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 ...

 

[gamerk316]

^^ Umm...I don't care how it happens, the stack pointer [ok, ANY POINTER] being incorrect is not a pretty sight. When I get more time, I'm going to read up on this one...

 

[sonoran]

quote>>

"AMD has taken your example and also analyzed the segmentation fault and
the fill_sons_in_loop code. We confirm that you have found an erratum
with som e AMD processor families. The specific compiled version of
the fill_sons_in_loop code, through a very specific sequence of
consecutive back-to-back pops and (near) return instructions, can
create a condition where the process or incorrectly updates the
stack pointer."

<<endquote

We exchanged a few emails to try to come up with a good test case.
Owing to the difficulty of reproducing the bug I constructed a
fully bootable DFly operating system & test case USB image and
verified that the bug was present on my test box using that image.
AMD was then able to reproduce the bug using that image on their own
machines. Over the last few months they have been working through
the possibilities and today emailed me the confirmation that it was,
indeed, a cpu bug.

So they had to have some special operating system this guy wrote to reproduce the error? While this could have been a major disaster, it sounds like it's so unlikely to occur in real life as to be pretty much a non-issue. I suppose some program, somewhere, might repeatedly crash when run on these machines (due to executing this specific set of instructions). But I haven't heard any reports of such a program - has anyone?

 

[saint19]

Too simple for be GTX 6xx series

 

[palladin9479]

The real question is, is it something that can be fixed via microcode patch or does it require another stepping revision?

 

[nforce4max]

The real question is, is it something that can be fixed via microcode patch or does it require another stepping revision?

Isn't worth it, they are all old Phenom 2 era or so that has been posted unless the same bug is in BD then likely people will forget about it.

I did have unusual blue screens with my p2 x4 820 some months back.

 

[-Fran-]

Problem is, that some Opterons involved as well, so I guess it doesn't matter if its EOL'ed. They must find a way to fix it.

I don't remember having any issues so far with the 965 at full load, so maybe it's a very specific set of instructions in a specific order that's causing the problem.

Cheers!

 

[nforce4max]

Too simple for be GTX 6xx series

Are you are sure about that? There is already a short review of the GT 650M out there atm. Pretty small die and current drivers do work. 384 shaders and got 11710 in 3DMark pared with a i3 2350m.

 

[gamerk316]

quote>>

"AMD has taken your example and also analyzed the segmentation fault and
the fill_sons_in_loop code. We confirm that you have found an erratum
with som e AMD processor families. The specific compiled version of
the fill_sons_in_loop code, through a very specific sequence of
consecutive back-to-back pops and (near) return instructions, can
create a condition where the process or incorrectly updates the
stack pointer."

<<endquote

We exchanged a few emails to try to come up with a good test case.
Owing to the difficulty of reproducing the bug I constructed a
fully bootable DFly operating system & test case USB image and
verified that the bug was present on my test box using that image.
AMD was then able to reproduce the bug using that image on their own
machines. Over the last few months they have been working through
the possibilities and today emailed me the confirmation that it was,
indeed, a cpu bug.

So they had to have some special operating system this guy wrote to reproduce the error? While this could have been a major disaster, it sounds like it's so unlikely to occur in real life as to be pretty much a non-issue. I suppose some program, somewhere, might repeatedly crash when run on these machines (due to executing this specific set of instructions). But I haven't heard any reports of such a program - has anyone?

A segmentation fault? Wow, I suspect this bug might have been around for some time then...

I rank this at about the same level as the Intel FDIV bug from way back when. Of course, Intel lost a few hundred million when all was said and done...

 

[saint19]

Are you are sure about that? There is already a short review of the GT 650M out there atm. Pretty small die and current drivers do work. 384 shaders and got 11710 in 3DMark pared with a i3 2350m.

Not sure, but if it's real that series would be the new GTX 4xx

 

[triny]

It wouldn't surprise me to see Samsung get some foundry work from AMD Should glofo
continue to bungle up the job

 

[r0ck3tm@n]

From Intel Xeon E5-2600: Doing Damage With Two Eight-Core CPUs:

Although we’ve seen AMD’s share of workstation CPU market grow to nearly five percent (in 2006, according to Jon Peddie Research), it’s now essentially zero. We’ve repeatedly invited AMD to participate in our workstation-oriented coverage, but it concedes that it’s no longer a player in this space.

In my view, this is sad and unnecessary. AMD doesn't care about workstations? That's not right. I'm guessing that maybe too many people were let go in the course of following someone's business plan and now the engineering team just isn't there. I don't know that I can consider AMD a serious player in the CPU business if they don't care about workstations. Really, that means to me they don't care about servers either. Tell me I'm wrong about that.

Perhaps Piledriver will get them back into the workstation market?

Ok, here is my Piledriver news for today -->[ ].

Right now at this moment I'm smiling that my new build is based on the i7-2600K. I do think that I should have gone with the i7-3820 but as I remember that wasn't actually on the market yet when I ordered. Don't know why anyone would not go this route especially if they wanted lots of PCIe lanes (palladin9479.)

Ok well that's it. Can we get some insider/leaked Piledriver specs or benchmarks?

 

[truegenius]

just saw toms gaming cpu chart,
why they said that 8150 is top in amd gaming cpu
are they joking?
Seems like they are joking.!

 

[de5_Roy]

just saw toms gaming cpu chart,
why they said that 8150 is top in amd gaming cpu
are they joking?
Seems like they are joking.!

may be because 8150 uses the gfx cards better than other fx cpus and least likely to bottleneck cfx/sli configs - especially true for amd gfx cards.

