AMD Bulldozer Thread

[R101]

The only problem is that it takes so much die space that it SHOULD be an 8-core.

[SKYMTL]

I find it a bit ridiculous that JF is getting all this flack.

Most of his posts (particularly those dealing with the "Bulldozer 20 questions" blogs) were factual and really didn't over hype the architecture or the performance thereof. You guys are feeding him to the wolves before the HPC processor he talked about (Interlagos) has even been extensively benchmarked. Granted, there were a few posts here and there that could be picked apart by detractors but for the most part, the information he brought to the table was often insightful and lent some sanity to discussions populated with people who didn't know what they were talking about. Most forums would kill to have someone like him around and yet here he gets crucified.

Even more to the point, you guys seem to have forgotten that some of XS's premier members (and supposedly those who should have been "in the know") and certain websites were the ones who propagated the vast majority of erroneous information to members. To be honest, most of the extreme amount of hype came from people who were doing nothing but musing about what Bulldozer COULD be. I won't point fingers here but if you're butthurt about being misled, this witch hunt is going after the wrong individual.

[ajaidev]

It's not quite an 8 core. This is more or less AMD's stab at hyper-threading. It's essentially 4 full cores that can handle 8 threads. A few review sites touched on that fact, but most ignored it.

No this is not like HT at all, this has more benefits than HT SMT but it also takes up more space. CMT will also be adopted by Intel in the future and with a beefed up shared areas dont expect performance to suffer. From what i have been told a further HT SMT implementation on the CMT is possible and the added sizer would only be around 2-4% of the twin core CMT and presto you get 4 threads out of 2 CMT cores.

This AMD way is the future too bad AMD is super bad in execution of the idea. Phenom had a wonderful arc. and so does BD but both failed where it mattered the most not to mention their power consumption figures...

[eRazorzEDGE]

Saying "AMD is super bad in execution of the idea" about the design of Bulldozer isn't really true. I believe they're anticipating trends in the computing requirements, obviously focusing more on server side than client side. Bulldozer can work well if the software is there to properly feed X amount of threads through Y amount of compute units.

It's really the same thing as multiple cores and the software side adoption of multi-threaded apps. It will take time for a new architectural change to show improvement if the software is the same old same old off the self code. You can't make changes to increase a programs performance if the hardware behind it doesn't know how to handle it, and hence an architectural change is required first.

Take a look at SuperPi... single threaded program. Move now to (can't think of names at the moment) multi threaded Pi programs. The multi-threaded programs can achieve higher performance, but until the first multi-core CPU, the performance will stay the same. Same with BD and future iterations of it.

[highoctane]

Bulldozer can work well if the software is there to properly feed X amount of threads through Y amount of compute units.

x264 should have been the application BD totally dominated it is merely enough, we're talking 8 cores vs 4 cores 8 threads for intel.

Needless to say a vast majority of folks don't use any software that requires 8 threads or cores at any one time so it's not heading in a direction that will improve the user experience at all if overall performance is sacrificed to gain those extra cores or threads.

At any rate I think it becomes more clear as to why there has been such a big management shakeup at AMD over the past year.

[ajaidev]

Saying "AMD is super bad in execution of the idea" about the design of Bulldozer isn't really true. I believe they're anticipating trends in the computing requirements, obviously focusing more on server side than client side. Bulldozer can work well if the software is there to properly feed X amount of threads through Y amount of compute units.

It's really the same thing as multiple cores and the software side adoption of multi-threaded apps. It will take time for a new architectural change to show improvement if the software is the same old same old off the self code. You can't make changes to increase a programs performance if the hardware behind it doesn't know how to handle it, and hence an architectural change is required first.

Take a look at SuperPi... single threaded program. Move now to (can't think of names at the moment) multi threaded Pi programs. The multi-threaded programs can achieve higher performance, but until the first multi-core CPU, the performance will stay the same. Same with BD and future iterations of it.

Well Intel also focuses on server market with most new architectures this includes SNB or Haswell. Ya i know SNB-E is for the heavy work but to get SNB-E you need a base that you can be worked on. Whenever a new design is proposed parts of the chip is simulated this is called virtual chip emulation. Single thread performance was seen reduced in preliminary emulations for Nehalem arc the way out was to get a system like Intel Turbo Boost implemented. AMD 's turbo is not as effective as Intel's not to mention that since the Bulldozer has pretty much linear performance a proper turbo implementation on a module could have lead to a very substantial gain.

Now for the multi thread thing yes BD is good for multi threads but its not good in power efficiency. In fact when loaded it eats up the same amount of power as a QX6xxx a quad core made in 65nm.


EDIT: Forgot to include that i also think that BD will be good in servers because of the parallel nature of the arc. but there is a possibility of resource scarcity.

