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 ...

 

[davemaster84]

Should be below 300 US to be competitive with the 3500k , somewhere around 280 or maybe 250

 

[alpha_c]

The problem is the i5-3570k is $190 or so, and the i5-2500k is $160 or so (on Microcenter) and come with $50 off motherboards. They're $220 or so on Newegg. The FX-8150 is $180, FX-8120 is $140, FX-6100=$110, FX-4100=$90.

Piledriver also has to compete with the Phenom II X4 965BE and Phenom II X6s which can be found for ~$100. Unless it outperforms Thuban in everything and at least outpaces a low level i5 like the i5-2300, there is no reason to buy AMD Piledriver if you have anything Core2Quad or newer. If it can't outpace an i3-2100 it shouldn't even exist.

Also, for gaming i3s do better than the FX-4100 and Phenom IIs (this was the case even with the i3-530 so it's not surprising).

http://www.tomshardware.com/reviews/gaming-fx-pentium-apu-benchmark,3120-10.html
"we're almost-shockingly left without an AMD CPU to recommend at any price point."

Where AMD needs to aim is non-gaming purposes where multi-core actually is useful (i.e. where the i7 is). Bulldozer/Piledriver is the AMD version of hyperthread but the performance needs to match i7s if they want to price it similarly. AMD processors still give more cores per dollar but the number of places that's useful is slim. If the quad core i7s with hyperthread are $300, then they need to put their FX 8xxx on roughly i5-2500k price (~$190) because that's where the majority of the market is headed. The Piledriver/Bulldozer architecture is a side effect of software not supporting the full amount of cores of Phenom IIs.

The tiers should be FX-8xxx= between i5-2xxx and i7-2xxx price , FX-6xxx and FX-4xxx= i3-2xxx price because they are likely not up to par with Ivy Bridge.

The other day I saw a pre-built deal on Costco with the Intel i7-3770 and a crap Pegatron B75 mobo for $600 (no graphics board though), which means AMD has to *really* get their act together. For non-gamers (compute), that blows away pretty much anything AMD has to offer. See http://www.costco.com/Browse/Product.aspx?Prodid=11771214

Haswell is coming 2013 and if Intel gets their IGPs up to par with the AMD APUS, then AMD will be in even deeper trouble.

 

[esrever]

Where AMD needs to aim is non-gaming purposes where multi-core actually is useful (i.e. where the i7 is). Bulldozer/Piledriver is the AMD version of hyperthread but the performance needs to match i7s if they want to price it similarly. AMD processors still give more cores per dollar but the number of places that's useful is slim. If the quad core i7s with hyperthread are $300, then they need to put their FX 8xxx on roughly i5-2500k price (~$190) because that's where the majority of the market is headed. The Piledriver/Bulldozer architecture is a side effect of software not supporting the full amount of cores of Phenom IIs.

The tiers should be FX-8xxx= between i5-2xxx and i7-2xxx price , FX-6xxx and FX-4xxx= i3-2xxx price because they are likely not up to par with Ivy Bridge.

You just described bulldozer's performance and pricing compared to sandybridge at launch...

 

[gamerk316]

I'm going to keep repeating this over and over again: More cores helps when running more applications. Within the confines of a single application however, you WILL be limited in core scaling.

Case in point, actually open up Task Manager, add the "threads" tab, and look how many threads applications already use. Thats the lowest level developers code to. So when we already have applications running 80+ threads, and you don't get scaling beyond 4 or so cores, I got news for you: Adding more cores won't help performance much (if any).

 

[esrever]

I'm going to keep repeating this over and over again: More cores helps when running more applications. Within the confines of a single application however, you WILL be limited in core scaling.

Case in point, actually open up Task Manager, add the "threads" tab, and look how many threads applications already use. Thats the lowest level developers code to. So when we already have applications running 80+ threads, and you don't get scaling beyond 4 or so cores, I got news for you: Adding more cores won't help performance much (if any).

most of those threads are dependent and coherent, a majority of them do nothing and is just pooled threaded when applications launch. Also you'd never notice the core scaling simply due to the performance of the modern cpu compared to light threads that are used by programs like chrome and itunes.

 

[noob2222]

I'm going to keep repeating this over and over again: More cores helps when running more applications. Within the confines of a single application however, you WILL be limited in core scaling.

