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 ...

 

[sarinaide]

AMD have given no indication of moving beyond 8 cores, but they have dedicated continual IPC improvements through its current and subsequent road map, what if SteamRollers complete re-model of modular architecture is the golden touch and there is a instant 30% IPC gain.

There are things to consider, are intel moving this IPC goal post at the rate people are led to believe, my 990X will more than copiously handle and sometimes manhandle intels current line up, by statistics given IB should be 20% faster IPC's which to me is the biggest lie under the sun, almost as bad as the 30-50% numbers claimed over AMD chips. Those kinds of disparities manifest into real noticeable difference. I had done a FX 6XXX build for a engineer using CAD and Revit based programs, it replaced his old i7 930, accordingly the verdict was the FX felt more fluid and faster.

I can take my Pentium D, take a 3960X and will see noticeable changes, from general fluidity and responsiveness to overall performance, similarly a 1100T over a Athlon II x4 640 shows a marked real world difference.

 

[fazers_on_stun]

When I hear Global Foundries, I see Hector Ruiz and I see me sticking knives in him with a great big smile on my face.....

Well AMD basically wrote off GloFo debts which whacked them hard but splitting from GloFo is the best decision in a long while.

AFAIK AMD still has a payment or two to go, but not being locked into GF anymore has got to be a good thing for them.

The extra 5% reduction in workforce should help their bottom line too, although hopefully they will not be further cutting the engineering staff..

 

[fazers_on_stun]

8 core sandybridge will have TPD of like 160W, its still a double edged sword. Ivy might shave off some but even then it would be unlikely that the cpu will be below 130W with the same frequency. I don't know about you but most workloads that can be threaded can be subdivided very well across more and more cores. Amdual's law doesn't kick in until more than 64 cores. By the time AMD offers 16 core consumer CPUs, AMD would have hoped to have pushed HSA to the point where they won't have to do it. Higher IPC also requires more transistors and die area as well as other things, don't think its "free" like you assume.

Probably why Intel stuck with six cores plus SMT for their consumer-class E series. Software just not there yet to make use of anything more.

I wouldn't be surprised to see Bridgewell with E CPUs having 8 cores plus SMT in a couple years, by which time more software might make use of the extra horsepower (oops, yet another car analogy to derail this thread 😛).

 

[sarinaide]

AFAIK AMD still has a payment or two to go, but not being locked into GF anymore has got to be a good thing for them.

The extra 5% reduction in workforce should help their bottom line too, although hopefully they will not be further cutting the engineering staff..

Yeah a 20m payment left, but its been written off against investments. The last cuts and redundancies were of non-engineering staff.

 

[NoUserBar]

Since Global unveiled it's 14xm up to 20nm architectures. Does anybody think that AMD is going to go down that path?

 

[Dangi]

I don't think Intel can slap 8 ivy Bridge cores together, the same way AMD couldn't put together more than 6 Phenom II, the arquitecture isn't design for that. You have to rework almost everything

Returning to car analogies, if you increase the horsepower it's meaningless unless you can't improve it's traction (that's why american cars suxx ) why have a 700CV in a car that can't take a curve over 100Km/H

 

[jdwii]

700mhz extra and they said they could not get the system over 200 watts!

http://www.techpowerup.com/reviews/AMD/FM2_APU_Review/8.html

 

[esrever]

Looks like a 4.4 - 4.6 OC is to be expected, probably closer to 4.4 for most chips. The graphics OC is where the APU really shines. 6670 level performance from and integrated chip is pretty amazing.

 

[jdwii]

Looks like a 4.4 - 4.6 OC is to be expected, probably closer to 4.4 for most chips. The graphics OC is where the APU really shines. 6670 level performance from and integrated chip is pretty amazing.

I'd say once you overclock both you will have 6670 and I3 performance which is pretty darn good for 130$. I'm hoping the L3 cache doesn't add too much power consumption.

 

[obsama1]

AMD is dominating the lower end. Why get the i3 when you get the A10 which is quad-core and has 10x better graphics? And when you factor in dual-graphics, the A10 is a much better value than the i3.

 

[jdwii]

AMD is dominating the lower end. Why get the i3 when you get the A10 which is quad-core and has 10x better graphics? And when you factor in dual-graphics, the A10 is a much better value than the i3.

I'd say more like 30-50% better but the A10 and the I3 is real close in CPU tasks way more then Llano and the I3 was.

 

[-Fran-]

Problem is, the power envelope.

It's overall good, I agree, but HTPC wise or laptops, it's not an OC friendly environment for Trinity. What matters is stock performance in the low end. You all know OC does little to nothing in the low end for volumes sales.

