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 ...

 

[BaronMatrix]

I so called it. AMD is having 32nm yield issues across the board, not just Llano. Llano has a bit more mainly due to the GPU, probably not the CPU.

Looks like low supply for BD until GF gets 32nm mature.



As I said before, I would assume 10% increase if Trinity doesn't follow BDs suit. The 30% GPU I would imagine would be due to higher clocks.

Wish AMD would put more info out there. Kinda annoying to announce something with very little info on it.

God himself could tell you you're wrong and you'd disregard it. The problem for BD is the amount of 32nm wafers WASTED, not BD yields. That also affects Phenom. Are there yield problems with Phenom?

 

[dogman_1234]

Welcome back Baron.

 

[gamerk316]

AMD just needs to get more support for XOP and FMA. If people aren't doing it even though Intel will have it means the shenanigans continue. But PD should be the third rev of the Bx series BD. PD is supposed to have bit manipulation instrs that may help with a lot of things. Have to research it.

Two points:

One, you are NOT going to see devs re-compile already released software to take advantage of ANY new processor, then ship it out again.

Two, only about 10% of devs use Intels compiler; most use MSVC [which is one of MSFTs best products]. So the "Intel doesn't optimize for AMD" argument is a bit of an overstatement.

BTW: Intel has the same exact problems with getting new instruction sets supported; the P4 wasn't that much faster then a P3, but when SSE was used, it was significantly faster. Some logic on display here.

 

[BaronMatrix]

Some of us - Jimmysmitty, myself, I think Chad and several others - tried pointing these potential problems with GF out for the last couple of years here, but many of the AMD diehard fans dismissed it as never gonna happen, due to the deep ties between AMD and GF. Unfortunately nothing, including ties between companies, last forever. I don't think AMD is going to let past ties influence what should be a purely business decision as to which foundry to use. So if they get better results from TSMC, then that's whom they should go with.

As for Intel fabbing AMD CPUs, might not be so far-fetched if the FTC steps in due to competition concerns. Given the onslaught of Intel products - Knights Corner, SB-E and Xeon, IB next year, IB-E late next year and then Haswell in 2013, AMD seems to be facing a large number of big challenges in the next 18 months. If Interlagos and PD/Trinity don't generate a lot of income during that timeframe, I can see AMD dumping server and high-end DT entirely, which might get the FTC's attention. Interlagos is already bargain-basement priced from what I can see, so not a lot of room to drop prices for a huge chip like that.

Finally, all those AMD diehard fans should start listening to Jimmysmitty - his crystal ball is working at 1080P, whereas some other guy's is working at maybe 10P 😛..

But what you are in effect sayign is that Jimmy could tell you what would have happened with 32nm IF AMD had NOT spun-off. And it is not a point of this thread. I can't say what Project Win is, but with BullDozer, AMD could slay by putting money into XOP\AVX and FMA.

PileDriver will have FMA3 which is compatible with the Haswell implementation. If AMD can land CS5 or Maya for PURE AVX\XOP FMA, they will have a year lead before their 3RD gen FMA which would come with the PD FX. Trinity is PD and would be the most powerful CPU for OpenCL and Win8.

CTI is a trademark and will be used for anything GF does. Certain people act like ATIC threw out all of the AMD Fab equipment and workers - one case where those at the top were sacrificed(Doug Grose et al). By the time PileDriver comes out GF should be flush with 28nm orders and plenty of profits so the naysayers will see that building a business isn't just a simple 1,2,3.

Because BD shipped at B2 and the oft-compared Phenom shipped at C2, I feel that means AMD will have a "true" 95W 8xxx FX in time for Win8. They can still try to get an interim patch in for the current rev for Win7, but it has to not step on future HW changes. MS likes to throw drivers out of Windows builds. CPUs are the only Kernel mode devices though so there's less risk.

Anyway, I would say that AMD may demo Trinity and PD at the delayed Analyst Day. That means mid-Q1 for tweaks - they don't have to send the samples out then. Another key for PD is that AMD still created MS' x64 implementation and have consistently increased perf\watt. People look at "certain" benchmarks and treat them as GOSPEL but reviews are designed to inform buyers of the total cost for the experience they want - something reviewers are currently failing horribly at.

