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 ...

 

[viridiancrystal]

It's really too bad that bulldozer did so bad, AMD provides such convenience in many other regards (namely not changing sockets every 8 months), i think they have all the potential, but BD just doesn't preform.

 

[gamerk316]

Okay, im going to jump into this tread here and figure some stuff out, im good with HW when it comes to knowing whats fast and whats not, so can someone help explain WHY bulldozer has a bad IPC and what AMD is trying to/needs to do to make it better?

Design tradeoff. Deeper pipelines are easier to achieve higher clock speeds with, but can have significant downsides to how long an individual instruction takes to execute. [See: Pentium 4 v Athlon for reference]. Combine that with a large, shared L2 cache [two things that kill IPC], and theres why IPC is so bad, even compared with Phenom II.

I said from day 1: AMD went for clock speed over IPC, and got burned by it. As many of us predicted, in the best case [software that scales to all of BD's modules], BD is faster then SB. But thats a very limited percentage of software, as most software is developed with one or two heavy threads doing most of the processing, favoring a processor with stronger cores.

 

[jimmysmitty]

Design tradeoff. Deeper pipelines are easier to achieve higher clock speeds with, but can have significant downsides to how long an individual instruction takes to execute. [See: Pentium 4 v Athlon for reference]. Combine that with a large, shared L2 cache [two things that kill IPC], and theres why IPC is so bad, even compared with Phenom II.

I said from day 1: AMD went for clock speed over IPC, and got burned by it. As many of us predicted, in the best case [software that scales to all of BD's modules], BD is faster then SB. But thats a very limited percentage of software, as most software is developed with one or two heavy threads doing most of the processing, favoring a processor with stronger cores.

I would assume though that even if it could use the stronger cores it may not always beat Sandy bridge. The IPC tradeoff could be enough to offset any multithreaded advantage it may have, as we have seen in some server applications where a 6 core/12 threaded Intel Xeon can keep up with or beat a 12 core AMD Opteron.

Of course people said nay and that IPC is pointless but every part of a CPU, old and new has a very real purpose. Changing the way a CPU performance is measured, to me anyways, is a sign of a fan trying to change the rules.

Either way, I think BD has potential IF AMD can cut some of the fat and redesign a few things inside, possibly take your suggestion of making the BIOS state the second core in every module is a SMT type core instead of a real core, even with the extra parts thrown in it. But I doubt that will happen. I think AMDs' marketing is running the company more than the engineeres are, who used to considering what we got with the Athlon 64.

And as for what piesquared said (a now deleted post), i wish I got paid to post. Would be nice to have some extra cash to spend on new PCs. But allas I don't. I have to work a job fixing and building PCs. Its not all that bad. I love the stuff so it works out.

 

[Reynod]

Jimmy haven't you been getting the cheques?

Jeepers ... I'd complain.

I'm getting paid just to shut up ... evidently all of the regulars in the forum are chipping in ...

 

[viridiancrystal]

you know, with piledriver being released Q1 2012 (correct if if im wrong), so soon after bulldozer, it makes me think that bulldozer was just a test to see how their new arch would preform in the enthusiast market.

 

[Chad Boga]

you know, with piledriver being released Q1 2012 (correct if if im wrong), so soon after bulldozer, it makes me think that bulldozer was just a test to see how their new arch would preform in the enthusiast market.

What do you mean?

 

[noob2222]

you know, with piledriver being released Q1 2012 (correct if if im wrong), so soon after bulldozer, it makes me think that bulldozer was just a test to see how their new arch would preform in the enthusiast market.

full PD is q3-q4. Trinity is PD cores without l3 cache. Q1 we will see B3 steppings (8170FX)

 

[viridiancrystal]

What do you mean?

Bulldozer is only going to be in the AMD spotlight for about 5-6 months, so they may have released it just to see how it preformed from most peoples perspectives, so they could make adjustments for Piledriver. What we call a test.

 

[noob2222]

or so you hope... 😗

have something to add or just trying to start something?

 

[viridiancrystal]

seriously...?
"released it just to see how it preformed from most peoples perspectives"
you really think that.

not entirely, but the fact that they have two releases of BD and PD cores in Trinity within a few months make it a reasonable theory. I am more or less trying to find a reason behind BD's performance.

 

[noob2222]

start something.?
are you really that insecure.?

AMD in-case you haven't noticed isn't exactly good at keeping promises or dates.
and how do you know Q1 will be the release of B3 stepping, and are you certain and wanting to bet money.?
C2 Phenom II's were about 9 months before the C3 stepping as an example..

so my comment to your comment was ' or so you hope so"
meaning what you said, would be nice if it was to be certain..

c2 to c3 timeframe is pretty irrelevant since it was a mature cpu stepping

B2 to B3 phenom was 4 months, November to March

and if you haven't noticed, a lot of people here love to start flamewars, so just adding a comment without any input usually leads to that happening.

