AMD FX 8-Core Processor Black Edition promo vid.

[LOL_Wut_Axel]

I'm seriously hoping for a big IPC increase.

Core (45nm) has ~10% higher IPC than K10.5, Nehalem has ~10% higher IPC than Core, and Sandy Bridge has ~15% higher IPC than Nehalem. This is not counting HyperThreading (it's not part of the IPC equation), so add 20% for that in multi-threaded apps.

What you end up with is with the fact that Intel has ~35% higher IPC than AMD currently. AMD got a 30% performance boost in heavily multi-threaded apps from the Phenom II X6 in comparison to the X4, so in those it can either match or be slightly slower than the Core i5 2500. It's still 20% behind in them in comparison to the 2600 because of HyperThreading. They can only match the Core i5 in very multi-threaded apps, but at the expense of performance in anything that's not.

In the end, they need a 20% IPC boost in combination with high stock clock speeds to be able to not lose by a high amount in single-threaded apps while being able to have a considerable gain in multi-threaded apps, which is the main point of them making 8-core CPUs for consumers.

So, consider this scenario: AMD FX-6100 has a 20% gain over the Phenom II X6 1100T because of higher IPC. That means it can almost match the 2600 in very heavily threaded apps (3.4GHz vs 3.3GHz=2% advantage), but loses by ~17% in single-threaded ones. Then comes the two other ones: the FX-8150 and FX-8100. They both have two more cores than the 6100, which should give it an advantage of ~20% in multi-threaded apps. The FX-8100 has a clock speed of 2.8GHz, giving it a deficit of ~15% in comparison to the FX-6100's 3.3GHz. That means it should only be 5% faster in multi-threaded apps, or 3% faster than the i7 2600, but also 32% slower in single-threaded apps. But then they have the FX-8150, which not only benefits from the two added cores, but also higher stock clock speed. In comparison to the FX-6100, it should benefit from higher clock speeds in single-threaded apps by 8%, and in multi-threaded apps from both the higher clock speed and also by 20% from the additional cores. That means it would be ~9-10% slower than the i7 2600 in single-threaded apps and 28% faster in multi-threaded apps, both stock. If performance at the same clocks is the method of comparison, then it would obviously be ~15% slower in single-threaded apps. It would also mean that it would beat the Core i7 990X overall by a very slight margin (though it doesn't compete with that).

Of course, this is all taking into account Nehalem IPC. If it's 5% lower or higher, then obviously everything I said changes by that same percentage. Also, don't take everything I said as fact, since obviously IPC can be different, and there could be other factors to take into account. The main point that I was trying to bring up is that if it has Nehalem-level IPC it's pretty much the exact same CPU in terms of performance at the same clocks as Gulftown, which benefited by 20% because of HyperThreading, while Bulldozer would find that 20% benefit from the additional two cores.

 

[frozentundra123456]

The problem being in the real world, business doesn't work that way. In fact, more things are sold because of better marketing rather than outright being better than all other competing products. And this is not just in the IT industry. It is true everywhere.

In fact, even if they had a poor performer, shutting up is exactly what they don't want to do. Poor product (meaning no word-of-mouth following will even happen) + poor marketing effort will ensure pitiful sales. If that was going to be the battle plan, why even bother?

You don't have to believe me. There are thousands of MBA's around who are willing to give you the same advice for a generous consulting fee.

If anything, AMD should learn from Intel that great marketing (like the Intel tune and the Blue Man Group thing) needs to accompany a great product, and up their marketing some more. Seeing as to how a good number of people probably can recognize the Intel brand, sometimes even the tune, while far less know AMD or that it is even an American company, their marketing is still not enough.


And now that I've dealt with the meat of the issue, let's return to the peripheral issue of:

1.) Not kidding.
2.) I do have a sense of humor, but your post was so unfunny that it did not register.
3.) I do understand sarcasm, and while I did realize you were trying to use sarcasm to make a point, I found the point to be significantly wrong and out of place that I decided on expending the effort on pointing out how ridiculous it is.

