Fudzilla: Bulldozer performance figures are in

[bridito]

That is what marketing is here for...selling to the willfully ignorant.

Ignorant because they don't know jack about the products they are purchasing, willfully ignorant because in this day and age with internet access and forums such as Anandtech you would pretty much have to go out of your way to remain ignorant of the products you are purchasing.

In this case I would argue that being ignorant is choice, making that choice is a matter of luxury.

I am willfully ignorant of the quality of meat that most of my favorite restaurants use in their steaks and hamburgers, but I am inclined to remain blissfully ignorant and keep on thinking that the meat is top notch and not purchased out of the back of some dude's van in an alleyway.

My family is made up of a vast number of willfully ignorant computer consumers. They have the time to change that, but it simply is not a priority to them. So off to best buy they go, happy as a lark to come back with whatever some dude in a blue shirt convinced them was the "deal of the decade".

The point is that people don't always get what they pay for, but they never get what they don't pay for.

Folks like you and I prioritize spending our time arming ourselves with data such that we can make data-driven purchasing decisions to maximize the chances of getting what we pay for.

Not everyone has the same list of priorities, sometimes it is unintentional but other times it is intentional.

Either way though it is not really fair to blame the for-profit business for seeking out ways to enhance their marketshare amongst the willfully ignorant demographic.

I could not agree with you more. Having spent my entire career in media I have been a mercenary and have accepted payment for work from legitimate companies which I specifically know was inherently misleading but rationalized it with my ethics by convincing myself that I didn't generate the copy and was only placing a strategy into action which was not determined by me... and I have a baby to take care of so gimme the money. I am quite convinced that many consumers on the receiving end of those campaigns purchased on the basis of what I placed in there... and were in the strictest sense of the word: scammed.

Watch Ramsay's Kitchen Nightmares and you might eat only at home from now on!

My family as well wouldn't know a video card from a hard drive, and if we're discussing blissfully ignorant... my sister in law and husband just bought a house through my other sister in law who is a real estate agent. It was only after the purchase that they finally realized she had been hoodwinked by her own sister in law. They bought a house for $330K that had appraised for $250K. The sister-agent was in cahoots with the buyer and actually forged documents to show her! Nice lady, huh? But if the buying couple had any concept whatsoever of what a real estate transaction involved they would have been aware of the forgery immediately. But stupid is as stupid does, and stupid just blew $80K.

So if we can't trust our own sisters in law, can we really trust any computer hardware manufacturer?

 

[drizek]

Either way though it is not really fair to blame the for-profit business for seeking out ways to enhance their marketshare amongst the willfully ignorant demographic.

It is when they tell people out and out lies.

 

[podspi]

It is when they tell people out and out lies.

Yes, but that is illegal. At least in the United States.

Unless you are saying that calling a BD core a 'core' is literally fraudulent.

You could make the argument, but I wouldn't agree with it.

 

[Accord99]

Each Bulldozer module has 2x 128bit FP FMACs (Fuzed Multiply Accumulate), that meas it can do 2x 128bit ADD and 2x 128bit MUL per cycle.

So, a 4 module Bulldozer can do 8x 128bit FMACs, that is 8x 128bit ADDs and 8x 128bit MULs, that's double what the SB can do.

Only if using a FMAC instruction, which may not be common. Otherwise, a 4 module Bulldozer can do 8x128 bit instructions, which can be split between ADDs or MULs.

Excuse me but I had to come back for a re-read and I need some more clarification.

What exactly do you mean? SB can do :

2) 4x128 bit Add + 4x128 bit Multiply per cycle with SSE OR 4x256 bit Add + 4x256 bit Multiply per cycle with AVX

This is correct for QC Sandy Bridge.

 

[drizek]

No, I was referring more to the people in the blue shirts. I can almost guarantee you that one of them referred to an i7 with HT as having "8 cores" at some point in time.

When I go to Best Buy, I hear so much bullshit from the salespeople that I begin to question my own sanity.

 

[Phynaz]

I doubt people that wear uniforms make for 90% of the world population that buy PCs.

Citation?

 

[podspi]

No, I was referring more to the people in the blue shirts. I can almost guarantee you that one of them referred to an i7 with HT as having "8 cores" at some point in time.

