Windows 8: Does AMD's Bulldozer Architecture Benefit?

[proffet]

I would like to see if the ARM can out perform AMD BD

nice haircut.. :/

 

[Guest]

I just got a fx-8120 system.
I feel ripped off.
I could take it back and get an intel i5-3450
problem with that is, 8gb less memory and its an hp system
what should i do??

 

[Crashman]

[citation][nom]Anonymous[/nom]I just got a fx-8120 system.I feel ripped off.I could take it back and get an intel i5-3450problem with that is, 8gb less memory and its an hp systemwhat should i do??[/citation]I'd use it, these aren't bad processors.

The article started as a test of AMD's claims towards Windows 8 performance improvements, and ended up showing those expectations unrealistic. The CPU after Windows 8 is just as good as it was before Windows 8.

Unless you bought your system based on those claims, you shouldn't feel ripped off.

 

[jeffredo]

I don't think anyone was really surprised by the result (and it kind of bugs me that AMD used this "its not all our fault!" excuse last year).

 

[hannibal]

Well there was guite a lot problems last year, but they have managet to fix most of those allready. Of course it does not make bulldoser good CPU, but as we can see from multi threath test, this can be really good CPU with right kind of programs!

 

[rds1220]

[citation][nom]hannibal[/nom]Well there was guite a lot problems last year, but they have managet to fix most of those allready. Of course it does not make bulldoser good CPU, but as we can see from multi threath test, this can be really good CPU with right kind of programs![/citation]

You missed the whole point of the article. You shouldn't need software to make a CPU great it should be great from the start. Yes Bulldozer had a lot of problems but Piledriver still isn't great, they still haven't compleatly fixed the IPC and the overall micro architecture still is slow and ineffcient compared to Intel. In single threaded programs the Piledriver still falls flat on it's face. Also I think you are giving Piledriver to much credit in heavily threaded programs. Yea it's better and it beats out the I7 in SOME programs but again the I7 is still faster and more effcient in most cases.

 

[blazorthon]

[citation][nom]rds1220[/nom]You missed the whole point of the article. You shouldn't need software to make a CPU great it should be great from the start. Yes Bulldozer had a lot of problems but Piledriver still isn't great, they still haven't compleatly fixed the IPC and the overall micro architecture still is slow and ineffcient compared to Intel. In single threaded programs the Piledriver still falls flat on it's face. Also I think you are giving Piledriver to much credit in heavily threaded programs. Yea it's better and it beats out the I7 in SOME programs but again the I7 is still faster and more effcient in most cases.[/citation]

Bullshit. AMD's greatest problems are their cache, not the micro-architecture. Also, please stop saying IPC because you don't seem to even know what it means. FYI, it's not performance per Hz. Furthermore, Piledriver isn't a less efficient architecture than anything from Intel. Compare the Sandy i5s to the Trinity A10s with their IGPs disabled and this becomes obvious where the A10s are actually more efficient. AMD's problems are mostly their crap cache and other front end issues, some of which, sure, are architectural, but that doesn't make it a bad architecture, it just means that it isn't perfect . Sandy/Ivy Bridge isn't perfect either and most certainly isn't superior, it's simply built with superior cache and process technology, something made obvious by the fact that they're architecturally very similar to Core 2 and Nehalem.

Also as proof of IPC not being what you seem to think it is, you should be aware of the fact that Bulldozer and Piledriver have exactly identical IPC despite not being the same micro-architecture and having different performance/power efficiency characteristics. Athlon II (AM3) and Phenom II (AM3) also have exactly identical IPC and have exactly identical micro-architectures too, but their different caches (Phenom II, excluding the 840 and 850, having an L3 cache of varying capacity but usually 6MiB and Athlon II not having an L3 cache as well as having varying L2 cache per core between different SKUs) gives them different performance.

 

[DJDeCiBeL]

Sandy/Ivy Bridge isn't perfect either and most certainly isn't superior, it's simply built with superior cache and process technology, something made obvious by the fact that they're architecturally very similar to Core 2 and Nehalem.

That's kind of contradictory, isn't it?

Also, Sandy and Ivy most certainty are superior, in the purest form of the word. Superior doesn't have to mean the best thing ever created, it just says that something is better than everything else in its class (no matter if it's a small difference, or not). I think Sandy and Ivy can both take claim to that definition quite easily.