Let's even for conjecture's sake say that the ENGR who bemoaned the choice to use automation is wrong because then you can use the SW to make the circuits better with less time spent on engineering and more on enhancements and future designs. Even the most successful company has to make choices.

Can YOU say BING!

The total negative vibe makes it impossible to look at the engineering aspects and any "reasons" are "excuses" rather than reflecting the complexity of 2B transistor chips (SB-E is 2B ) which try to unload the "legacy baggage."

Even if we just look at B3 which will - in terms of manufacturing - be included in the PD revs there should be 3-4 months of CTI at plat which should lower current load power. But considerign that the Anand Opteron review showed that new systems need new settings - maybe THAT'S why you have to Flash the BIOS - and Win2008 vs. ESX showed that software does matter as the B2 Opteron idle was comparable to Xeon with only 6C\12T.

As soon as other obligations are covered, FX is my upgrade. I'll get enough f a boost from SSD that I'll be blissful waiting for software - or writing my own.

 

[BaronMatrix]

Also remember that the Scheduling issue starts at at Windows Startup such that every PROCESS that spawns more than two THREADS may end up on a module not in the L2 range of the PROCESS. That type of cascade effect could easily drain 10-15% off of total processing efficiency in a situation with 100s of PROCESSES and 1000s of THREADS. Or Windows 7. Time slices also play a large role as context switches with "misses" are even more costly.

Not sure what MS is planning for time slices with Win8 or Win7 SP2. It will all roll up into Win8 Server so AMDs future is bright with the only real DX11 in 2012 for Win8 - Tegra uses ARM and won't play CrySis.

 

[gamerk316]

Also remember that the Scheduling issue starts at at Windows Startup such that every PROCESS that spawns more than two THREADS may end up on a module not in the L2 range of the PROCESS. That type of cascade effect could easily drain 10-15% off of total processing efficiency in a situation with 100s of PROCESSES and 1000s of THREADS. Or Windows 7. Time slices also play a large role as context switches with "misses" are even more costly.

Not sure what MS is planning for time slices with Win8 or Win7 SP2. It will all roll up into Win8 Server so AMDs future is bright with the only real DX11 in 2012 for Win8 - Tegra uses ARM and won't play CrySis.

Baron, MS isn't going to re-code its entire scheduler because AMD refuses to call a logical core a logical core. AMD is quite capable of fixing this issue itself by setting a single bit in the CPUID field.

 

[fazers_on_stun]

I would say it's because they may want to have a six core APU next year.

I thought Trinity was dual-module (or 4-cores in AMD-speak). Three modules coupled with a higher performance GPU might be a bit much to fit into a 130 watt TDP, given the current 8-core 8150 load power consumption..

 

[fazers_on_stun]

God himself could tell you you're wrong and you'd disregard it. The problem for BD is the amount of 32nm wafers WASTED, not BD yields. That also affects Phenom. Are there yield problems with Phenom?

Isn't "yield" defined as the # of good chips out of the total produced? I don't see much difference between "wafers WASTED" and yields, except that you are saying an entire wafer is trashed instead of just individual chips from that wafer.

Anyway, if a fab can only move so many wafers per month, then bad wafers and/or bad chips are a yield problem that reduces the total output available for all the different chips produced by that fab. In the Dresden fab's case, it also allegedly affects AMD's remaining 45nm chips since that line supposedly shares some equipment with the 32nm line. Which also neatly explains why AMD EOL'd their 45nm CPUs in a hurry..

 

[fazers_on_stun]

But what you are in effect sayign is that Jimmy could tell you what would have happened with 32nm IF AMD had NOT spun-off.

No, I'm saying that Jimmy correctly identified POTENTIAL problems with the spinoff, such as AMD and GF moving apart as separate companies and having diverging priorities due to that separation, and possibly communications problems as well. That is just the nature of divestiture, and you really don't have to look too far to see precedent - look at the AT&T split and the Baby Bells created out of that. They became cutthroat competitors soon after the split.

And it is not a point of this thread. I can't say what Project Win is, but with BullDozer, AMD could slay by putting money into XOP\AVX and FMA.

