AMD Bulldozer Eng. Sample leaked, benched

[seronx]

Is that 50% more per core or...?
---
It's called speculating and discussing

50% more per cpu.... for Donhambier<--I am just going to call this guy the Swede yeesh

He is captain obvious
6 cores -> 8 cores = 50% more cores!!
4 cores/8 threads -> 8 cores = I don't even know....

Thats not really a prediction, thats a ballpark quote, a 25% - 75% performance increase statement (or prediction) pretty much tells me that noone has any idea of how much it will be better and is only marginally better than saying 0 - 100%, which in the end pretty much does make it speculative.

Had to tidy this up again, if some of you guys cant squabble with class then don't squabble at all.

No one knows but there was a Cray benchmark

32 core BD 1.8GHz 2P vs 24 core MC 1.9GHz 2P/48 core MC 1.9GHz 4P

and bulldozer was .6x-1.3x better

and it averaged .9x better (aka it performed worst)

http://arstechnica.com/business/new...enchmarks-may-give-glimpse-of-amds-future.ars

http://www.realworldtech.com/page.cfm?ArticleID=RWT033011040021

But these were Server workloads not desktop/gaming workloads

 

[repman244]

No one knows but there was a Cray benchmark

32 core BD 1.8GHz 2P vs 24 core MC 1.9GHz 2P

and bulldozer was .8x-1.21x better

and it averaged .8x better (aka it performed worst)

Do you maybe know what exactly does Cray benchmark measure? I'm a bit rusty with the benchmarks.

EDIT: Reading the second link I saw this:

The STREAM triad benchmark showed particularly poor memory system performance for Interlagos: ~6GB/s per socket, compared to ~27GB/s for Magny-Cours

And a bunch of other constraints, maybe the same thing is happening with these Zambezi ES leaks (very crippled chips), but that's just a guess.

 

[seronx]

Do you maybe know what exactly does Cray benchmark measure? I'm a bit rusty with the benchmarks.

Himeno: Effect of cache on performance and calculation size

Parallel BZIP2: SMP Data Compression Software

C-ray: Ray Tracing

Fast Fourier Transform: Linpack somewhat http://en.wikipedia.org/wiki/Fast_Fourier_transform

Jacobi SOR: I dunna know

Monte Carlo Pi: Area of a circle...?

the other two dunna know..

Sparse Matrix Multiply

Dense LU Factorizing

Bulldozer was better somewhat in...
Himeno, Bzip2, C-ray, and Pi

EDIT: Reading the second link I saw this:

The STREAM triad benchmark showed particularly poor memory system performance for Interlagos: ~6GB/s per socket, compared to ~27GB/s for Magny-Cours

And a bunch of other constraints, maybe the same thing is happening with these Zambezi ES leaks (very crippled chips), but that's just a guess.

ES are crippled...but what is crippled we dunna know

They made predictions of what is crippled but are there more crippled components?

 

[repman244]

Himeno: Effect of cache on performance and calculation size

Parallel BZIP2: SMP Data Compression Software

C-ray: Ray Tracing

Fast Fourier Transform: Linpack somewhat http://en.wikipedia.org/wiki/Fast_Fourier_transform

Jacobi SOR: I dunna know

Monte Carlo Pi: Area of a circle...?

the other two dunna know..

Sparse Matrix Multiply

Dense LU Factorizing

Bulldozer was better somewhat in...
Himeno, Bzip2, C-ray, and Pi

Thank you for that info, as you can see in my edited post above I think that is the reason for the weird scores.
And we don't know what else has been crippled and how good was the MB support for the chips.

It's very hard to say anything about BD performance based on those numbers.

 

[seronx]

Thank you for that info, as you can see in my edited post above I think that is the reason for the weird scores.
And we don't know what else has been crippled and how good was the MB support for the chips.

It's very hard to say anything about BD performance based on those numbers.

Especially desktop BD Performance since we do not know the clock speeds for desktop Bulldozers...

 

[Horrux]

And given that Cray benches are designed for server or supercomputer purposes, where memory bandwidth can be everything, it is obvious that the dual channel memory design of the upcoming desktop parts are crippled compared to the quad channel server parts. That is no surprise at all.

Still, saying that BD is Phenom II somewhat tweaked and shrunk is pure misinformation. You can't do that to a CPU. Well maybe in1980 you could but not today. When making such changes as the cores to modules that AMD is making, a complete redesign is required. I don't know much, but I know timings are one important issue at current scales and frequencies.

