Rumour: Bulldozer 50% Faster than Core i7 and Phenom II.

[purefun1965]

I dont think that chip you got is final stepping.

http://semiaccurate.com/forums/showpost.php?p=109662&postcount=74

that makes sense thanks for the link

 

[purefun1965]

that makes sense thanks for the link

as i said this is a test to kick the tires. we were told we have new samples coming friday.

 

[purefun1965]

i called my boss he said we got the wrong ones. he said as gimpr stated we will get a1 stepping and a new bios for the boards

 

[purefun1965]

sorry for the confusion. I will retest friday if all goes well.

 

[purefun1965]

well look at it this way it did damn good considering a bo stepping and 2 semi working bioses. I wish i could show you guys the benchmarks. they weren't as bad as i thought they would be.

 

[Tuna-Fish]

Any chance you can do a little test for us?

    .text
.globl main
    .type	main, @function
main:
    xorq    %r11, %r11
    movq    %rbp, -8(%rsp)
    movq    %rsp, %r10
.ALIGN 32
.L3:
    incl    %eax
    decl    %ebx
    addl    $1, %ecx
    addl    $1, %edx
    incl    %esi
    decl    %edi
    addq    $1, %r9
    addq    $1, %r8
    
    incl    %eax
    decl    %ebx
    addl    $1, %ecx
    addl    $1, %edx
    incl    %esi
    decl    %edi
    addq    $1, %r9
    addq    $1, %r8
    
    incl    %eax
    decl    %ebx
    addl    $1, %ecx
    addl    $1, %edx
    incl    %esi
    decl    %edi
    addl    $1, %esp
    addl    $1, %ebp
    
    incl    %eax
    decl    %ebx
    addl    $1, %ecx
    addl    $1, %edx
    incl    %esi
    decl    %edi
    addq    $1, %r9
    addq    $1, %r8
    
    incl    %eax
    decl    %ebx
    addl    $1, %ecx
    addl    $1, %edx
    incl    %esi
    decl    %edi
    addq    $1, %r9
    addq    $1, %r8
    
    incl    %eax
    decl    %ebx
    addl    $1, %ecx
    addl    $1, %edx
    incl    %esi
    decl    %edi
    addl    $1, %esp
    addl    $1, %ebp
    
    incl    %eax
    decl    %ebx
    addl    $1, %ecx
    addl    $1, %edx
    incl    %esi
    decl    %edi
    addq    $1, %r9
    addq    $1, %r8
    
    incl    %eax
    decl    %ebx
    addl    $1, %ecx
    addl    $1, %edx
    incl    %esi
    decl    %edi
    addq    $1, %r9
    addl    $1, %r11d
    
    incl    %eax
    decl    %ebx
    addl    $1, %ecx
    addl    $1, %edx
    incl    %esi
    decl    %edi
    addl    $1, %esp
    addl    $1, %ebp
   
    
    incq    %rax
    decq    %rbx
    addl    $1, %ecx
    addl    $1, %edx
    incq    %rsi
    addq    $1, %rbx

.L2:
    cmpl	$499999999, %r11d
    jle	.L3
    movq    %r10, %rsp
    movq    -8(%rsp), %rbp
    xorl	%eax, %eax
    ret

It's gnu assembly for linux x86_64. Save it as test.s
compile with: gcc bench.s
run with: time ./a.out
on an otherwise idle system, and report the time it output and the clockspeed the machine is running on.

It's pretty useless for evaluating performance, so running it might be kosher. The only thing it'll tell us is the execution width of the processor when running (inc dec add add)

 

[Idontcare]

charts tell the whole story do they?

Of course they don't, I was asking a question and merely using the chart as a speaking tool.

The chart does add more to the discussion versus just making open-ended unsupported statements like:

It is designed with the server market in mind. AMD'S bread and butter.

 

[RyanGreener]

If what purefun1965 is saying is true, then this is definitely intriguing....

 

[Martimus]

Is the server market really AMD's bread and butter? I thought their market share in serves was a mere 7%.

Trefis has the server segment as contributing only 20% to AMD's valuation.

https://www.trefis.com/company?hm=AMD.trefis&from=widget:forecast&ovd_urlid=451774#

While it is true that the Server market probably isn't AMDs current bread and butter, it is the low-hanging fruit they can pick to get more marketshare, since as you state, they only have a 7% marketshare. In fact the link you provided predicts AMD will take between 12% and 17% of the marketshare by years end, which will double their current marketshare. Combine that with the expected 10% growth in server hardware this year and their best case prediction gives AMD a 267% growth in server sales this year.

 

[Idontcare]

While it is true that the Server market probably isn't AMDs current bread and butter, it is the low-hanging fruit they can pick to get more marketshare, since as you state, they only have a 7% marketshare. In fact the link you provided predicts AMD will take between 12% and 17% of the marketshare by years end, which will double their current marketshare. Combine that with the expected 10% growth in server hardware this year and their best case prediction gives AMD a 267% growth in server sales this year.