According to S/A, Project WIN was just Read's putting lipstick on the layoff pig, for internal AMD consumption, to take the remaining employee's minds off their apparent lack of job security.

PileDriver will have FMA3 which is compatible with the Haswell implementation. If AMD can land CS5 or Maya for PURE AVX\XOP FMA, they will have a year lead before their 3RD gen FMA which would come with the PD FX. Trinity is PD and would be the most powerful CPU for OpenCL and Win8.

IOW, same-o, same-o. "Please ignore the fact that our current CPU that doesn't do so good, just WAIT FOR THE NEXT ONE". You realize this tactic only works once or twice. Take a look around the various BD & PD threads here - a lot of AMD customers are leaving as they are tired of AMD promising this & that and then not delivering.

CTI is a trademark and will be used for anything GF does. Certain people act like ATIC threw out all of the AMD Fab equipment and workers - one case where those at the top were sacrificed(Doug Grose et al). By the time PileDriver comes out GF should be flush with 28nm orders and plenty of profits so the naysayers will see that building a business isn't just a simple 1,2,3.

By the time PD comes out I wouldn't be surprised to see AMD giving all its 28nm orders to TSMC, since it appears TSMC is ahead of GF on that node.

Because BD shipped at B2 and the oft-compared Phenom shipped at C2, I feel that means AMD will have a "true" 95W 8xxx FX in time for Win8. They can still try to get an interim patch in for the current rev for Win7, but it has to not step on future HW changes. MS likes to throw drivers out of Windows builds. CPUs are the only Kernel mode devices though so there's less risk.

So instead of 5 years for BD, AMD should have waited yet another instead of trying to recoup some of the R&D expenses right now as they are trying to save $$ for next year with the layoffs? I should also point out that by the time Win8 ships, Intel will likely have its second stepping of 22nm Ivy Bridge out too, and possibly lower than the 75W TDP the first one will have.

Even if we just look at B3 which will - in terms of manufacturing - be included in the PD revs there should be 3-4 months of CTI at plat which should lower current load power. But considerign that the Anand Opteron review showed that new systems need new settings - maybe THAT'S why you have to Flash the BIOS - and Win2008 vs. ESX showed that software does matter as the B2 Opteron idle was comparable to Xeon with only 6C\12T.

Has AMD even identified where all the power usage problems with BD arise? CTI = continuous TRANSISTOR improvement IIRC, so if the problems are mostly design-related then CTI won't help much with load power usage. And for idle power, AMD appears to be using power gating like Intel does, so I would say that is also hardware for the most part, as long as the OS recognizes the proper C-states.

As soon as other obligations are covered, FX is my upgrade. I'll get enough f a boost from SSD that I'll be blissful waiting for software - or writing my own.

Well I'd ask you for your own review after you get one, but I suspect the results are pre-ordained already. 😀.

 

[jimmysmitty]

God himself could tell you you're wrong and you'd disregard it. The problem for BD is the amount of 32nm wafers WASTED, not BD yields. That also affects Phenom. Are there yield problems with Phenom?

You just proved my point. If there are bad wafers then there are bad yields. The whole point to yields is how many good chips you get per wafer and if more wafers are being cast as bad and thrown out, that lowers yields. GF has admited to having yield issues with 32nm as a whole, not 45nm and not just the APUs. The APUs had a issue because of the GPU as well so their yields are probably worse than BD but if 32nm as a whole is having yield issues then it will affect BD>

It will NOT affect Phenom II because Phenom II is a 45nm part that was rather mature and GF will still be using the same equipment that AMD was using before the spinoff. 32nm however was after the GF spinoff so its all on GF.

I never said AMD would have yield issues with 32nm, I would have said they would have low yields to start just like 45nm did because the one downside to SOI besides a bit hgiher costs is lower starting yields. But after a bit of time, yields pick up pretty fast.

And as Fazers said, I was talking about GF moving on to their own world after the spinoff. GF is a seperate company and they will go with whatever has their best interest in it. I would imagine that there will be a point where GF will produce more chips for other companies than they do AMD as they need to make money to survive. Do you think TSMC would only do one company? They do nVidia and ATI GPUs along with many others. Its all about the bottom line. In this industry, there are no loyalties. The only company who makes just for one is Intel as they build, research and pay for their own FABS. But I would not be suprised if Intel does at some point move to opening their FABs to others to recoup more costs than just selling the old equipment would.