When I go to Best Buy, I hear so much bullshit from the salespeople that I begin to question my own sanity.

Actually, most people I've spoken to with HT-enabled CPUs believe they have twice as many cores as they actually do. Probably because of those blue shirts

But again, I believe that the module approach AND HT are both legitimate technologies. Most consumers are fine with a modern dual-core, as frightening as that may sound to many of us here D:

 

[ilkhan]

No you talk to intel and they will tell you straight out . They as a producer of CPUs as has yet to release its highend SB platiform Enthusiast if you insist . Until we see AMDS best against intels best your making a sensely debate . Apples to apples based on performance not cost . Since it is in FACT the performance that sets the PRICE. I asssure you Z68 SB 1155 is not Intels best. The 2011 socket and chipset are intels best . So many say lets wait for real results and thats what I will do wait for real results against AMDs best VS Intels best 2011 socket. Or aren't the reviewers going to bench the 2011 platiform. I may be annoying but I sure as hell won't overlook the fact that the bar for AMD was lower to mid high for AMD . Annoying yes, To some but true,

Damn. I missed trying to decypher Nemesis' posts.

I'll wait for real results with real retail chips.

 

[bryanW1995]

Your "uninformed buys Intel" statement is ludicrous. You have no statistical basis for that. Back it up with a linky please.

His reasoning behind it is pretty solid though isn't it? If 80% of the computers sold in the past have an intel cpu, 80% of the computers available at B & M stores are intel, and intel spends 80% of the total cpu marketing (closer to 100% but we'll just use 80% for the sake of argument) then it's reasonable to assume that a casual pc buyer is more apt to purchase an intel rig than an amd rig.

According to interbrand, the world's largest brand consultancy, in 2010 intel had the 7th best global brand in the world with an approximate rank of $32 billion. Burberry was #100 at $3.1 billion, and AMD didn't make the list at all.

 

[Idontcare]

It is when they tell people out and out lies.

That is what lawyers and class-action lawsuits are for

 

[bryanW1995]

I could not agree with you more. Having spent my entire career in media I have been a mercenary and have accepted payment for work from legitimate companies which I specifically know was inherently misleading but rationalized it with my ethics by convincing myself that I didn't generate the copy and was only placing a strategy into action which was not determined by me... and I have a baby to take care of so gimme the money. I am quite convinced that many consumers on the receiving end of those campaigns purchased on the basis of what I placed in there... and were in the strictest sense of the word: scammed.

Watch Ramsay's Kitchen Nightmares and you might eat only at home from now on!

My family as well wouldn't know a video card from a hard drive, and if we're discussing blissfully ignorant... my sister in law and husband just bought a house through my other sister in law who is a real estate agent. It was only after the purchase that they finally realized she had been hoodwinked by her own sister in law. They bought a house for $330K that had appraised for $250K. The sister-agent was in cahoots with the buyer and actually forged documents to show her! Nice lady, huh? But if the buying couple had any concept whatsoever of what a real estate transaction involved they would have been aware of the forgery immediately. But stupid is as stupid does, and stupid just blew $80K.

So if we can't trust our own sisters in law, can we really trust any computer hardware manufacturer?

All real estate agents are in it for themselves. Their entire job is to get the buyer and seller to both agree to terms that everybody can live with. Just because the house appraised for $250k doesn't mean that it wasn't worth $330k or $500k for that matter, it just reflects the unbelievably intense govt scrutiny currently on anything to do with home mortgages. More than that, it could very likely reflect a market with a lot more distressed/repo sales recently than traditional sales. in many neighborhoods/communities, especially upper middle class areas, these sales have an enormous deleterious effect upon the appraisal. Look at it this way: no appraiser is going to be investigated by the govt on a loan that goes bad if he underestimated the home value by $50k, but if he OVERestimated by $5 then the govt might go after his license, go for jail time, bring out the tar and feathers, etc etc etc. Everybody in the real estate industry is scared shitless right now of making any sort of mistake. And do you know what the #1 deal-killer is on home purchases? The appraisal. Even with 20% down on a $330k home that appraises for $250k your sister-in-law would still be financing well over the appraised value; she's lucky that she got the loan on the house at all.

