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]

AMD (processors) is actually green too. AMD (Radeon) is red still though.

 

[dogman_1234]

 

[4745454b]

I'm sorry if this has been covered already, but I'm not reading 101 pages of past stuff.

I consider myself a reasonably smart guy. Sometimes however it takes my brain awhile to crunch stuff and let me know things. This is one of those times. PD will be a lot better then BD, and better then the "stars" Phenom II arch. I base this on the Trinity article that Toms did.

http://www.tomshardware.com/reviews/a10-4600m-trinity-piledriver,3202.html

I'm assuming as a "quad core" this is a dual module PD based APU. As before the L3 cache has been removed to make room for the GPU/IGP. The thing that finally clicked for me was they compared this to a Llano based APU, which uses the stars arch. Llano also removes the L3 cache so this is similar to comparing it to an Athlon x4. Check out the last page in the article.

http://www.tomshardware.com/reviews/a10-4600m-trinity-piledriver,3202-18.html

Ignore the graphics stuff as they get a boost due to the better IGP. If you look at the content creation part of the graph you'll see that the Trinity does indeed do better then Stars. Almost 20% better? And compared to the slower BD, PD should be nearly 30% faster. Looks like AMD might actually be able to pull off the claimed improvements with PD and make it a better CPU. It won't catch IB or even SB, but at least it will be faster then BD and finally the Phenom II.

 

[dogman_1234]

@47: Hmmm...never caught onto that. why do you suggest that there will be a 20-30% gain in overall performance or is this aggregate percentage?

 

[jimmysmitty]

I'm sorry if this has been covered already, but I'm not reading 101 pages of past stuff.

I consider myself a reasonably smart guy. Sometimes however it takes my brain awhile to crunch stuff and let me know things. This is one of those times. PD will be a lot better then BD, and better then the "stars" Phenom II arch. I base this on the Trinity article that Toms did.

http://www.tomshardware.com/reviews/a10-4600m-trinity-piledriver,3202.html

I'm assuming as a "quad core" this is a dual module PD based APU. As before the L3 cache has been removed to make room for the GPU/IGP. The thing that finally clicked for me was they compared this to a Llano based APU, which uses the stars arch. Llano also removes the L3 cache so this is similar to comparing it to an Athlon x4. Check out the last page in the article.

http://www.tomshardware.com/reviews/a10-4600m-trinity-piledriver,3202-18.html

Ignore the graphics stuff as they get a boost due to the better IGP. If you look at the content creation part of the graph you'll see that the Trinity does indeed do better then Stars. Almost 20% better? And compared to the slower BD, PD should be nearly 30% faster. Looks like AMD might actually be able to pull off the claimed improvements with PD and make it a better CPU. It won't catch IB or even SB, but at least it will be faster then BD and finally the Phenom II.

This is actually in line with most guesstimates I have seen. It will catch up to Nehalem, but not SB.

PD anyways. Still its not enough for me to really care as if the new CEO gets his way we may never see another K8 from AMD again.

 

[truegenius]

Nehalem is better than phenom 2
means you are accepting that pd will be better than phenom 2 in terms of ipc

 

[anxiousinfusion]

This is actually in line with most guesstimates I have seen. It will catch up to Nehalem, but not SB.

PD anyways. Still its not enough for me to really care as if the new CEO gets his way we may never see another K8 from AMD again.

Meh, good enough :/ .

And AMD have been pretty quiet about PD. That, coupled with the downplaying from the new CEO, would make for a pretty funny scenario if AMD where to pull an unbelievably beastly CPU out of nowhere. /pipedream

 

[jdwii]

Meh, good enough :/ .

And AMD have been pretty quiet about PD. That, coupled with the downplaying from the new CEO, would make for a pretty funny scenario if AMD where to pull an unbelievably beastly CPU out of nowhere. /pipedream

Amd is not even saying this, actually they say 15% better that's it. I say 15-20% better performance with 10-20% better efficiency on average over the 8150fx and i don't think its going to beat the Phenom II in IPC or performance per clock and its definitely not going to come any where near the original I7.

 

[jdwii]

http://www.xbitlabs.com/news/cpu/display/20120524182802_AMD_Rumoured_to_Start_Production_of_Next_Gen_FX_Chips_in_Q3.html.

 

[4745454b]

@47: Hmmm...never caught onto that. why do you suggest that there will be a 20-30% gain in overall performance or is this aggregate percentage?

From the content creation graph. Llano is at 58.1%, PD is at 77.5. If you go to productivity then you move from 65.7% to 83.5. Just shy of 20% improvement over Stars. If BD is ~7% slower then stars then PD should be 25-30% faster then BD. Honestly I'm surprised by this as I was expecting PD to be just faster then stars, or about a 10-15% improvement over BD.