AMD needs volumes sales and the A10 and A8 could do that in the 65W sweet spot, where most OEM MoBos feel good because of productions costs around them.

100W+ Trinities won't be popular in most places, I'm sure. The K series are just a tease of what we can expect out of PD, but not much more IMO. They're not very practical APUs for what they target: the lowest Intel CPU.

The K series represent a big V8 engine in a car targeted for the city (city-car for short), like the old Mini or a 442. We all think it's cool and sweet, but you know no one will get one.

Cheers!

 

[jdwii]

Problem is, the power envelope.

It's overall good, I agree, but HTPC wise or laptops, it's not an OC friendly environment for Trinity. What matters is stock performance in the low end. You all know OC does little to nothing in the low end for volumes sales.

AMD needs volumes sales and the A10 and A8 could do that in the 65W sweet spot, where most OEM MoBos feel good because of productions costs around them.

100W+ Trinities won't be popular in most places, I'm sure. The K series are just a tease of what we can expect out of PD, but not much more IMO. They're not very practical APUs for what they target: the lowest Intel CPU.

The K series represent a big V8 engine in a car targeted for the city (city-car for short), like the old Mini or a 442. We all think it's cool and sweet, but you know no one will get one.

Cheers!

100 watts isn't that bad and on idle it uses less power consumption than Intel's offer. I'd say there will be plenty of A10's i know i would never go lower than that on a laptop which i see online for 700$(a little to much) On desktops you might be right but its dumb when it only uses 100 watts vs Intel's I3 at the end they cost the same to put into the Desktop and OEMs pay the same either way and with the A10 they can advertise good graphics or what not.

 

[truegenius]

I don't think Intel can slap 8 ivy Bridge cores together

8 core sandybridge will have TPD of like 160W, its still a double edged sword. Ivy might shave off some but even then it would be unlikely that the cpu will be below 130W with the same frequency.

http://wccftech.com/intels-leaked-roadmap-shows-ivy-bridgeepen-processors-12-cores-30mb-cache-130w-tdp/
12 core 130w ivy leaks

 

[jdwii]

I don't think Intel can slap 8 ivy Bridge cores together

8 core sandybridge will have TPD of like 160W, its still a double edged sword. Ivy might shave off some but even then it would be unlikely that the cpu will be below 130W with the same frequency.

http://wccftech.com/intels-leaked-roadmap-shows-ivy-bridgeepen-processors-12-cores-30mb-cache-130w-tdp/
12 core 130w ivy leaks

With what clock speed? Cost? If we thought a 6 core was a lot what about this one.

 

[noob2222]

Probably why Intel stuck with six cores plus SMT for their consumer-class E series. Software just not there yet to make use of anything more.

I wouldn't be surprised to see Bridgewell with E CPUs having 8 cores plus SMT in a couple years, by which time more software might make use of the extra horsepower (oops, yet another car analogy to derail this thread 😛).

AMD is working to break Intel's straglehold on the market. How many dual core I3s would sell if more programs were designed for quad core and above? Its in Intel's best interest to keep programmers believing that its impossible to code for multiple cores. They are the only ones still trying to sell dual core cpus for over $100.

IMO HSA is a make or break deal with AMD. If they can break the Intel strangle on the market, they will make it, if not they will be broke. Intel doesn't want HSA, they have too much market share to lose. Thats why all you hear is IPC IPC IPC IPC. Their gpu sucks, their method for HT only works sometimes and haswell has even more of that. A single core can only go so fast and do so much, and thats what Intel is banking on, pushing that one core to the absolute max. IPC IPC IPC.

Once AMD gets back on track with SR, they will start to push that one core harder, right now its just jogging along, tripping a lot, but mostly just waiting. As long as AMD starts to feed that core that wants more work and has plenty of room to get more work done, they will catch up much faster considering Intel's IPC is capping.

http://www.anandtech.com/show/6355/intels-haswell-architecture

seems to me pretty much this entire article is all about reducing power. Why is Intel going after power use so hard now? Im sure someone will show up soon and let you konw.

 

[mayankleoboy1]

AMD is dominating the lower end. Why get the i3 when you get the A10 which is quad-core and has 10x better graphics? And when you factor in dual-graphics, the A10 is a much better value than the i3.

AMD has the chip that will dominate the lower end notebook market, IF OEM make AMD notebooks AND IF people actually buy them.
AMD seriously needs some tasty marketing.

 

[de5_Roy]

http://www.anandtech.com/show/6355/intels-haswell-architecture

seems to me pretty much this entire article is all about reducing power. Why is Intel going after power use so hard now? Im sure someone will show up soon and let you konw.

would have debated feasibility of dualcore desktop cpus a year ago but now dualcore cpus are running on borrowed time imo.
i thought the same thing about haswell after reading that article and the idf info. intel has been trying to improve power efficiency for a while now, not just because of haswell. looks like they're trying so hard this time because they consider mobile arm socs a legitimate threat unlike in amd desktop sector. that is not a bad thing for amd.