And you're right, IPC isn't the only issue with BD and PD, but it certainly doesn't help, when the rest of the architecture has problems too.

The module design was a leap of faith that has yet to pan out (way ahead of its time, IMO, and therefore not very impressive, as of yet). It needs a lot of work and refinement. At least if you disable a core in each module and overclock, you can actually make a BD or PD chip perform well in lightly threaded uses, but it's idiotic that that's even needed.

 

[blazorthon]

That's kind of contradictory, isn't it?

Also, Sandy and Ivy most certainty are superior, in the purest form of the word. Superior doesn't have to mean the best thing ever created, it just says that something is better than everything else in its class (no matter if it's a small difference, or not). I think Sandy and Ivy can both take claim to that definition quite easily.

And you're right, IPC isn't the only issue with BD and PD, but it certainly doesn't help, when the rest of the architecture has problems too.

The module design was a leap of faith that has yet to pan out (way ahead of its time, IMO, and therefore not very impressive, as of yet). It needs a lot of work and refinement. At least if you disable a core in each module and overclock, you can actually make a BD or PD chip perform well in lightly threaded uses, but it's idiotic that that's even needed.

The architecture and the cache and process technology are not the same things and I don't see how they could be confused as such. Architecturally, you have no way of proving which is superior just by looking strictly at performance between FX and Sandy/Ivy because the architectural differences aren't nearly the only major difference. I can prove that Piledriver is superior architecturally by comparing Trinity with an IGP disabled (not the FM2 Athlon II SKUs that are poorly binned, the APUs with their IGPs manually disabled) to Sandy Bridge.

For example, the A10-5800K, with its IGP disabled, uses less power than even the Ivy i3s at idle (heck, it manages to use less at idle even with its IGP enabled) and can use about the same at load as the Sandy i3s despite its own many hardware disadvantages such as poor design methods, far inferior cache, and an otherwise still bottle-necking front end in each module.

It even has similar average gaming performance to those Sandy i3s and quite the advantage in many other things. An architecture that can manage similar average gaming performance and power consumption to the very good Sandy Bridge architecture on the same process node (yet still a superior process technology) despite having higher latency, lower bandwidth L2 cache, no L3 cache, and insufficient L1 cache capacity as well as its other bottle-necks such as too few x86 decoders and more is obviously superior.

The problem with the modular design is nothing like what you seem to think it is. The problem is simply a front end blunder (admittedly, I don't know how AMD could have been so stupid as to screw it up so badly, but still), nothing more. AMD's modular concept isn't ahead of its time (despite AMD and others making such claims, these claims really just excuses for how badly they screwed it up IMO), it was simply poorly executed. It doesn't need too much work and refinement, it simply needs the front end to be expanded, which is slated for being done in Steamroller around late 2013 or early 2014.

 

[DJDeCiBeL]

The architecture and the cache and process technology are not the same things and I don't see how they could be confused as such.

It's a generalization of the term. Ease up. When 99% of everyone thinks that's the correct usage, it doesn't even matter anyway. That's not even the point. You're getting argumentative and lost over semantics.

The point (and the basic, unsaid point of this article) is that that BD and PD are not great CPU's. They're just not. They can be "good", but they will never be defined as great, no matter what software "optimizations" are made.

And I still believe the module design was the wrong path for AMD to work towards, but they're stuck with it now, so hopefully, they'll make the best of it.

 

[rds1220]

[citation][nom]blazorthon[/nom]Bullshit. AMD's greatest problems are their cache, not the micro-architecture. Also, please stop saying IPC because you don't seem to even know what it means. FYI, it's not performance per Hz. Furthermore, Piledriver isn't a less efficient architecture than anything from Intel. Compare the Sandy i5s to the Trinity A10s with their IGPs disabled and this becomes obvious where the A10s are actually more efficient. AMD's problems are mostly their crap cache and other front end issues, some of which, sure, are architectural, but that doesn't make it a bad architecture, it just means that it isn't perfect . Sandy/Ivy Bridge isn't perfect either and most certainly isn't superior, it's simply built with superior cache and process technology, something made obvious by the fact that they're architecturally very similar to Core 2 and Nehalem.Also as proof of IPC not being what you seem to think it is, you should be aware of the fact that Bulldozer and Piledriver have exactly identical IPC despite not being the same micro-architecture and having different performance/power efficiency characteristics. Athlon II (AM3) and Phenom II (AM3) also have exactly identical IPC and have exactly identical micro-architectures too, but their different caches (Phenom II, excluding the 840 and 850, having an L3 cache of varying capacity but usually 6MiB and Athlon II not having an L3 cache as well as having varying L2 cache per core between different SKUs) gives them different performance.[/citation]

EDIT: I typed a big response but you are clueless.

