Fudzilla: Bulldozer performance figures are in

[lol123]

I agree with Lonbjerg that everything JF-AMD posts is AMD PR and affects the perception of AMD as a company. If you look at the post at Overclock.net, you will see that he even has a "HARDWARE REP" banner below his nick.

However the second paragraph of my post, in case you missed it, was "The comic that was released some time ago, and which actually was an official piece of PR, is in the same vein." I absolutely believe that AMD's major problem in the market is lack of recognition and credibility as a viable option, as opposed to Intel being perceived as the safe choice. The vast majority of people considering an AMD processor or a computer with one don't want to give Intel a smack in the face, they want a processor that works and a PR strategy that consists of attacking the competition and defensively invalidating the concerns of buyers (as JF-AMD to some degree did in his post) undermines their reputation as a professional company, one which AMD used to have when that was built on their engineering capability and reflecting that through PR and word of mouth, and not on appealing to the populist anti-Intel views of some AMD fanboys.

I know that this forum is not exactly a place of vigorous debate and differing opinions but I really did mean what I wrote as constructive criticism and nothing else. If people want to attack me for it, then that really speaks volumes for itself. All we can know for sure and agree on is that AMD's current strategy, PR and otherwise, is not working.

LOL, yeah I'm sure he's really worried about how professional his personal, non-company, opinion is communicated or received:

Are you gonna critique his grammar and spelling too? Maybe his font choice and ratio of bold to regular type text is non-professional as well?

Failing to see the forest for the trees here.

He should be worried, because it's his company and his employer that he is representing. He is writing and signing his posts as a Director of Producting Marketing at AMD. Were you even really serious about this post?

Edit: Correction, he is not the Chief Marketing Officer but the Director of Product Marketing for Server/Workstation products. I am sorry for the mistake.

 

[Dresdenboy]

It's an easy thing to write...but I don't buy it.

We are talking PR here...the profession of smooth words.

Have you seen my latest blog entry on Bulldozer ES performance on http://citavia.blog.de/ ? I covered a similar topic. In essence any not faked BD ES result is a lower bound for expectable performance, while upside is unknown.

Sent from my GT-I9000 using Tapatalk

 

[Nemesis 1]

Well, when I read his post at overclock.net it dosn't look like a post from a private individual, but like he is doing his job...AMD PR.

That you then try to advocate he is not doing AMD PR...just boogles my mind.

I call it like I see it...and I see PR.

Ya I was going to reply to his blog post . But since I already debate him more than I should I waited to see who stepped up . I new you would be one . THe newer member above also stepped up . That blog to me as I read it is 100% defensive in nature .
The main point being If you don't have lots of cores your not in the game . and if your after single threaded performance ya need to buy a single core. Ya 8 cores is good for servers . But for the desktop IPC is the most important element . A quad on the desktop is more than enough.
I was a little surprised by IDC stance here . The blog reads like a cring towel and IDC has good reading comprehension so that surprized me . Other than that the BLOG was amusing marketing meant for the mindless.

 

[lol123]

Well, when I read his post at overclock.net it dosn't look like a post from a private individual, but like he is doing his job...AMD PR.

That you then try to advocate he is not doing AMD PR...just boogles my mind.

It really surprises me too for several reasons:

1. He posts as "JF-AMD". Everything he posts will be associated with AMD as a company, and there is simply no way around that.
2. The post is structured as a systematic, point-to-point explanation of AMD strategy to the enthusiast community. There is nothing wrong with that, but how could that possibly be construed as his "personal opinion and nothing else"?
3. He requests that his thread be closed immediately because he is "not going to answer questions here". That very clearly sounds like someone speaking from a company platform, and it is definitely not the behavior of just anyone starting a new thread on a hardware site.

So Idontcare, do you seriously claim that nothing of what he posts represents AMD for the single reason that he signs his posts with the standard disclaimer "everything I post is my opinion and not that of my company"?

 

[exar333]

Speaking of all of the crap flying, this should answer some of it:

http://www.overclock.net/amd-cpus/1107646-bulldozer-pre-launch-faq.html

This post has a lot of words but says nothing. Let me re-cap:

BD speed is unknown
BD release is unknown
You will know when it's available

Awesome. All our questions are cleared-up now...

Edit: I apologize, but I don't see the point of linking this. You could just have easily said that in a sentence or two. (You may already have - this thread is rediculously long!)

 

[Arkaign]

Oh boy. That post by JF-AMD has some rather worrying things towards the end, whereby he basically says that IPC by itself is meaningless, and then says if you're worried about single-thread performance to go buy a single-core. Ugh.

