AMD Piledriver rumours ... and expert conjecture

[Reynod]

We have had several requests for a sticky on AMD's yet to be released Piledriver architecture ... so here it is.

I want to make a few things clear though.

Post a question relevant to the topic, or information about the topic, or it will be deleted.

Post any negative personal comments about another user ... and they will be deleted.

Post flame baiting comments about the blue, red and green team and they will be deleted.

Enjoy ...

 

[EXT64]

The Piledriver desktop chip should use the same AM3+ socket, and the PD Server chip should continue to use the current G34. I think though that these will be the last chips for both platforms.

 

[$hawn]

to me, it looks like 20% via ipc and 50% less power usage for same performance (thus at least 50% more performance per watt. in comparison to bd)

bro, do u realize how how massive a 50% savings in power is? looks like AMD might finally get some share back on the server side
In fact, i think servers running BD mite find it a good enof reason to upgrade to PD by the end of the year 😀😀

 

[skitz9417]

cool so they will use am3 when will they be releasted

 

[truegenius]

bro, do u realize how how massive a 50% savings in power is? looks like AMD might finally get some share back on the server side
In fact, i think servers running BD mite find it a good enof reason to upgrade to PD by the end of the year

yup that's massive performance per watt increase.
as you can see that trinity got 10-15% more performance at same clock (estimated) and use only 60-70W thus (95w/70w=1.35) 1.35 ratio for performance per watt (excluding ipc), now add l3 cache and ipc gain to it then we have 1.35+0.15+0.05=1.55=~50% more performance per watt than bd

cool so they will use am3+ when will they be released

 

[$hawn]

as you can see that trinity got 10-15% more performance at same clock (estimated) and use only 60-70W thus (95w/70w=1.35) 1.35 ratio for performance per watt (excluding ipc), now add l3 cache and ipc gain to it then we have 1.35+0.15+0.05=1.55=~50% more performance per watt than bd

Actually there's a catch Adding an 8MB L3 cache will drain more power
So a more realistic calc wud be , 1.15 form IPC, and another 5% form L3 and finally 10% from cyclos tech......adds upto to 30% as the best case scenario

 

[skitz9417]

does any know when they are going come out

 

[amdfangirl]

Who knows... depends on AMD really.

 

[BBN8]

I'm still using a 955 be and really want to upgrade my cpu soon and would like to see what amd can offer before I switch to intel.I haven't really followed PD at all can someone kinda catch me up? I'd much appreciate it I don't really want to read through 100's of pages. Thanks alot!

 

[amdfangirl]

I'm hoping that PD will be decent, if not I'll shoot on a Phenom II X6.

 

[skitz9417]

it would good to see amd cpu come back to the best cpu and put intel to shame

 

[$hawn]

I'm still using a 955 be and really want to upgrade my cpu soon and would like to see what amd can offer before I switch to intel.I haven't really followed PD at all can someone kinda catch me up? I'd much appreciate it I don't really want to read through 100's of pages. Thanks alot!

Well PD is already tested to be 15% faster than BD at the same clocks......on top of that PD shall be clocked higher than BD at the same TDP level....so atleast a 20% faster than BD is as good as guaranteed

 

[EXT64]

I think AMD has a good shot at beating Intel in fully parallel Interger calcs, not a hope of catching intel in serial tasks, and a some chance of getting near intel in fully parallel float (if they add a 5th module, if not, not so much).

 

[gamerk316]

Well PD is already tested to be 15% faster than BD at the same clocks......

Just like PD was tested to be upwards 50% faster then SB before it released? Beware of *best case* testing. I suspect in non-threaded situations, your max performance gain will be closer to 5%.

 

[$hawn]

Just like PD was tested to be upwards 50% faster then SB before it released? Beware of *best case* testing. I suspect in non-threaded situations, your max performance gain will be closer to 5%.

Hmmm....its still ok BD cud hold its ground against the i5's and i7's in highly threaded apps....its single threaded performance that needed improving
And i got the 15% figure from tom's hardware review of the Trinity desktop chips

 

[gamerk316]

And what are you benching against? BD, or llano? Because llano is based off the Phenom II chip, and NOT BD.

 

[amdfangirl]

Here's your 15%.

