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 ...

 

[Reynod]

I'm wondering if messing around the way your implying with high end parts is technically bestiality or not ...

 

[king smp]

^ Mal +1 LMAO

Cmon cheap Thuban BEs on Ebay Baby
I am hoping within year that I can lay my hands on a <$150 Thuban BE
Papa needs a 4ghz 6 core
where are all the junkies on Craigslist selling cheap? LOL

 

[king smp]

I'm wondering if messing around the way your implying with high end parts is technically bestiality or not ...

Mechanality?

I have bled,sweat and cried over my towers
many have demanded a blood sacrifice to work LOL

 

[MU_Engineer]

Cleaned up the stuff about gold and GPUs. I do have to admit, this thread has stayed pretty well on topic for being >100 pages long 😀

 

[king smp]

A big Congrats to MU_Engineer
I am really glad to see you get a Mod position
You definitley deserve it

 

[truegenius]

if you were speaking with my date with the Cray Supercomputer CX1.?
http://www.cray.com/Products/CX1/Systems.aspx

well after a few weeks and then finally getting her to come over for a while.
(told her to bring toothbrush)
we finally parted ways after she robbed me for all my parts and hardware.
she did let me open her up and feel all inside, what an experience.

I did get her this as a parting gift:
http://www.newegg.com/Product/Product.aspx?Item=N82E16817151109
(most powerful PSU I have ever laid my hands on.)

I must have been pretty good because after she left me I got a nice envelope with several $100 bills in it..
(this model had only 4 blades in it so I didn't have to dig in too deep.)

I'm such a swinger...

:lol:
i got your [strike]both meanings[/strike] every swings 😗

 

[dogman_1234]

A big Congrats to MU_Engineer
I am really glad to see you get a Mod position
You definitley deserve it

Suck Up, Brown noser!

 

[king smp]

Suck Up, Brown noser!

hey MUengineer I have known since my earliest days on Toms two years ago
awesome very knowledgable posts and the first member I ever PMd LOL
remember the smp in King smp stands for symmetric multi processing and every rig including his HTPC has duallies in it
I used to have a dual P4 Xeon rig years ago so I love the concept of dual CPUs

 

[jimmysmitty]

If I paid for a Corvette and drove within 0 - 60 mph 99% of the time, should I be angry if my Corvette did not go faster than 60 mph? I think so. Would you tell me it makes no sense for me to buy a Corvette or would you just say that I had too much money to spend?

This was the issue with BD as well, and its not suprising if PD has the same setup as its not a major overhaul of the arch

Do you really know this? Why does the "voice of reason" always so anti-AMD with you?

We do not overclock systems we build, that includes memory. If the customer wants to thats fine but again, using 1.65v RAM viods the warranty on the CPU and since its still overclocking, we do not do it. Its policy. Even if we put the best water cooling on the system we wouldn't overclock it, thats the customers choice if they want to mess with it.

Besides you didn't have to deal with the guy. Was a jerk. Trust me. Wanted to cancel the build and when we told him the system was already built and almost done (was doin updates) he didn't trust us at all. And I have a feeling that he will be a problem in the future as well.

And sure. I am anti-AMD. Thats why I just spent more on AMD (almost $500 GPU) than I did on Intel ($200 CPU).

Bulldozer had the same issue, where they could only fill so many instructions at once. A bottleneck. PD is BD with the clock mesh tech, higher clocks and some tweaks. I doubt they somehow in a very short period of time drastically alterd BD to a newish arch. Intel takes 2 years between new arch (actually more but we don't know much more) and the actual major change from Barcelona (Deneb was a die shrink and based on the same arch) to a new arch was almost 5 years.

Even changing a arch majorly takes time and money. Instead AMD can include the power saving feature, ramp up clocks to what they wanted BD to launch at or higher and have a bit better performance so they can sell a bit higher and make a bit more money.

 

[noob2222]

I doubt buying 2133 ram would even make a difference(except for APU's). Not to mention Intel gets more bandwidth from 1333 then Amd does from 1866.