 

[warmon6]

At first glance your post is sarcastic, however, did you mean paid just to shut up or paid to delete posts that don't tow the line?

Here it is again....

One of the many relevant points, which speak volumes when jim calls it pointless. If this is pointless, the real enthusiast market is dead.

The researchers from Canada and China say paying people to post comments is an "interesting strategy in business marketing" but it is not a benign activity.

"Paid posters may create a significant negative effect on the online communities, since the information from paid posters is usually not trustworthy," they wrote.

http://www.bbc.co.uk/news/technology-15869683

Seriously, you need to make that a topic somewhere else instead of linking too the same place over and again.

Also, the reason why Jimmy is calling it pointless due to it not being remotely related to the discussion of the thread.

There no amd cpu or even a gpu in the link. There no mention of AMD in general. Last there nothing even about computer hardware directly... So is it pointless for this thread? Yes.


Now, is it a pointless topic? No. I read it and it certainly something people of the web can talk about as it can hurt forums like this as a whole.

 

[jimmysmitty]

Jimmy haven't you been getting the cheques?

Jeepers ... I'd complain.

I'm getting paid just to shut up ... evidently all of the regulars in the forum are chipping in ...

I wish I was getting some nice big extra cheques. Right now I just get paid my salary to fix PCs, build PCs and well today was a great example, be over run with questions and such. Sad to say, I have more knowledge than most at my job only because I try to keep up with whats new. But they are all good guys.

Funny thing is I would have probably made a pretty good sales person too. I just hope my boss doesn't add that to my job description Then it would be "Head technician and Sales guy".

you know, with piledriver being released Q1 2012 (correct if if im wrong), so soon after bulldozer, it makes me think that bulldozer was just a test to see how their new arch would preform in the enthusiast market.

I doubt BD was just a test. As for the relase, I remember Q3 2012 more precisely, pending any delays thanks to issues with the revised arch or good ol GF.

Boy what a great idea GF turned out to be.....

full PD is q3-q4. Trinity is PD cores without l3 cache. Q1 we will see B3 steppings (8170FX)

I have been saying Q3 2012 for PD. Some thought Q1 2012 but I doubt AMD would push a new arch "revision" so fast. Would negate their sales of BD and force them to sell at or below cost to get rid of them or to the low end OEMs like eMachines.

Kinda off topic but a customer at work the other day told me that eMachines had bought out Asus. I kinda laughed at that idea.

c2 to c3 timeframe is pretty irrelevant since it was a mature cpu stepping

B2 to B3 phenom was 4 months, November to March

and if you haven't noticed, a lot of people here love to start flamewars, so just adding a comment without any input usually leads to that happening.

The timeframe is very relevant. If you look at it in the proper way. Phenom II was a decent CPU. It wasn't quite the Intel smasher but it was a boost from Barcelonas failure and helped AMD move back into a better market. They didn't have to worry about the new steppings so much since it was going to sell more than Phenom did.

BD on the other hand is not the same. Its weaker than the Intel competition by enough that AMD is pressured to get a new stepping out, one that either fixes the mess ups (like B3 did for Phenom I) or lowers TDP enough to clock it higher at stock to help it somewhat compete with Intels IB, which I am going to doubt it will.

Of course it all remains to be seen.

As for flame wars, not many here try to encite flame wars. Its much betetr than when I first got here. But I can say there are those who do. And then there are those who will post something and someone twists it to make it seem like a flame post.

Honestly, give up on the link. No one here gets paid to post anything here. Hell even the few that do get paid by Intel post that they do work for Intel which negates anything. And the ones that we know do are actually very well balanced in opinion.

 

[noob2222]

The timeframe is very relevant. If you look at it in the proper way. Phenom II was a decent CPU. It wasn't quite the Intel smasher but it was a boost from Barcelonas failure and helped AMD move back into a better market. They didn't have to worry about the new steppings so much since it was going to sell more than Phenom did.

BD on the other hand is not the same. Its weaker than the Intel competition by enough that AMD is pressured to get a new stepping out, one that either fixes the mess ups (like B3 did for Phenom I) or lowers TDP enough to clock it higher at stock to help it somewhat compete with Intels IB, which I am going to doubt it will.

Of course it all remains to be seen.

If you looked at what that statement was in refrence to, you would see that its pretty much what I said.

C2 to C3 is irrelevant because it was on a mature cpu, 9 months in between because it didn't really matter that much, it was a minor tweak.

B2 to B3 will bring the same stepping as BD and Phenom I. 4 months. AMD would be stupid to sit on current BD for 9 months, then push PD out 3 months after revising BD.

Wich time frame is more relevant? its not c2 to c3.

 

[Mousemonkey]

http://hexus.net/business/news/corporate/32779-amd-make-move-mobile/

 

[gamerk316]

http://hexus.net/business/news/corporate/32779-amd-make-move-mobile/

Duh? Isn't that why they fired their previous CEO?