 

[blazorthon]

It's a generalization of the term. Ease up. When 99% of everyone thinks that's the correct usage, it doesn't even matter anyway. That's not even the point. You're getting argumentative and lost over semantics.

The point (and the basic, unsaid point of this article) is that that BD and PD are not great CPU's. They're just not. They can be "good", but they will never be defined as great, no matter what software "optimizations" are made.

And I still believe the module design was the wrong path for AMD to work towards, but they're stuck with it now, so hopefully, they'll make the best of it.

Architecture, cache, and process technology aren't the same. At that point, it's not even a generalization of the terms, it's a confusing mistake and most people are not calling them the same unless they are people who are ignorant of the subject and that should be told that they aren't the same (perhaps giving an explanation of the differences) if they want to talk about that subject. That's not something that I'll ease up on. It's not semantics.

Whether or not they are great is not the point and even if it was, you seem to have missed it. Depending on the usage, some of the FX CPUs most certainly can be great. That's especially true about Trinity which can be great in very many ways compared to its competition without necessarily needing extra work done to get proper performance.

The point of the article was to show that Windows 8 did not improve optimization for Bulldozer and Piledriver. It most certainly did not prove that optimization is impossible. Furthermore, software optimization most certainly can be extremely impacting on performance. For example, optimizing to be able to efficiently scale x86 performance across many cores can and does change the situation for AMD drastically. In the case of the i3s and below, making good use of some floating point instructions also gives AMD's lower end options a huge advantage for some workloads. Heck, simply looking at graphics drivers shows how software optimization (in this case, drivers, which nowadays are often so complex for graphics that they're almost OSs in of themselves) can make huge differences.

What could possibly be wrong with the modular path? Once AMD fixes scaling in Steamroller, it'll basically let you get two cores to take up the die area of a little more than one core while having the same x86 performance as two otherwise identical individual cores. It's an incredibly effective method of scaling core count in CPUs without increasing die area by too much.

 

[blazorthon]

EDIT: I typed a big response but you are clueless.

Hilarious. You say edit there, but according to the forums, you didn't actually edit that post at all nor any post before it. Furthermore, I gave plenty of explanation and evidence, but you offer nothing more than an insult to counter my claims. You're really not making a convincing case.

 

[DJDeCiBeL]

What could possibly be wrong with the modular path?

I never said anything was particularly wrong with the core of the module design, just AMD's implementation of it.

I just think it was stupid of AMD to take the risk when they could least afford to. And it bit them. They are forever scarred from it, no matter if they come out with the best CPU ever made because of it. Nobody will care. The boy who cried wolf...

 

[blazorthon]

I never said anything was particularly wrong with the core of the module design, just AMD's implementation of it.

I just think it was stupid of AMD to take the risk when they could least afford to. And it bit them. They are forever scarred from it, no matter if they come out with the best CPU ever made because of it. Nobody will care. The boy who cried wolf...

Alright, that's definitely not something that I can't agree with, although I wouldn't say that nobody would care if they managed to solve this mess. A lot of people would care.

 

[rds1220]

[citation][nom]blazorthon[/nom]Hilarious. You say edit there, but according to the forums, you didn't actually edit that post at all nor any post before it. Furthermore, I gave plenty of explanation and evidence, but you offer nothing more than an insult to counter my claims. You're really not making a convincing case.[/citation]

I know I deleted it before I posted and just typed that. You offer no real explanation, you try to be objective but you're not fooling anyone you are an AMD fanboy and it shows with your responses where you fanatically defend AMD at every chance you get.