When I bought my house I knew better than to allow the realtor negotiate for me, so I did it myself. I found out after the deal was over that the seller of that home had put a $15k spiff for selling the house, but he ended up pulling it out b/c I negotiated him down so far. In my experience most realtors that specialize in private home sales are good with people and poor at getting their clients the best deal, whether as the buying or selling agent.

Sorry for the 100% OT reply, but I've spent a lot of time over the past few years interviewing/hiring former real estate agents who were driven out of business by the enormous changes that market has endured over the past few years. I'm sure that your sisters-in-law don't need my relationship advice here, but I suspect that there is more to this situation than the "taken advantage of" sister-in-law realized.


edit: by the way, I am enormously glad that this same govt scrutiny did not go after the auto industry. Auto loans are now back at a reasonable level, and most deals that make sense to a bank are getting done again. In home loans, traditionally 6 in 10 were approved. During the run-up to the recession that number spiked to more like 9 in 10, with an inordinate amount of them subprime adjustable rate arms of 5-7 years. What's the highest risk loan, prime or subprime? subprime obviously. What's the highest risk type of loan, adjustable rate or standard rate? Adjustable, especially with extremely low interest rates at the time of inception. And which type of loan is more apt to result in a default, a conventional 20 or 30 year note or a 5-7 year interest only loan? Yep, the mortgage industry inflicted much of this pain upon themselves. All of these factors created a perfect storm, so much so that now deals that traditionally would have gotten done (say a 30 year fixed rate loan, 20% down, 700 trans union) are taking months instead of hours, and often the buyer loses interest/finds another use for the money/decides to hold off/etc etc. Heck, I'm a 769 credit score with a good job, equity in my home, etc etc, and it literally took me a month and a half to complete a refi last year. My data is a few months old now, but last I heard we were at more like 3 in 10 loans getting approved now.

 

[AtenRa]

Only if using a FMAC instruction, which may not be common. Otherwise, a 4 module Bulldozer can do 8x128 bit instructions, which can be split between ADDs or MULs.

FMACs are hardware logic and not an instruction set like AVX. That means they don’t need special programming in order to be used.
So a 4 module Bulldozer can issue, execute and retire 8x 128bit FMACs anytime

So, if BD can do 8x 128bit FMACs, does that mean that it can do 4X 256bit FMACs as well??

Im not 100% sure about that, but since 256bit AVX is 2x 128bit then yes BD can do 4x 256bit FMACs.

 

[Idontcare]

FMACs are hardware logic and not an instruction set like AVX. That means they don’t need special programming in order to be used.

Not disagreeing with you, because I really don't know, but wasn't there something about FMAC's not being IEEE 754 compliant and because of this it requires the program to essentially "opt in" to using FMAC computations since the end-user may or may not be aware that the resultant output will not be IEEE 754 compliant?

 

[AtenRa]

FMA (Fuzed Multiply Add) or FMAC is part of the latest IEEE 754

http://ieeexplore.ieee.org/Xplore/l...487502.pdf?arnumber=5487502&authDecision=-203

 

[GammaLaser]

Not disagreeing with you, because I really don't know, but wasn't there something about FMAC's not being IEEE 754 compliant and because of this it requires the program to essentially "opt in" to using FMAC computations since the end-user may or may not be aware that the resultant output will not be IEEE 754 compliant?

Well, there's the two 128-bit FMAC units in a Bulldozer module, which does all the floating-point computations, and then there's the FMA3/FMA4 instructions which require recompilation of the application.

 

[Accord99]

FMACs are hardware logic and not an instruction set like AVX. That means they don’t need special programming in order to be used.
So a 4 module Bulldozer can issue, execute and retire 8x 128bit FMACs anytime

But the current x86 instruction set doesn't have a FMA instruction so there is no existing code that will do a FMA. That's why extensions to x86 are needed:

http://en.wikipedia.org/wiki/FMA_instruction_set

and also programmers/compilers must generate code that uses the FMA instructions in order to do a FMA. And then you also need an operation that requires the FMA in order to reach the max theoretical FP capabilities of Bulldozer.