[Dainas]

Netburst got off to a rougher start and still improved, but I'm a gamer not a bench-marker and that architecture was always a turd to me till its final literation. I think BD will end upa little bit better; like K10, underwhelming till finally okay to excellent value, its unfortunately not going to shake intels tree in the high end as I was hoping. I just hope all those who threw a hissy fit over the blatant writing on the wall take a piece of humble pie and shut up next time someone is less than impressed by team red.

[eRazorzEDGE]

x264 should have been the application BD totally dominated it is merely enough, we're talking 8 cores vs 4 cores 8 threads for intel.

Needless to say a vast majority of folks don't use any software that requires 8 threads or cores at any one time so it's not heading in a direction that will improve the user experience at all if overall performance is sacrificed to gain those extra cores or threads.

At any rate I think it becomes more clear as to why there has been such a big management shakeup at AMD over the past year.

But that's just it, all processors, even ARM, are heading towards more cores. Just because software hasn't caught up to multi-core processors doesn't mean that CPU engineers should go back to making single core chips. The engineers have decided that more cores at a certain clock is faster and more efficient to engineer, manufacture, and power/operate than a single core at a really high clock.

Also, I believe the FX-8150 is a four core processor. They may tout it as an 8-core, but I believe it's really supposed to be a physical solution to Intel's logical hyper-threading.

[nn_step]

But that's just it, all processors, even ARM, are heading towards more cores. Just because software hasn't caught up to multi-core processors doesn't mean that CPU engineers should go back to making single core chips. The engineers have decided that more cores at a certain clock is faster and more efficient to engineer, manufacture, and power/operate than a single core at a really high clock.

Also, I believe the FX-8150 is a four core processor. They may tout it as an 8-core, but I believe it's really supposed to be a physical solution to Intel's logical hyper-threading.

There is exactly one major reason why we are moving to multiple cores:

One key problem is that the complex multiple-issue dispatch logic scales somewhere between quadratically and exponentially with the issue-width. That is, the dispatch logic of a 5-issue processor is almost twice as big as a 4-issue design, with 6-issue being 4 times as big, 7-issue 8 times and so on.

Humans have significant problems when it comes to reasoning about parallel execution and thus depend more and more on language level abstractions. Then comes the inevitable source of the problem :

Humans generally are single taskers aka we only play one game at a time or work on a single document at a time.

Thus anything depending on human input is largely limited by the serial nature of the human at the keyboard.

Now if you go into servers or scientific calculations, then the limit is not the human but rather the number of GFLOPs the hardware can crank out in a second.

Thus for the desktop the goal is not more cores and higher performance but rather more integration and lower cost and lower power.

For servers and scientific calculations the goal is more cores, faster cores and lower power.

[Sparky]

There is exactly one major reason why we are moving to multiple cores:

One key problem is that the complex multiple-issue dispatch logic scales somewhere between quadratically and exponentially with the issue-width. That is, the dispatch logic of a 5-issue processor is almost twice as big as a 4-issue design, with 6-issue being 4 times as big, 7-issue 8 times and so on.

Humans have significant problems when it comes to reasoning about parallel execution and thus depend more and more on language level abstractions. Then comes the inevitable source of the problem :

Humans generally are single taskers aka we only play one game at a time or work on a single document at a time.

Thus anything depending on human input is largely limited by the serial nature of the human at the keyboard.

Now if you go into servers or scientific calculations, then the limit is not the human but rather the number of GFLOPs the hardware can crank out in a second.

Thus for the desktop the goal is not more cores and higher performance but rather more integration and lower cost and lower power.

For servers and scientific calculations the goal is more cores, faster cores and lower power.

... and what I got from this is I need to clone myself, so I can double use my PC

[NovoRei]

... and what I got from this is I need to clone myself, so I can double use my IPC

IPC/IPS/CPI etc.

[jazkat]

I find it a bit ridiculous that JF is getting all this flack.

Most of his posts (particularly those dealing with the "Bulldozer 20 questions" blogs) were factual and really didn't over hype the architecture or the performance thereof. You guys are feeding him to the wolves before the HPC processor he talked about (Interlagos) has even been extensively benchmarked. Granted, there were a few posts here and there that could be picked apart by detractors but for the most part, the information he brought to the table was often insightful and lent some sanity to discussions populated with people who didn't know what they were talking about. Most forums would kill to have someone like him around and yet here he gets crucified.

Even more to the point, you guys seem to have forgotten that some of XS's premier members (and supposedly those who should have been "in the know") and certain websites were the ones who propagated the vast majority of erroneous information to members. To be honest, most of the extreme amount of hype came from people who were doing nothing but musing about what Bulldozer COULD be. I won't point fingers here but if you're butthurt about being misled, this witch hunt is going after the wrong individual.

i 1000000000% agree with you my friend, you are spot on.

some of the top guys here on this forum who most look upto were the ones saying BD was a win, but they have had no grilling and are invisible even tho they gave incorrect info, JF gets it all and i think its unfair.

there followers wont admit it anyway.