 

[$hawn]

And what are you benching against? BD, or llano? Because llano is based off the Phenom II chip, and NOT BD.

i noe man...chill they did a test of the FX against Trinity desktop....both at same 3.8GHz clock.....to compare improvement in IPC

pls read,
http://www.tomshardware.com/reviews/a10-5800k-a8-5600k-a6-5400k,3224-2.html

 

[$hawn]

@AMDFangirl thanx

@gamerk316 Oh yeah, n btw that 15% improvement was WITHOUT the 8MB L3 cache that the BD enjoys

 

[truegenius]

http://media.bestofmicro.com/E/Z/341387/original/per core itunes.png
http://media.bestofmicro.com/E/X/341385/original/per core 3dsmax.png

Here's your 15%.

it is 4t VS 8T!!!
is it still 15%

😗 oh, it is single threaded task so 1 vs 1

 

[$hawn]

it is 4t VS 8T!!!
is it still 15%

😗 oh, it is single threaded task so 1 vs 1

yup ejjactly 😉

 

[muy]

I don't see any performance benefit gained wrt more channels ... look at Nehalem and SB ... they don't seem to need (benefit) from the increased bw.

the gpu would probably take advantage of a 4 channel memory config even if it would make little difference for the cpu. doubling memory channels from 64-128-256-512 has always greatly increased gpu memory bandwith in the past.

<cough>and now give me a 32 nm phenom II, i want a nice gaming cpu to go with my 6970 + crosshair V <cough>

 

[gamerk316]

http://media.bestofmicro.com/E/Z/341387/original/per core itunes.png
http://media.bestofmicro.com/E/X/341385/original/per core 3dsmax.png

Here's your 15%.

Both benchies are also very well threaded. So you have 15% overall performance improvements based sorely on IPC...in best case conditions. This infers that PD will have the same problems that BD had with non-scaling SW.

 

[$hawn]

Both benchies are also very well threaded. So you have 15% overall performance improvements based sorely on IPC...in best case conditions. This infers that PD will have the same problems that BD had with non-scaling SW.

itunes? Well threaded??!!! Am i missing something here??

Yes 15% only due to IPC, cyclos tech will possibly allow for even higher clocks at the same TDP rating, adding perhaps maybe another 10% more
And regarding highly threaded loads, like i said above, BD already manages to stay between the i5 and i7 performance boundaries

 

[jdwii]

Both benchies are also very well threaded. So you have 15% overall performance improvements based sorely on IPC...in best case conditions. This infers that PD will have the same problems that BD had with non-scaling SW.

Do you really hate Amd this much to say something untrue? Itunes is one of the worst software programs in the world (and by no means "well threaded") made by such a huge company.

Piledriver will be between 7-15% faster when compared to Bulldozer per clock on average probably around 7-10%. On top of this clock mesh will help improve performance even more while maintaining power consumption at a reasonable level.

Some of you guys really need to look at the Bulldozer vs the I5 2500K(Again) in most benchmarks besides games it does come close to beating it and some times it even comes next to the I7 2600K. The Ivy I5 is only 5% faster compared to Sandy, I'd say with a 20% performance increase a 8 core Piledriver will meet or beat the I5 ivy and come close to their I7 4 Core under most multithreaded apps. Its still going to be slower per core but more and more programs are beginning to use 4+ cores.

One of the main issues we saw with Bulldozer was Power consumption(the whole point of this design was to improve power consumption?) this has most likely been resolved with Piledriver. Improve this and we improve Heat and in return we can crank the clocks up a bit.

 

[gamerk316]

Do you really hate Amd this much to say something untrue? Itunes is one of the worst software programs in the world (and by no means "well threaded") made buy such a huge company.

Piledriver will be between 7-15% faster when compared to Bulldozer per clock on average probably around 7-10%. On top of this clock mesh will help improve performance even more while maintaining power consumption at a reasonable level.

Some of you guys really need to look at the Bulldozer vs the I5 2500K(Again) in most benchmarks besides games it does come close to beating it and some times it even comes next to the I7 2600K. The Ivy I5 is only 5% faster compared to Sandy, I'd say with a 20% performance increase a 8 core Piledriver will meet or beat the I5 ivy and come close to their I7 4 Core under most multithreaded apps. Its still going to be slower per core but more and more programs are beginning to use 4+ cores.

One of the main issues we saw with Bulldozer was Power consumption(the whole point of this design was to improve power consumption?) this has most likely been resolved with Piledriver. Improve this and we improve Heat and in return we can crank the clocks up a bit.

As I have said on MANY occasions: In applications that scale beyond 4 cores, PD > SB. But for the majority of SW, that is not the case.