Given that he presumably knows more than we do, I'd surmise that indeed as many expect : BD will handle loads that can heavily multithread fairly decently and maybe even better than most of the current Intel setups, but will still fall behind in IPC and per-core performance.

 

[exar333]

Oh boy. That post by JF-AMD has some rather worrying things towards the end, whereby he basically says that IPC by itself is meaningless, and then says if you're worried about single-thread performance to go buy a single-core. Ugh.

Given that he presumably knows more than we do, I'd surmise that indeed as many expect : BD will handle loads that can heavily multithread fairly decently and maybe even better than most of the current Intel setups, but will still fall behind in IPC and per-core performance.

I think I will need a day off work just to read the huge amount of reviews that will be all posted at the same time following the official launch.

 

[Abwx]

But for the desktop IPC is the most important element . A quad on the desktop is more than enough.

By your own words, provided it s good enough , IPC is not enough
if so much cores are necessary.
Even a dual core is already an evidence that IPC is not enough,
even if very good, since doubling the core count will yield way
more performance than the traditionnal almost simple digit percentages
of IPC improvements...

 

[Abwx]

Oh boy. That post by JF-AMD has some rather worrying things towards the end, whereby he basically says that IPC by itself is meaningless, and then says if you're worried about single-thread performance to go buy a single-core. Ugh.

Given that he presumably knows more than we do, I'd surmise that indeed as many expect : BD will handle loads that can heavily multithread fairly decently and maybe even better than most of the current Intel setups, but will still fall behind in IPC and per-core performance.

This IPC saga has been stretched as much as the current states debts...

Some reminder for the skeptics :

Michael Golden presented the paper, and his quote about the clock characteristics of these tightly knit units is as follows: The out-of-order scheduler must efficiently pick up to four ready instructions for execution and wake up dependent instructions so that they may be picked in the next cycle. The execution units must compute results in a single cycle and forward them to dependent operations in the following cycle. All of this is required so that the module gives high architectural performance, measured in the number of instructions completed per cycle (IPC).

http://www.pcper.com/reviews/Proces...ocessors/Integer-Scheduler-and-Execution-Unit

 

[Nemesis 1]

By your own words, provided it s good enough , IPC is not enough
if so much cores are necessary.
Even a dual core is already an evidence that IPC is not enough,
even if very good, since doubling the core count will yield way
more performance than the traditionnal almost simple digit percentages
of IPC improvements...

So your saying that BD will beat SB in the majority of review benchmarks . Its a possiability only if the reviewes change benchmarks . Anthing more than 4 cores on desktop is a waste to masjority of users. IPC is the real deal always has been always will be . Except to a small amd fanboy base who want a 100 core cpu to run 4 threaded apps . LOL

 

[Abwx]

IPC is the real deal always has been always will be . Except to a small amd fanboy base who want a 100 core cpu to run 4 threaded apps . LOL

So say the guy that did unrelentlessly praise the P4.....

 

[arredondo]

I don't have a preference, but don't understand one thing about the debate - is there a qualitative difference for the user between having a few cores or a lot of cores if the output-per-dollar spent is roughly the same?

For example, if I buy a SB processor from Intel that comfortably gives me 4.5 Ghz for $200 after OC'ing it, and buy a BD processor from AMD that achieves the same OC speed at the same price, does it matter how either one does it? Assume 99% of the games I play with those two chips produce the same frame rates, and that the motherboard I bought with them also were comparable in price to each other.

What qualitative difference would it make as far as how many cores were involved to achieve that price-per-Ghz result if my games perform equally well with either? This is an honest question, because maybe there is something obvious I'm missing about the tech involved. If not, then it seems obvious to me that if I want a 12" pizza, it shouldn't matter much if I get it sliced into eight medium slices or twelve thinner slices - it is still the same amount of food.

 

[LOL_Wut_Axel]

IPC IS important, but it's not the only thing to take into account. IPC enhances both single and multi-threaded performance, while adding cores only enhances multi-threaded performance. At the same time, single-threaded performance nowadays is less relevant given the fact that most software is made to take advantage of two or more threads.

Games in general take advantage of three cores, with some only using two, and some four, and a minority using six. Given what we've seen this year and the past one, games are going multi-threaded. Then we have things that have been multi-threaded for a long time like image editing, video encoding, content creation, 3D rendering, file compression/decompression, compiling, and math calculations (Excel). So, then, what are we left with? Audio encoding. As of now, it's the only application that takes advantage of a single thread.