My point is that both Llano and Bulldozer have been delayed repeatedly,and Llano on the deskop is a low end chip at best, having a 2 generation old CPU architecture and graphics that, while better than intel integrated, is still beaten by a 50.00 discrete card. I will grant you that marketing plays a role in a business, but eventually you have to produce a superior or at least competitive product. Currently AMD does not do this except in price at the low end, as far as the CPU lineup is concerned. And I cant believe that all the delays bode well for the performance of Bulldozer.

 

[Riek]

I'm seriously hoping for a big IPC increase.

Core (45nm) has ~10% higher IPC than K10.5, Nehalem has ~10% higher IPC than Core, and Sandy Bridge has ~15% higher IPC than Nehalem. This is not counting HyperThreading (it's not part of the IPC equation), so add 20% for that in multi-threaded apps.

What you end up with is with the fact that Intel has ~35% higher IPC than AMD currently. AMD got a 30% performance boost in heavily multi-threaded apps from the Phenom II X6 in comparison to the X4, so in those it can either match or be slightly slower than the Core i5 2500. It's still 20% behind in them in comparison to the 2600 because of HyperThreading. They can only match the Core i5 in very multi-threaded apps, but at the expense of performance in anything that's not.

In the end, they need a 20% IPC boost in combination with high stock clock speeds to be able to not lose by a high amount in single-threaded apps while being able to have a considerable gain in multi-threaded apps, which is the main point of them making 8-core CPUs for consumers.

So, consider this scenario: AMD FX-6100 has a 20% gain over the Phenom II X6 1100T because of higher IPC. That means it can almost match the 2600 in very heavily threaded apps (3.4GHz vs 3.3GHz=2% advantage), but loses by ~17% in single-threaded ones. Then comes the two other ones: the FX-8150 and FX-8100. They both have two more cores than the 6100, which should give it an advantage of ~20% in multi-threaded apps. The FX-8100 has a clock speed of 2.8GHz, giving it a deficit of ~15% in comparison to the FX-6100's 3.3GHz. That means it should only be 5% faster in multi-threaded apps, or 3% faster than the i7 2600, but also 32% slower in single-threaded apps. But then they have the FX-8150, which not only benefits from the two added cores, but also higher stock clock speed. In comparison to the FX-6100, it should benefit from higher clock speeds in single-threaded apps by 8%, and in multi-threaded apps from both the higher clock speed and also by 20% from the additional cores. That means it would be ~9-10% slower than the i7 2600 in single-threaded apps and 28% faster in multi-threaded apps, both stock. If performance at the same clocks is the method of comparison, then it would obviously be ~15% slower in single-threaded apps. It would also mean that it would beat the Core i7 990X overall by a very slight margin (though it doesn't compete with that).

Of course, this is all taking into account Nehalem IPC. If it's 5% lower or higher, then obviously everything I said changes by that same percentage. Also, don't take everything I said as fact, since obviously IPC can be different, and there could be other factors to take into account. The main point that I was trying to bring up is that if it has Nehalem-level IPC it's pretty much the exact same CPU in terms of performance at the same clocks as Gulftown, which benefited by 20% because of HyperThreading, while Bulldozer would find that 20% benefit from the additional two cores.

First of all stop the core logic. counting core and multiplying the performance and vice versa is long gone when BD is released.
BD core perform worse when 2threads are run on it. (or BD core perform better when only one thread is run in the module).
For some years we have turbo boost determining the scene for low threaded parograms. in this case when intel 2500 or 2600 is running around >3 threads they barely gain anything. If BD has the same turbo frequency over 3cores as over 1 (which would be logical given their turbo2,0 explanation) the difference might be completely in favor of AMD with your numbers. As long as we don't know the frequencies (which we don't) and we don't know where BD will really shine and we don't know when turboboost will be applied and in what circumstances. we really can't say a thing. in above example if turboboost only works on 1 thread or over 3threads can give a difference of 18% (3,3 to 3,9). (most applications are somewhere between 2 and 4 threads)

My point is that both Llano and Bulldozer have been delayed repeatedly,and Llano on the deskop is a low end chip at best, having a 2 generation old CPU architecture and graphics that, while better than intel integrated, is still beaten by a 50.00 discrete card. I will grant you that marketing plays a role in a business, but eventually you have to produce a superior or at least competitive product. Currently AMD does not do this except in price at the low end, as far as the CPU lineup is concerned. And I cant believe that all the delays bode well for the performance of Bulldozer.

