Rumor: AMD "Piledriver" FX CPU production to begin Q3 2012

[guskline]

Got my hand up in the air here (I have no pride left)- hate to admit it but late last summer I returned a 2500K that I got for $150 and a Gigabyte Z68 board.

Now running a 990FX Sabertooth and a 960T Zosma (I have a FX6100 sitting in a drawer somewhere). Not a bad setup really for my use, but still....

What did Ali call this maneuver?? - oh yea - 'rope a dope'.

Don't feel bad sequoia464, I bought a Sabertooth 990FX also believing the Bulldozer would be the UBER chip. Fotunate for me I could put together Intel SB 2500k rigs to feel what true gaming power is. Thus my comments above that no way I'm getting a Piledriver chip until ANANDTECH and other reputable reviewers put it through it's paces.

 

[podspi]

Don't feel bad sequoia464, I bouight a Sabertooth 990FX also believing the Bulldozer would be the UBER chip. Fotunate for me I could put together Intel SB 2500k rigs to feel what true gaming power is. Thus my comments above that no way I'm getting a Piledriver chip until ANANDTECH and other reputable reviewers put it through it's paces.

You think that's bad? I bought an 890 board since I thought Bulldozer would be UBER and compatible!

Of course, at that time it was the 1055t vs. i5-750. 1055t was better for my workload so... I have no real regrets. Still...

 

[dagamer34]

I'm still wondering how AT readers expect AMD to pull something magic out of their hat when Intel is outspending them on R&D by 10:1. Continuing to play the same game and hoping for a different result is foolish.

 

[nismotigerwvu]

The constant negativity and childishness is getting really old on this forum in general. Yes Bulldozer came up short, yes Phenom II hung around a year or so too long, but it does look like Piledriver is bringing something to the table here. We're not going to know if it improves enough until almost Christmas, why are we bickering about this in May? Does it make a lot of the members here feel warm and fuzzy to accuse every AMD employee top to bottom of being a bumbling idiot incapable of tying their own shoes? It's just really lame, especially considering Trinity looks like it's primed to make crappy entry level notebooks decidedly less crappy. Yes, the CPU core is more Nehalem level, but the iGPU benefits ARE tangible and a large portion of the market will appreciate it (and a greater slice if openCL does finally take off, in which case the compute level would exceed the intel offering in these applications, but most certainly lag behind in legacy applications).

 

[AtenRa]

Remember you are a tiny minority. And the IGP can still be used in conjunction with a discrete card.

And Intel do allow you to get 6 cores on the desktop, if you pay for it.

IB quad core Chip is 160mm2 including the iGPU. The Quad Core 8 Threads Core i7 3770K is priced at $340. Remove the iGPU, install two more cores and 2MB more L3 and sell me a 6 core 12 threads ~160mm2 CPU for the same money.
I dont have to pay more. But you know what ?? Intel will loose precious GPU market share by doint that. They dont want you to have more cores in Desktop. They want you to have their GPUs because by stagnating the desktop you seam to believe that Quad core CPUs are enough for today's and future apps.

It is ok to have a $100-150 APU (Pentiums and Core i3 could do the job nicely) but nobody will use the iGPU on a $200-300 CPU.

 

[ShintaiDK]

IB quad core Chip is 160mm2 including the iGPU. The Quad Core 8 Threads Core i7 3770K is priced at $340. Remove the iGPU, install two more cores and 2MB more L3 and sell me a 6 core 12 threads ~160mm2 CPU for the same money.
I dont have to pay more. But you know what ?? Intel will loose precious GPU market share by doint that. They dont want you to have more cores in Desktop. They want you to have their GPUs because by stagnating the desktop you seam to believe that Quad core CPUs are enough for today's and future apps.

It is ok to have a $100-150 APU (Pentiums and Core i3 could do the job nicely) but nobody will use the iGPU on a $200-300 CPU.

People needing a 6 core can for example buy the i7-3930K or a Xeon. The option for 6 cores is there.

And software is stagnating your "core dream" on the desktop. Its pretty silly you keep pretending its Intels fault because you dont like the company. Kinda silly since it wasnt Intel who put the current joke in your painted fan chassis

 

[Skurge]

People needing a 6 core can for example buy the i7-3930K or a Xeon. The option for 6 cores is there.

And software is stagnating your "core dream" on the desktop. Its pretty silly you keep pretending its Intels fault because you dont like the company. Kinda silly since it wasnt Intel who put the current joke in your painted fan chassis

Wouldn't the software follow if the hardware is there?

 

[Ajay]

Yes, but thats a very outdated slide compared to the other. And it makes no sense to launch IB-E if HW-E is something like 3-6 months after it. And I honestly wouldnt want to run SB-E/IB-E against a HW-DT on socket H3.