Case in point, actually open up Task Manager, add the "threads" tab, and look how many threads applications already use. Thats the lowest level developers code to. So when we already have applications running 80+ threads, and you don't get scaling beyond 4 or so cores, I got news for you: Adding more cores won't help performance much (if any).

for sure until programmers quit crying and figure out how to come up with a new style and quit using old code since its too easy to just copy/paste or put a call to the previous code.

 

[-Fran-]

for sure until programmers quit crying and figure out how to come up with a new style and quit using old code since its too easy to just copy/paste or put a call to the previous code.

Hey, it's not a programmers fault that companies don't want to optimize code to be threaded =/

I can thread and do some very decent scaling software in Java, but if the management doesn't want to move on with legacy code, we'll be using little threading approaches most of the time to fit the old code or libraries.

Personally, I like to think problems in a threaded manner, since they're almost different problems altogether, but if they don't want the threaded approach, there's nothing I can do. So, if you really want better threaded software, blame the managers and (some) architects that don't like the idea and call the shots 😛

Cheers!

 

[esrever]

I think the problem with things is stagnation. Nobody really cares enough to make things threaded because almost everything already run well enough. They'd rather make their software run well on as many machines as possible than push for new code paths.

 

[mayankleoboy1]

and not every type of a problem can be divided into parallel threads.
Like web browsers, where most of the objects and transitions are sequential.

 

[viridiancrystal]

I think the problem with things is stagnation. Nobody really cares enough to make things threaded because almost everything already run well enough. They'd rather make their software run well on as many machines as possible than push for new code paths.

I hope that the new generation of consoles will change that.

 

[mayankleoboy1]

:O :O AMD to stop making performance CPU for desktop after Vishera platform. It has given up the fight with intel

http://www.obr-hardware.com/2012/08/amd-fx2-piledriver-is-here-last-amd.html

Today I bring you two important news from the world of AMD. Both are related to Piledriver CPU architecture. On the VR-Zone magazine appeared article, where the author speculates that AMD desktop Piledriver CPU architecture (FX2-8300) will not be launched, and AMD will wait for architecture Steamroller. Its total bullshit! FX2-8300/6300 processors are in the final phase of ES production and will go to the market on schedule. In one fact is VR-Zone correct - CPU performance FX2-8300 is only a few percent over the FX-8100. Performance is the same *** as Bulldozer. Piledriver (FX2-8300) = Bulldozer C0 revision!


The second news is shocking. I have secret internal AMD document here, and there is something big: AMD FX2-8300/6300 (Piledriver Vishera) desktop processor is the last performance desktop CPU ever! Yes, it's true you read it correctly! All next processors in desktop will be only APU. AMD completely ends production of performance desktop CPUs! FX2-8300/6300 processors for socket AM3 + are the last competitors against Intel-DT platform. AMD gives up the fight against Intel here, and in the future will launch only APU with mainstream performance! Vishera + AM3+ are the last representative of an AMD performance CPU platform! Buy it, someday it will be a valued historical artifact.

 

[-Fran-]

Well, by AMD's definition what Intel has been selling since Westmere are APUs... And Intel is as serious as AMD in the APU road, like it or not.

Seems like we'll be back to buying server class CPUs to be really enthusiasts... /sigh

Cheers!

 

[de5_Roy]

earlier i asked what piledriver's 'attraction' might be... looks like obrhardware's answer is 'historical artifact'. 😛

Well, by AMD's definition what Intel has been selling since Westmere are APUs... And Intel is as serious as AMD in the APU road, like it or not.

Seems like we'll be back to buying server class CPUs to be really enthusiasts... /sigh

Cheers!

imo amd took the scenic route. made their users wait for 2-3 years, then launched an underwhelming cpu lineup, publicly admitted giving up chasing intel for cpu performance, [strike]fired a bunch of people[/strike] restructured, vaguely mentioned am3+ being the last cpu platform, announces new platform for apus, then announced a new socket for apus, so on....
compared to that intel just switched to lga1156 and put an igpu on the cpu die and went from there. meanwhile they started selling rebadged xeons at jacked up prices in the name of 'extreme performance' to 'enthusiasts'.
both intel and amd are aiming for the same thing. they're just taking different routes to get there. :3

 

[noob2222]

All next processors in desktop will be only APU.

We already knew this. That doesn't mean steamroller will be complete junk, it will just have an IGPU since they are pushing heterogenous computing with it. About the only thing that obr article may have clarified is that there will be no AM3+ Steamroller cpu.