The delays are mostly immaginary. Most of the launch dates weren't from AMD, you can't hold AMD responsible for public misinformation by someone. (e.g. OBr for example). AMD only recently started to campaign bulldozer. Yes bulldozer was delayed, but not in the way that some people describe here. I actually believe its beensheduled in Q3 for a long time... then rumours started it would be released in Q2 etc and everbody started to use those rumours as reality)

And you compete on price. Wether it is in the low end or not. All the delays have nothing to do with performance. Its about bugs and process technology.

 

[Vesku]

AMD said Q2 for desktop, officially pushed back to Q3.

 

[Arzachel]

BD core perform worse when 2threads are run on it. (or BD core perform better when only one thread is run in the module).

Wrong. Threads should be run on as few modules as possible to get the most out of turbo. I can dig up a quote from JFAMD if you want.

 

[Riek]

Wrong. Threads should be run on as few modules as possible to get the most out of turbo. I can dig up a quote from JFAMD if you want.

Where did i mention it should be run on different modules?
hence where did i mention i was talking about more threads?

Fact: a module has shared logic that does performance decrease.
Assumption: an FX8 can run more modules at the max frequency.

 

[soccerballtux]

you have the power to change that...

Let it rest in piece. There was too much in the one thread anyhow, making it almost impossible to keep track of what had been talked about, etc. Towards the end it was just going around in circles repeating things that had already been discussed 30 pages ago. At least I don't think it was ever Godwined, surprisingly enough.

meant to say that tongue in cheek with an emoticon or something. I wasn't actually saying he should unlock it. I don't care either way. I was just pointing out, he was saying "well it was locked" as if there were nothing he could do about it

 

[LOL_Wut_Axel]

First of all stop the core logic. counting core and multiplying the performance and vice versa is long gone when BD is released.
BD core perform worse when 2threads are run on it. (or BD core perform better when only one thread is run in the module).
For some years we have turbo boost determining the scene for low threaded parograms. in this case when intel 2500 or 2600 is running around >3 threads they barely gain anything. If BD has the same turbo frequency over 3cores as over 1 (which would be logical given their turbo2,0 explanation) the difference might be completely in favor of AMD with your numbers. As long as we don't know the frequencies (which we don't) and we don't know where BD will really shine and we don't know when turboboost will be applied and in what circumstances. we really can't say a thing. in above example if turboboost only works on 1 thread or over 3threads can give a difference of 18% (3,3 to 3,9). (most applications are somewhere between 2 and 4 threads)

Yeah, no. Pretty sure you don't have a good idea of what you're talking about. And I'm not taking Turbo into account, just the stock frequencies. The frequencies and models were leaked.

Most mainstream apps take advantage of three or four threads. We're not in 2005-2007 anymore. About the only thing you'll find now that only takes advantage of one to two threads is audio encoding. Most recent games take advantage of three to four threads, and things like video encoding, 3D rendering, content creation, and file compression (in 7-zip, that is) are heavily multi-threaded. Even web browsers like Chrome and IE 9 can take advantage of multi-threading.

 

[frozentundra123456]

[From Riek:

And you compete on price. Wether it is in the low end or not. All the delays have nothing to do with performance. Its about bugs and process technology.[/QUOTE]


How do you know the delays have nothing to do with performance. Do you work for AMD and have access to insider information?

I honestly hope Bulldozer comes out soon and has excellent performance. I am just amazed that you so steadfastly refuse to consider that Bulldozer may be further delayed and that it will be disappointing when it does come out. And as to process technology, how long has Intel been producing 32 nm chips???