It's and older slide, yes, but Intel essentially made a promise. They could fulfill that promise with HW-E, if it's LGA 2011 compatible. AMD broke their promise with the 4x4 platform and earned the disdain of the enthusiast community. I really think that Intel will want to avoid this - how, we'll just have to wait and see.

In the slide you just posted, it just says >= i7-3960X, we don't really know what that means just yet. I wish we had some info on when HW-E will be out (if it will be out). If you have info on this, I'd be really interested in seeing it. Obviously, with two 256 AVX(2) units, the computational power of Haswell will be outstanding (and apparently the L2$'s performance has been improved to help sustain high throughput - so it should truly be awesome).

Some of us want more cores, like myself, for F@H and other highly threaded applications, and are willing to pay the price for it (more expensive CPU and Mobo).

 

[ShintaiDK]

Wouldn't the software follow if the hardware is there?

The hardware is already there.

People tend to have some illusion, that its just about making multithreaded code. Not about if its even possible or the benefits of doing so.

As I said before as well. You would be surprised how much serverside software thats essentially singlethreaded. But lives on the concept of concurrent users to fix its "multithreading". The same cant be applied for mobile/desktop. Just try and count the games that supports more than 2 threads. I´m sure you can do that on your hands.

 

[Ajay]

Amdahl's law has been broken. One way is by transactional memory, there are other hardware features that break it as well, I just can't recall which, ATM.

 

[ShintaiDK]

In the slide you just posted, it just says >= i7-3960X, we don't really know what that means just yet.

>= always refers to speedbins. That should be no surprise. Same applies for the IBs in Q3. Its simply a note about they could release a speedbin higehr if needed or if the processnode works out for it.

 

[ShintaiDK]

Amdahl's law has been broken. One way is by transactional memory, there are other hardware features that break it as well, I just can't recall which, ATM.

Could you show me where?

Transactional memory doesnt break Amdalhs law. Its just better at handling some of the issues with multithreaded performance penalties on active memory sharing code.

 

[AtenRa]

The hardware is already there.

People tend to have some illusion, that its just about making multithreaded code. Not about if its even possible or the benefits of doing so.

As I said before as well. You would be surprised how much serverside software thats essentially singlethreaded. But lives on the concept of concurrent users to fix its "multithreading". The same cant be applied for mobile/desktop. Just try and count the games that supports more than 2 threads. I´m sure you can do that on your hands.

I suggest you read the following two,

http://www.futurechips.org/thoughts-for-researchers/parallel-programming-gene-amdahl-said.html

http://www.futurechips.org/thoughts...l-programming-amdahls-law-gustafsons-law.html

Also, Multi-core CPUs are needed for Multi-Tasking as well. The notion that we dont need more than 4 cores must end. We need more than 4 cores and we need more IPC from SIMDs like AVX etc.

 

[LOL_Wut_Axel]

The constant negativity and childishness is getting really old on this forum in general. Yes Bulldozer came up short, yes Phenom II hung around a year or so too long, but it does look like Piledriver is bringing something to the table here. We're not going to know if it improves enough until almost Christmas, why are we bickering about this in May? Does it make a lot of the members here feel warm and fuzzy to accuse every AMD employee top to bottom of being a bumbling idiot incapable of tying their own shoes? It's just really lame, especially considering Trinity looks like it's primed to make crappy entry level notebooks decidedly less crappy. Yes, the CPU core is more Nehalem level, but the iGPU benefits ARE tangible and a large portion of the market will appreciate it (and a greater slice if openCL does finally take off, in which case the compute level would exceed the intel offering in these applications, but most certainly lag behind in legacy applications).

The CPU core isn't Nehalem level. Not even close. All they're doing is bringing performance back up to what it was in the Phenom II days, and we all know the Nehalem Core i7 was faster than that.

 

[LOL_Wut_Axel]

It's and older slide, yes, but Intel essentially made a promise. They could fulfill that promise with HW-E, if it's LGA 2011 compatible. AMD broke their promise with the 4x4 platform and earned the disdain of the enthusiast community. I really think that Intel will want to avoid this - how, we'll just have to wait and see.

In the slide you just posted, it just says >= i7-3960X, we don't really know what that means just yet. I wish we had some info on when HW-E will be out (if it will be out). If you have info on this, I'd be really interested in seeing it. Obviously, with two 256 AVX(2) units, the computational power of Haswell will be outstanding (and apparently the L2$'s performance has been improved to help sustain high throughput - so it should truly be awesome).

Some of us want more cores, like myself, for F@H and other highly threaded applications, and are willing to pay the price for it (more expensive CPU and Mobo).

For f@h what you want is some used 12-core Magny Cours Opterons on a SuperMicro board so you can overclock. Unprecedented performance/watt.

 

[ShintaiDK]

I suggest you read the following two,

http://www.futurechips.org/thoughts-for-researchers/parallel-programming-gene-amdahl-said.html

http://www.futurechips.org/thoughts...l-programming-amdahls-law-gustafsons-law.html

Also, Multi-core CPUs are needed for Multi-Tasking as well. The notion that we dont need more than 4 cores must end. We need more than 4 cores and we need more IPC from SIMDs like AVX etc.

