AMD Processors
Decrease font size
Increase font size
Topic Title: Bulldozer Postmortem
Topic Summary:
Created On: 06/01/2012 01:56 PM
Status: Read Only
Linear : Threading : Single : Branch
Search Topic Search Topic
Topic Tools Topic Tools
View similar topics View similar topics
View topic in raw text format. Print this topic.
 06/01/2012 01:56 PM
User is offline View Users Profile Print this message

Author Icon
Mime
Troll Hunter

Posts: 8517
Joined: 10/06/2003

Johan De Gelas has a look at what went wrong with Bulldozer, and why the newer Piledriver cores in Trinity didn't seem to help much. There may be light at the end of this tunnel after all, but we'll have to wait a while to see it.

http://www.anandtech.com/show/...h-delving-even-deeper/

-------------------------
Containment Breach

Do not meddle in the affairs of archers, for they are subtle and quick to anger.
 06/02/2012 06:36 PM
User is offline View Users Profile Print this message

Author Icon
technic58
Minister of Spare Parts

Posts: 13953
Joined: 07/25/2004

Gotta love waiting!

-------------------------
My GT430 Is Folding For AMD Team 34106
FOR A GOOD CAUSE---Accepting Donations of New, Used, Working, Nonworking PC Components
Thanks to KLEETUS for his generous donation of components!!
 06/03/2012 12:30 PM
User is offline View Users Profile Print this message

Author Icon
Immortal Lobster
Resident Crustacean

Posts: 12033
Joined: 01/26/2005

Ive been waiting.... Primarily because of financial reasosn... but patiently waiting

-------------------------

The opinions expressed above do not represent those of Advanced Micro Devices or any of their affiliates.
 06/04/2012 05:23 AM
User is offline View Users Profile Print this message

Author Icon
Xajel
Case Modder

Posts: 1554
Joined: 10/08/2003

I've been waiting, because I love AIO but I don't see them that much Enthusiastic, eg.. no customization or at least no better option if you need high-end GPU !!... but I love them specially the newely uncovered ASUS Transofrmer AIO, i'm just in love with it !! perfect as a living room system and as a home tablet...

But I love High-end systems, I love picking my parts and build it my own !! I love high-end GPU's and CPU's !!

That's why I'm waiting...
 06/12/2012 07:17 AM
User is offline View Users Profile Print this message

Author Icon
jeepsteer
Newbie

Posts: 27
Joined: 07/24/2009

I think we will be waiting a long time.
Intel made a huge boost in IPC from Pentium 4 to Core 2, then got another boost
when ditching FSB in Nehalems. But now Intel are only making small improvements
e.g. Ivy Bridge. So if Intel struggles to score a few percent faster benchmarks, and AMD
can go 10% faster, AMD will catch up in a decade?
 06/12/2012 10:35 AM
User is offline View Users Profile Print this message

Author Icon
Canis-X
The Frozen One

Posts: 4142
Joined: 01/19/2009

Well, I just got tired of waiting! For me what was really sad with the BullDozer release was the fact the my 1090T could still outperform the FX8150 in just about every aspect, coupled with the fact that with my phase change I could run my 1090T at 4.6Ghz (24/7 stable) there was absolutely zero reason for me to move to the FX lineup. Really sad!!!! So, I got tired of my 1090T bottlenecking my GTX590's and moved on to Intel. I really hated doing it, because I really don't like Intel and I really REALLY didn't want to give them any of my money. I really hope that AMD will be able to pull a rabbit out of their hat some day and I will be able to move back over to them again.

-------------------------
The opinions expressed above do not represent those of Advanced Micro Devices or any of their affiliates.
 06/12/2012 01:10 PM
User is offline View Users Profile Print this message

Author Icon
Mime
Troll Hunter

Posts: 8517
Joined: 10/06/2003

If they can increase Bulldozer's clock speed, then that would help a lot with its relatively weak single threaded performance. These days it doesn't seem like AMD is as committed to the performance crown for desktops in general though, so... wait and see I guess.

-------------------------
Containment Breach

Do not meddle in the affairs of archers, for they are subtle and quick to anger.
 06/12/2012 03:40 PM
User is offline View Users Profile Print this message

Author Icon
Canis-X
The Frozen One

Posts: 4142
Joined: 01/19/2009

True, but with that increased clock speed comes a lot more volts and excess heat. Refining the IPC, IMHO, is the only solution to AMD/BD's issues.....once that is accomplished, then they can work on the scheduler fix with the major OS developers to help with the heavy multi-threaded stuff.

-------------------------
The opinions expressed above do not represent those of Advanced Micro Devices or any of their affiliates.
 06/16/2012 02:22 AM
User is offline View Users Profile Print this message

Author Icon
MU_Engineer
Dr. Mu

Posts: 1837
Joined: 08/26/2006

Originally posted by: Canis-X

True, but with that increased clock speed comes a lot more volts and excess heat. Refining the IPC, IMHO, is the only solution to AMD/BD's issues.....once that is accomplished, then they can work on the scheduler fix with the major OS developers to help with the heavy multi-threaded stuff.


