Why Bulldozer's spotty performance is good news.

[twilyth]

This is really pretty obvious if you think about it. What was BD designed to do? Run 8 threads in real time doing integer calculations. Which benchmarks does it perform best at? Multi-threaded integer applications. What types of work loads do servers mostly deal with? Yup . . . you guessed it.

www.pcworld.com/businesscenter/article/241961/amds_bulldozer_disappoints_why_thats_good_news.html

AMD's latest-and-greatest chip may lag slightly behind Intel’s competing Core i5, as initial PCWorld performance-testing indicates. But these disappointing results hide benefits that AMD's "Bulldozer" FX CPU will likely offer, especially for cost-conscious small businesses.

AMD BulldozerThe issue is that most CPU-performance tests don't reflect the potential computational power offered by FX, which has up to eight cores, depending on the version. Sure, computationally-wise, preliminary synthetic tests, such as PCMark 7 and Cinebench, reflect real-world computing performance and indicate that the FX lags in comparison with Intel’s Core i5. That's what PCWorld's tests showed after running the four-core FX-4100 through the paces.

Why can’t the FX’ multi-core design crank past Intel’s Core i5 in these tests? Most of these tests are largely geared for CPUs with two or fewer cores. Software makers also have yet to bring to market applications that will take advantage of FX multi-core design for multi-threading tasks.

The server equivalent of the FX, code-named "Interlagos"--meant to launch in a few weeks--already takes advantage of the eight cores to a greater extent than the desktop equivalent of the FX does, AMD says.

“AMD FX and Bulldozer CPU technology was optimized for multi-processing and multi-threaded applications,” Dina McKinney, corporate vice president, design engineering, for AMD said via email.

The eight cores also benefit from AMD’s Turbo Core feature, which automatically boosts the clock speed of different cores when others are not in use above and beyond their normal speeds. When Turbo Core kicks in, the standard clock speed of the FX-8150, the highest-end version of the FX, can speed from 3.6GHz to 3.6GHz.

Turbo Core also does this while monitoring power consumption and will lower the processing speed if overheating occurs (Intel’s Turbo Boost has a similar functionality).

So in the future, look out for potential video editing, engineering, and other software that might harness what eight cores and Turbo Boost can offer both in the desktop space. While it is has yet to be proven, the FX with its eight cores could very well be ahead of its time.

More at link.

edit: I guess this is just the sort of rationalization you would get from a fanboy, so maybe there's no point in denying it. It does make sense to me but then the chick you pick up at last call never really looks all that bad either.

 

[HTC]

speed from 3.6GHz to 3.6GHz.

There's a typo there, but it seems it's from the source.

Interesting.

Bulldozer and Interlagos are supposed to be the same processor but different targets: the 1st being desktop and the latter being server, correct? This being the case, if Interlagos turns out to be very good, what will this mean for Bulldozer?

 

[twilyth]

Probably not much I'm afraid. People who are knowledgeable will look at benchmarks that reflect how they use their computers. So until more software is intensively multithreaded, I don't think most users are going to see any reason to pick BD.

However as the article points out, software will continue to move in that direction, so if you plan to hold on to a rig for 3 or 4 years it might be a consideration. Right now though, the main draw will price/performance which is about on a par with the 2500.

 

[lilhasselhoffer]

It can go from 3.6 GHz to 3.6 GHz. Wow, that's a revelation...


Joking aside, duh. Our processor designed for multi-threaded applications runs multi-threaded applications well. Give that marketting chimp a cookie.

If I buy a consumer chip it should perform well in a consumer environment. The converse is true in a server/work station environment.


The failure isn't the architecture, it's the chimps who didn't think about the target before releasing the project............... All of this seems convenient, given the management shake-up and damage control mode at the top of the company.

 

[hat]

But BD isn't a server chip, it's a desktop chip. AMD has had server chips with loads of cores before BD, like the "Magny-Cours" chip. I think this is a moot point for the majority of users, since most software still isn't highly compatible with multithreading yet.