Why would you assume that? Because Intel didn't get any benefit from faster memory? Maybe my own tests are invalid for some reason?

http://www.madshrimps.be/articles/article/1000220/AMD-FX-8150-Bulldozer-CPU-Review/4#axzz1w00JeNBc +5% from 1600 to 2133 on farcry 2 and mafia 2
http://vr-zone.com/articles/amd-fx-8150-memory-scaling-investigation--feeding-the-bulldozer/13704.html +4% in video encoding, another 5% on L4D

BD Speed > timings.

$20 investment to guarantee your not bottlenecking your system with cheap memory, depends on whether or not you value your overall system performance per dollar.

Amd really needs to work on this going forward. Graphics cards need high bandwidth Look at the difference in performance between a 6670 DDR3 and a DDR5 6670. At times its as high as 20%. Another thing Amd needs to do is include a small buffer for their graphics. They should look at these items when they shrink their APU to 28nm.

Noob2222 i'm sure you can agree and admit Amd needs to improve their performance per core. When their 12 core Opteron beats their 16 core Opteron their is problems.

Rory read is in idiot and should not be in control of a technology company when he says things like "Good enough performance". Its like Apple when they made a CEO in charge on their company when all he did is sell Pepsi a product that never changes. Technology changes so fast and we can't have someone who says good enough. I miss Dirk Meyer.

Just saying

I still question wether or not the comment by rory is taken from a discussion on cloud computing.

Besides, if memory speed right now is "good enough" then why are we trying to move to DDR4? wouldn't it be pointless?

 

[noob2222]

Bulldozer had the same issue, where they could only fill so many instructions at once. A bottleneck. PD is BD with the clock mesh tech, higher clocks and some tweaks. I doubt they somehow in a very short period of time drastically alterd BD to a newish arch. Intel takes 2 years between new arch (actually more but we don't know much more) and the actual major change from Barcelona (Deneb was a die shrink and based on the same arch) to a new arch was almost 5 years.

Even changing a arch majorly takes time and money. Instead AMD can include the power saving feature, ramp up clocks to what they wanted BD to launch at or higher and have a bit better performance so they can sell a bit higher and make a bit more money.

Would be funny to see history repeat itself.

http://www.anandtech.com/show/2702/9

I don't think all the benefit came from a die shrink.

~2-27% was huge for phenom to phenom II and everyone cheered. Time between releases? 10 months.

~5 - 15% from bd to PD is the same ... but this time its not resonable to assume it can happen?

 

[truegenius]

remember the smp in King smp stands for symmetric multi processing

i knew that smp means something related to multiprocessing
also x6 😗

wow i got a great logic system inbuilt, but needs to feed it with data from old style cd drive and thus it can process small data with speed (which is in cache, ram is already filled with complete os and other main background apps) 😀

 

[king smp]

how's his 'fruit of the looms' smell... 😗
geez.

ROFLMFAO!

Bite me LOL

 

[king smp]

okay back on topic

AMD not trying to compete in high performance desktop segment is just smart cost effective management
trying to catch up to Intel in that market would cost much more than AMD can afford in fab and R&D
and for what?
lets face it
enthuisast custom rigs are not the huge volume money makers
it is low powered servers,SOCs and mobile solutions followed by budget OEM pcs
also in my new job we selL HP systems to many shool systems and governement institutions
I just imaged and delivered 60 AMD SFFs early in the week
plenty of AMD laptops going out too
we did 250 AMD laptop images in 3 hours for a school system two weeks ago

I would say about almost half are systems going out the door in the past three weeks were AMD

enthuisiast performance desktops sad to say are the minority

 

[jdwii]

okay back on topic

AMD not trying to compete in high performance desktop segment is just smart cost effective management
trying to catch up to Intel in that market would cost much more than AMD can afford in fab and R&D
and for what?
lets face it
enthuisast custom rigs are not the huge volume money makers
it is low powered servers,SOCs and mobile solutions followed by budget OEM pcs
also in my new job we selL HP systems to many shool systems and governement institutions
I just imaged and delivered 60 AMD SFFs early in the week
plenty of AMD laptops going out too
we did 250 AMD laptop images in 3 hours for a school system two weeks ago

I would say about almost half are systems going out the door in the past three weeks were AMD

enthuisiast performance desktops sad to say are the minority

Your Amd's top friend!

I still can't believe Amd has 43% of the desktop market 😱
If they can get that on desktops they can sure beat that on laptops with their APU's. If they have all that why the heck are they worth crap(market cap went down since last year before Llano) heck they have less money then Nvidia and Arm.