And yes, a company does compete on price, I granted you that already. But you also need to be competitive in the high end. But the main thing that annoys me is that AMD is always claiming to be so technologically advanced, but their products never seem to measure up. For instance, their hype that Phenom was a "true quad core" chip while intel was not. Well Intel kicked their a** and still is. And how about "the future is Fusion"? Granted fusion has a place in the mobile world, but it is nowhere the earthshaking advance that AMD wanted us to expect. And actually, even though it has low performance, Intel beat AMD to the on-die graphics, despite all the fusion hype.

So as I said before: AMD, just get out a superior product and let the product speak for itself.

 

[Ken g6]

As long as it reaches decent performance characteristics, we won't really care if they reach it through improved IPC or just mega clocks. This is a contrived and completely exaggerated example, but just to illustrate, so what if BD launches at 4.5GHz but offers a minimum 20% better performance than a 3.3GHz SB, and at the same price? IPC will be terrible in comparison, but out of the box performance is better anyway, and that's all users will notice. (of course, unless it has better OC headroom, us enthusiasts will be bothered, but OEMs won't be, and that's really all AMD and Intel are after)

Enthusiasts aren't the only ones who can overclock. :sneaky: I expect if this happens, within a few weeks we'll see Intel release an i7 at 4GHz or something stock (Turbo to 4.2 or 4.3). It'll have a high TDP, but BD surely has that problem too.

 

[jvroig]

Enthusiasts aren't the only ones who can overclock

Absolutely true. Most people can overclock, but the overwhelming "normal user" population don't bother, or don't really factor it in their buying decisions (mostly prebuilts like Dell / HP / Acer / etc), hence "enthusiasts like us will be bothered, but OEMs won't be".

I expect if this happens, within a few weeks we'll see Intel release an i7 at 4GHz or something stock (Turbo to 4.2 or 4.3). It'll have a high TDP, but BD surely has that problem too.

I would say quite possible given that SB has the reputation of OCing easily to 4.2 - 4.3. I wouldn't really know how likely it is, but I certainly wouldn't bet against Intel being able to release a higher performing SKU if they need to by Sept / Oct. They probably have a model number reserved for it, or an SKU they are originally planning to release further down the line but won't exactly break their hearts to release earlier if needed.

 

[notty22]

To add to that line of reasoning.

Intel Finally Has a Real 4 GHz CPU
Intel has quietly introduced its first CPU that is capable of running at 4 GHz clock speed off the shelf.
An updated spec sheet reveals that the recently introduced Xeon E3-1290 runs at 3.6 GHz with four cores, but the chip's turbo boost will scale the clock speed to 4.0 GHz when running on only one core. What makes this processor particularly interesting is the fact that it is closely related to the i7-2000 Sandy Bridge series, which would indicate that Intel could be launching a 4 GHz desktop processor as well.

 

[jvroig]

Very cool, can't believe I missed that. Nice find, thanks, notty22 :thumbsup:

Perhaps their impending SB-E product line has one of those already.

 

[Idontcare]

All the delays have nothing to do with performance. Its about bugs and process technology.

I remember when the same line of thinking was stated over and over again in these forums some four years ago, only it was about the delayed and impending launch of Phenom.

 

[AnandThenMan]

If anything, AMD should learn from Intel that great marketing (like the Intel tune and the Blue Man Group thing) needs to accompany a great product, and up their marketing some more. Seeing as to how a good number of people probably can recognize the Intel brand, sometimes even the tune, while far less know AMD or that it is even an American company, their marketing is still not enough.

Let's be clear here, AMD can in no way afford to out market Intel, they simply don't have the funds. AMD already puts as much money as possible into R&D, there is nothing left over. And Intel has succeeded in LARGE part due to "other forms" of marketing, especially when AMD had a much better product. It's easy to sit back and say hey, AMD is hardly known by the general public, they have failed as a company to market themselves properly. In a vacuum, that's true, but in reality they simply cannot compete with Intel financially. In fact in x86 no one can, how many competitors does Intel have? It's really easy to become incredibly rich and pour billions into new Fabs when you don't any real competition.