But still, prove me wrong that BD is not having yield issues due to the 32nm having yield issues. It was released 10/12 and in a matter of hours was sold out on the largest vendors (Tiger Direct, Newegg). It has gone in and out of stock multiple times since its release. So either AMD is holding it back, although thats pointless since they still make the same amount per CPU and the extra pricing is due to low supply, or they have too low amounts to keep the supply channels full. ANd no, Cray buying a buttload would not have affected that. You don't release a CPU with a small amount, AMD has never done that and Cray tends to buy AMDs new CPUs right away yet Phenom II never had these low supply issues.

 

[BaronMatrix]

Two points:

One, you are NOT going to see devs re-compile already released software to take advantage of ANY new processor, then ship it out again.

Two, only about 10% of devs use Intels compiler; most use MSVC [which is one of MSFTs best products]. So the "Intel doesn't optimize for AMD" argument is a bit of an overstatement.

BTW: Intel has the same exact problems with getting new instruction sets supported; the P4 wasn't that much faster then a P3, but when SSE was used, it was significantly faster. Some logic on display here.

Please. How many apps do you have that aren't updated over the Internet?

 

[BaronMatrix]

Also remember that the Scheduling issue starts at at Windows Startup such that every PROCESS that spawns more than two THREADS may end up on a module not in the L2 range of the PROCESS. That type of cascade effect could easily drain 10-15% off of total processing efficiency in a situation with 100s of PROCESSES and 1000s of THREADS. Or Windows 7. Time slices also play a large role as context switches with "misses" are even more costly.

Not sure what MS is planning for time slices with Win8 or Win7 SP2. It will all roll up into Win8 Server so AMDs future is bright with the only real DX11 in 2012 for Win8 - Tegra uses ARM and won't play CrySis.

Baron, MS isn't going to re-code its entire scheduler because AMD refuses to call a logical core a logical core. AMD is quite capable of fixing this issue itself by setting a single bit in the CPUID field.

What does any of your comment have to do with my statements? I explained how the scheduling issue affects everything and you say what MS won't do. I said Win 8 will fix it on the OS side. Nothing about logical cores. I still think you should all just hang out on Intel threads. I tire of the negativity based on a lack of understanding.

 

[BaronMatrix]

You just proved my point. If there are bad wafers then there are bad yields. The whole point to yields is how many good chips you get per wafer and if more wafers are being cast as bad and thrown out, that lowers yields. GF has admited to having yield issues with 32nm as a whole, not 45nm and not just the APUs. The APUs had a issue because of the GPU as well so their yields are probably worse than BD but if 32nm as a whole is having yield issues then it will affect BD>

It will NOT affect Phenom II because Phenom II is a 45nm part that was rather mature and GF will still be using the same equipment that AMD was using before the spinoff. 32nm however was after the GF spinoff so its all on GF.

I never said AMD would have yield issues with 32nm, I would have said they would have low yields to start just like 45nm did because the one downside to SOI besides a bit hgiher costs is lower starting yields. But after a bit of time, yields pick up pretty fast.

And as Fazers said, I was talking about GF moving on to their own world after the spinoff. GF is a seperate company and they will go with whatever has their best interest in it. I would imagine that there will be a point where GF will produce more chips for other companies than they do AMD as they need to make money to survive. Do you think TSMC would only do one company? They do nVidia and ATI GPUs along with many others. Its all about the bottom line. In this industry, there are no loyalties. The only company who makes just for one is Intel as they build, research and pay for their own FABS. But I would not be suprised if Intel does at some point move to opening their FABs to others to recoup more costs than just selling the old equipment would.

But still, prove me wrong that BD is not having yield issues due to the 32nm having yield issues. It was released 10/12 and in a matter of hours was sold out on the largest vendors (Tiger Direct, Newegg). It has gone in and out of stock multiple times since its release. So either AMD is holding it back, although thats pointless since they still make the same amount per CPU and the extra pricing is due to low supply, or they have too low amounts to keep the supply channels full. ANd no, Cray buying a buttload would not have affected that. You don't release a CPU with a small amount, AMD has never done that and Cray tends to buy AMDs new CPUs right away yet Phenom II never had these low supply issues.