Anyways, I see them trying to push a reduced Trinity, rather then licensing ARM. I also predict, like all "low power" X86 offerings, it will be a failure in the mobile market. [X86 simply does not scale down very well...it wasn't designed with power efficency in mind.] Hence, I'm prediciting a disaster of epic proportions for AMD going forward.

 

[truegenius]

i think amd cam make a comeback by launching
1). ph3 (32nm) or ph4 (22nm) with more powerful cores (as 4 or 6 cores are enough and many application can only use one core or 2,3 or 4 max).
2). by finding a technology some what like intel's hyper threading.
3). (if possible) by finding a technology with which 2 cores can act as 1 core (i.e, reverse of htt of intel),by which bd will become a better cpu
(think positive, be futuristic) 😀

 

[jimmysmitty]

Duh? Isn't that why they fired their previous CEO?

Anyways, I see them trying to push a reduced Trinity, rather then licensing ARM. I also predict, like all "low power" X86 offerings, it will be a failure in the mobile market. [X86 simply does not scale down very well...it wasn't designed with power efficency in mind.] Hence, I'm prediciting a disaster of epic proportions for AMD going forward.

I agree. But if anything I think Intel will be the first to push a decent x86 arch into the mobile sector with 22nm. Plus this is awesome:

"We're at an inflection point ... We will all need to let go of the old 'AMD versus Intel' mind-set, because it won't be about that anymore."

SO I guess Intels no longer a target? Is he basically saying screw the CPU market? Because last I checked thats where they can make the most money.

 

[de5_Roy]

i think amd cam make a comeback by launching
1). ph3 (32nm) or ph4 (22nm) with more powerful cores (as 4 or 6 cores are enough and many application can only use one core or 2,3 or 4 max).
2). by finding a technology some what like intel's hyper threading.
3). (if possible) by finding a technology with which 2 cores can act as 1 core (i.e, reverse of htt of intel),by which bd will become a better cpu
(think positive, be futuristic) 😀

1. llano. they have lower frequency though. afaik amd's next apus will be 28 nm.
2. bulldozer.
3. hmm, this could help amd.. but i think bulldozer actually incorporates something between hyperthreading and the thing you mentioned.

by bulldozer i mean the architecture, not the cpus (i.e. zambezi, valencia etc)

 

[truegenius]

llano is apu, but i mean a pure cpu

is bd is using htt

 

[truegenius]

did you mean A8-3850 which is a four core apu can process 8 thread

 

[lilcinw]

did you mean A8-3850 which is a four core apu can process 8 thread

Llano and BD are two different archs, don't confuse them. BD uses SMT which places two threads on the same 'module', which many argue equate more closely to traditional cores and the FX-8xxx is really a 4c/8t design.

 

[-Fran-]

http://hexus.net/business/news/corporate/32779-amd-make-move-mobile/

This is a very interesting quote:

"We're at an inflection point ... We will all need to let go of the old 'AMD versus Intel' mind-set, because it won't be about that anymore."

Their APU road is going to be pressuring on Intel at the bottom for now, but the real target seems to be ultra portables (smartphones and tablets) within ARM territory. That's going to be very interesting. I wonder how they'll make that strategy work out.

Also, about AMD making RAM modules... That's something very interesting too. If they can design stuff to make their integrated solutions even better, damn... AMD might make a real comeback pressuring ARM and Intel on both ends ('cept enthusiast, maybe). I really wonder how deep they'll get in the RAM business... Could they make RAMBUS their enemy/friend? Not that RAMBUS has a very good track record, but they DO own a truck load of RAM related technology patents.

Cheers!

 

[gamerk316]

1). ph3 (32nm) or ph4 (22nm) with more powerful cores (as 4 or 6 cores are enough and many application can only use one core or 2,3 or 4 max).

Not happening. PII is more or less a dead-end arch. I'm surprised AMD was able to take the K8 as far as it did. Nevermind the development costs in getting a P3 up and running...

2). by finding a technology some what like intel's hyper threading.

They do; its called CMT, and they use it in Bulldozer. Basically, its a more powerful version of HTT, but at the cost of more die space. [We had this discussion to DEATH in the BD roumer thread...]

3). (if possible) by finding a technology with which 2 cores can act as 1 core (i.e, reverse of htt of intel),by which bd will become a better cpu
(think positive, be futuristic)

Waste of money. This would only offer significant performance benifits if you could, at the same time, ramp up clock speeds by upwards of 50%, which is unrealistic.

Also remember the OS is CONSTANTLY swapping threads around, so the CPU configuration could, in the worst case, flip from single to dual core on a clock by clock basis, which would in all likelyhood significantly impact the processors lifespan.

 

[gamerk316]

Not that RAMBUS has a very good track record, but they DO own a truck load of RAM related technology patents.

Their main one just got struck down though, and RAMBUS stock tanked like 60% immediatly afterward. I fully expect them to go under within the next year or so as a result.