 

[Idontcare]

FMA (Fuzed Multiply Add) or FMAC is part of the latest IEEE 754

http://ieeexplore.ieee.org/Xplore/l...487502.pdf?arnumber=5487502&authDecision=-203

Doh! Somewhere out there is a FUD article that had me convinced otherwise.

Oh well, that's the number one reason I come to these forums, to have my ignorance vanquished! Thanks ever so much for obliging me :thumbsup:

 

[Tuna-Fish]

FMACs are hardware logic and not an instruction set like AVX. That means they don’t need special programming in order to be used.
So a 4 module Bulldozer can issue, execute and retire 8x 128bit FMACs anytime

No. FMAC is not the same operation as multiply followed by an add, because there is no intervening rounding. The cpu cannot ditch the rounding, because it would, among other things, immediately cause all synced simulation multiplayer games (like supreme commander) to desync. There is an AMD patent related to faking the rounding in a FMAC merged by the cpu, but based on all we know, BD won't do that.

So the only places where FMAC helps is on software that has been recompiled to use the instruction. Until Intel's implementation arrives with Haswell, the only real piece of software that uses it will likely be the Radeon drivers. Sometimes it just sucks to be the underdog -- 3dnow! was also awesome at it's time.

(there is another stealth benefit to FMAC -- it means that a single thread can start two FPADDs or two FPMULs per clock, when before they could only do one of each. But the advantage of that will likely not be very common.)

 

[AtenRa]

well we all learn a new thing every day

I must say i have to read more about BDs FP unit again

 

[nonameo]

No. FMAC is not the same operation as multiply followed by an add, because there is no intervening rounding. The cpu cannot ditch the rounding, because it would, among other things, immediately cause all synced simulation multiplayer games (like supreme commander) to desync. There is an AMD patent related to faking the rounding in a FMAC merged by the cpu, but based on all we know, BD won't do that.

So the only places where FMAC helps is on software that has been recompiled to use the instruction. Until Intel's implementation arrives with Haswell, the only real piece of software that uses it will likely be the Radeon drivers. Sometimes it just sucks to be the underdog -- 3dnow! was also awesome at it's time.

(there is another stealth benefit to FMAC -- it means that a single thread can start two FPADDs or two FPMULs per clock, when before they could only do one of each. But the advantage of that will likely not be very common.)

It really is a sad catch-22. Because they lack in process tech(and other things)AMD needs special instructions to get ahead of intel. However, those instructions(generally) don't take off without implementation from intel. And who is going to do what intel does best? Intel.

But you know, intel had IBM behind them when they started, and AMD was only in the game as a backup. It really is pretty amazing what they've done despite having poor leadership at times. However, I can't really see AMD having another big swing like they had with the athlon and athlon 64. R&D is too expensive and process tech is much more important as people move towards mobility and away from computational power. Graphics is a nice stab, but TBH the only thing the extra power that the APU has is useful for is gaming, and once again, gaming is and has been either disappearing or moving towards consoles(well, I guess the latest trend is moving towards cell phones...)

AMD needs another market, something besides graphics and CPUs. I wonder if that will ever happen. ( I mean something big, I know they've had some other small things...)

 

[beginner99]

Nice example, i will use this one to show people that AMD Quad Core Phenom II 955 ($120) with a SATA-3 SSD is faster than an Intel 6 core 12 Threads $999 CPU with 1 TB 7200 rpm HDD.

I will use exactly your words,
"I think we all know which system will be perceived as being faster by a normal, clueless user for doing a little MS Word and facebook. They will start the test with booting. "



You got to be kidding right ??

A little bit yes. But you just proof my point. All this multi-core APU crap is a marketing joke. As you said only Gamers or some rare hard-core photoshop or video editing people need to care about tons of CPU and GPU power.

Video de- and encode can be done most efficient with fixed function hardware. No need for a beefy "GPU" or CPU (smartphones...). For webpage rendering: single-core IPC is what matters.
Biggest bottleneck however is IO namely HDD.

For average-Joe a high IPC single-core + SSD would be the best. What I mean is he better spend the money on an SSD rather than a better CPU or GPU.