 

[seronx]

And given that Cray benches are designed for server or supercomputer purposes, where memory bandwidth can be everything, it is obvious that the dual channel memory design of the upcoming desktop parts are crippled compared to the quad channel server parts. That is no surprise at all.

Still, saying that BD is Phenom II somewhat tweaked and shrunk is pure misinformation. You can't do that to a CPU. Well maybe in1980 you could but not today. When making such changes as the cores to modules that AMD is making, a complete redesign is required. I don't know much, but I know timings are one important issue at current scales and frequencies.

I am talking about the Core....mainly the core -> Integer/Memory specifically

Bulldozer as a Module doesn't look like Phenom II
but,
Bulldozer cores as a Integer Core compared to Phenom IIs integer core

is basically the change from
VLIW5 to VLIW4
Same architecture, just one is easier to code(edit: or should I say more efficient?) for *cough*VLIW4*cough*

Bulldozer 2C-> 3 Unified Pipelines over Phenoms II 1C 6 Dedicated Pipelines any day

6GB/s is pretty bad vs 27GB/s

 

[Thatguy]

I am talking about the Core....mainly the core -> Integer/Memory specifically

Bulldozer as a Module doesn't look like Phenom II
but,
Bulldozer cores as a Integer Core compared to Phenom IIs integer core

is basically the change from
VLIW5 to VLIW4
Same architecture, just one is easier to code(edit: or should I say more efficient?) for *cough*VLIW4*cough*

Bulldozer 2C-> 3 Unified Pipelines over Phenoms II 1C 6 Dedicated Pipelines any day

6GB/s is pretty bad vs 27GB/s

Bulldozer is not PII Stop saying it is.

 

[seronx]

Bulldozer is not PII Stop saying it is.

I'm going to be blunt....

Bulldozer is K7....

It has a lot more stuff going for it than just being another dirty K7 clone like

K8->K10->K10.5

It didn't help AMD having an architecture going from 1999 to 2011, that had no optimization in core design

Netburst -> Core2 -> Core i7 -> Core i7 2nd gen

Were all basically improved core designs....

K7 vs Netburst, Loss
K8 vs Core 2, Loss
K10/K10.5 vs Nehalem, Loss
K15 vs Sandy Bridge = Loss?

The nature of things are that the AMD will lose regardless....

And since the redesign still has the image of the K7, well...

It's a lot more efficient than K7 dropped that lousy 3rd ALU/AGU and 3rd FPU for
A 2ALU/AGU design + 1FPU that can address 2 cores



But, it still has that lousy AMD spirit, "If isn't broken, don't fix it"

 

[repman244]

I'm going to be blunt....

Bulldozer is K7....

It has a lot more stuff going for it than just being another dirty K7 clone like

K8->K10->K10.5

It didn't help AMD having an architecture going from 1999 to 2011, that had no optimization in core design

Netburst -> Core2 -> Core i7 -> Core i7 2nd gen

Were all basically improved core designs....

K7 vs Netburst, Loss
K8 vs Core 2, Loss
K10/K10.5 vs Nehalem, Loss
K15 vs Sandy Bridge = Loss?

The nature of things are that the AMD will lose regardless....

And since the redesign still has the image of the K7, well...

It's alot more efficient than k7 dropped that lousy 3rd ALU/AGU and 3rd FPU unit for
A 2ALU/AGU design + 1FPU that can address 2 cores

Wasn't it K8 vs. Netburst, where Netburst (Pentium 4 and the horrible Pentium D) got raped by Athlons?

 

[seronx]

Wasn't it K8 vs. Netburst, where Netburst (Pentium 4) got raped by Athlons?

1999 K7
2000 Netburst
2003 K8
2006 Core 2
2007 K10
2008 Nehalem
2009 K10.5
2011 Sandy Bridge
2011 Bulldozer
2013 Haswell

Well actually this makes it looks even more tragic

 

[renq]

Netburst -> Core2 -> Core i7 -> Core i7 2nd gen

You forgot 1st gen Core CPUs, that, though being only for laptops etc, brought Intel out of the slump/dead end Netburst had lead it to.

 

[Red_Machine]

Netburst -> Core2 -> Core i7 -> Core i7 2nd gen

Actually, the Core series is basically a die-shrunk Pentium M, which is basically a die-shrunk Pentium III with the Netburst FSB bolted on. The Netburst architecture itself was discontinued after the Pentium D.