The same tools are used for 45nm and 32nm so YES, Phenom and BD are affected. If you have to run 2X the wafers for volume Llano that's wafers AND machines that can't be used for BD and Phenom. AMD even said this in a CC.

 

[dogman_1234]

Please. How many apps do you have that aren't updated over the Internet?

Pretty much all of mine cannot be updated online.

 

[lilcinw]

The problem with Baron is that he thinks in terms of workstation/productivity apps (and when did apps become the vernacular for the desktop space BTW? /rant).

Typically such products are much quicker to support the latest hardware trends (CUDA, multi-thread, AVX) than consumer products. The problem I still have is that BD was positioned as a consumer grade product, which unfortunately is where it doesn't perform as well as I expected.

PS: Trying to run 4+ Exchange server VMs on consumer grade HW makes me want to tear my hair out. I would love to be rocking an FX-8xxx with 16 GB of DDR3 right now.

 

[jimmysmitty]

The same tools are used for 45nm and 32nm so YES, Phenom and BD are affected. If you have to run 2X the wafers for volume Llano that's wafers AND machines that can't be used for BD and Phenom. AMD even said this in a CC.

So AMD, who has very little in the fabrication part now and little to no control over GF knows the exact setup being used.

I honestly doubt Llano is causing BDs low supplies. I still say its the issues with 32nm yields. There is no way one part on the same process is not affected by bad yields on said process. As I said, Llano is a bit worse because the GPU was not designed around 32nm and so the change was a bit harder for GF to grab but still, if GF said they are having 32nm yield issues it affects everything that is using that process tech.

 

[ctbaars]

The same tools are used for 45nm and 32nm.

So AMD, who has very little in the fabrication part now and little to no control over GF knows the exact setup being used

you have to run 2X the wafers for volume Llano that's wafers AND machines that can't be used for BD and Phenom

I honestly doubt Llano is causing BDs low supplies

OK. So I know nothing. I'm calling it all bunk. I don't think any of you know what you're talking about. But you're all very witty, making it very entertaining.

Spoiler

I think an AMD inspector without a mask sneezed in the GF clean area.

I love how fazers_on_stun baited you both into this :lol: especially you jimmysmitty.

 

[jimmysmitty]

OK. So I know nothing. I'm calling it all bunk. I don't think any of you know what you're talking about. But you're all very witty, making it very entertaining.

Spoiler

I think an AMD inspector without a mask sneezed in the GF clean area.

I love how fazers_on_stun baited you both into this :lol: especially you jimmysmitty.

We are talking about process manufacturing. Intel builds and owns their FABs, while AMD used to. They had two at one point but sold one off and kept the one in Dresden Germany. They then sold off their entire Fabrication unit to GF (Guilf Foundries) who is owned now by ATIC which was heavily invested in by the Abu Dubai, whol also invested $7Billion into Citi Bank a few years ago.

My point is that GF recently claimed they were having issues with their 32nm process yields, yeild meaning how many good CPUs you get per wafer. AMD has been using SOI which tends to have a low starting yield but ramps up nicley enough to help negate that. But because AMD now has no control over the Fabrication process and is no longer a majority holder in ATIC/GF, they are having low supply issues. Thats why the FX series of CPUs has been in and out of stock recently, hell we finally got a few FX-8120s last week as none of our vendors could even get their hands on them but we never have an issue with Phenom/Athlon II or Sandy bridge when it was released.

The only benefit I can see from AMDs perspective is two: 1. they don't have to research the new process tech or buy the equipment which save a lot of money, I think Intel said it was $13 Billion (yes a B) to upgrade their 32nm plant in Chandler, AZ to 22nm. And 2. under the current contract, AMD does not have to pay for any bad chips/wafers, only the good ones. But since there will be a point in time where AMD will release any holding they have and ATIC/GF will become an independant Fabrication company, much like TSMC who is doing both nVidia and ATI GPUs, its hard to say if that will stay true as I am sure its hurtting the GFs bottom line. At that point AMD might look to other Fabrication giants like TSMC to do their CPUs.