I´m not sure you actually read your links. They basicly just say its even worse with multithreaded aka parallel code. Its no secret that Amdalhs law is based on perfect code.

I dont disagree we need higher IPC. But currently we dont have a need for more cores. Adding more cores currently is the biggest waste possible. You would be better off simply adding bigger caches if you must waste the silicon.

And multitasking by itself doesnt require multiple cores. It simply depends on the workload. Having 200 notepads open wont run faster on a 16 core CPU vs a singlecore.

 

[grimpr]

The current state of things is that the windows software ecosystem slowly rots away, killer applications are nowhere to be found and cloud based computing along with windows 8 metro apps will give the finishing blow to the classic PC desktop, hardcore enthusiasts are living in a fantasy world just like audio fidelists with golden ears and expensive equipement.

 

[blastingcap]

The current state of things is that the windows software ecosystem slowly rots away, killer applications are nowhere to be found and cloud based computing along with windows 8 metro apps will give the finishing blow to the classic PC desktop, hardcore enthusiasts are living in a fantasy world just like audio fidelists with golden ears and expensive equipement.

Wow, stunning insight there; we haven't ever heard that before. *yawn*

Aside from certain types of productivity software (rendering, digital image manipulation), there are also PC games, particularly games that do not lend themselves to cloudification due to latency issues (see, e.g., OnLIVE).

But for everyday websurfing, etc., for those unconcerned about the cloud's privacy and security risks, a thin client is enough.

 

[Ajay]

For f@h what you want is some used 12-core Magny Cours Opterons on a SuperMicro board so you can overclock. Unprecedented performance/watt.

I'll also use it as my desktop (including gaming), so I want an enthusiast board. But, you make a good point visa vi folding.

 

[Smartazz]

Isn't saying today's laptops are good enough akin to Gates' comment about 640KB of RAM?

 

[pelov]

Isn't saying today's laptops are good enough akin to Gates' comment about 640KB of RAM?

Not if he's right, no. Unlike Gates' famous 640K remark, Read's comment have some standing. Most people are transitioning away from the need for modern hardware in desktops/laptops for their everyday tasks. We have been for years actually and in fact we've gone backwards (see specs of any smartphone/tablet). The overwhelming majority of users can nowadays get by on a tablet + keyboard and most of their interaction with things PC-related tend to be with smartphones and tablets anyway. Cutting edge PCs/laptops are unnecessary but for a small percentage of users, namely gamers and those who use their PCs as workstations.

The comment seems a bit rash but it's actually well founded when you think about it.

 

[nismotigerwvu]

The CPU core isn't Nehalem level. Not even close. All they're doing is bringing performance back up to what it was in the Phenom II days, and we all know the Nehalem Core i7 was faster than that.

Reading comprehension FTL there bud. Not only was that not the even the point I was making, but you just continued to cement the one that I was. That said, describing something as "more Nehalem level" doesn't mean equal to, it means in the general range, so give or take say 5 to 10 percent. I'll agree with you, the Piledriver cores going into Trinity do appear to be shy of last gen Intel desktop performance (which is not equal to IPC, we are talking about absolute performance) but these aren't desktop chips and that is a very apples to oranges comparison. AMD has also mentioned (which by no means guarantees) additional core improvements and we can assuredly expect double the cores and almost double the clocks. Looking at Anand's preview of Trinity, the 2.3gHz base/3.2gHz turbo 4 thread A10 lagged the 1.7gHz base/2.8gHz turbo SB based 4 thread i7 by about 15 to 20% in single threaded performance and ~7% in multithreaded (looking mostly at cinebench, which isn't perfect, but we don't have a full review to go on yet). I'd feel comfortable putting that in "Nehalem class" compute and the iGPU outperforms the Intel chips by a larger percentage as a whole (there are some cases where HD4000 catches up though). Applications amendable to gpGPU may actually have greater performance on the trinity chips as well, but these are currently nowhere to be found and as such should be treated as hypotheticals. Even without improvements beyond clocks/modules, there should be a niche for these chips, definitely not on the high end, but they aren’t trying to compete there. You can chose to be negative about this as much as you want, but I for one tend to view notable improvements in a more positive light.

 

[Arzachel]

Cinebench is the worst case scenario for Trinity. Purely FP load.

 

[Schmide]

Cinebench is the worst case scenario for Trinity. Purely FP load.

Actually it's even worst than pure FP. it's hierarchy transversal, transformation, culling, material/texture lookup, composition, etc. All very cache/memory intensive.

Trinity would probably do better at pure FP.

 

[Soulkeeper]

any speculation on a chipset update ?

anyone think we will get these features in the next 6 months ?
-pcie 3.0
-ddr4

or will these be things reserved for the socket after AM3+ ?