AMD Piledriver rumours ... and expert conjecture

Page 138 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Status
Not open for further replies.
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 ...
 

Globalfoundries said Thursday that it had begun to install a special set of production tools to create through-silicon vias (TSVs) in semiconductor wafers processed on the company's leading-edge 20nm technology platform at Fab 8. The TSV capabilities will allow customers to stack multiple chips on top of each other, providing another avenue for delivering the demanding requirements of future electronic devices.

TSVs (vertical holes etched in silicon and filled with copper) enable communication between vertically stacked integrated circuits. For example, the technology could allow circuit designers to place stacks of memory chips on top of an application processor, which can dramatically increase memory bandwidth and reduce power consumption, a key challenge for designers of the next generation of mobile devices such as smartphones and tablets.

At leading-edge nodes, the adoption of 3D stacking of integrated circuits is increasingly being viewed as an alternative to traditional technology node scaling at the transistor level. However, as new packaging technologies are introduced, the complexity of chip-package interaction is going up significantly and it is increasingly difficult for foundries and their partners to be able to deliver end-to-end solutions that meet the requirements of the broad range of leading-edge designs.

"To help address forthcoming challenges on new silicon nodes, we are engaging early with partners to jointly develop packaging solutions that will enable the next wave of innovation in the industry. Our approach is broad and collaborative, giving customers maximum choice and flexibility, while delivering cost savings, faster time-to-volume, and a reduction in the technical risk associated with developing new technologies. With the installation of TSV capabilities for 20nm technology in Fab 8, we are adding an important capability that will be supplemented by our joint development and manufacturing partnerships with companies across the semiconductor ecosystem, from design to assembly and test," said Gregg Bartlett, chief technology officer of Globalfoundries.

That does seem to be the wave of the future. However given GF's problems with HKMG on SOI (which they appear to be abandoning at 22nm/20nm according to a BSN article - will be on strained silicon only), I just wonder how successful they will be with TSVs and 3D stacking.
 
I have a problem with "top stacking" on CPUs... Why in hell are they doing it on top first instead of just making it package with interconnects in another layer?

I'm sure is not THAT hard to do so and it's been done before (different stuff in the same CPU package). Is it a cost thing? Marketing thing?

Cheers!
 
What does "optimized for Intel" mean then? Easiest thing Intel can do to optimize for their cpu is offer the dev their compiler with a little help to use it.

http://www.intel.co.uk/content/www/uk/en/gaming/overclocking/games-optimized-for-intel.html

If absolutely everyone uses MSVC then why does Intel even have their own compiler?
Because Intel don't want their fate to be in other people's hands, so they do their best to minimise this happening.

By having their own compiler, it puts pressure on Microsoft and other compiler makers to constantly be improving their compilers.

This is part of the reason why x86 has prevailed over many superior architectures.

If you refuse to believe that Intel's agenda is to get everyone to believe how slow AMD is then there isn't any more we can discuss.

How is this either relevant or significant?

Every company tries to downplay their rivals products, yet once again we see this crazy suggestion that Intel is the Great Satan for doing the normal thing.
 
8120 now for 165, isn't it good



Not bad at all really its where it should of been(maybe even a little more like 179.99$) on release its kinda late with ivy coming out. This is the only thing Amd can do for now is lower prices, Makes me want to buy a 8150fx for the fun of it just to see if i can overclock the crap out of it, and if i can get great temps with my 212+ and maybe even get a small boost here or their. But... i'm going to wait until Piledriver and then upgrade If its priced decently.




gamerk316

When i first came on here i never found much of what you were saying to be logical(same as chad) but now i do? i wonder if its me or you? 😗
 
I have a problem with "top stacking" on CPUs... Why in hell are they doing it on top first instead of just making it package with interconnects in another layer?

I'm sure is not THAT hard to do so and it's been done before (different stuff in the same CPU package). Is it a cost thing? Marketing thing?

Cheers!
Actually, traveling accross the di is slow, as opposed to just to be ro reach up, so to speak.
You set too many layers, each has to have its insulation, if you will, this way, with an interposer, that insulation is there, the distances are shortened.