I agree that servers were, once upon a time, AMD's bread and butter.

And if you asked me 5 months ago I would have agreed that AMD's bread and butter in the future was highly likely to once again be the server market.

But that expectation changed, for me at least, in Jan when AMD's BoD decided Dirk was not the kind of leadership that AMD needed for the future.

The future is fusion. It is things like bobcat and llano, tablets and so on. Or so they seem to be convinced as evidenced by the leadership change-ups of late.

But I do not follow revenue and margin breakouts for semi-companies any more, hence my surprise (and request for more definitive info) at the comment that server "is" AMD's bread and butter in the present-tense. If this is the case then that is/was news to me. That was all I was trying to communicate.

 

[Topweasel]

One thing in mind is Margin. Even at a smaller portion of company Dollars, it could be more profitable then their other segments (like Zacate) where pricing competition of both the market segment and competition have them sold at slightly higher then cost. If they can sell a desktop chip at 20 million units at a $200-$400 and 5 million $600-$800 with both of them costing $100 a chip, that would be 4 billion in profit vs 3 billion in profit on half as much money and a quarter of unit sales. That's on two different market segments sharing the same basic CPU also the two with the highest ASP.

Zacate and Llano while great techs need a lot more additional development and both are targeted at market segments with slim margins. If I remember originally the Atom was being sold at $8 a CPU.

The same if not more so would apply to AMD's current sales, well maybe pre 2011. Its the bread and butter because AMD is able to leverage their desktop large market CPU for higher ASP's better profits. It's why we got Thuban and its probably why AMD switched to this flexible Module design, they can in the server market pretty much just toss on "cores" at will. On the other end the can remove them at will to fill market segments that they might otherwise had trouble filling.

Before something like the Athlon II X2 might have taken a full rework of the core to reduce die size for that market segment.

 

[formulav8]

do you really think this? AMD wasted years and money to under perform their current arch? LET alone not be competitive with intel?

No, what I really think is that this BD design is VERY different from all past Intel/AMD designs that AMD wouldn't have much of a baseline to determine exactly how well this combination would perform or, maybe the other combination would be better, or ect. AMD could only rely on human logic and simulations. They don't exactly have anything to go by in the real world per-say, like you would if building upon a previous arch. Most current/past Intel/AMD chips have been built on previous generation chips. So while I agree that AMD knows much more than anyone of us on this forum, it just wouldn't have been 100% possible to know the outcome for some of the big decisions they made, do to BD being such a radical design change. BD looks very good and interesting on paper for sure. AND it probably Will perform just fine in all around. I am just the cynical type and always have been. So I just said I hope it doesn't end up being a P4. Its nothing more than a little opinion.

 

[Nemesis 1]

I agree that servers were, once upon a time, AMD's bread and butter.

And if you asked me 5 months ago I would have agreed that AMD's bread and butter in the future was highly likely to once again be the server market.

But that expectation changed, for me at least, in Jan when AMD's BoD decided Dirk was not the kind of leadership that AMD needed for the future.

The future is fusion. It is things like bobcat and llano, tablets and so on. Or so they seem to be convinced as evidenced by the leadership change-ups of late.

But I do not follow revenue and margin breakouts for semi-companies any more, hence my surprise (and request for more definitive info) at the comment that server "is" AMD's bread and butter in the present-tense. If this is the case then that is/was news to me. That was all I was trying to communicate.

I agree . Intel should go after Dirk he would be a great addition to intel .

 

[Topweasel]

No, what I really think is that this BD design is VERY different from all past Intel/AMD designs that AMD wouldn't have much of a baseline to determine exactly how well this combination would perform or, maybe the other combination would be better, or ect. AMD could only rely on human logic and simulations. They don't exactly have anything to go by in the real world per-say, like you would if building upon a previous arch. Most current/past Intel/AMD chips have been built on previous generation chips. So while I agree that AMD knows much more than anyone of us on this forum, it just wouldn't have been 100% possible to know the outcome for some of the big decisions they made, do to BD being such a radical design change. BD looks very good and interesting on paper for sure. AND it probably Will perform just fine in all around. I am just the cynical type and always have been. So I just said I hope it doesn't end up being a P4. Its nothing more than a little opinion.

Keep in mind that AMD did that once. They created a processor that under the different compromises they made did not perform well or did not have a great out look. They shelved it and in turn reworked the K7/K8 cores for the K10. Considering the markets that AMD is moving into the same year as the release of this new design it would have been easier for them to make tweaks (even major ones) to the current design and completely gives up the "performance sector". Much like the K10 they could have done this in 2009 when it would have been easy to see the limitations of the design and be ready by 2013 for a different architecture. They didn't. Which to me means that I doubt that its slower at all.