Nehalem is better than phenom 2
means you are accepting that pd will be better than phenom 2 in terms of ipc

Not really. It could be better then Nehalem just because its clocked faster. I actually didn't bother to work out if Trinity is faster then Llano C4C. I honestly don't really care either. If AMD can give us a 95W quad core CPU that is clocked at 6.5GHz (or a 125/140W Octo core) that can beat Stars or most other chips then I'm ok with that. C4C or a high IPC is good, but even the P4 was able to keep up a bit with the x2 chips. (just had that high heat issue...)

i don't think its going to beat the Phenom II in IPC or performance per clock and its definitely not going to come any where near the original I7.

I think passing PhII's IPC will be task dependent, but I think it will be a lot closer then you give it credit for. Looking at the Trinity review I also think the original i7 might be within reach.

 

[jdwii]

@47: Hmmm...never caught onto that. why do you suggest that there will be a 20-30% gain in overall performance or is this aggregate percentage?

From the content creation graph. Llano is at 58.1%, PD is at 77.5. If you go to productivity then you move from 65.7% to 83.5. Just shy of 20% improvement over Stars. If BD is ~7% slower then stars then PD should be 25-30% faster then BD. Honestly I'm surprised by this as I was expecting PD to be just faster then stars, or about a 10-15% improvement over BD.

Nehalem is better than phenom 2
means you are accepting that pd will be better than phenom 2 in terms of ipc

Not really. It could be better then Nehalem just because its clocked faster. I actually didn't bother to work out if Trinity is faster then Llano C4C. I honestly don't really care either. If AMD can give us a 95W quad core CPU that is clocked at 6.5GHz (or a 125/140W Octo core) that can beat Stars or most other chips then I'm ok with that. C4C or a high IPC is good, but even the P4 was able to keep up a bit with the x2 chips. (just had that high heat issue...)

i don't think its going to beat the Phenom II in IPC or performance per clock and its definitely not going to come any where near the original I7.

I think passing PhII's IPC will be task dependent, but I think it will be a lot closer then you give it credit for. Looking at the Trinity review I also think the original i7 might be within reach.

Maybe but also remember that Llano was clocked at 1.5Ghz and trinity was clocked at 2.3Ghz that's a 50% clock speed advantage with around a 20% boost in CPU Performance. See why i think PD wont really beat the Phenom clock per clock. Plus Bulldozer is already clocked at 3.6Ghz to have a 50% clock advantage it would need to be clocked at 5.4Ghz which is not going to happen unless amd are wizards.

 

[truegenius]

^^+1 😉

 

[jimmysmitty]

@47: Hmmm...never caught onto that. why do you suggest that there will be a 20-30% gain in overall performance or is this aggregate percentage?

From the content creation graph. Llano is at 58.1%, PD is at 77.5. If you go to productivity then you move from 65.7% to 83.5. Just shy of 20% improvement over Stars. If BD is ~7% slower then stars then PD should be 25-30% faster then BD. Honestly I'm surprised by this as I was expecting PD to be just faster then stars, or about a 10-15% improvement over BD.

Nehalem is better than phenom 2
means you are accepting that pd will be better than phenom 2 in terms of ipc

Not really. It could be better then Nehalem just because its clocked faster. I actually didn't bother to work out if Trinity is faster then Llano C4C. I honestly don't really care either. If AMD can give us a 95W quad core CPU that is clocked at 6.5GHz (or a 125/140W Octo core) that can beat Stars or most other chips then I'm ok with that. C4C or a high IPC is good, but even the P4 was able to keep up a bit with the x2 chips. (just had that high heat issue...)

i don't think its going to beat the Phenom II in IPC or performance per clock and its definitely not going to come any where near the original I7.

I think passing PhII's IPC will be task dependent, but I think it will be a lot closer then you give it credit for. Looking at the Trinity review I also think the original i7 might be within reach.

Something like 6.5GHz is not going to happen very soon. The main reason why I don't see AMD getting to that first as well is because they will rely on TSMC/GF/whoever to produce a process that is able to get that high without causing massive amounts of leakage, power draw and heat dissipation. Intel has the best FABs and the best process right now so I think if anything they will hit that first. I also don't see Intel getting there very soon which is why IB OCing didn't suprise me much.

As for the IPC, I would hope that IPC wise PD is better than Deneb. But it can still be better just by having higher stock clocks and not IPC gain. That just worries me as its not something usefull if they hit a GHz ceiling (which they will) as the IPC never tends to scale linearly with clock speed and if they can't get improvements via clock speed due to a ceiling, then they will have to go with IPC which is not as easy to do.

Thats the reason why intel does a two step process with theirs. One year is new process with old arch, maybe some tweaks (or major with IB IGP) then use the mature process to get the new arch out. Fix issues in new arch on new die shrink and rinse and repeate. That way Intel can see what will and will not work on what process or what arch.

 

[sarinaide]

Why are people making out that Phenom II IPC is so unattainable, all things considered it is ridiculously old and innefficient by modern standards, why suggest AMD will not beat its IPC with concurrent chips?