 

[noob2222]

Hey, it's not a programmers fault that companies don't want to optimize code to be threaded =/

I can thread and do some very decent scaling software in Java, but if the management doesn't want to move on with legacy code, we'll be using little threading approaches most of the time to fit the old code or libraries.

Personally, I like to think problems in a threaded manner, since they're almost different problems altogether, but if they don't want the threaded approach, there's nothing I can do. So, if you really want better threaded software, blame the managers and (some) architects that don't like the idea and call the shots 😛

Cheers!

True that, busineses only want to do it as cheap as possible, and that means no new code if old stuff will work.

 

[sarinaide]

Wait, someone to a blog essentially and took it as binding source......my god people are becoming stupid.

http://www.obr-hardware.com/2012/08/amd-fx2-piledriver-is-here-last-amd.html

Biggest box of testicles ever concieved. I mean wasn't Trinity effectively the end of APU's and Steam Roller the first big step in CPU/GPU fusion, Steam Roller is intended to be the big marker in the quest for heterogeneous computing, this entire blog is a load of sweaty balls.

 

[truegenius]

AMD Abu Dhabi server parts could launch next month
http://www.fudzilla.com/home/item/28472-amd-abu-dhabi-server-parts-could-launch-next-month

^ it will draw 1250W, not 1200W for 80%
efficiency

😗
let x be the power input and y be output and z is efficiency
we have
x=?
y=1000w
z=80%
according to the defiation of effeciency we have
z=(y/x) * 100
thus
80 = (1000/x) * 100
=>x=(1000*100)/80
=> x = 1250
thus if power outpu is 1000w and effeciency is 80% then the required input power is 1250w 😛
btw you are a veteran 😀

 

[fazers_on_stun]

Wasn't OBR the guy who got ahold of some AMD internal documents about this time last year, showing that BD would be far short of what expectations were before its launch?

 

[de5_Roy]

budget quads might have a new member - fx4130 a.k.a. lol
http://www.tomshardware.com/news/amd-zambezi-bulldozer-fx-4130-cpu,17096.html
amd preps business class apus(trinity):
http://www.cpu-world.com/news_2012/2012082301_AMD_business-class_Trinity_APUs_are_in_the_works.html

http://www.fudzilla.com/home/item/28471-darren-grasby-takes-helm-of-amd-emea-sales

 

[sarinaide]

"Super Secret" documents.

 

[truegenius]

"Super Secret" documents.

 

[gamerk316]

Hey, it's not a programmers fault that companies don't want to optimize code to be threaded =/

I can thread and do some very decent scaling software in Java, but if the management doesn't want to move on with legacy code, we'll be using little threading approaches most of the time to fit the old code or libraries.

Personally, I like to think problems in a threaded manner, since they're almost different problems altogether, but if they don't want the threaded approach, there's nothing I can do. So, if you really want better threaded software, blame the managers and (some) architects that don't like the idea and call the shots 😛

Cheers!

But as I've explained: The embarrasingly parallel stuff has already been offloaded to the GPU [rendering and physics]. The remaining parts of code is MUCH harder to make parallel, especially outside of specific subsystems.

For example, I can easily make a parallel AI engine, where every in game AI object is fully independent of the next. Thats simply to accomplish. But then you get into interactions between AI's, which requires a LOT of locking between each AI object [or each AI thread], which kills performance. Then I have to worry about processing environmental factors into my AI routine needs [the rendering equation has to finish creating the geometry so the AI can do a LOS check, audio engine needs to process audio cues the AI uses, etc]. So even though each AI in theory could be processed on a different core, because of low level locking of threads, you won't process more then one or two at a time.

On top of that, developers NEVER select what cores to load threads on. Thats the job of the OS scheduler. The best we can do is give "hints" what threads can be run in parallel, but its up to the OS to load them in an optimal manner and schedule them in the most optimal manner.

 

[sarinaide]

There is a lot of Ethan Hawk going on here, did he ab-sail down from the ventalation systems and steal a document housed inside a glass showpiece titled "super secret duck sauce" using laser cutters?

 

[truegenius]

There is a lot of Ethan Hawk going on here, did he ab-sail down from the ventalation systems and steal a document housed inside a glass showpiece titled "super secret duck sauce" using laser cutters?

 

[mayankleoboy1]

thus if power outpu is 1000w and effeciency is 80% then the required input power is 1250w 😛

thats what i wrote..