 

[king smp]

Your Amd's top friend!

I still can't believe Amd has 43% of the desktop market 😱
If they can get that on desktops they can sure beat that on laptops with their APU's. If they have all that why the heck are they worth crap(market cap went down since last year before Llano) heck they have less money then Nvidia and Arm.

it is servers where they will win or lose
high volume combined with high profit
need low TDP with many cores of reasonable IPC
the Opterons have to dominate
the clock mesh tech could be a big winner in that market
they even bought a server company recently to get ahold of their tech involving
communication using the PCIe I think
I have to look that up
I still have to do my server certs in the next few months LOL
(got a HP student ID for free and low cost proctored exams through HP from work)
I am sure some of the more expert members could explain that tech better than me LOL

if you compare their retail sales of CPUs to their enterprise sales and OEM sales of CPUs it would be minor in comparison

 

[Haserath]

I'm wondering if messing around the way your implying with high end parts is technically bestiality or not ...

Electronality.

It gets those electrons flowing when you see a barebone system in all its glory. 😍

 

[Haserath]

Would be funny to see history repeat itself.

http://www.anandtech.com/show/2702/9

I don't think all the benefit came from a die shrink.

~2-27% was huge for phenom to phenom II and everyone cheered. Time between releases? 10 months.

~5 - 15% from bd to PD is the same ... but this time its not resonable to assume it can happen?

The 27% and 16% from Excel and Winrar look like they were mostly boosted from the extra L3...

PD shouldn't notice such a benefit from added cache.

 

[jimmysmitty]

Why would you assume that? Because Intel didn't get any benefit from faster memory? Maybe my own tests are invalid for some reason?

http://www.madshrimps.be/articles/article/1000220/AMD-FX-8150-Bulldozer-CPU-Review/4#axzz1w00JeNBc +5% from 1600 to 2133 on farcry 2 and mafia 2
http://vr-zone.com/articles/amd-fx-8150-memory-scaling-investigation--feeding-the-bulldozer/13704.html +4% in video encoding, another 5% on L4D

BD Speed > timings.

$20 investment to guarantee your not bottlenecking your system with cheap memory, depends on whether or not you value your overall system performance per dollar.

Amd really needs to work on this going forward. Graphics cards need high bandwidth Look at the difference in performance between a 6670 DDR3 and a DDR5 6670. At times its as high as 20%. Another thing Amd needs to do is include a small buffer for their graphics. They should look at these items when they shrink their APU to 28nm.

Noob2222 i'm sure you can agree and admit Amd needs to improve their performance per core. When their 12 core Opteron beats their 16 core Opteron their is problems.

Rory read is in idiot and should not be in control of a technology company when he says things like "Good enough performance". Its like Apple when they made a CEO in charge on their company when all he did is sell Pepsi a product that never changes. Technology changes so fast and we can't have someone who says good enough. I miss Dirk Meyer.

Just saying

I still question wether or not the comment by rory is taken from a discussion on cloud computing.

Besides, if memory speed right now is "good enough" then why are we trying to move to DDR4? wouldn't it be pointless?

DDR4 is a different approach. No longer will it have multiple DIMMs per channel but rather one DIMM per channel each connected to the IMC individually. It will give up to 2x the speeds.

That said, its like all real technology changes. They start in the server, then come to us. DDR4 will benefit servers a lot. Not so much the majority of desktop users. But there will be a time when DDR4 will have larger sticks and cost less per GB than DDR3 and it will take over.

Would be funny to see history repeat itself.

http://www.anandtech.com/show/2702/9

I don't think all the benefit came from a die shrink.

~2-27% was huge for phenom to phenom II and everyone cheered. Time between releases? 10 months.

~5 - 15% from bd to PD is the same ... but this time its not resonable to assume it can happen?

5-15% but my guess has been that it will mostly be from clock speed gains rather than actual IPC.

Phenom I was pretty bad to start. Low clocks did not help. Top end was 3GHz. Phenom II was able to reach much higher clock speeds, not to mention overclock much higher than Phenom.

Thos factors plus L3, allowed for better than expected results as Phenom II was still just a die shrink overall. Same base arch.