[duron]

BSOD?
http://scalibq.wordpress.com/2011/10...et-even-worse/

[Mechromancer]

BSOD?
http://scalibq.wordpress.com/2011/10...et-even-worse/

Dang...

[mongoled]

Dang...

Could be a good thing, force MS hand to issue a patch for Bulldozer!

[R101]

I think there is a good chance it is fixable with Bios update..

First Phenom I had also BSOD-ed randomly from time to time.

[Olivon]

AMD Appoints Mark Papermaster as Senior Vice President and Chief Technology Officer - AMD

SUNNYVALE, Calif. —10/19/2011
AMD (NYSE: AMD) announced today that Mark Papermaster, 50, has joined as the company’s senior vice president and chief technology officer. He will report to President and Chief Executive Officer Rory Read and will oversee all of AMD’s engineering, research and development (R&D), and product development functions as the head of the newly-formed Technology and Engineering Group. Papermaster, who was most recently vice president of Silicon Engineering at Cisco, will be responsible for establishing and executing the company’s technology and product roadmaps, integrated hardware and software development, and overseeing the creation of all of AMD’s products.

[haylui]

AMD Appoints Mark Papermaster as Senior Vice President and Chief Technology Officer - AMD

could AMD possibly have "antenna gate" ?

[gallag]

BSOD?
http://scalibq.wordpress.com/2011/10...et-even-worse/

If true, this would cement bd as the worst cpu ever, i thought intels p4 would hold that title for a long time.

[ilkkahy]

Maybe amd should take panic plan-B into action and start making 32nm Phenoms instead. They have had plenty of time with Bulldozer but it simply sucks so bad in practice and i dont think there will be any miraclous improvements with Piledriver.

[Alex-Ro]

Yeap,Bulldozer is the worst fail of this decade so far.Never ever in history a company would release a model that's worse than it's predecessor.

[Lanek]

I find it a bit ridiculous that JF is getting all this flack.

Most of his posts (particularly those dealing with the "Bulldozer 20 questions" blogs) were factual and really didn't over hype the architecture or the performance thereof. You guys are feeding him to the wolves before the HPC processor he talked about (Interlagos) has even been extensively benchmarked. Granted, there were a few posts here and there that could be picked apart by detractors but for the most part, the information he brought to the table was often insightful and lent some sanity to discussions populated with people who didn't know what they were talking about. Most forums would kill to have someone like him around and yet here he gets crucified.

Even more to the point, you guys seem to have forgotten that some of XS's premier members (and supposedly those who should have been "in the know") and certain websites were the ones who propagated the vast majority of erroneous information to members. To be honest, most of the extreme amount of hype came from people who were doing nothing but musing about what Bulldozer COULD be. I won't point fingers here but if you're butthurt about being misled, this witch hunt is going after the wrong individual.

really thanks for this words sky .............

but i can imagine this is internet and sadly we all know how it work.

[tajoh111]

Maybe amd should take panic plan-B into action and start making 32nm Phenoms instead. They have had plenty of time with Bulldozer but it simply sucks so bad in practice and i dont think there will be any miraclous improvements with Piledriver.

problem is AMD is too small and their branding is not strong enough to make money off of bad processors. They need bulldozer to succeed because it takes too long for them to pump out a new architecture. Staying with phenom would seal their fate because the architecture has reached a dead end. Adding more cores is the only way AMD seems to be adding performance lately and doing this with phenom will add very little return in performance. Adding 2 more cores to the t1xxx series, to make octocores will not get them very far.

What AMD will need to do is hang onto their niche netbook/ ultra budget laptop line to survive(a problem with this market is it appears to be eroding due to tablets), unless BD is fixable. BD has to succeed if they want to be for longer than 4 years. AMD is in for some pain for the short term though and their stock price reflects this.

I have a feeling performance won't improve that much with Piledriver either. Even if it does require special coding to succeed, AMD does not enough pull to make companies code for it enough to sway the market.

[Olivon]

AMD apologizes about wrong 2500K/980X R11.5 score - AMD Official

Recently AMD posted a video comparing the performance of the AMD FX processor to both the Intel Core i5 and Core i7 processors. Unfortunately, during the production of this video, the graphics were not correctly labeled. Some of you noticed this error and pointed it out to us. As a result, we immediately took down the video, corrected the mislabeled graphics, and we’ve reposted the updated video

[Maxforces]

Unfortunately, during the production of this video, the graphics were not correctly labeled.