If AMD needs to "learn" anything from Intel, it's how to bribe, threaten, and strong arm their partners. Not to mention crippling compilers, rigging benchmarks, buying off review sites. So yes AMD does have a lot to learn.

 

[jvroig]

Let's be clear here, AMD can in no way afford to out market Intel, they simply don't have the funds

They don't have to, and I'm not saying their goal should be to out-spend Intel.

All I said was their marketing needs to improve, and that doesn't necessarily mean "spend more than Intel". It just literally means improve their marketing, and that takes into account not just ads/promos, but the whole she-bang:
-the 4 P's (or the Marketing Mix): Product, Price, Place, Promotion
-Extended Marketing Mix: People, Process, Physical evidence

Given a finite set of funds, there are several ways to accomplish your goal (in this case, "marketing") and design a marketing strategy, depending on what factors (the different P's) you put emphasis on over others.

 

[frostedflakes]

To add to that line of reasoning.

Only took them about 7 years to reach that elusive 4GHz barrier.

 

[drizek]

Prescott must be so proud.

 

[drizek]

I remember when the same line of thinking was stated over and over again in these forums some four years ago, only it was about the delayed and impending launch of Phenom.

To be fair, Phenom did have bugs as well.

 

[Riek]

I remember when the same line of thinking was stated over and over again in these forums some four years ago, only it was about the delayed and impending launch of Phenom.

Well phenom had a bug that caused issues and had a process that wasn't doing what it should be doing. Because of the TLB bug they needed to focus on fixing the bug instead of improving the process. Which caused phenom to perform bad due to low clocks. which was a process problem.

In this case the problem isn't BD design it is bugs of working modules/turbo and a process that isn't doing that great. Does that affect performance? yes.. Does that make BD having a performance problem? no it has a bug an process problem. They will eventually have a performance problem if they need to release having bugs or having process problems. But then it is not the design that is at fault.

Yeah, no. Pretty sure you don't have a good idea of what you're talking about. And I'm not taking Turbo into account, just the stock frequencies. The frequencies and models were leaked.

Most mainstream apps take advantage of three or four threads. We're not in 2005-2007 anymore. About the only thing you'll find now that only takes advantage of one to two threads is audio encoding. Most recent games take advantage of three to four threads, and things like video encoding, 3D rendering, content creation, and file compression (in 7-zip, that is) are heavily multi-threaded. Even web browsers like Chrome and IE 9 can take advantage of multi-threading.

If you don't take turboboost into account you can pretty much throw your whole post into the trashcan. from the models leaked, BD has higher turbo frequencies than intel and previous qmd lineup. According to the turbospec running some sort of turbo can occur with all cores in use.

If the only insult you can give is that you are sure that i don't know what i'm talking about and directly afterwards convince me i'm right in saying most taks in the mainstream use indeed 2-4 threads... i'm pretty sure you missed a corner somewhere.

 

[MentalIlness]

I have usually been an AMD fan...but the last 4 years I have had Intel only. Because it performed better, and was a good price at the time. Even overclocked to 4.0 *E7200* and I am still using it with the 6870.

Am I bottlenecked ? Yes. Do I care ? No. I plan on a system overhaul when Bulldozer is released because I want to go back to AMD for a while....but isn't because I have anything against Intel. Quite the opposite really......I like them both.

We all have high hopes for Bulldozer. I already know what the Intel chips "currently" are capable of, Just need some kind of performance test for BD.

I am still rocking this E7200 @ 4.0 for the last few years. Abit getting a little impatient now waiting on AMD.

What *IF* AMD does outperform Intel with BD ? And what *IF* AMD BD outperforms IB ? Im hoping it does, of course dont see it happening. But what *IF* it does ? In my opinion, if it *does* or even comes close, ill be building a 2600K machine along side the Bulldozer, since prices will more than likely lower.