 

[Lionheart]

I feel like AMD bulldozer has hidden potential, I guess we just have to wait and see

 

[twilyth]

It can go from 3.6 GHz to 3.6 GHz. Wow, that's a revelation...


Joking aside, duh. Our processor designed for multi-threaded applications runs multi-threaded applications well. Give that marketting chimp a cookie.

If I buy a consumer chip it should perform well in a consumer environment. The converse is true in a server/work station environment.


The failure isn't the architecture, it's the chimps who didn't think about the target before releasing the project............... All of this seems convenient, given the management shake-up and damage control mode at the top of the company.

Yup. I'm sure that's part of it too. Plus I read someplace that BD was really supposed to have launched a couple of years ago. If it had, that would have been something special. Now it's like buying a 2008 car and trying to pretend it's the latest and greatest.

 

[hat]

Sure, pretty much any 2011 tech just out the door would have been something special back in 2008.

 

[Goodman]

I couldn't give a rat ass about future , would you still use A64 single core on socket 754 or 939 (future ready back then) because now we can all use "good" software & Windows Vista/7 64bit?

Sorry! but what comes out now is got to be good to use right now , not in 2 years when it will get close to be obsolete or replace by something better...

Beside Interlagos is not doing much better then Opteron base CPU's as far as i heard
Anyhow no point of denied BD is 70% fail , long live PII

 

[lilhasselhoffer]

Sure, pretty much any 2011 tech just out the door would have been something special back in 2008.

Development started in 2006, with targetted release of 2008. Delays, fighting Intel, and money crunches crippled AMD.

Now in 2011 the product is released. Pit a C2Q against bulldozer, and you've got a hell of a competition. The threading issue is still there, but it would have been a level playing field. SB eats C2, and you can understand where that leaves BD.

I really wanted BD to live up to what AMD said, but it doesn't. What it is can be described in one word, disappointment.

 

[Inceptor]

Well... pit a C2Q against an AM3 Phenom II x4, and you've got a hell of a competition, in fact, the Phenom comes out on top.
Overclock an AM3 Phenom II x4, and it can match a first gen i5 in some benchmarks, not all, just a few.

So... pit a BD against a first gen quad core i7, and you've got a hell of a competition, in fact, the BD comes out on top or equal in many benchmarks.
Overclock a BD, and it can match or overtake a second gen i5 in many benchmarks.

Objectively, the only problems it has are its power draw when overclocked, and the instruction scheduling errors from the OS side.

Are you guys seeing a pattern here? I am.
It may be a public relations and marketing fiasco with Enthusiasts, but it seems they're still just as far from Intel performance as they were in the past.
This is obviously the fall-back position. They were gunning for earlier release, but something happened, whatever it was doesn't matter, so they had to delay, but, at least in hindsight, they could afford to delay and still release the thing and not be any further away in performance from Intel than usual.

I don't know... that's just my guess as to what they were thinking.

 

[theeldest]

I feel like AMD bulldozer has hidden potential, I guess we just have to wait and see

Agreed. I'd actually like to see how this works in a VMware ESXi cluster. I've seen in a couple reviews where BD got decent performance increases from Windows 8 because it has better thread management.

ESXi is all about resource scheduling in massively multithreaded environments. Historically Intel has won the VMware benchmarks but I think BD can compete here.

 

[Altered]

I must be confused as hell. Ive read on this site it needs Windows 8 to get full potiential.
I also read it was to be released at least a year ago.

Had the chip been released a yr ago we would have had to wait 2 years to see the chip run as it should? Just what was a guy to think?

Well... pit a C2Q against an AM3 Phenom II x4, and you've got a hell of a competition, in fact, the Phenom comes out on top.
Overclock an AM3 Phenom II x4, and it can match a first gen i5 in some benchmarks, not all, just a few.