 

[nonameo]

Nice example, i will use this one to show people that AMD Quad Core Phenom II 955 ($120) with a SATA-3 SSD is faster than an Intel 6 core 12 Threads $999 CPU with 1 TB 7200 rpm HDD.

I will use exactly your words,
"I think we all know which system will be perceived as being faster by a normal, clueless user for doing a little MS Word and facebook. They will start the test with booting. "



You got to be kidding right ??

He has a point, the problem is that he hasn't dug deep enough yet. Dual core atom/bobcat is sufficient for average users while single core of either is not. An SSD with either of them(the duals) would be "better" for the average user.

 

[podspi]

It really is a sad catch-22. Because they lack in process tech(and other things)AMD needs special instructions to get ahead of intel. However, those instructions(generally) don't take off without implementation from intel. And who is going to do what intel does best? Intel.

But you know, intel had IBM behind them when they started, and AMD was only in the game as a backup. It really is pretty amazing what they've done despite having poor leadership at times. However, I can't really see AMD having another big swing like they had with the athlon and athlon 64. R&D is too expensive and process tech is much more important as people move towards mobility and away from computational power. Graphics is a nice stab, but TBH the only thing the extra power that the APU has is useful for is gaming, and once again, gaming is and has been either disappearing or moving towards consoles(well, I guess the latest trend is moving towards cell phones...)

AMD needs another market, something besides graphics and CPUs. I wonder if that will ever happen. ( I mean something big, I know they've had some other small things...)

This is why it is so important (and good for AMD) that ARM has embraced OpenCL, and Apple has somehow managed to strongarm Intel into using OpenCL. AMD's APUs (or CPU+IGP if you prefer) are better at executing OpenCL than Intel's, and it is hard to see that changing in the near-future. This gives them quite an advantage. Intel, for their part, is going to have a hard time because

1) They want those large margins


2) They don't want to risk losing the CPU crown, and so will be hesitant to dedicate "too many" transistors to the GPU.


I expect Trinity will have more die space dedicated to the GPU than the CPU, and for most users this will be just fine.

 

[Mr Vain]

His reasoning behind it is pretty solid though isn't it? If 80% of the computers sold in the past have an intel cpu, 80% of the computers available at B & M stores are intel, and intel spends 80% of the total cpu marketing (closer to 100% but we'll just use 80% for the sake of argument) then it's reasonable to assume that a casual pc buyer is more apt to purchase an intel rig than an amd rig.

According to interbrand, the world's largest brand consultancy, in 2010 intel had the 7th best global brand in the world with an approximate rank of $32 billion. Burberry was #100 at $3.1 billion, and AMD didn't make the list at all.

Leszinske said Intel faces "more competitors than we have had in recent memory",

"there is a huge risk with a lot more competitors but we think we have a lot of capabilities that will allow us to be successful".

http://www.theinquirer.net/inquirer/news/2096921/intel-competes-qualcomm-arm

Huge risk, what a joke, poor Intel, we may need to buy more of their stuff to keep them afloat guy’s lol.

 

[Blitzvogel]

Leszinske said Intel faces "more competitors than we have had in recent memory",

"there is a huge risk with a lot more competitors but we think we have a lot of capabilities that will allow us to be successful".

http://www.theinquirer.net/inquirer/news/2096921/intel-competes-qualcomm-arm

Huge risk, what a joke, poor Intel, we may need to buy more of their stuff to keep them afloat guy’s lol.

The risk is more related to ARM than anything, though there is always the possibility of AMD or even VIA smacking Intel in the face, hence why Intel has always pushed their advertising very hard. When I saw Captain America the other night, there was an Intel ad just before the previews.

The dominance in the x86 market is there, but x86 itself is at risk in portable computing and Intel needs to make sure people know who they are so when they buy a computer, they go Intel because they've heard of them. This will be especially important when we start seeing "real" ARM based laptops.

It's too bad AMD doesn't have the advertising budget to make commercials and Youtube ads, etc. While Brazos and Ontario Fusion took off excellently thanks to good pricing (which in a way is also deceptive since most people don't understand how weak those parts are), Llano could absolutely obliterate low end Fusion. I just think most of those Llano based machines are priced too high.