The reason why the 2500K is so similar to the 2600K in gaming is that games do not make use of HyperThreading, and most games, like I said earlier, only take advantage of two-four cores. This is where IPC matters. But once you get to applications that use more than four threads, it starts to matter less since by then software is highly parallel in its execution.

If you only care about gaming, you should be forgetting about the FX-6100 and above already. Likewise for Intel, where buying anything higher than a 2500K for gaming makes no sense. At the same time, unless you're doing audio encoding all day, single-threaded performance isn't the end-all, be-all. If you're gaming, IPC is still important because software is only mildly multi-threaded and four faster cores will do better than six slower ones. But then we have all the multi-threaded applications, which make up the majority (in number of applications available, not users necessarily).

What am I getting to? Well, here's what you should expect: Bulldozer will certainly not match Sandy Bridge in IPC. It also won't match it single-threaded performance unless they can do 5GHz on average. That means it'll be slower than the Core i5s in gaming since they only take advantage for the most part of two-four cores. Taking into account the same IPC as Llano, that means the FX-8000s would match up to the Core i7s in multi-threaded programs. Taking into account 20% higher IPC than Llano, or comparable to Nehalem, they would lose slightly in gaming and be much faster in most else, like the Core i7 990X.

 

[Plimogz]

What about the server space, though? Afterall, this is being fuelled by an overclock.net forum post from JFAMD. Who, as he's in the habit of reminding us on occasion, is an (The?) AMD server guy.

So what if IPC is low in the absolute, if the thing clocks to high heaven? And who's an average desktop gamer to say that anything over 4 cores is excessive or useless? when there are obviously enterprise users out there with deep-pockets and an appetite for as many cores as they can fit within already limited rack-space / thermal envelope.

I hope that the 2 module / 4 core model is all harvested failed die and AMD is going ahead with another round of core unlocking shenanigans, that I might sup on the crumbs which fall from the enterprise giants' table.

 

[GaiaHunter]

The thing with multiple cores is that it either use them or it doesn't.

Additionally it also depends on what can be split into the different threads - which means adding 2 more cores might just add 30% more performance, for example.

The higher core (slower) count vs lower core (faster) count debate basically boils to people not liking things that are faster sometimes but are slower other times.

Just look at the E8X00 vs Q6600 - when the programs use 2 or less threads the E8x00 is the best choice, but when the program use 3 or more threads, the Q6600 is better.

Of course this is all pointless until we know the speed of BD in 1-4 threads and eventually programs will make use of more threads.

So, in the case you mentioned both SB and BD delivered the same performance, you would be silly to not either by the cheaper or the one with more cores.

 

[LOL_Wut_Axel]

What about the server space, though? Afterall, this is being fuelled by an overclock.net forum post from JFAMD. Who, as he's in the habit of reminding us on occasion, is an (The?) AMD server guy.

So what if IPC is low in the absolute, if the thing clocks to high heaven? And who's an average desktop gamer to say that anything over 4 cores is excessive or useless? when there are obviously enterprise users out there with deep-pockets and an appetite for as many cores as they can fit within already limited rack-space / thermal envelope.

I hope that the 2 module / 4 core model is all harvested failed die and AMD is going ahead with another round of core unlocking shenanigans, that I might sup on the crumbs which fall from the enterprise giants' table.

Unlocking is, and will always be, a hit or miss. Some of the CPUs end up with less cores to fill demand in lower price points; some end up with less cores to both fill demand in lower price points AND because they have no other choice due to some part being defective. As of now unlocking is around 60% fail/40% succeed from what I can see.

In any case, the FX series of Bulldozer is not meant for servers. It's for clients, hence the high clock speeds and unlocked CPU multipliers.

 

[Riek]

So your saying that BD will beat SB in the majority of review benchmarks . Its a possiability only if the reviewes change benchmarks . Anthing more than 4 cores on desktop is a waste to masjority of users. IPC is the real deal always has been always will be . Except to a small amd fanboy base who want a 100 core cpu to run 4 threaded apps . LOL

Err... sure thing boss...
Care to tell me why ipc is more dominant than clockspeed in those applications?
4 threads -> SB 3,4-3,5.
4threads FX8 (rumoured clocks) 4,2GHz (full turboboost when half modules are in use).
Thats a +20% clock advantage. so how is ipc more dominant than lets say clock speed?


I will give you the answer, it isn't!
Ipc (which isn't even a fixed number...) is just as relevant as clockspeeds in applications. Yoiu can't determine the performance with only one metric for a given application.


Also this is again a dumb discussion.... we have no id of official clocks nor about ipc/application.