So... pit a BD against a first gen quad core i7, and you've got a hell of a competition, in fact, the BD comes out on top or equal in many benchmarks.
Overclock a BD, and it can match or overtake a second gen i5 in many benchmarks.

And then I read this and the best I remember the Core 2 Quad beat up the Phenom II x4 up pretty good unless your only counting gaming, where most of these quad-core chips perform so similarly that it isn’t worth factoring in the tenths of a frame.
As far as BD against an i7. AMD’s fastest offering isn’t able to match i7 chip from what I have seen.

I really wanted to see this BD take the crown but I am either missing something or "spotty performance" is not good news.

 

[Inceptor]

If you read carefully, you'll see I said AM3 Phenom II x4. Not the older AM2+ models and not any x3 or x2 models.
And I said first generation quad core i7, not current second generation i7, not first generation Gulftown i7.

:shadedshu

 

[John Doe]

If you read carefully, you'll see I said AM3 Phenom II x4. Not the older AM2+ models and not any x3 or x2 models.
And I said first generation quad core i7, not current second generation i7, not first generation Gulftown i7.

:shadedshu

Regardless, your original statement is highly unlikely to be true. Besides, pretty much the only difference between AM3 and the AM2+ Phenom 2 was the DDR3 controller. As you can see, Deneb has worse per-clock performance than Yorkfield:

http://www.hardwarecanucks.com/charts/index.php?pid=61,76&tid=3

With that aside, Deneb OCs worse than Yorkfield. So what AMD did was to target it towards budget unlike Intel who milked Core 2 Quads. So from here, you can not expect BD to perform like a Nehalem. At most, I would expect it to perform between Yorkfield and Ibex Peak (old i5). As somebody who has owned both an i7 870 and a X4 975, I have not seen a big difference between the two. But if you were to talk about pure performance; then C2Q, i5, i7, Sandy they all beat the Phenom 2, and changes are most of them will beat BD, whether both chips are OCed or not.

 

[billcat479]

To be released in 2008 and now? What is the type of software does it work with today? Just because it wasn't released way back then didn't mean they quit working on it or somethings are just out of whack.
Have to take a time machine back to see what their roadmap really was in 2006 because they would have started at least by then to start the design work on it. I sure don't remember hearing a whole lot about it back then but that's just my bad memory maybe. But seeing anything like this design posted back then, I doubt it very much. Please post a link so we can take a look at this.
From the testing it is pretty clear it's been built or totally reworked for a new set of software specs that were not out then. Or is not out even now so take your time machine back to figure out this perplexing line of logic please?
Past or present or future what is out is out, what it's built to do is still being done and going to be done more in the future. So either Microsuck or AMD or both are totally out of phase with each other and their software releases and compilers and optimizations.
AMD has put some head scratching cpu's out when there was not any present need for what they did. It started with the the 64 bit cpu when 64bit home OS software that wasn't out yet but is now. They just kept the 32bit part working well till it started to come together. Started adding more core's to their designs.
And back then when people had time to work with them AMD was praised for their design. I think a bit of that has to be put into this new cpu. It's really hard to know just what it can really do.
Kind of like making a product to influence the future. Intel said what and why bother and then copied AMD while still saying why bother until their duel sort of cores came out. Funny.
AMD's timing has a lot of question marks, they are stuck with a smaller R&D budget, has to wait on the fabs to get projected dies working and then make them. It is different when they split with their own production and have to work with what came out of that.
They don't have it as easy as intel so they try to predict ahead of time what way the computer will be used and have long range plans worked out before they finish and produce. Risky in short term but if right cheaper in the long run. They have to think over new designs a lot more carefully than Intel does, that part is obvious.