Or, as some have said, if Intel decides to open thier plants to it maybe Intel will also become one. I would say that would be a good idea as everyone would jump to have the latest process tech since Intel always has the most advanced process technology, 22nm will be out beginning next year.

BTW this is what I was talking about:

Built a AMD system with a FX-4100 at work and this has to be, out of maybe 5 BD builds, the 4th one with this. Its annoying as sometimes the plastic wipes off the stock TIM. Luckily this one didn't but AMD needs to get GF to do better QC because this is just rediculous.

 

[BaronMatrix]

Pretty much all of mine cannot be updated online.

Then you are being ripped off. I bought software without a box that has online updating. Some only post new packages. More expensive software - or even Flash - updates without a reboot.

Is Steam a Pro app?
Real Player?
Movie Outline?
Kindle For PC?
Photoshop Elements?
Icon Workshop?

Then as usual you live vicariously through an abusive company that cares nothing about you.

 

[BaronMatrix]

So AMD, who has very little in the fabrication part now and little to no control over GF knows the exact setup being used.

I honestly doubt Llano is causing BDs low supplies. I still say its the issues with 32nm yields. There is no way one part on the same process is not affected by bad yields on said process. As I said, Llano is a bit worse because the GPU was not designed around 32nm and so the change was a bit harder for GF to grab but still, if GF said they are having 32nm yield issues it affects everything that is using that process tech.

Interlagos is available on Newegg as is every FX except 8150 which sells out as soon as it comes in. There are only OEM A8- 3800s while the mobile versions are where the real volume for Llano is.

But remember that Phenom, Athlon and Turion are being replaced by Llano so they have to have the volume of every AM3+ chip for Llano. It'll take awhile even without GPU yield issues.

 

[BaronMatrix]

Built a AMD system with a FX-4100 at work and this has to be, out of maybe 5 BD builds, the 4th one with this. Its annoying as sometimes the plastic wipes off the stock TIM. Luckily this one didn't but AMD needs to get GF to do better QC because this is just rediculous.

That would actually be the AMD packaging plants in Malaysia. GF ships hermetically sealed DIES, not ceramic packaging.

 

[Chad Boga]

Then as usual you live vicariously through an abusive company that cares nothing about you.

Could you please provide a list of companies that do care about you.

 

[BaronMatrix]

Could you please provide a list of companies that do care about you.

All of those who don't crap on their partners. The list is long. All of those who don't have anti-trust issues everywhere they do business.


Don't get me started.

But at any rate, things are looking good and will look better. TSMC has fab issues and there are no conspiracy theories. I'm confident the engrs are working hard to fix every issue and improve the perf and power.

 

[Chad Boga]

All of those who don't crap on their partners.

That would obviously rule out companies who left their retail partners high and dry when they started supplying Dell. 😉 😀

 

[jimmysmitty]

Interlagos is available on Newegg as is every FX except 8150 which sells out as soon as it comes in. There are only OEM A8- 3800s while the mobile versions are where the real volume for Llano is.

But remember that Phenom, Athlon and Turion are being replaced by Llano so they have to have the volume of every AM3+ chip for Llano. It'll take awhile even without GPU yield issues.

That is true but my work place still had a hard time coming buy an FX series until almost a month after they were released. We just finally got a decent stock in and a few 8120s.

Tht still shows low supply, especially if the top end CPU keeps selling out too fast.

Built a AMD system with a FX-4100 at work and this has to be, out of maybe 5 BD builds, the 4th one with this. Its annoying as sometimes the plastic wipes off the stock TIM. Luckily this one didn't but AMD needs to get GF to do better QC because this is just rediculous.

That would actually be the AMD packaging plants in Malaysia. GF ships hermetically sealed DIES, not ceramic packaging.

Then AMD needs to start smacking some heads there as thats pretty bad. Before BD, I have never seen anything like this from AMD or Intel. Laziness tends to show.

As for the issues I see with the discolored markings on CPUs, is that a GF issue or AMD?

That would obviously rule out companies who left their retail partners high and dry when they started supplying Dell. 😉 😀

I rememebr when AMD did that. A lot of the small OEMs were pretty mad. I would have been too.

But to some, AMD can do no wrong and will cuddle with you every night while also checking under the bed for big bad monsters.