AMD has the chip that will dominate the lower end notebook market, IF OEM make AMD notebooks AND IF people actually buy them.
AMD seriously needs some tasty marketing.

intel left the low end mobile pc market as soon as smartphones and tablets started to take off, that's when amd swooped in with brazos apus. amd won't be able to make money off that market for long because those buyers have started to shift their interests towards tablets. that's why tablet-friendly apus are needed. it'd be easy for amd to undercut intel but harder to undercut cheap tablets based on arm socs. it'll only get harder if glofo and tsmc can't deliver decent amount of yields (remember llano and apple) and if haswell is indeed capable of scaling down for tablets. marketing comes after that.
future would be nothing short of interesting.

 

[viridiancrystal]

Problem is, the power envelope.

It's overall good, I agree, but HTPC wise or laptops, it's not an OC friendly environment for Trinity. What matters is stock performance in the low end. You all know OC does little to nothing in the low end for volumes sales.

AMD needs volumes sales and the A10 and A8 could do that in the 65W sweet spot, where most OEM MoBos feel good because of productions costs around them.

100W+ Trinities won't be popular in most places, I'm sure. The K series are just a tease of what we can expect out of PD, but not much more IMO. They're not very practical APUs for what they target: the lowest Intel CPU.

The K series represent a big V8 engine in a car targeted for the city (city-car for short), like the old Mini or a 442. We all think it's cool and sweet, but you know no one will get one.

Cheers!

I still haven't seen power numbers on the A10-5700, which is a 65w piece. assuming it stays close to that power draw, then it would be better competition for an i3 for those who want to keep power in mind.

 

[fazers_on_stun]

Since Global unveiled it's 14xm up to 20nm architectures. Does anybody think that AMD is going to go down that path?

GF has a long way to go before producing and ramping 22 or 20nm, let alone 14nm, so if AMD is smart they'll take a wait & see approach. Ditto for using TSMC as their foundry.

There was speculation in the Haswell thread, linked by me, where some financial analysts think Intel's 22nm ramped far better than Intel expected (better than 32nm apparently, which had excellent yields), to the point that Intel's 4th 22nm fab is nowhere near capacity. Of course, that could be due to the continuing global recession, but in the meantime Intel could use that excess fab capacity for stuff like fabbing Apple's A6 ARM CPU for the iPhone & iPad at 22nm instead of the current Samsung 32nm. So that would fit right in with Apple's desire to move entirely away from Samsung, plus give them an entire node advantage..

OTOH there's lots of rumors that Apple wants to get away from using Intel CPUs for their macbooks, so maybe not. I'm sure Intel would want to tie fabbing the A6 to continued purchase of i5 & i7 CPUs . Dunno if Apple actually wants to get into fabbing their own CPUs anyway, but I doubt they would want to be dependent on GF and/or TSMC forever, given both companies ramp problems..

 

[fazers_on_stun]

I don't think Intel can slap 8 ivy Bridge cores together, the same way AMD couldn't put together more than 6 Phenom II, the arquitecture isn't design for that. You have to rework almost everything

Returning to car analogies, if you increase the horsepower it's meaningless unless you can't improve it's traction (that's why american cars suxx ) why have a 700CV in a car that can't take a curve over 100Km/H

Considering there have been 10-core/20-thread 32nm Sandy Bridge Xeons out for some time now, dunno where you get the idea that Intel can't do 8-core/16-thread Ivy CPUs. Both SB and IB (and future CPUs like Haswell & Bridgewell) are modular in design, similar to Bulldozer...

 

[fazers_on_stun]

I don't think Intel can slap 8 ivy Bridge cores together

8 core sandybridge will have TPD of like 160W, its still a double edged sword. Ivy might shave off some but even then it would be unlikely that the cpu will be below 130W with the same frequency.

http://wccftech.com/intels-leaked-roadmap-shows-ivy-bridgeepen-processors-12-cores-30mb-cache-130w-tdp/
12 core 130w ivy leaks

And the 10-core/20-thread variants drop to 90W TDP, from Sandy's 130W..

 

[Chad Boga]

I don't think Intel can slap 8 ivy Bridge cores together, the same way AMD couldn't put together more than 6 Phenom II, the arquitecture isn't design for that. You have to rework almost everything

Returning to car analogies, if you increase the horsepower it's meaningless unless you can't improve it's traction (that's why american cars suxx ) why have a 700CV in a car that can't take a curve over 100Km/H

:lol:

Ivy Bridge-EP/EN Processors with 12 Cores

Up to now, most sources stated the upcoming 22 nm replacement for Xeon E5 and Core i7 39XX to have 10 cores per die. Post IDF, a very credible source says it was 12 all the while, just the yields weren't sure of.