For PD, there is no die shrink, no L3 being added just the clock mesh along with possibly a more mature 32nm to help performance/power usage. Its easier to determine that it wont be a major change.

 

[dogman_1234]

So, could BD be a Phenom I and PD or SR be PII?

 

[sarinaide]

okay back on topic

AMD not trying to compete in high performance desktop segment is just smart cost effective management
trying to catch up to Intel in that market would cost much more than AMD can afford in fab and R&D
and for what?
lets face it
enthuisast custom rigs are not the huge volume money makers
it is low powered servers,SOCs and mobile solutions followed by budget OEM pcs
also in my new job we selL HP systems to many shool systems and governement institutions
I just imaged and delivered 60 AMD SFFs early in the week
plenty of AMD laptops going out too
we did 250 AMD laptop images in 3 hours for a school system two weeks ago

I would say about almost half are systems going out the door in the past three weeks were AMD

enthuisiast performance desktops sad to say are the minority

+1 AMD are targeting the commercial and mainstream segment. Also the fact that AMD have a 40+ market share is because a AMD processor is still more than enough to do what 40~% needed it to do.

On the comments that the X% gains from BD to PD are clockspeed orientated, for what has been said about Piledriver is that it will have lower base clockspeed and turbo frequencies than BD, it is for all intents and purposes a IPC gain, again the "AMD cannot improve IPC" rhetoric is getting tedious.

 

[jdwii]

DDR4 is a different approach. No longer will it have multiple DIMMs per channel but rather one DIMM per channel each connected to the IMC individually. It will give up to 2x the speeds.

That said, its like all real technology changes. They start in the server, then come to us. DDR4 will benefit servers a lot. Not so much the majority of desktop users. But there will be a time when DDR4 will have larger sticks and cost less per GB than DDR3 and it will take over.



5-15% but my guess has been that it will mostly be from clock speed gains rather than actual IPC.

Phenom I was pretty bad to start. Low clocks did not help. Top end was 3GHz. Phenom II was able to reach much higher clock speeds, not to mention overclock much higher than Phenom.

Thos factors plus L3, allowed for better than expected results as Phenom II was still just a die shrink overall. Same base arch.

For PD, there is no die shrink, no L3 being added just the clock mesh along with possibly a more mature 32nm to help performance/power usage. Its easier to determine that it wont be a major change.

Your some what wrong Amd also made other changes that should improve IPC Such as their Branch prediction, L2 cache speeds as well as the pre-fetcher plus a Larger L1 cache and they improved the scheduling for the Integer and Floating point units.
All of this is sure to improve IPC or CPI, my guess is IPC will go up by around 7-10% on average. Clock mesh will allow a higher clock speed plus they will have a more advanced 32nm die(like you said). I'm guessing the clock speed will be 15% higher then the 8150 which is probably going to add 7-10% more performance on average as well. In summery with the IPC improvements as well as the new clock mesh technology plus anything else that's been improved by Global foundries Piledriver should add 15% more performance while being 10-20% more efficient on average. Rumors are stating that Amd will start production around the 3rd Quarter of this year which hint at a early 4th Quarter release probably around the same time Bulldozer came out last year i personally expect Piledriver out in October.

Why would you assume that? Because Intel didn't get any benefit from faster memory? Maybe my own tests are invalid for some reason?

http://www.madshrimps.be/articles/ [...] z1w00JeNBc +5% from 1600 to 2133 on farcry 2 and mafia 2
http://vr-zone.com/articles/amd-fx [...] 13704.html +4% in video encoding, another 5% on L4D

BD Speed > timings.

$20 investment to guarantee your not bottlenecking your system with cheap memory, depends on whether or not you value your overall system performance per dollar.

Sorry but you need to know that Acer/HP/Dell and so on put cheap slow ram into their machines which means Amd needs to work on their memory controller. Take a look here

Both are using the same ram but Trinity had half the bandwidth.

But then again, like the Phenom II, it's what AMD should have released......
kinda makes you think if AMD is *insert swear word* if they have to release a crappy product between 2 OK products everytime

Microsoft and Amd are playing the same game then
Windows ME-XP-Vista-windows 7-windows 8-
Phenom I...Phenom II...Bulldozer...Piledriver

On the comments that the X% gains from BD to PD are clockspeed orientated, for what has been said about Piledriver is that it will have lower base clockspeed and turbo frequencies than BD, it is for all intents and purposes a IPC gain, again the "AMD cannot improve IPC" rhetoric is getting tedious.