It has been stated in pretty much all tests that this design is not fit too well for today's software but for tomorrows.
Which is not that good sounding for the people with a lot of money to waste buying a new system every year to say at least on paper my computer is faster than yours now so why buy something that I could care less if it runs faster .5 to 1 year from now.
If people buy a computer to last 2 years and most of them do then AMD makes a lot of sense MAYBE. The maybe part is the hard one to buy into unless you know how it will work with optimized software for it's design. All cpu's depend on this aspect to do their best.
So do you think the folks at AMD are that stupid?
Do they turn a blind eye to current and coming software uses? That is an easy no.
In some aspects they might have taken a better route. It was their first Athlons that made the jump to smaller pipelines and more efficient cpu's while intel was still trying the mhz is best first super long pipeline cpu that could not be cooled.
AMD has sort of taken this path in some of this new cpu which is not good.
Maybe this is why they have a new CEO now while the last one could not sit for what they were doing?
Who knows about that. Lots of maybes here and there.
It can only be looked at for what it can do and why and if it is a cpu that will get faster as it ages like their first Athlon run did.

The really funny part of all this is in a way we are back in time, AMD has put out a CPU that needs newer software which is already showing up in programs and future OS's and will continue to be better coded with better enhancements that could, could maybe, make their cpu get faster as time catches up with it. It's a gamble.
The sites that took this approach did show how well this cpu can be. With the right coding it's a pretty fast cpu.
In a way it's nice to see products come out besides wine that get better with age. Make's it easy to skip a upgrade for the next 2 years buying into this new cpu.
Most computer users don't hack up there computers every 6 months, only the benchmark led fanatics do that an it's a very small percent of computer owners that are in this category.

This is really what it boils down to isn't it. What you see when you use the computer and what you THINK you should see because of the black and white world of benchmarks while most any current brand are so fast now it doesn't really matter what cpu your running because you can run any game, or app and at home and you would never know the difference between them until you looked inside.
This is the most funny and insane part of the computer benchmark world where it is more applicable for non standard software use then for home systems. Saying yeah, I bought this cpu because I only run 2 programs on it and it's twice as fast in doing it. That is funny. And not realistic.
It's exciting to take a look at new things except when you can't see all of what the new thing can do and a redesigned cpu can fit in this category.
Like watching a Ferrari being tested on a 100ft. chunk of roadway.
Pointless. I'm going to give it more time and keep a open mind about it while it's testing can be ironed out. It's way to early for this new of a design to make final opinions. So far it isn't that bad but needs more tests and more info on what software optimized programs can really do with it. So far they have only shown a small part of what it can do or will never do well.
AMD's timing really sucks..

 

[Fourstaff]

Two things to note:

Bulldozer excels at multithreaded loads, guess what server load is?

Bulldozer sucks at singlethreaded loads, guess what consumer load is?

Bulldozer looks almost like its geared to server use, and APUs like Llano for normal desktop use, leaving a rather big hole in the middle for people like us, who demand both excellent single threaded and multithreaded applications, or at least not a bad compromise.

 

[BarbaricSoul]

Well... pit a C2Q against an AM3 Phenom II x4, and you've got a hell of a competition, in fact, the Phenom comes out on top.

A PII at the same clock speeds as my C2Q DOES NOT beat my C2Q. Oh, and don't mention OC'ing, my C2Q can do 4.5ghz on air. Let's see a PII beat a Q9*50 at +4ghz. I'll dismiss the rest of your statement as fanboism

 

[John Doe]

lol. Though, keep in mind that S775 is EOL with it's still jacked up prices. Denebs and especially Thubans give solid price/performance. If you are on tight budget, there is no other option than AMD for a quad on the cheap, well unless you buy second hand.

 

[qubit]

I feel like AMD bulldozer has hidden potential, I guess we just have to wait and see

I think it has the same kind of "potential" that the HD2900XT had against the 8800 GTX. Very good specs on paper, but it never did beat it.

 

[BarbaricSoul]

lol. Though, keep in mind that S775 is EOL with it's still jacked up prices. Denebs and especially Thubans give solid price/performance. If you are on tight budget, there is no other option than AMD for a quad on the cheap, well unless you buy second hand.