 

[blazorthon]

[citation][nom]rds1220[/nom]I know I deleted it before I posted and just typed that. You offer no real explanation, you try to be objective but you're not fooling anyone you are an AMD fanboy and it shows with your responses where you fanatically defend AMD at every chance you get.[/citation]

I offered plenty of explanation. Tom's, Anand, and Tech Report (as well as others) have shown us many aspects of Trinity, Sandy Bridge, and Ivy Bridge and I explained some of them. I also explained how AMD's main problem isn't even their architectures even beyond that point and that their cache is their main problem. Hell, Anand and Tom's show us how Intel's L3 cache has lower latency and higher bandwidth than AMD's L2 cache, let alone AMD's L2 cache compared to Intel's L2 cache and Intel's L3 cache compared to AMD's incredibly weak L3 cache. After all of that, I gave examples of how IPC isn't even the problem.

I'm not being fanatic nor a fanboy about this, I'm simply being reasonable. You're the one who isn't being reasonable here, not I. It seems that you even went as far as thumbing down all of my posts just because you disagree with me.

 

[blazorthon]

[citation][nom]lostmyclan[/nom]yup l3 l2 is the problem... i want a 6350 but for now ? the mother board still hard to belive in prices 6350 + 990fx i can get z77 asrock xtreme 6 and a 3570k. Less wattage Less heat =D why the hell i get a piledriver!?[/citation]

What's your point?

 

[jrazor247]

I only run one program at a time on my eight core processor and I don't know how to write a script that launches multiple apps and initiates background I/O, so this review is extremely helpful to me so I clicked your ads. Props to this benching methodology.

 

[maverick knight]

No one will go wrong with either Intel or AMD. Bulldozer is not a bad chip, it just that AMD promised something better but in the end that was not the case. BD is still impressive.

I have an i5 2500K and I'm happy that AMD did not make my CPU obsolete. The facts of the matter is that no matter how you put it, build it, paint it, or splice it Intel chips are, for the most part, the logical choice. If in 2014 the power switches to AMD I will build one instead.

I know people are trying to defend their side of their choice but its hard to argue with facts. Just because Intel is slightly faster by just a couple of seconds (or fractions of a second) does not make AMD a slow chip. We should all get over this feud.

 

[blazorthon]

No one will go wrong with either Intel or AMD. Bulldozer is not a bad chip, it just that AMD promised something better but in the end that was not the case. BD is still impressive.

I have an i5 2500K and I'm happy that AMD did not make my CPU obsolete. The facts of the matter is that no matter how you put it, build it, paint it, or splice it Intel chips are, for the most part, the logical choice. If in 2014 the power switches to AMD I will build one instead.

I know people are trying to defend their side of their choice but its hard to argue with facts. Just because Intel is slightly faster by just a couple of seconds (or fractions of a second) does not make AMD a slow chip. We should all get over this feud.

Even if AMD's 81350 was twice as fast as the i5-2500K in every way, how would that make the i5-2500K any more or less obsolete? It isn't obsolete just because there are faster options. I don't see good reason in calling something obsolete so long as it can get the job done properly.

 

[warhead99]

tbh a cpu can only affect FPS to an extent untill you start running into factors such as memory and GPU over the years of overclocking, and purchasing parts for upgrade, i have seen that cpu affects roughly about a ten to fifteen percent increase in fps by overclocking a few GHZ and a complete upgrade generally does the same, so for a software change to give a 2-3 percent increase in some games i do believe its a success. Yes we know the Bulldozer are inferior to the Sandy bridge, and no i do not believe that its all about blaiming 'microsoft' for 'bad' OS architecture. I have had my FX8120 since January and what can i say im not dissapointed, ive had a hell of a ride overclocking it to 4.9 Ghz (been close to tears with frustration), i can run every game available at a higher frame than the eye can recognize anyway, and its a fraction of the price of the enthusiast dominant CPU's that intell are selling. i also agree that AMD bull(dozer)shat it up (pardon the pun) but its a mid to high end processor that gets the job done just dont be expecting the highest points on 3Dmark. nd just recently i damaged my graphics card downgrading to a 4890 from an unlocked 6950 that got better fps than an average 6970 (gutted) but i deffinately say a massive dip in fps, if your bothered about FPS spend abit less on cpu and more on graphics.

 

[warhead99]

sorry if that dosnt make sence its 5 am

 

[Thomas Creel]

I wouldn't have expected it anyways, new processor =/= best.

 

[dozerman]

techreport.com/review/21865/a-quick-look-at-bulldozer-thread-scheduling

Something isn't right here. Techreport clearly shows a good single-threaded bump in performance in software with less than five threads when affinity is manually scheduled.