Not true Even Trinity is supposed to have a 3.8Ghz clock speed as seen here
http://www.fudzilla.com/home/item/26204-top-trinity-is-a10-5800k-at-38ghz

And here

http://www.techpowerup.com/162843/AMD-A10-5800K-quot-Trinity-quot-APU-Tested.html

Plus read this

Cyclos and AMD didn't go into too much detail about Piledriver, though they did say it will consist of a 4GHz+ x86-64 core built on a 32nm CMOS process.

So if Trinity will have a 3.8Ghz clock speed with 384 radeon cores with a TDP of 100watts i'm more then sure Piledriver will release with a 4.0+Ghz clock speed with around 125 watt TDP. If Amd is smart they will also be able to make a 3.6-3.8Ghz 95 watt 8 core on top of a 6 and 4 core Piledriver with a 65 watt version maybe a 45 watt for their 4 core.

 

[noob2222]

Sorry but you need to know that Acer/HP/Dell and so on put cheap slow ram into their machines which means Amd needs to work on their memory controller. Take a look here

http://media.bestofmicro.com/A/Z/337355/original/Sandra mem.png

Both are using the same ram but Trinity had half the bandwidth.

I doubt buying 2133 ram would even make a difference(except for APU's).

So why would you say there is no benefit from going with 2133 ram? yes, AMD should work on thier memory controller but thats not what I said. With 2133 memory I am getting 22+ gb/sec on sandra tests. Changing the subject doesn't change the fact that 2133 (and 1866 wich BD and PD fully supports) ram makes a difference at least with AMD cpus.

Instead all test are done with what makes Intel look better, slow 1333 memory. If IPC is conditional on components, is it fair to test the cpu with slower than supported hardware?

As for dell/hp/whoever, why do you think most of us build systems. They are crap in comparison, cheapest memory, cheapest harddrives, cheapest motherboards, ect.

 

[sarinaide]

Fair enough.

The complaint with 4ghz + is down to efficiency, but honestly how many concern themselves with that, considering BD halved Thubans efficiency numbers and still made older generations look ludicrously power consuming. At the end, people here said they would rather CPU manufacturers focus on clock speed et al and put less focus on power numbers. To me power efficiency is at the level that is tollerable, I am more concerned about performance numbers.

As for the RAM issue, I to have found that AMD benefits more from higher frequency RAM, more so than Intel, Intel stops responding beyond 1600mhz, latency timings have little effect too. With the FX having higher native frequencies, the advantage is there in some synthitics, but overall its not really a major factor as of yet.

 

[jdwii]

I doubt buying 2133 ram would even make a difference(except for APU's).

So why would you say there is no benefit from going with 2133 ram? yes, AMD should work on thier memory controller but thats not what I said. With 2133 memory I am getting 22+ gb/sec on sandra tests. Changing the subject doesn't change the fact that 2133 (and 1866 wich BD and PD fully supports) ram makes a difference at least with AMD cpus.

Instead all test are done with what makes Intel look better, slow 1333 memory. If IPC is conditional on components, is it fair to test the cpu with slower than supported hardware?

As for dell/hp/whoever, why do you think most of us build systems. They are crap in comparison, cheapest memory, cheapest harddrives, cheapest motherboards, ect.

Its true in the real world(People who buy Computers not build them 95+% of everyone) people wont notice much difference. But when gaming we would notice a difference i agree that high speed ram needs to be put into the machines but this does not happen often and Intel will do better then Amd on slower ram. Anyways if these benchmarks are right Amd would need 2600 ram to equal 1600 from Intel not good not good at all.

DDR4 ram is going to make a big difference if its 2 times as fast that's going to make gaming benchmarks look 15-20%(on Average) better on even Intel graphics.

Plus stop saying Ram speeds are so Important for IPC because its not. If that was the case gaming would be much better on faster ram then slower ram and its been stated over and over it makes less then a 5% difference(Unless were talking about integrated graphics).

I'm not saying the memory controller will make a big enough difference for the CPU but it will make a good difference on their APU.