The only possible trade off, heat, but even this may be better comparatively
 
MU, thanks for the post. Lots of info. Most people forget that GPUs tend to be the major bottleneck in games these days while most CPUs are more than enough and under utilized.

There are a few games that are not coded well that will need a faster CPU per say and thats where Intel may shine but normally Intel and AMD do the same in gaming at high resolutions.

Except for overclockers, IB is not a step sideways or even backwards. The 3770K is a bit cheaper than the 2700K, slightly outperforms it in CPU benchmarks and greatly outperforms it in QS and GPU. Plus there's USB 3.0 and PCIe 3.0. Besides, the Z77 mobos did not appear and then just sit around for 6 months waiting for IB to release, unlike AM3+ and BD.

IIRC SB wasn't that great an overclocker either when it first released - I think the average review pegged it around 4.6 or 4.7GHz. Most people expect IB to similarly improve with time.

While IB for desktop does not make sense as an upgrade for anyone with SB, it does for those of us still using C2Q's 😛..

Thats true. SB was about 4.5-4.7GHz on air, the higher if lucky. There was no stepping improvement but the mobos and the BIOS did get some upgrades (Z68) which may have helped. There were very few SB CPUs at 5GHz on air cooling and when they started to appear, they were using modified BIOS, mostly on Gigabyte mobos if I remember, which gave the extra boost. Some were getting 5.6GHz on air but I doubt temps were within reasonable.

As I said before, there will be a GHz barrier we wont overcome for a few more gens of CPUs. It took us quite a while to get past 4GHz on air, Pentium D was the first to get very close and if I remember correctly it was Core 2 that hit 4GHz air first then first gen Core i and Phenom II. SB was really the first arch to go beyond 4GHz on air.

I think even PD with that new tech will have a hard time getting to 5GHz on air.

Random +1 to valve. 9 year old games from them push 4 of my cores equally. I love good programming.

On topic; I think that saying BD should have been clocked higher is true. Most every FX-8xxx can reach 4.2Ghz or more. I think AMD left the clocks lower is because of the already excessive power use. That being said, the fact that my FX-8120's voltage supposedly goes up to 1.55v under max turbo blows my mind. I could undervolt the chip to 1.1-1.2v with little problems probably. Power use must drop some by doing that, so why doesn't AMD do it?

It depends on the VALVe game but any of them using the newer L4D engine, which will have multicore rendering in the video options, will do this. I remember when VALVe implemented it in L4D and my Q6600 would hit 80% usage on all 4 cores.

I haven't checked usage of TF2 or any game on my 2500K but I doubt its 80% on all 4. But that is the benefit to creating an amazing modular engine and just upgrading it over time instead of just throwing a brand new engine out that takes more time and money to develop and possibly more bugs/issues. I would say that Source is probably one of the most stable engines out right now.

As for BD, the reason why they don't undervolt them or set them to 1.v-1.2v is because they probably wont have as many stable CPUs. There are always a few CPUs of any arch that will allow for lower than stock voltage and be stable. My Q6600 did 3GHz on less than stock (1.325v) and even less than the VID (1.275v) at 1.25v and was 100% stable. Never had a BSoD. I had a QX6850 that hit 3.5GHz using 1.275v, less than the VID of 1.325v. My 2500K is doing 4.5GHz at 1.3v (most need about 1.35v to hit 4.5GHz stable) and idles 1.6GHz at .986v. I seem to be lucky in that manner..

But that said, AMD goes for what is mostly stable instead of the lowest stable. So does Intel.

Because Intel don't want their fate to be in other people's hands, so they do their best to minimise this happening.

By having their own compiler, it puts pressure on Microsoft and other compiler makers to constantly be improving their compilers.

This is part of the reason why x86 has prevailed over many superior architectures.

If you refuse to believe that Intel's agenda is to get everyone to believe how slow AMD is then there isn't any more we can discuss.

How is this either relevant or significant?

Every company tries to downplay their rivals products, yet once again we see this crazy suggestion that Intel is the Great Satan for doing the normal thing.