 

[Dribble]

I agree that servers were, once upon a time, AMD's bread and butter.

And if you asked me 5 months ago I would have agreed that AMD's bread and butter in the future was highly likely to once again be the server market.

But that expectation changed, for me at least, in Jan when AMD's BoD decided Dirk was not the kind of leadership that AMD needed for the future.

The future is fusion. It is things like bobcat and llano, tablets and so on. Or so they seem to be convinced as evidenced by the leadership change-ups of late.

I think it's safe to safe BD was designed with Dirk's direction in mind being as it takes many years to design a processor and Dirk was only fired a few months ago.

If it were a desktop part then it would try to provide 4 cores with the highest IPC possible, and have on-chip graphics. That's clearly not the case - from what I have read it seems we have lots of low IPC cores (up to 16 it seems). imo it's a server part first and foremost. Llano is the desktop part.

 

[Martimus]

I think it's safe to safe BD was designed with Dirk's direction in mind being as it takes many years to design a processor and Dirk was only fired a few months ago.

If it were a desktop part then it would try to provide 4 cores with the highest IPC possible, and have on-chip graphics. That's clearly not the case - from what I have read it seems we have lots of low IPC cores (up to 16 it seems). imo it's a server part first and foremost. Llano is the desktop part.

I think that is what IDC is trying to say. Dirk got fired because he prioritized developing a new server part (Bulldozer) over a new mainstream and portable part (Llano and Brazos). Both were being developed, but the BoD believes that there is more money to be made in the latter than the former. I can't say that I agree with them, if only because they have been proven wrong often in the past.

edit: The other possibility is that he let the server portion of the company lose so much market share, and that AMD was being buoyed by the lower margin products. It does seem that AMD has lost more and more server market share since Dirk Meyer took over as CEO. Of course, it also has shown a profit nearly every quarter he has been CEO, unlike his predecessor. The bottom line looks much better than the shrinking marketshare.

 

[HW2050Plus]

Any chance you can do a little test for us?

It is not so good to use 32 Bit operations in 64 Bit code since that might cause false dependencies, though you spread the operations widly over registers.

Changed to 64 Bit only, no false dependencies, I also changed the immediate operations to register operations to not run into decoding issues by larger immediate values.

.text
.globl main
.type main, @function
main:
xorq %r11, %r11
movq %rbp, -8(%rsp)
movq %rsp, %r10
movq $1, %r15
movq $499999999, %r14
.ALIGN 32
.L3:
incq %rax
decq %rbx
addq %r15, %rcx
addq %r15, %rdx
incq %rsi
decq %rdi
addq %r15, %r9
addq %r15, %r8
 
incq %rax
decq %rbx
addq %r15, %rcx
addq %r15, %rdx
incq %rsi
decq %rdi
addq %r15, %r9
addq %r15, %r8
 
incq %rax
decq %rbx
addq %r15, %rcx
addq %r15, %rdx
incq %rsi
decq %rdi
addq %r15, %rsp
addq %r15, %rbp
 
incq %rax
decq %rbx
addq %r15, %rcx
addq %r15, %rdx
incq %rsi
decq %rdi
addq %r15, %r9
addq %r15, %r8
 
incq %rax
decq %rbx
addq %r15, %rcx
addq %r15, %rdx
incq %rsi
decq %rdi
addq %r15, %r9
addq %r15, %r8
 
incq %rax
decq %rbx
addq %r15, %rcx
addq %r15, %rdx
incq %rsi
decq %rdi
addq %r15, %rsp
addq %r15, %rbp
 
incq %rax
decq %rbx
addq %r15, %rcx
addq %r15, %rdx
incq %rsi
decq %rdi
addq %r15, %r9
addq %r15, %r8
 
incq %rax
decq %rbx
addq %r15, %rcx
addq %r15, %rdx
incq %rsi
decq %rdi
addq %r15, %r9
addq %r15, %r11
 
incq %rax
decq %rbx
addq %r15, %rcx
addq %r15, %rdx
incq %rsi
decq %rdi
addq %r15, %rsp
addq %r15, %rbp
 
 
incq %rax
decq %rbx
addq %r15, %rcx
addq %r15, %rdx
incq %rsi
addq %r15, %rbx
 
.L2:
cmpq %r14, %r11
jle .L3
movq %r10, %rsp
movq -8(%rsp), %rbp
xorq %rax, %rax
ret

 

[Edgy]

I think that is what IDC is trying to say. Dirk got fired because he prioritized developing a new server part (Bulldozer) over a new mainstream and portable part (Llano and Brazos). Both were being developed, but the BoD believes that there is more money to be made in the latter than the former. I can't say that I agree with them, if only because they have been proven wrong often in the past.

