a new architecture really. There are already SuperPi results out for Trinity, extrapolating them towards a direct comparison for Llano isn't even an option imo.

There isn't much time left until Trinity is released, so all these speculations and trying to use SuperPi as a metric should be history shortly...

[xdan]

Anyway you put it if BD would have been better in Super Pi than Phenom instead of worse it would have been the cpu which all people expected.

[RussC]

Exactly,
The benchmark is not useless, sorry. Its completely informative. Too me, AMD can't produce a processor that can touch INTCs. Untill AMD makes a processor that execute x87 code as well as INTCs, they wont have the performance of INTCs and be the red-headed bastard child that gets kicked to the curb. Sorry. This benchmark tells me alot.

RussC

Anyway you put it if BD would have been better in Super Pi than Phenom instead of worse it would have been the cpu who all people expected.

[Smartidiot89]

If BD performed better in SuperPi I wouldn't have been any less disappointed with it. x87 executions meens absolutely nothing today. AMD doesn't need to be able to execute SuperPi (sorry I meen the "all-important" x87) well. Sandy Bridge (2600K) is about twice as fast than Bulldozer (FX-8150) in SuperPi, the difference i single-threaded performance isn't that huge. SuperPi is a useless metric, which is very entertaining for overclockers and benchers but that's just what it is.

[informal]

Stock vs stock, 2600K has 37% lower single thread runtime in THG benchmark suite that measures single thread efficiency. These are hard facts.
http://www.tomshardware.com/reviews/...0x,3075-8.html

2600K , 8:11
8150, 11:13
673s/491=1.37 or 37% lower runtime. So approx. on average intel 2600K is 37% faster in single threaded tasks. In MT the runtime is just 12% better for 2600K (again stock vs stock).

If we look at the 3960X and normalize for clockspeed,intel achieves 1.5x lower single thread runtime than AMD with FX. Some may think this gap cannot be overcome with any soon to come core but it's not quite the case. What AMD needs to achieve parity with SB-E in single thread runtime? They need a 2 successive IPC jumps of approx 10% each(BD->PD 10%, PD->SR 10%) and they need a clock speed uplift with SR core @28nm to 4.5Ghz/5Ghz stock/Turbo. This would push AMD's single core performance right in the league of SB-E,again stock vs stock: 1.1 x 1.1 x 5Ghz/4.2Ghz~=1.44x. Runtime of SR core in the same same suite THG used would land around 7:48 .

Now to achieve 2 successive IPC jumps of ~10% is very hard and although SR may do it over PD,I doubt PD will be 10% faster than BD. Clock speed uplift on the other hand is possible since not only FX launched on sucky 32nm process from GloFo, both PD and SR will incorporate new technological solutions that will push the clock speed higher. PD (and SR) will have resonant clock mesh,while SR will be done on smaller node (28nm) so there will be some natural power draw reduction,die area reduction and clock speed gain.Pipeline itself is designed for higher clockspeeds so I think AMD knows how high can they clock this thing,provided power is under control.

[tbone8ty]

http://www.nordichardware.com/news/6...n-folding.html

so what folding benchmark did they use? my girlfriend has an A8-3500m HP laptop wanna see if they are similar scores...

[Kristoferr]

Need to know 4500M clock speed, to be able to make any sort of perfomance predicions

[informal]

4500M is supposedly 1.6Ghz/2.2Ghz ,def/turbo.
http://www.sibvez.ru/shop/UID_13697.html

[Kristoferr]

Well, this is very good then! Looks promising for Piledriver to be a lot better than Bullzoder too.

[Mechanical Man]

Well, trinity could be in higher turbo state for longer time than llano. So that doesn not really tell much.

[BeepBeep2]

4500M is supposedly 1.6Ghz/2.2Ghz ,def/turbo.
http://www.sibvez.ru/shop/UID_13697.html

[Tomasis]

Well, trinity could be in higher turbo state for longer time than llano. So that doesn not really tell much.

well if it does in the same thermal , its still good!

[Tomasis]

what kind of programs make use of increased Integer speed?

Educate me

[Darakian]

what kind of programs make use of increased Integer speed?

Educate me

Video/Still image editing and a number of server apps (web servers, ftp servers, etc...)

[Macadamia]

FP throughput is still worryingly problematic... but good to hear that Turbo's actually working now! (gasp!)

In Llano desktop/NB it was hyper-conservative and felt like a scam- most OEMs sold the CPUs at turbo speed marketed but the chips themselves almost never, ever reached them (unless you did some tweaking)

[BeepBeep2]

FP throughput is still worryingly problematic... but good to hear that Turbo's actually working now! (gasp!)

In Llano desktop/NB it was hyper-conservative and felt like a scam- most OEMs sold the CPUs at turbo speed marketed but the chips themselves almost never, ever reached them (unless you did some tweaking)

It might have to do with the architecture...I'm just guessing around though. Phenom II X6 turbo never really worked either while BD turbo worked very well.

It certainly made BD look better than it was compared to Phenom II...

[superpyton]

Need to know 4500M clock speed, to be able to make any sort of perfomance predicions

A6-4400m will run at 2,6 / 3,2 GHz (source), so A8-4500m should be clocked higher (maybe 2,8 / 3,4 GHz?)...

[Kristoferr]

Not too bad but i expected more from CPU side.

The new Trinity based quad-core runs at 3.8GHz, but it overclocks to 4.2GHz on Turbo Core. It features 4MB of L3 cache and powerful HD 7660D graphics, all packed in a 100W TDP envelope.

http://www.fudzilla.com/home/item/26...enches-are-out

So can we predict that ICP of desktop chip will be around 20-25% slower than i5? Looks like with Ivy, the cap between them widens some more.

[hwgeek]

http://www.legitreviews.com/news/12666/

finally I can buy slim case pc with good GPU! 4 3d programs and some games:-)

[Kristoferr]

Thats 11% or more CPU points with 3800/4200mz vs 2900mhz on 3850K.
Faildozer all over again.

[demonkevy666]

Not too bad but i expected more from CPU side.

http://www.fudzilla.com/home/item/26...enches-are-out

So can we predict that ICP of desktop chip will be around 20-25% slower than i5? Looks like with Ivy, the cap between them widens some more.

that's funny trinity was support to have 25% high cpu vs bulldozer. but bulldozer was 15% slow then phenom II and abuot 35-45% slower then sandybridge.

[Oliverda]

Thats 11% or more CPU points with 3800/4200mz vs 2900mhz on 3850K.
Faildozer all over again.

It's an APU.

[flyck]

Thats 11% or more CPU points with 3800/4200mz vs 2900mhz on 3850K.
Faildozer all over again.

So same process
same TDP

resulted in:

better cpu performance
better idle power consumption
better gpu performance

clearly a fail...

Trinity is a major step forward compared to llano.

[Kristoferr]

GPU is pretty much the only thing whats improved. Not CPU which is actually performing worse per mhz than Llano is. Yes, TDP is the same, but that due better GPU and perhaps little better optimized base voltage of CPU and GPU.
Yes overall perfomance is fine. I am more thinking about Vishera than Trinity here. Hopefully L3 will make desktop Piledriver cores perform better, but i doubt it will make big enough difference.