AMD Piledriver rumours ... and expert conjecture

Page 280 - 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 ...
 


Sorry, posted the Wiki def for those others here who think Senior Notes are a tune being hummed by some old guy 😛..

As you say, the devil is in the details of the contract. But if I were a bank buying and reselling said senior notes for AMD, I'd make sure I had some serious leverage in the contract..
 


IMO, not only management but execution as well (as clearly stated in the last 6 or 7 quarterly conference calls).

A short history review: AMD's 65nm had problems initially (IIRC there were some 90nm CPUs that outperformed the same design on 65nm). Yields and performance far under what AMD predicted - Barcelona debuted with clock speeds around 2GHz instead of the targeted 3GHz or so. IIRC AMD was already at 2.4 GHz or maybe 2.6GHz with their mature 90nm process. So IMO, this plus the cost of developing for future nodes as well as the huge losses in 2008-2009, is what made AMD spin off their fabs to ATIC in the GloFlo partnership.

Now where management made some costly mistakes was in cutting the deal sweet enough to convince the UAE company to agree to it. IIRC AMD got around $800M - even at the time, didn't seem like AMD was getting that much out of it, but they had no choice after the ATI purchase and the huge losses in the 2 years before the deal. The big benefit, according to AMD and GF, would be lower fab prices due to GF spreading the costs of those future nodes over more than one customer. But even so, those wily Arabs 😛 imposed new terms giving AMD just around 35% ownership of GF from the initial 50/50 split, even before the company took off, due to AMD's stock price taking a tumble.

And when you look at the details of the various fab deals since then, between GF and AMD, it seems clear that once ATIC & GF had AMD committed to the deal, they went for the jugular with all sorts of terms and conditions in exchange for that one-year 'good wafers' contract, and then the exclusivity waiver when it turned out GF had yield problems on 32nm.

And the kicker is, I believe GF still has only one customer for 32nm SOI - AMD. Even back in 2009, SOI wafers amounted to just 5% of the global market: http://www.theinquirer.net/inquirer/news/1025681/why-amd-set-fabless

So the R&D plus capex savings with multiple customers never materialized. Which explains why AMD is moving to strained silicon on 22/20 nm with TSMC and GF and maybe Samsung according to the rumors.
 


Well, they might have done what I recommend almost a year ago: Set the HTT bit so Windows scheduler handles BD's modules properly... I suspect fake benchies anyway, but that would be interesting if true...
 


The workforce question is a simple one: Look at the total cost of the employee. Look at how much revenue they bring in to the company.

If negative, fire them.

Layoffs, with very few exceptions, translates too "We F'd up, so we're going to make investors happy (less loss/more profit) though workplace reductions". Which works great for a year or too, until your next product miss. Rinse and repeat until bankrupt.

Hence the downside to corporations: Short term focus rules all. No investor wants to hear "We're going to bleed cash and share for two years while our new kickass product is being designed. Just wait until then!"; that gets the CEO fired. Oh wait...
 
I heard that windows 8, however bad it's UI, has big improvements in its scheduler for bulldozer. Any news on that yet, and what it means for performance?

I would like to see a update-able article that benches different cpu's across different OS's (windows xp 7 8 linux mac) with the exact same other components (as applicable). Just to see how much OS has changed in handling CPUs.
 


You know how your iTunes music costs 99c+/song well, we pay $1.69-$2.19/song despite our AUD being beyond parity 😛 Go fish.



Oh noes! I have an AM3+ motherboard 😛, AMD can't do an FM1 to me.



I for one am sick of cooling Northbridges, Southbridges and etc. I welcome a move to smaller integrated motherboards.



They'll start selling engineers to Intel. 😛
 
AM3+ has been around since what? 2008? I do agree that AMD shouldn't pull another FM1 ever again, but Intel seems to be in the habit of giving a socket form a year more or less before dropping it (they discontinued a handful of LGA1155 CPUs already).
 

I know all about the Australian market, it's beyond screwed up, they charge a fortune for video games- such as Call of Duty: MW2 for $59 CDN/USD, but it was $99 AUD, I don't even want to mention the price of the "hardened edition".
 
First ES samples of FM2+ Kaviri based APU's should be available by February possibly March, number brandishing is going round but it goes on to echo the sentiments of many here that APU's have a future while FX seems to be rather pointless. Not only is the evolution of the APU got greater promise the FM2 parts are able to match Phenom II's and FX 1 parts in discrete graphics performance. Piledriver is better and notable enough but the APU offers evolving iGPU solutions with integrated 7700 GCN parts expected to be confirmed.

The other interesting prospect being mentioned is asymmetrical crossfire support may be extended to two discrete cards, the downside is that despite having GCN mainstream iGPU cores asymmetrical crossfire is only supported by low level graphics cards yet to be announced but watered down 7700 cards but on the GCN architecture.

I mentioned it before but AM3+ goes against what the "future is fusion" mantra implies so to me its the odd chip out, its also very outdated feature wise so to me its the kicking of a dying horse.
 

having two discreet cards for asymmetrical cfx sounds nice... but amd will have to build a reasonably powerful cpu able to drive two cards or three gpus without bottlenecking them. even if those are low end. i wonder if kaveri's cpu would be that powerful....
i doubt that there willl be gcn cards lower than 7750. amd's igpus themselves cannibalized low end discreet gpus. amd still making low end retail cards sounds contradictory.
it'd be very nice if kaveri's igpu cfxes with 7750 or 7770. i dream.
 