Because Intel is evil. Yet strangley in all of their marketing and all of their performanc estimates, they never compare their upcoming product to AMD, just their previous product while AMD always compares themselves to Intel in their marketing.

Still I think this topic should saty back where it was, 10 pages + ago. It tends to fire people up.
 
True and the Intel fanbois stirring things up are welcome to go back to the Intel threads if they have nothing of value to add here.

I believe I mentioned this several times.

Post stuff relevant to the topic of PileDriver ...or failing that Trinity is fine ...since we didn't put up a specific sticky for that.

If you want to just stick the boots in to AMD ... then move along.



 
I have a problem with "top stacking" on CPUs... Why in hell are they doing it on top first instead of just making it package with interconnects in another layer?

I'm sure is not THAT hard to do so and it's been done before (different stuff in the same CPU package). Is it a cost thing? Marketing thing?

Cheers!

Latency & bandwidth, for one thing. Going vertical means you can have a really wide, low latency connection to such things as LDDR for the GPU. Going horizontal might require yet another layer of connections to add to the overall capacitance and typically means fewer in number and longer in length, depending on the physical layout of the two devices. IIRC there are quite a few layers on modern CPUs already. Of course, heat is an issue but according to S/A, the reduction in allowed power is much less significant than the benefits.
 
http://www.brightsideofnews.com/news/2012/3/6/amd-ditches-soi-kaveri-goes-bulk-at-gf2c-more-details-from-the-new-wsa.aspx

AMD Ditches SOI: Kaveri goes Bulk at GF

On the conference call regarding the Wafer Supply Agreement (WSA) amendment covering 2012, AMDs CFO Thomas Seifert delivered a few interesting statements that reveal some small details regarding the roadmap going forward. We will also go into a few details on the impact of the new WSA on AMD.

There is one statement made in the Q&A section of the call on the new WSA that caught our attention:

"So with respect to SOI (Silicon On Insulator), we made statements that on 28-nanometer, all of our products will be bulk."

Also he basically confirmed, that Kaveri, the successor to Trinity is going to be manufactured at GlobalFoundries at the 28nm node:

"If you look at the roadmaps that we have presented at Financial Analyst Day, there is a 28-nanometer successor product to Trinity on the roadmap that we will ramp next year, and that is also manufactured at GLOBALFOUNDRIES."

Kaveri will integrate 2-4 Steamroller CPU cores (successor to Piledriver cores in Trinity, 3rd Generation Bulldozer core) and a GPU based on the GCN architecture. It will also provide support for HSA, which means integration of third party IP into the same die. The APU is scheduled for a 2013 launch.

Departure from SOI
At the Morgan Stanley Technology, Media and Telecom Conference Thomas Seifert last week gave a bit more detail on the matter:

"We said that at the 28nm node we are going to be on bulk silicon across all products, not only graphics but also CPUs. And We have made no statement beyond that. But for 28[nm] we will be on bulk for all products."

He also added:

"There are always tradeoff decisions. But the flexibility that we gain moving in that direction... the flexibility across foundry partners, across design tools outhweigh that by far, the benefits of SOI."

Seems to be a smart move on AMD's part, since that opens up other foundries like TSMC to fab more of their CPUs. I would not be surprised to see all of AMD's CPUs going to bulk Si eventually.
 
Latency & bandwidth, for one thing. Going vertical means you can have a really wide, low latency connection to such things as LDDR for the GPU. Going horizontal might require yet another layer of connections to add to the overall capacitance and typically means fewer in number and longer in length, depending on the physical layout of the two devices. IIRC there are quite a few layers on modern CPUs already. Of course, heat is an issue but according to S/A, the reduction in allowed power is much less significant than the benefits.

My problem with that is they still have the alternative to do it horizontal, yet they don't. I'm pretty sure an additional interconnect layer on the package of the CPU is not that expensive in heat nor cost to do, but yet they don't. I know it won't be as fast as having L4 on die or "on top", but it will still be way faster than having DDR3 or even DDR4 put outside (latency wise at least).