AMD will release reports or review sites will get hold of paper spec information such will say how they have refined the architecture to better handle single and multithreaded instructions, improved efficiency, updated and improved hardware and firmware....yet all of this and it will de facto be slower than Phenom II #roll eyes#

It seems like we can only take worst case scenario

 

[4745454b]

Why are people making out that Phenom II IPC is so unattainable, all things considered it is ridiculously old and innefficient by modern standards, why suggest AMD will not beat its IPC with concurrent chips?

Because they didn't beat Stars with BD. BD clock for clock is slower then Phenom. Considering that Stars is a touch slower then even Core 2, this is really said.

Were the clocks that far off? I was hoping/assuming they were closer then that. A 50% bump in clock speed would more then explain the increase in performance.

 

[sarinaide]

Still the issue is overlooked, what if the refinement to the BD architecture is spot on, the bottlenecks are reduced, wouldn't that have the effect of improving IPC? Or are we just going to assume that PD is anything like BD?

 

[JAYDEEJOHN]

OK, let me quote a guy who didnt want to be quoted:
11% better than Husky

 

[4745454b]

The other advantage PD had was the faster ram, 1600 vs 1333. Due to the dynamic nature of turbo however we wont' really know how much more PD spent ~1GHz higher then stars.

 

[sarinaide]

I will still go with the not better than SB/IB, but close enough to tickle a fancy.

 

[4745454b]

No, not near SB/IB. Nehalem however will feel that tickle.

 

[truegenius]

OK, let me quote a guy who didnt want to be quoted:
11% better than Husky

i quote 😗
how it is 11% better than husky
that looks like overhype, my assumption is 3ghz husky equals 3ghz pd but will use only 50% power than that of daneb

and if it is near nehalem in per clock performance for per core, then 5m/10c pd (if is in the list) can beat a i7 4c

 

[viridiancrystal]

Maybe but also remember that Llano was clocked at 1.5Ghz and trinity was clocked at 2.3Ghz that's a 50% clock speed advantage with around a 20% boost in CPU Performance. See why i think PD wont really beat the Phenom clock per clock. Plus Bulldozer is already clocked at 3.6Ghz to have a 50% clock advantage it would need to be clocked at 5.4Ghz which is not going to happen unless amd are wizards.

Someone earlier on the forum posted benchmarks of his/her FX-8120 with 4 cores @2.3GHz (same clocks as Trinity) enabled and put them up against Tom's Trinity review. Those showed that Trinity was ~10% faster than a similar BD. Factor in L3 cache, and whatever the tweaks AMD will do between Trinity and PD, and I think this statement may be rather believable.

It is interesting to note that the Donanimhaber indicates that instruction per clock (IPC) performance of Vishera will be 15% higher compared to that of current-generation Zambezi thanks to Piledriver micro-architecture as well as some other tweaks. Earlier it was widely believed that FX "Vishera" chips will only bring 10% speed improvement at the same clock-speed compared to the currently available chips.

I may be wrong, but there is no "best case scenario" for IPC. There's no specific test that will show more IPC improvement than others. (at least by much). So if AMD gets their 15% (that big "if" is important) along with higher clocks, and much better power usage (Trinity's showed well), we may have a good set of chips on the way.

 

[4745454b]

Those showed that Trinity was ~10% faster than a similar BD. Factor in L3 cache, and whatever the tweaks AMD will do between Trinity and PD

L3 cache for AMD historically hasn't meant much. The Athlon x4 were how much slower then their Phenom II brothers? 5%? Tacking on L3 cache to trinity might bump them from 10% faster to 15-17%. I can't believe I overlooked clock speed. (or that toms would test such massively different CPUs.)

There's no specific test that will show more IPC improvement than others.

LAME or iTunes is the best I can think of. Single thread only, so set CPUs at the same clock speed and the chip with the highest IPC will get done first.

 

[sarinaide]

Mal, I just hope you don't someday regret it.

 

[jdwii]

Those showed that Trinity was ~10% faster than a similar BD. Factor in L3 cache, and whatever the tweaks AMD will do between Trinity and PD

L3 cache for AMD historically hasn't meant much. The Athlon x4 were how much slower then their Phenom II brothers? 5%? Tacking on L3 cache to trinity might bump them from 10% faster to 15-17%. I can't believe I overlooked clock speed. (or that toms would test such massively different CPUs.)

There's no specific test that will show more IPC improvement than others.

LAME or iTunes is the best I can think of. Single thread only, so set CPUs at the same clock speed and the chip with the highest IPC will get done first.

Your spot on, Also don't forget Cincebench! When i test single core performance i also use the Fritz chess Benchmark but only run it on one core.

For fun here is some benchmarks that i did on my Phenom II x6 OC 3.9Ghz


CineBench
Single core: 1.18
All cores: 6.91
Fritz Chess Benchmark
Single Core
Relative Speed: 5.06
Kilo nodes per second: 2430
All Cores:
Relative Speed: 28.28
Kilo nodes per second: 13573
wPrime
32M: 7.02
1024M: 219.167


Be cool if Piledriver could beat that at stock! Whats sad is my friends Laptop with a I7(sandy) which is clocked at 2.2Ghz gets a 1.21 on Cincebench on just one core.