 

[truegenius]

Anything about Piledriver/Trinity in the last 100 post?

here is something to munch upon
www.fudzilla.com/home/item/26204-top-trinity-is-a10-5800k-at-38ghz

wow, 65w a10-5700 with 3.4ghz base (4ghz turbo) + 7660d

to me, cpu side (of a10 5700) seems 5-15% better than llano (a8-3870k) (considering that core of trinity is bd not e-bd which will give more performance)

 

[palladin9479]

From Intel Xeon E5-2600: Doing Damage With Two Eight-Core CPUs:

Although we’ve seen AMD’s share of workstation CPU market grow to nearly five percent (in 2006, according to Jon Peddie Research), it’s now essentially zero. We’ve repeatedly invited AMD to participate in our workstation-oriented coverage, but it concedes that it’s no longer a player in this space.

In my view, this is sad and unnecessary. AMD doesn't care about workstations? That's not right. I'm guessing that maybe too many people were let go in the course of following someone's business plan and now the engineering team just isn't there. I don't know that I can consider AMD a serious player in the CPU business if they don't care about workstations. Really, that means to me they don't care about servers either. Tell me I'm wrong about that.

Perhaps Piledriver will get them back into the workstation market?

Ok, here is my Piledriver news for today -->[ ].

Right now at this moment I'm smiling that my new build is based on the i7-2600K. I do think that I should have gone with the i7-3820 but as I remember that wasn't actually on the market yet when I ordered. Don't know why anyone would not go this route especially if they wanted lots of PCIe lanes (palladin9479.)

Ok well that's it. Can we get some insider/leaked Piledriver specs or benchmarks?

"Workstations" are an incredibly small market when compared to OEM / OA Desktop (most office machines) and servers.

I'd be more concerned with their server offerings which right now is rather lacking. We'll see how things go in the future, but do not use "workstations" as a gauge for anything.

 

[viridiancrystal]

here is something to munch upon
www.fudzilla.com/home/item/26204-top-trinity-is-a10-5800k-at-38ghz

wow, 65w a10-5700 with 3.4ghz base (4ghz turbo) + 7660d

to me, cpu side (of a10 5700) seems 5-15% better than llano (a8-3870k) (considering that core of trinity is bd not e-bd which will give more performance)

Trinity's cpu cores are based off of PD cores, thus all the discussion in the PD thread.

2 things about this:
1. Specs don't lie, by which i mean: AMD cannot just say a products TDP is 65 and have it be close to 100.

2. If a Quad core bulldozer FX is faster than llano, than a higher clocked PD quad core will be faster, unless AMD found a way to go backwards on the same arch.

Whatever the case, AMD has found a way to save power, which looks good for a PD improvement.

 

[truegenius]

trinity = e-fx4 + gpu = but still 65w for whole, and thus aprox 35-50w for cpu
fx4 is 95-125w

means they managed to fit e-fx4 derivative + gpu in under 65w,

that means lot of power saving (in pd). An probabily higher clocks.

 

[Cazalan]

AMD should focus 50% on Fusion line and 50% on GCN. Leave the workstation/server CPU market. Sadly they just don't have the money to compete in that space being up against Intel, IBM and Oracle.

They have less competition in the GPGPU market with mainly just Nvidia, until Intel Knights Ferry comes online.

Trim the fat and focus on what you can do well.

 

[g4114rd0]

TSMC suddenly halts 28nm production complete stop..."

http://semiaccurate.com/2012/03/07/tsmc-suddenly-halts-28nm-production/

 

[palladin9479]

my guess is AMD hasn't, but Glo-Fo has.... :lol:

AMD bought power recovery technology recently.

http://www.tomshardware.com/news/amd-piledriver-cyclos-clock-mesh-kers,14766.html

Must remember, "heat" is just electricity that's been wasted due to ohmic heating and current leakage. If you could make a CPU out of superconducting material then it would have zero or near zero heat output and could clock into the terrahertz (until quantum physics gets in the way). Scientists and engineer use these principles when conducting experiments with plasma physics or smashing atoms.

So if you can find a way to recover the wasted electricity before it becomes heat then you can both lower temperature and reduce electrical consumption at the same time.

 

[truegenius]

like regenerative breaking in fastest car (i saw it in a show of discovery), i may install something like this in my home made car (but unfortunately currently i only have a bicycle)

 

[palladin9479]

TSMC suddenly halts 28nm production complete stop..."

http://semiaccurate.com/2012/03/07/tsmc-suddenly-halts-28nm-production/

You don't do that unless something very very bad happened, or could happen. It's too sudden to be process related, those take too long to do. Possibly safety / security / regulatory reasons?

Moderator Edit: Watch the language please

 

[fazers_on_stun]

AMD today announced it has invested in Nuvixa..."

http://www.techpowerup.com/161820/AMD-and-Nuvixa-Bring-New-Immersive-Dimension-to-Telepresence.html

Hmm, sounds interesting. My company is increasingly reliant on telecommunication tools although at the moment we are sorta stuck on WebEX. However bandwidth is something of a problem at the moment. My telecommuting equipment consists of a VOIP connection to my office phone, over the VPN connection from home and since we already have maybe 5000 telecommuters it's a strain on the office bandwidth. My ISP is Verizon FIOS and the 15/5 is more than sufficient bandwidth, but the office doesn't like any non-work related stuff like Youtube or music streaming due to the bandwidth..

 

[fazers_on_stun]

Anything about Piledriver/Trinity in the last 100 post?

Yes. You just mentioned it in your post! 😛..

 

[fazers_on_stun]

OMG...

Crap. Guess it'll be much longer now until we see any 6xx or 7xx GPUs from NV..