Bulldozer is actually pretty decent. The biggest issues with it is that Windows benchmarks don't like it very much and that the desktop units actually draw their full TDP using Turbo CORE. The heat production is even worse on Intel CPUs as they actually exceed their TDPs until they overheat and then Turbo Boost dials things back. The only post-P4 high-end Intel CPUs that actually ran somewhat cool were the Core 2 Duos. The C2Qs ran hot, Nehalem ran really hot, and anybody who's turned the screws to an SB or IB knows they get hot too- especially IB. With the exception of IB, Intel chips got a pass and only AMD got nailed for making hot chips.

The "Bulldozer is slow" bit is also not entirely true. Many Windows benchmarks are compiled with Intel's ICC compiler which runs poorly-optimized code on AMD CPUs...despite Bulldozer supporting every SIMD extension Intel's current CPUs do. Windows 7 itself is poorly optimized for thread scheduling on Bulldozer's modular architecture. It's essentially the same kinds of issues that Windows had with HyperThreading when it first came out, and the issues they had with the Opterons' NUMA they had for a LONG time. Linux however treats the Bulldozers a lot better as recent Linux kernels schedule for Bulldozer decently well and GCC 4.6 and later have specific Bulldozer optimizations that can speed up execution, particularly with GCC 4.7.0 and later. My current single Opteron 6234 is pretty much on par with my previous *two* Opteron 6128s in multithreaded performance and about 40% ahead in single-threaded tasks. It also idles considerably cooler (thanks to core gating/CC6) and runs maybe 2-3 degrees warmer under full load with similar fan settings. So the "Bulldozer is junk" is bunk.

-------------------------
 06/16/2012 02:45 AM
User is offline View Users Profile Print this message

Author Icon
Canis-X
The Frozen One

Posts: 4142
Joined: 01/19/2009

Well, I'm not all that sure what to say tbh. For the amount of time that it took to get BD out on the market it would have been a good idea for AMD to have worked more with the major OS developers to get the environments a little more friendly to the new arch. Doesn't make much sense to release something new to the consumer space with no prep work in the consumer space. Win8 doesn't seem to handle BD that much better, granted the benchmarks were probably earmarked for Intel's word. Sill, you can't really overlook the power draw of a OC'd 8150 at load.....WOW (not saying that my 3930k is any better, because I know it's not, but it handles my 4 GPU's much better). I hope that in the near future AMD can get some developers to write code that is more amicable to AMD or even neutral would be nice.

I do know one thing though......from reading your posts on here....I know very very little compared to you, so I respectfully bow out of this conversation before I make myself look completely stupid....LOL. Cheers MU!!

-------------------------
The opinions expressed above do not represent those of Advanced Micro Devices or any of their affiliates.
 06/17/2012 12:29 AM
User is offline View Users Profile Print this message

Author Icon
Mime
Troll Hunter

Posts: 8517
Joined: 10/06/2003

Yes, it's not all bad news especially when you look at server benchmarks as well. Those were much more a mixed bag than the initial desktop reviews which were almost entirely disappointing. If the chip can do better then that's great, but most people will continue to view the chip as a disappointment until it actually does do better. ICC was borking benchmarks long before Bulldozer made its appearance, so(just like everything else) that's only part of the story. If AMD wants to ride this deal out and not cave under pressure, then that's great also, but a chip that doesn't do so well in the current build of windows can't really be called a success either.

-------------------------
Containment Breach

Do not meddle in the affairs of archers, for they are subtle and quick to anger.

Edited: 06/17/2012 at 12:45 AM by Mime
 06/17/2012 02:05 AM
User is offline View Users Profile Print this message

Author Icon
MU_Engineer
Dr. Mu

Posts: 1837
Joined: 08/26/2006

Originally posted by: Canis-X

Well, I'm not all that sure what to say tbh. For the amount of time that it took to get BD out on the market it would have been a good idea for AMD to have worked more with the major OS developers to get the environments a little more friendly to the new arch.


Bulldozer was designed as server chip. Linux is the major server OS and it has good OS and compiler support. Windows has a much slower release cycle that Linux does- a new OS version comes out every 3-5 years instead of about every 6 months for many Linux distributors. Windows also has a pretty poor track record of supporting newer technologies well, hence my comments about HyperThreading and NUMA. That is one of the many reasons Windows is a minor player in the server arena.

Doesn't make much sense to release something new to the consumer space with no prep work in the consumer space.