 

[mayankleoboy1]

I wonder if Apple would be better off with buying GloFo. They already are designing their own Ax CPU's. Then they can even fabricate them internally. Maybe add 'feature' that would prevent software iOS jailbraking and integrated user tracking + reality distortion field generator.
Apple has shitloads of money. They can pump a few billion dollars into GloFO and easily make it competitive with TSMC, if not Intel.

Apple already hates Samsung. Intel are not willing to go into general fabrication business. TSMC have conflicting reports on whether it wants Apple's business or not. Apples way to deal with these kinds of supply chain issues is to buy the company.

If Intel is smart, they wont ever fabricate mobile chips for Apple. TSMC/Samsung has the similar nodes for Apple/Tegra/Krait/Exynos SoC lines. This means similar performance/power figures for all SoC.
If Intel fabricates for Apple, Intels node advantage over TSMC/Samsung will mean Apple Ax SoC's will have inherent advantage over the competition. That would probably mean Android will get killed quickly. And if Android gets killed, what will Intel Atom/Medfield++ run on ? Surely not iOS.

 

[fazers_on_stun]

AMD is working to break Intel's straglehold on the market. How many dual core I3s would sell if more programs were designed for quad core and above? Its in Intel's best interest to keep programmers believing that its impossible to code for multiple cores. They are the only ones still trying to sell dual core cpus for over $100.

IMO HSA is a make or break deal with AMD. If they can break the Intel strangle on the market, they will make it, if not they will be broke. Intel doesn't want HSA, they have too much market share to lose. Thats why all you hear is IPC IPC IPC IPC. Their gpu sucks, their method for HT only works sometimes and haswell has even more of that. A single core can only go so fast and do so much, and thats what Intel is banking on, pushing that one core to the absolute max. IPC IPC IPC.

Once AMD gets back on track with SR, they will start to push that one core harder, right now its just jogging along, tripping a lot, but mostly just waiting. As long as AMD starts to feed that core that wants more work and has plenty of room to get more work done, they will catch up much faster considering Intel's IPC is capping.

http://www.anandtech.com/show/6355/intels-haswell-architecture

seems to me pretty much this entire article is all about reducing power. Why is Intel going after power use so hard now? Im sure someone will show up soon and let you konw.

Yep, read that article yesterday, and it speculates that the reason why Intel is extending Haswell's power range is to move the lowest TDP down into tablet range to go after ARM. Basically a two-pronged approach between 14nm Atoms and Bridgewell eventually to put the squeeze on ARM..

Never heard any allegations before, that Intel was trying to influence programmers (or software devs) into "believing that its impossible to code for multiple cores", seeing as how Intel sells 10-core/20-thread server Xeons as well as 6-core/12-thread desktop Sandy's . It would be rather silly of Intel to discourage coding for multiple cores, seeing the huge profits they make off those Xeons and E-class desktop CPUs..

I sorta suspect Intel will come out with an alternative approach to HSA - they seem to have a lot of hubris and want to set the standards themselves, which is why 3DNnow and FMA4 and who knows what else didn't see much adoption in the marketplace. About the only thing I can think of where AMD prevailed was X64, catching Intel asleep at the wheel (yet another car analogy) or more likely wanting to push Itanium as the 64-bit CPU instead of the P4.

As for IPC, it's important for those threads that can't be parallelized past a certain point, and clearly AMD realizes the same now that they intend to bump up BD's IPC by providing the resources it should have had, with Steamy. I noticed that while Haswell basically retains the same front end as Sandy & Ivy, with improved branch prediction etc, they have increased the computational back end width by 33% by going to 8 ports instead of the current 6.

And I suspect that once AMD gets into similar 'approaching asymptotically' IPC territory as Intel currently is, they will have a more difficult task of obtaining further gains than Intel has. Intel has had the best X86 branch prediction in the business for decades now, their R&D budget is huge compared to AMD's, and they spend $$ working with the software devs to help them improve their code while at the same time getting invaluable experience as to how all that code can benefit from architectural changes, also a form of R&D. The larger transistor budget available to Intel due to being on at least one lower node at any given time means that they can increase performance at the same die size and lower manufacturing costs as well.

Finally, given AMD's statement that they don't intend to compete at the high end, I suspect that Steamy or Exxy will fall well short of whatever Intel has at the time to compete, and stick with their "good enuff" approach.