There's still space left in the CPU package they could use, so it's really confusing for me. Why haven't they even tried it? Maybe I'm wrong and they actually did, but it was a flop or something?

http://www.brightsideofnews.com/news/2012/3/6/amd-ditches-soi-kaveri-goes-bulk-at-gf2c-more-details-from-the-new-wsa.aspx

Seems to be a smart move on AMD's part, since that opens up other foundries like TSMC to fab more of their CPUs. I would not be surprised to see all of AMD's CPUs going to bulk Si eventually.

This is all in favor of the "Execution" Mr. Read promised to accomplish I guess. And I agree as well, smart move; maybe they'll get more money delivering chips instead of waiting for GF to get their act together.

Cheers!
 
😱
😱 90 for 960t + board 😱
😱

I know right !!! I paid $99.99 for my 960T alone, add another $115 on my mobo. Microcenter has a 960T plus an extreme 4 (mine only extreme 3) for 159.98 !

I'm not really sure why they are slashing their prices so much on the Phenoms, the FX's are a side grade at best (IMHO) And I'm pretty sure we can all agree the good Phenom's like the C3 Deneb's/Thuban's and Zosma's are actually quite in demand at the moment for a power AMD machine. Oh well I'm not complaining, get em while you can. I know I will !
 
IMHO with a AM3 board the choice between AM3+ mobo and PD or going with a Thuban BE and same mobo is a no brainer
unless you need the features of a new mobo then a Thuban or Deneb is the only way to go
especially if you dont mind buying used
really CPUs rarely fail and they either work or dont work
so as long as you have a decent return policy or can try it in your unit first then going used for a CPU is the way to go
I have a chance for a used 955be for $50 USD
on craigslist and at that price I will buy it blind and take my chance
I have gambled $50 on stupider bets LOL
 
As for BD, the reason why they don't undervolt them or set them to 1.v-1.2v is because they probably wont have as many stable CPUs. There are always a few CPUs of any arch that will allow for lower than stock voltage and be stable. My Q6600 did 3GHz on less than stock (1.325v) and even less than the VID (1.275v) at 1.25v and was 100% stable. Never had a BSoD. I had a QX6850 that hit 3.5GHz using 1.275v, less than the VID of 1.325v. My 2500K is doing 4.5GHz at 1.3v (most need about 1.35v to hit 4.5GHz stable) and idles 1.6GHz at .986v. I seem to be lucky in that manner..

But that said, AMD goes for what is mostly stable instead of the lowest stable. So does Intel.

Because if you sold the same cpu at a lower power envelope (using better binned cpu's) then no one would want the same chip with higher power reqierments. The highest demand for the chip would fall on the model with the lowest supply.

Makes sense.
 

Same case as that of mine 🙁
(during assembling of mine brain in GOD's shop) looks like i choosed a 64GB ssd instead of a 2TB mechenical drive,
and thats why i need to install all programs from cd/dvd (books) every time i want to use them 🙁
and thats the cause which is why i am a £¤¤$€® now 🙁
 
Heck there are mornings I am loading off of audio data cassettes LOL
does anybody remember loading programs from magnetic audio cassettes?
I had a trs-80 that used them

so according to that link IB HD 4000 IGP can match up now to Llano APU in graphics performance but they are saying that Trinity APU with 7000 series graphics will be %50 better than Llano so AMD should have the lead again in IGP onboard graphics

 
Heck there are mornings I am loading off of audio data cassettes LOL
does anybody remember loading programs from magnetic audio cassettes?
I had a trs-80 that used them

so according to that link IB HD 4000 IGP can match up now to Llano APU in graphics performance but they are saying that Trinity APU with 7000 series graphics will be %50 better than Llano so AMD should have the lead again in IGP onboard graphics

thats why i am leaning towards ivy for no ndgpu build, but i wan max cpu performance but without crossing tjmax.

but according to this http://ehow.com/m/info_8482421_radeon-hd-4290-specs.html
hd4290 is having 700mhz core clock, also i can overclock my hd4250 to 750 so maybe it can beat hd2500k
 
Status
Not open for further replies.