The real consumer chips are the APUs and AMD sells them hand over fist. Even AMD's own guys concede that the enthusiast market is tiny and the costs to make a CPU that can beat Intel's high-end CPUs in current Windows Intel-optimized benchmarks is far too high. Releasing a server chip as a desktop chip is absolutely the right move. What is AMD going to do, spend billions in R&D and manufacturing for a chip that still won't likely decisively beat Intel's latest and greatest, and then get only a handful millions of dollars in revenue? The FXes perform well enough for gaming. For workstation tasks, if the FXes aren't fast enough, you can get yourselves some Opterons that will embarass i7s for the same cost. Selling server chips as desktop enthusiast chips has been going on for a long time- remember the first Athlon 64 FX? It even used server registered memory for crying out loud. Agena (Barcelona) was a server chip, Thuban (Istanbul) was a server chip, as was the first Pentium 4 "Emergency Edition," the Core 2 Extreme QX9775 (that even came in server LGA771 packaging), and all of the Core i-series high-end chips.

Win8 doesn't seem to handle BD that much better, granted the benchmarks were probably earmarked for Intel's word.


So, running a chip that the benchmarks are known to not be optimized for on an OS with a long history of poor support for new CPU technologies yields poor peformance. Whoda thunk?

Sill, you can't really overlook the power draw of a OC'd 8150 at load.....WOW (not saying that my 3930k is any better, because I know it's not, but it handles my 4 GPU's much better). I hope that in the near future AMD can get some developers to write code that is more amicable to AMD or even neutral would be nice.


You run four GPUs yet complain about the CPU drawing a lot of power? Id' hazard a guess that your four GPUs probably draw three times the power as a highly overclocked FX-8150. The CPU's power consumption is a drop in the bucket in that setup. And about the temps/power draw you even admit that overclocking an i7-3930K draws just as much power. Yes, Bulldozer does draw a bit of power, especially at high clocks, and AMD is working on that. No word if Intel is working on bringing IB's high power draw at high clocks down.

The whole code neutrality thing was supposed to have been fixed with the antitrust lawsuit a couple of years ago. What Intel did was make the "disable all optimizations every time there is a non-GenuineIntel CPU present" bit disable-able. It is still on by default and people much more familiar with ICC than I am say that it is rarely disabled.

-------------------------
 06/18/2012 06:22 PM
User is offline View Users Profile Print this message

Author Icon
Mime
Troll Hunter

Posts: 8517
Joined: 10/06/2003

I can understand the reaction to Bulldozer being average, adequate, tolerable, or whatever lukewarm adjective might actually describe it. I agree it's the right move not to chase after Intel for the absolute fastest desktop chip available, but being the best at being ordinary(decent, suitable, sufficient, or passable) isn't really a cause for praise. It's like how an interior decorator can describe a gratified, moldy, broken-down, rat infested, former crackhouse as having "potential". Bulldozer is a chip that might perform well as a desktop processor at some point in the future once software learns what to do with it... yay?

But to some degree that almost always happens. It happened more this time because there's more new stuff in Bulldozer. Whatever happens in software, there are some things to do still in hardware which affect both desktop and server performance. That's the kind of stuff I was interested in because even the server benchmarks weren't entirely positive and ranged from a verdict of disaster to BEST CHIP EVAR!. Something wasn't right...

-------------------------
Containment Breach

Do not meddle in the affairs of archers, for they are subtle and quick to anger.
 06/19/2012 06:30 AM
User is offline View Users Profile Print this message

Author Icon
FrozenSteel
Ninja Zombie Killer

Posts: 6365
Joined: 06/11/2004

Just to throw the FX-8150 a bone, it does extremely well running the passmark benchmark against Intel's SB chips, surpassing the 2500k in its benchmarks and almost beating the 2600k.

-------------------------
Alienware m17x R4
17.3" 1920x1080 -- Intel Core i7-3820QM -- 16GB DDR3 1600MHz -- Intel 520 180GB SSD -- 750GB 7200RPM Data HDD -- Slot BD-ROM -- 2GB AMD 7970m -- Intel Centrino 6250+WiMax -- Mediacom 20 Mbps Internet Connection
 07/07/2012 06:56 AM
User is offline View Users Profile Print this message

Author Icon
FG Computers
Lurker

Posts: 13
Joined: 08/09/2011

The issue with BD is not that it's "slow" (it might be in some tests but not overall) it's inconsistent. Ie does very well in some heavy threaded applications, then does quite poorly in lighter threaded workloads. The previous AMD processors are more consistent with this overall. That is why I think many held off of updating to FX. I am AM3+ ready to roll but still with a PhII processor because I'm holding out for piledriver and improved performance/power consumption too.

I think the disappointment is down to the long time BD was in the works as such..people were expecting to be blown away. BS isn't a complete disaster but AMD must get it right with piledriver.
Statistics
112018 users are registered to the AMD Processors forum.
There are currently 0 users logged in.

FuseTalk Hosting Executive Plan v3.2 - © 1999-2014 FuseTalk Inc. All rights reserved.



Contact AMD Terms and Conditions ©2007 Advanced Micro Devices, Inc. Privacy Trademark information