 

[LOL_Wut_Axel]

So your saying that BD will beat SB in the majority of review benchmarks . Its a possiability only if the reviewes change benchmarks . Anthing more than 4 cores on desktop is a waste to masjority of users. IPC is the real deal always has been always will be . Except to a small amd fanboy base who want a 100 core cpu to run 4 threaded apps . LOL

Right...

You might want to know that there's more applications than gaming and audio encoding. If you're gaming only, anyway, you'd be a fool to get anything higher than a Core i5.

 

[RussianSensation]

Err... sure thing boss...
Care to tell me why ipc is more dominant than clockspeed in those applications?
4 threads -> SB 3,4-3,5.
4threads FX8 (rumoured clocks) 4,2GHz (full turboboost when half modules are in use).
Thats a +20% clock advantage. so how is ipc more dominant than lets say clock speed?


I will give you the answer, it isn't!
Ipc (which isn't even a fixed number...) is just as relevant as clockspeeds in applications. Yoiu can't determine the performance with only one metric for a given application.


Also this is again a dumb discussion.... we have no id of official clocks nor about ipc/application.

You are 100% correct that IPC on its own is not that important. However, most of us who keep bringing IPC have considered why in this specific instance it is important after taking certain points into consideration:

1) BD and SB = both built on 32nm
2) SB overclocks extremely well. It's unlikely that BD will overclock to 5.5-6.0ghz on 32nm node
3) Most programs do not benefit from more than 4 cores. In fact, most programs are still using 1-3 threads.

Therefore, under these 3 points, IPC suddenly becomes extremely critical. IPC wouldn't be relevant IF BD had some magical overclocking headroom, SB had mediocre overclocking, or 80% of programs used 6-8 threads, etc.

I want to point this out so that people don't continue to dismiss IPC as "irrelevant". In the context of BD vs. SB for us enthusiasts who overclock, IPC becomes extremely relevant. This isn't a situation of Pentium 4 vs. Athlon 64 where Pentium 4's massive IPC deficit was somewhat offset by its 80-100% higher frequencies. SB has 30-40% overclocking headroom well into 4.5-4.6ghz range. I don't expect BD to have any significant frequency advantage when comparing both of these processors in their overclocked states.

More so, any frequency advantage that BD may have out of the gate is likely to shrink when Intel launched IB with even higher clock speeds and even more overclocking potential. And again, the most important differentiators will become IPC and # of cores, NOT frequency. Imo, IPC can only be dismissed if the frequencies of the 2 processors in questions are very far apart, which is not the case between BD and SB.

 

[Dresdenboy]

Err... sure thing boss...
Care to tell me why ipc is more dominant than clockspeed in those applications?
4 threads -> SB 3,4-3,5.
4threads FX8 (rumoured clocks) 4,2GHz (full turboboost when half modules are in use).
Thats a +20% clock advantage. so how is ipc more dominant than lets say clock speed?


I will give you the answer, it isn't!
Ipc (which isn't even a fixed number...) is just as relevant as clockspeeds in applications. Yoiu can't determine the performance with only one metric for a given application.

To add to this:

SB has a clear advantage in IPC in single threaded mode (per core that is) compared to BD running one thread per module. But some (or more?) ppl discussing this topic seem to look at it this way:

A SB core has a higher IPC than a single FX core. And SB can run 8 threads too. So it must be faster overall, mustn't it? There is a fallacy: If a SB core runs 2 threads it likely achieves more like 50-70% of single threaded IPC per thread.

Sure, the core has a combined IPC of 100-140% of single threaded IPC in these cases, but these 8 threads would run at 4-5.6X the single threaded performance in this example. Turbo headroom would also be lower. So if 2 BD cores achieve e.g. 170% of single threaded performance (yes, we heard different numbers). For about the same performance in this example, a SB core has to be only 1.2X to 1.7X faster than a single BD core. Different scaling behaviour of different architectures at play.

We'll see how this will play out.

 

[Riek]

You are 100% correct that IPC on its own is not that important. However, most of us who keep bringing IPC have considered why in this specific instance it is important after taking certain points into consideration:

1) BD and SB = both built on 32nm
2) SB overclocks extremely well. It's unlikely that BD will overclock to 5.5-6.0ghz on 32nm node
3) Most programs do not benefit from more than 4 cores. In fact, most programs are still using 1-3 threads.

Therefore, under these 3 points, IPC suddenly becomes extremely critical. IPC wouldn't be relevant IF BD had some magical overclocking headroom, SB had mediocre overclocking, or 80% of programs used 6-8 threads, etc.