 

[repman244]

1999 K7
2000 Netburst
2003 K8
2006 Core 2
2007 K10
2008 Nehalem
2009 K10.5
2011 Sandy Bridge
2011 Bulldozer
2013 Haswell

Well actually this makes it look even more tragic

Lol what? K8 went vs Netburst, remember Athlon64, FX days? Pentiums were getting raped even later Pentium D was getting raped by Athlon x2's

 

[seronx]

Lol what? K8 went vs Netburst, remember Athlon64, FX days? Pentiums were getting raped even later Pentium D was getting raped by Athlon x2's

No it was the opposite...

In many tests the Athlon 64 FX62 performs better than the Core 2 Duo E6400. Still, the FX-62 is slower than the Core 2 Duo E6600, E6700 and X6800 microprocessors.

http://www.cpubenchmark.net/cpu.php?cpu=AMD+Athlon+64+FX-62+Dual+Core

http://www.cpubenchmark.net/cpu.php?cpu=Intel+Core2+Duo+E6600+@+2.40GHz

The Architecture is meant to compete downward not upward

K15 -> Haswell

 

[repman244]

No it was the opposite...



http://www.cpubenchmark.net/cpu.php?cpu=AMD+Athlon+64+FX-62+Dual+Core

http://www.cpubenchmark.net/cpu.php?cpu=Intel+Core2+Duo+E6600+@+2.40GHz

Read again what I wrote. I did not mention Core 2 duo's.
K8 was to counter Netburst AFAIK

 

[seronx]

Read again what I wrote. I did not mention Core 2 duo's.
K8 was to counter Netburst AFAIK

1999 K7
^
2000 Netburst
v
2003 K8
v
2006 Core 2
^
2007 K10
^
2008 Nehalem
^
2009 K10.5
^^
2011 Sandy Bridge
^^
2011 Bulldozer
????
2013 Haswell

look again



Were off topic again

K7 = Netburst
K8 = Core2
Then what the... AMD?!!?!?!
K10 = Netburst/Core2
K10.5 = Core2/Nehalem

AMD was competing downward while Intel was riding them

 

[repman244]

1999 K7
^
2000 Netburst
v
2003 K8
v
2006 Core 2
^
2007 K10
^
2008 Nehalem
^
2009 K10.5
^^
2011 Sandy Bridge
^^
2011 Bulldozer
????
2013 Haswell

look again



Were off topic again

I see that for 3 years Intel was getting raped, I don't know why are you saying K8 is vs Core2. It was the other way around, Intel released Core 2 to counter K8 (successfully I have to say)
You cannot deny AMD was the best at that time, and the only time (as long as I remember) they had chips at $1000.

 

[seronx]

I see that for 3 years Intel was getting raped, I don't know why are you saying K8 is vs Core2. It was the other way around, Intel released Core 2 to counter K8 (successfully I have to say)
You cannot deny AMD was the best at that time, and the only time (as long as I remember) they had chips at $1000.

I'm stopping until a production pilot Bulldozer leak

Edit: I am back and well

Athlon 64/K8 and FX/K15

Have a lot in common

OH LAWD IT BLEW UP!

 

[Horrux]

Am I the only one who feels seronx is spreading FUD and being a bit of a jerk?

 

[cadaveca]

Not everyone's sense of humour is the same.

 

[Velvet Wafer]

I smell severe Intel fanboyism, covered by what should look like real knowledge, in a thread, where it does not belong.

You crap this thread severely, so, please stay silent,as you have promised, until AMD has released some silicone, thats worth looking at,
which will give at least some useful benches, without myriads of bugs

Not everyone's sense of humour is the same.

Yeah, some people just have a very bad one.... thats simply provocative, without sense.

 

[xenocide]

I smell severe Intel fanboyism, covered by what should look like real knowledge, in a thread, where it does not belong.

You crap this thread severely, so, please stay silent,as you have promised, until AMD has released some silicone, thats worth looking at,
which will give at least some useful benches, without myriads of bugs

He made a few good points at first, but it has quickly diminished into speculation and opinion.

 

[Melvis]

I'm stopping until a production pilot Bulldozer leak

http://img39.imageshack.us/img39/603/dozer20fire.jpg

OH LAWD IT BLEW UP!

It just couldn't handle the Awesomeness

 

[TheoneandonlyMrK]

im now pondering when and what form the bulldozer enhanced will ship as they were rumoured to be not long after the originally earlier bulldozer launch BD Q2 Q3 enhanced Q4 -Q1 2012 but what now that BD has slipped the Q's .