If indeed Dirk got fired because of BD being focused on Server and the Board felt that future is all about renaissance of going small (smartphones/tablets) - I find that to be such a short-sighted egregious error in judgement on the Board's part in my opinion...

When I look at Atom and how excellent it was for Intel for the boom of netbook market - what I see is a product that pretty much had the booming market by itself with nary a competition product - yet, the return in terms of profits for Intel while were good, were not spectacular especially considering that it had no competition and dominated a booming netbook sector.

Hell, even Intel's CFO was quoted to referring the Atom as a "recession product" which I take it to mean, volume oriented low cost, relatively slim/flat margin.

What I see on smartphone/tablet/netbook areas is NOT the chipmakers making SPECTACULAR profits but rather the end-user product designers/manufacturers like Apple and Motorola (hell Droids even saved Motorola from going under) taking the lionshare of profits.

I think the market that chipmakers such as AMD and Intel traditionally make the "spectacular" margins are the Servers.

The math is ridiculously simple (this is an example of AMD's case in the past):
An Athlon dual core 2.8 GHz chip would sell for let's say $100.
Well, the same chip with couple more HT enabled & named Opteron would sell for $999 - a 4/8 way HT enabled chip can sell for $2000. I'm sure similar correlations exists today for Intel and to lesser extent for AMD.

Server market also has much less bias (quite a bit exists but less than desktops/mobiles) toward Intel chips which have made Opterons breaking into the market lot easier even during netburst era when Intel was throwing money at people to avoid using AMD chips.

And most importantly, in my opinion, the ROI in the server side is far greater even at increment gains in market share. Simply put, gaining mere 10% marketshare on servers can potentially double/triple AMD's entire bottom line profits whereas gaining 20-30% marketshare with Brazos CPU chips while boosting total volume quite a bit, may not bring in much profits for AMD.

So yes - I agree that Fusion is future for AMD and volume sectors are important and critical but the area where AMD can quickly turn itself around for instant profits towards its bottom line is on the server side.

I dare say Dirk got his priorities correct, the board is full of idiots looking for headlines such as "AMD in tablet/smartphone" for short-term stock boost.

 

[drizek]

What are you people talking about? Brazos is awesome for mobile. Should it have come out two years ago? Sure, but I don't thunk it is fair to say that and doesn't have a mobile strategy.

 

[Phynaz]

What are you people talking about? Brazos is awesome for mobile. Should it have come out two years ago? Sure, but I don't thunk it is fair to say that and doesn't have a mobile strategy.

Brazos is good for laptops, not mobile from what I am aware of.

Mobile is generally defined as pocketable, although tables have somewhat changed that definition.

Blackberries and iPhones are what companies mean when they say mobile.

 

[OCGuy]

The article up on Fudzilla falls in line with what I think will be the case. Still much weaker CPU performance than i7, but better graphics.

Which to me means nothing, because I would never us a IG solution.

 

[Martimus]

The article up on Fudzilla falls in line with what I think will be the case. Still much weaker CPU performance than i7, but better graphics.

Which to me means nothing, because I would never us a IG solution.

Bulldozer doesn't have Integrated Graphics. Are you talking about Llano?

Also, Fuado has proven time and again that he doesn't know anymore about what he is talking about than the average person making random guesses about these things. He has consistantly just made stuff up that he thinks sound right, and publishes it.

 

[drizek]

Brazos is good for laptops, not mobile from what I am aware of.

Mobile is generally defined as pocketable, although tables have somewhat changed that definition.

Blackberries and iPhones are what companies mean when they say mobile.

And so how is Intel any better? They sold off XScale, and their atom chip is inferior to Brazos.

Intels advantage in mobile, if they have one, is an advantage over GloFlo and TSMC with their 3D 22nm stuff, not AMD per se.

 

[drizek]

Bulldozer doesn't have Integrated Graphics. Are you talking about Llano?

Also, Fuado has proven time and again that he doesn't know anymore about what he is talking about than the average person making random guesses about these things. He has consistantly just made stuff up that he thinks sound right, and publishes it.

Llano +1 is going to be based on Bulldozer. Bulldozer is going to be used across the line, eventually, probably for a long time.

 

[podspi]

Brazos is good for laptops, not mobile from what I am aware of.

Mobile is generally defined as pocketable, although tables have somewhat changed that definition.

Blackberries and iPhones are what companies mean when they say mobile.

I personally believe AMD has less of a chance of getting into the mobile market than Intel does with Atom. We still don't know why Dirk got the ouster, and we may not know for a while (if ever), but if that turns out to be the reason why, the BoD are, well, not as bright as the CEO they kicked out to put it nicely...


I do however think Bulldozer is going to put up the good fight against the ARM incursion into servers. 64-cores in a 4P is pretty dense...