I want to point this out so that people don't continue to dismiss IPC as "irrelevant". In the context of BD vs. SB for us enthusiasts who overclock, IPC becomes extremely relevant. This isn't a situation of Pentium 4 vs. Athlon 64 where Pentium 4's massive IPC deficit was somewhat offset by its 80-100% higher frequencies. SB has 30-40% overclocking headroom well into 4.5-4.6ghz range. I don't expect BD to have any significant frequency advantage when comparing both of these processors in their overclocked states.

More so, any frequency advantage that BD may have out of the gate is likely to shrink when Intel launched IB with even higher clock speeds and even more overclocking potential. And again, the most important differentiators will become IPC and # of cores, NOT frequency. Imo, IPC can only be dismissed if the frequencies of the 2 processors in questions are very far apart, which is not the case between BD and SB.

I already indicated that, according to the rumours, there are clockspeed differences.

Out of the gate BD will clock up to 20% higher around 3 to 4 threads. Even for 1-2 threads BD will have a +12% clock advantage according to the rumoured clocks. (ofcourse we compare a 125W model now, butthose clocks will also be reached for the next batch).

For reference, if BD core reaches nehelam ipc it will be faster in 3-4 threads while trading blows in 1-2 threads.

Overclocking is irrelevant for the business. SB also exceeds its thermal envelop when overclocked. As seen with the rumoured SB-E quad core, which will lower the clockspeed advantage of BD FX8 by a 5% in 3-4 threads.

So Intel cannot launch much higher clockfrequencies either.

We have no id how good or bad BD will overclock.

 

[psolord]

Ah, I was like you... once... before my expectations of BD became subterranean. :'(

I'm still hoping for the best, but expecting a CPU which will be chewed up and spit out by SB-E in every benchmark under the sun. Whenever it is that they will finally get around to releasing it.

This may be the case, but prices should also be taken into consideration. How much will SB-E cost as a platform and how much a BD platform?

Also the unlocked state of all the FX processors should be taken into consideration, since from what was published recently, SB-E unlocked cpus will start at 500$+! You will probably be able to get the whole BD platform (+ram) with a FX-8150 for that kind of money.

 

[psolord]

You are 100% correct that IPC on its own is not that important. However, most of us who keep bringing IPC have considered why in this specific instance it is important after taking certain points into consideration:

1) BD and SB = both built on 32nm
2) SB overclocks extremely well. It's unlikely that BD will overclock to 5.5-6.0ghz on 32nm node
3) Most programs do not benefit from more than 4 cores. In fact, most programs are still using 1-3 threads.

Therefore, under these 3 points, IPC suddenly becomes extremely critical. IPC wouldn't be relevant IF BD had some magical overclocking headroom, SB had mediocre overclocking, or 80% of programs used 6-8 threads, etc.

I want to point this out so that people don't continue to dismiss IPC as "irrelevant". In the context of BD vs. SB for us enthusiasts who overclock, IPC becomes extremely relevant. This isn't a situation of Pentium 4 vs. Athlon 64 where Pentium 4's massive IPC deficit was somewhat offset by its 80-100% higher frequencies. SB has 30-40% overclocking headroom well into 4.5-4.6ghz range. I don't expect BD to have any significant frequency advantage when comparing both of these processors in their overclocked states.

More so, any frequency advantage that BD may have out of the gate is likely to shrink when Intel launched IB with even higher clock speeds and even more overclocking potential. And again, the most important differentiators will become IPC and # of cores, NOT frequency. Imo, IPC can only be dismissed if the frequencies of the 2 processors in questions are very far apart, which is not the case between BD and SB.

Bravo. Well said!

That's what I wanted to answer to our good friend JF (thanks for the OC post btw ) regarding his out of context IPC approach. It's not really out of context, since when we are talking about IPC, we have the SB cpus in mind along with all their technical aspects, including great clocks and great power draw. This should be clear and not needed to be mentioned in every IPC related post.

 

[Abwx]

2) SB overclocks extremely well. It's unlikely that BD will overclock to 5.5-6.0ghz on 32nm node

You should be cautious before making such statements...

AMD increased the frequency of their 45nm phenom2 by as much
as 23% at equal core count , so with a mature 32nm process we
can expect the max turbo frequency to be 20% higher than at
launch , wich is expected to be already at 4.2ghz , making
the 5ghz at stocks quite a certainity in future steppings.

 

[PreferLinux]

You should be cautious before making such statements...

AMD increased the frequency of their 45nm phenom2 by as much
as 23% at equal core count , so with a mature 32nm process we
can expect the max turbo frequency to be 20% higher than at
launch , wich is expected to be already at 4.2ghz , making
the 5ghz at stocks quite a certainity in future steppings.

Have you forgotten that the same applies to Intel/SB?