I know, that is *ALOT* of what if's. But doesn't hurt to ponder about it.

Get the chips out, plain and simple.

 

[LOL_Wut_Axel]

Well phenom had a bug that caused issues and had a process that wasn't doing what it should be doing. Because of the TLB bug they needed to focus on fixing the bug instead of improving the process. Which caused phenom to perform bad due to low clocks. which was a process problem.

In this case the problem isn't BD design it is bugs of working modules/turbo and a process that isn't doing that great. Does that affect performance? yes.. Does that make BD having a performance problem? no it has a bug an process problem. They will eventually have a performance problem if they need to release having bugs or having process problems. But then it is not the design that is at fault.





If you don't take turboboost into account you can pretty much throw your whole post into the trashcan. from the models leaked, BD has higher turbo frequencies than intel and previous qmd lineup. According to the turbospec running some sort of turbo can occur with all cores in use.

If the only insult you can give is that you are sure that i don't know what i'm talking about and directly afterwards convince me i'm right in saying most taks in the mainstream use indeed 2-4 threads... i'm pretty sure you missed a corner somewhere.

Except that introduces a variable that's dependent on current testing conditions, which is why it's best to compare at base frequencies. That, and if you're thinking of over-clocking, you'll pretty much kiss Turbo goodbye.

And now I'm confirming you don't have much knowledge about this topic again in your second paragraph. I already gave you several examples, and if an application takes advantage of more than one thread (going up to four threads), it's mildly multi-threaded. If you didn't read what I said before, again:

Single-threaded: audio encoding
Mildly multi-threaded: games, web browsers
Multi-threaded: image editing, video encoding, 3D rendering, content creation, file compression/decompression (7-zip, PAR2), compiling, folding, games and productivity.

I'm sorry, but most things are either already multi-threaded or switching to it. Most new games introduced this year are taking advantage of multi-threading, and older games from 08-10 were already mildly multi-threaded (used two to four threads). The only real case you can make for single-threaded applications here is audio encoding.

 

[Phynaz]

Let's be clear here, AMD can in no way afford to out market Intel, they simply don't have the funds.

It's not about the amount you spend, it's about the quality of your marketing. For example, everyone knows the Intel three chord chimes.

Intel:
Three chord chimes
Intel Inside
Bunny Men
BMG

AMD :
Can't think of a single marketing program.

 

[AnandThenMan]

It's not about the amount you spend, it's about the quality of your marketing. For example, everyone knows the Intel three chord chimes.

Intel:
Three chord chimes
Intel Inside
Bunny Men
BMG

AMD :
Can't think of a single marketing program.

Suppose you could think of an AMD marketing program. How many people would have seen it? You might have, but with a tiny fraction of Intel's marking budget, chances are most people would never be exposed to whatever marketing push AMD came up with. I don't know what Intel's marketing budget is currently, but it's probably higher than AMD's total revenue. How can AMD even begin to compete with that? The answer is you don't, you'd be tossing good money away.

I think what AMD is doing now is going after design wins aggressively, and pushing Fusion really hard in the big box stores and etailers. I took a look at the mobile offerings at Best Buy yesterday, and was surprised by two things. One, there was a lot of AMD hardware, E-350, A6 and A8 Fusion stuff. Two, in some cases AMD seems to be commanding a price premium for these units in comparison to similar Intel hardware. Atom has been relegated to the very bottom end and is being discounted.

 

[frozentundra123456]

If AMD needs to "learn" anything from Intel, it's how to bribe, threaten, and strong arm their partners. Not to mention crippling compilers, rigging benchmarks, buying off review sites. So yes AMD does have a lot to learn.

What they need to "learn" from Intel or somebody, is how to produce a superior product and get it to market on time.

While Intel may have used unfair practices in the past, they paid AMD a ton of money to settle that. And in these days of technical forums and readily available product reviews on the internet, I believe if AMD produced a superior product, people would be aware of it.