I never suggested spending the money on a new Q9650 now. Only way I would suggest that would be to someone with a C2D system that doesn't have the money for a full upgrade, and then I would only suggest a used Q9650 for $200 or less.

 

[Fourstaff]

A PII at the same clock speeds as my C2Q DOES NOT beat my C2Q. Oh, and don't mention OC'ing, my C2Q can do 4.5ghz on air. Let's see a PII beat a Q9*50 at +4ghz. I'll dismiss the rest of your statement as fanboism

Neither is comparing clock for clock, you have to measure it twice, once at stock speeds, and another at average max sustained overclock. The former for out of the box experience felt by non-techies, the latter for people like us. C2Q comes out tops in both metric I believe, but I don't like the way you jump straight to calling people fanboys while preaching a faulty metric. And then there is Price/Perf, which evens things out, depending on budget.

 

[John Doe]

I never suggested spending the money on a new Q9650 now. Only way I would suggest that would be to someone with a C2D system that doesn't have the money for a full upgrade, and then I would only suggest a used Q9650 for $200 or less.

Yeah, I know. That is what I would do, too. The Yorkfield is still a great chip. But if you are buying new, then your only option is AMD for an OCable, cheap quad, since non-K i5s are locked. They still give decent performance though while coming in more expensive.

 

[BarbaricSoul]

Neither is comparing clock for clock, you have to measure it twice, once at stock speeds, and another at average max sustained overclock. The former for out of the box experience felt by non-techies, the latter for people like us. C2Q comes out tops in both metric I believe, but I don't like the way you jump straight to calling people fanboys while preaching a faulty metric. And then there is Price/Perf, which evens things out, depending on budget.

How is my metric faulty. What I basically said is that a PII @ 3ghz does not beat amy C2Q @ 3ghz, it also does not beat my C2Q with both clocked @ 3.6ghz, or both clocked @ 4ghz. The actual speed they are clocked at doesn't matter to what I said, as long as both are clocked at the same speed. The fanboy comment was from the impression he gave me saying the PII does beat the C2Q, and that BD with OC'ing can beat 2nd gen I5 if the I5 isn't OC'ed. He isn't putting both platforms on equal terms. That's why I saw his comments as fanboism.

And no, I'm not a fanboy myself. Like everyone else, I want BD to be great, and I was considering using one to replace my C2Q, but now that the real performance of BD is known, no way I would buy one over a I5 or I7.

 

[Fourstaff]

How is my metric faulty. What I basically said is that a PII @ 3ghz does not beat amy C2Q @ 3ghz, it also does not beat my C2Q with both clocked @ 3.6ghz, or both clocked @ 4ghz. The actual speed they are clocked at doesn't matter to what I said, as long as both are clocked at the same speed. The fanboy comment was from the impression he gave me saying the PII does beat the C2Q, and that BD with OC'ing can beat 2nd gen I5 if the I5 isn't OC'ed. He isn't putting both platforms on equal terms. That's why I saw his comments as fanboism.

And no, I'm not a fanboy myself. Like everyone else, I want BD to be great, and I was considering using one to replace my C2Q, but now that the real performance of BD is known, no way I would buy one over a I5 or I7.

Because you don't run both at the same frequency, stock its 9550's 2.83Ghz vs 955's stock 3.2Ghz (for the non-overclocking croud), and then ~4ghz for 9550 and about the same (or a bit less) for the 955 in the overclocked performance. You don't go about electronically limiting your Mustang to 100mph and then claim that your Toyota Prius goes faster because it is not limited to 100mph: you have to take two readings, one with limiters on (whatever it is, in this case corresponding to stock), and then another without limiters (overclocked). Or limit the rev to 2000rpm, knowing that the Mustang's optimum rev is much higher than that, while Prius is closer to its optimum rev.

Edit: I am not saying that you are wrong, just that your view is skewed to better clock for clock performance, which died with P4 when Intel needed 1.5x freq to equal AMD's offering.