I don't see the reason to drop AM3+

What will we get with a new socket ??
Better phase power design something better than the actual 8+2 ??
And that's all

PCIe 3.0 is due to CPU and chipset, and there is no real need for PCIe 3.0 since there is no GPU that can use the full bandwich of 2.0

Better change socket when DDR4 becomes mainstream
 


Some time back it was mentioned on a techpowerup article that Kaviri would support HD7750's for the highest end part as the iGPU is some revision of the 7700 parts, the other mention was iGPU performance around or over 1.3-1.5 GFlops which is impressive for a iGPU. On the CPU side, with new architecture which many by now have seen the intended changes probably will be AMD's biggest CPU performance jump in many years, that said the IPC improvement on Piledriver alone many even hardened critics of AMD say is a mighty fine effort. I have found that the A10 5800K @3.8ghz is very capable of gaming on single high end GPU's and scarily enough get very close or match upper FX and Thuban parts, as for multi high end cards, wrong platform for that, but I did get good results with HD6850's and GTX560SE's. I honestly used to think very little of the APU's but really when you disect them its a very capable piece, not enthusiast level unless you enjoy overclocking notably extreme clocking, but as a platform it is capable of letting the budget user play titles like BF3 at low-med setting at 16x10 res, some titles max out at full HD but every single title is playable just have to find the blend, then on the other hand it can very well take a discrete card.

Feature wise A85 gives plenty connectivity, ASRock have lucid chipsets, Most are 8+2 or better Digi VRM control, come in ATX and M-ATX form factors and now support 2way CFX and SLI all under $130 so far as the most expensive, normally under $100. I am beginning to wonder what FX's purpose is.
 

AM3 came out in 2009, am3+ in 2011. the funny thing about amd sockets is their compatibility direction. all am3+ boards will support all the newest cpus along with the previous gen (AM3) AM2 cpus are not compatible because they don't have a DDR3 controller.

on the other hand, most AM2 boards can support up to the AM3 cpus. for the most part, all AMD cpus are downward compatible except BD wich actually used larger pin diameter, you can actually upgrade from an athlon 64 up to a phenom II without changing motherboards, provided the manufacturer supplied a bios update ... Wich is the reason I dropped MSI as a personal favorite, they decided not to update their 690 boards. Asus and Gigabyte pretty much have full support even on their older 6xx chipset AM2 boards.

longevity is never a bad thing unless your worried about making profits on motherboard sales with each cpu sold, although it would be interesting to see the 4170 difference on a 760g board against a 990fx board.
 

be happy minimal wage in AUS is almost 2x as much as it is in the states.
 


The Australian Minimum Wage Myth
http://www.foxbusiness.com/on-air/stossel/blog/2012/07/13/australian-minimum-wage-myth

Most people who earn minimum wage are young, unskilled workers. How are they doing in Australia?

In June, Australia's unemployment rate for workers age 15 to 19 was 16.5%.

Last December, 63% of all jobs lost were jobs for young, unskilled Australians.

It's simple: when the price of something goes up, people buy less. So when the price of labor goes up, employers hire less.

Read more: http://www.foxbusiness.com/on-air/stossel/blog/2012/07/13/australian-minimum-wage-myth#ixzz29xSTV2xn
 
yea. I wouldn't really trust anything from fox regarding anything. They probably made that just to keep the minimal wage from going up in the states.
 


Isn't min wage there 14$, and valued the same as USD?

If so, it takes less work hours to buy a game for 99$ at 14 dollars an hour compared to buying a game for 60$ at 7.75 dollars an hour.

7 hours of work gets you a game in Australia.
7:30 hours gets you a game in America.


Correct me if I'm wrong.

 
http://www.overclock.net/t/1318240/h-8350-leak-vs-3770k-ht/0_40
[H] 8350 Leak vs 3770K -HT

3770K with HT Disabled, 1866 memory with 9-10-9-28 timings

I don't know how to take those few benchmarks thanks a lot for posting them though back when BD was coming i was so excited and was looking and looking everywhere for info but now with Piledriver(and since i'm sticking with what i have) i pretty much already know what to expect.

One thing is for sure its going to have some great Price/Performance ratios but in reality its going to be only 30$ cheaper than the unlocked I5 and perform better under Multithreaded workloads by a percentage and perform worse in 4 cores or less benchmarks by a percentage by exactly will have to wait and see.

Under Fritz chess the 8350 got a score of 14184 when it was clocked at 4.5Ghz while my processor got a 13573 while clocked at 3.9Ghz. Under cinebench the 8350(While overclocked to 4.5Ghz) got a 7.53 with all cores while mine gets a 6.91 while clocked at 3.9Ghz under those two benchmarks it looks like Piledriver is still 5-10% slower per clock when compared to the Phenom. But its getting a higher clock by around 21% compared to my processor at stock and its getting a 8% higher clock when compared to the Phenom II x4 980 all of this under the same TDP. Not to mention under newer applications Piledriver will pull way ahead of the Phenom and probably under newer games as well. I'd say recommending the Phenom over Piledriver will not be a feasible option which is great news.
 


GCN is better but it's going to take more than that to rocket the APU speeds up. AMD already has bandwidth limitations with Piledriver seeing improvements up to DDR3 2400.

GCN on the desktop has the benefit of 150+GB/s of local GDDR5 memory.

AMD is running into scaling issues without adopting faster memory or on-die RAM/cache. DDR4 is still looking at late 2014 for mainstream so that's 3 chips away (SR, EX, ??? ).
 

PD will still suffer from the module scaling over phenom II. That part of the design hasn't changed, we will have to wait till SR to see any changes in that department.

I do see pd as a small improvement, but more of an implementation of RCM wich should reduce power usage.
 

Kaveri is the last APU that is going to have no on die memory Im guessing.
 
Status
Not open for further replies.