Bookmark and Share


Analysts and market observers have expected Advanced Micro Devices' code-named Bulldozer microprocessors to tangibly boost performance compared to the company's today's chips. But while from architectural standpoints Bulldozer looks impressive, the company itself does not make claims about extraordinary performance improvements. In fact, "per-core" performance of the new Bulldozer-based processors will be only slightly higher compared to contemporary chips.

"From a performance standpoint, if you compare our 16-core Interlagos to our current 12-core AMD Opteron 6100-series processors (code named “Magny Cours”) we estimate that customers will see up to 50% more performance from 33% more cores.  This means we expect the per core performance to go in the right direction - up," explained John Fruehe, the director of product marketing for server/workstation products at AMD.

While "per core" performance of Bulldozer may not be that impressive, the new chip designs may allow AMD to clock the forthcoming microprocessors higher without increase of power consumption and heat dissipation or to pack more cores into the next-generation processors without any running into thermal problems.

Based on the information provided by AMD during its annual Analyst Day last November, the first Bulldozer micro-architecture desktop/workstation chip code-named Zambezi (which belongs to Orochi family, according to the firm) will feature eight x86 processing engines with a multithreading technology, two 128-bit FMAC floating point units, shared L2 cache, shared L3 cache as well as integrated memory controller. AMD also states that the new CPU will feature “extensive new power management innovations”. The new chips that belong to Bulldozer family will also support Advanced Vector Extensions (AVX) that support 256-bit FP operations. Based on a diagram that AMD demonstrated in the past, the company intends to dramatically improve multithreading performance of its CPUs with the help of two INT schedulers, an FP scheduler and separate data caches for each of four cores should do the job very well.

In the second quarter of calendar 2010 the world's second largest supplier of central processing units taped out the first Bulldozer microprocessors. The "tape out" means that the artwork for the photomask of a circuit is sent to manufacturing. While it is unclear at this point whether AMD had already received samples of the Bulldozer processors back from manufacturing and assembling or at least manufactured wafers from Globalfoundries, given that the company on Tuesday made the first Bulldozer performance-related statements, it is likely that the firm has first samples at hands.

Tags: AMD, Bulldozer, Opteron, Maranello, Interlagos, 32nm, Orochi, Zambezi


Comments currently: 30
Discussion started: 08/03/10 04:32:44 PM
Latest comment: 08/18/10 03:52:16 AM
Expand all threads | Collapse all threads


This year, we delivered 80-120% greater performance with Magny Cours. Next year we will add another 50% on top of that.

In what world is that considered "not dramatic"?

0 2 [Posted by: JF-AMD  | Date: 08/03/10 07:58:25 PM]

Its dramatic or not, depends on whom you are competing with. if your performance is slightly higher or lower than the competition, you will not be noticed, you need to bring some dramatic performance improvement and new features, then you will be noticed.
Especially, we see some software vendors still charge for per core license (not per unit). So we will prefer better 8 core than slightly higher performance 16 core processor.

JF, I strongly believe that this statement will help your current server sales , so you made this statement . I will repharse the statement with " 50% performance improvement with ___ % reduction in power".
we will wait for the real numbers.
0 0 [Posted by: Kicha31  | Date: 08/03/10 09:56:36 PM]

BD will use two memory channels per 8 cores (four modules).
Is it enough to feed eight int cores + four fp cores with AVX?
Or BD freq won't be much higher than that of Magny-Cours?
0 0 [Posted by: Azazel  | Date: 08/03/10 10:41:55 PM]
- collapse thread

considering its DDR3-1600 that will be plenty.
ddr2-800 or ddr3-1333 hardly made a difference in the current quad-cores.

the 3 channels intel uses was overkill, but now everybody seems to think dual channel is obsolete.

and no its unlikely BD's frequencies will be much higher. even thou its made on 32nm we've reached what seems to be the optimum frequencies for CPU's. frequencies will only increase slowly with each new production process.
0 0 [Posted by: Countess  | Date: 08/04/10 02:56:58 AM]

Too early to say anything about Bulldozer perfomance per core,depend on test.In fact one Bulldozer core have two CPU core,so if u compare this with one CPU core, not Bulldozer
core u can use factor x 2 so increase in perfomance will be 100 % .,but this sounds interesting:

"the new chip designs may allow AMD to clock the forthcoming microprocessors higher without increase of power consumption and heat dissipation or to pack more cores into the next-generation processors without any running into thermal problems".

This is also from AMD blog:

Bulldozer will feature a new floating point unit that can support up to 256-bit floating point execution, which will boost the performance for technical applications that rely on floating point math.
What is mean perfomance per core? in fact one buldozer core have two CPU core,
0 1 [Posted by: Blackcode  | Date: 08/04/10 12:50:08 AM]

"50% more performance from 33% more cores" = +13% performance from a Bulldozer core at the same clock speed compared with an Opteron 6100 core, if I am not doing something wrong.

This is "not dramatic". It is just better and if you add lower power consumption, higher speeds, more cores and other minor optimizations you have a very competitive chip against i7 at the same clock speed.
But what happens if we see an i8 from Intel? Things are going to be almost the same like they are today. Only difference. We don't need much faster cpus today. Today's microprocessors are more than fast for today's software in most cases. The same thing with gpus. Today a 200$ gpu runs almost anything fast enough. A 100$ cpu again runs almost anything fast enough. Things are changing and speed is not going to be the main reason for buying a cpu in the near future.
0 0 [Posted by: john_gre  | Date: 08/04/10 01:35:25 AM]

AMD Bulldozer en exclusiva
Find Bulldozer performance comparisons below.. (in Spanish)
0 0 [Posted by: mdzakaria  | Date: 08/04/10 03:43:49 AM]
- collapse thread

Epic fail fake!
0 0 [Posted by: Tester128  | Date: 08/04/10 05:39:46 AM]
It's unscientific fantastic.
Interlagos freq won't be higher than 2GHz.
0 0 [Posted by: Azazel  | Date: 08/04/10 05:52:23 AM]

Power and thermals will be the same range as Magny Cours today, so if performance if 50% higher and the power is the same, you can do your own math on performance per watt.

You all have to keep in mind that customers don't buy cores, they buy processors. There will be a 50% increase in total processor performance. Doing the math and saying that per core vs. Magny Cours ignores the fact that there are 33% more cores.

If you do want to go down the path of comparing a 12-core Interlagos to a 12-core Magny Cours, the Interlagos will be faster AND will be less expensive because for Magny Cours 12-core is the top of the stack and for Interlagos, 12-core is at the bottom of the stack.

The fact that the Opteron 6100 beats the Xeon 5600 in integer and floating poing today AND will have a 50% increase in performance, means that intel will need well more than 50% increase with sandybridge just to catch up.
0 2 [Posted by: JF-AMD  | Date: 08/04/10 04:26:34 AM]
- collapse thread

SPEC_INT_rate_base for 2P systems:
1) IBM System x3690 X5 (Intel Xeon X7560) - 364,
2) ASUS RS700-E6 server system (Intel Xeon X5680) - 356,
3) Dell PowerEdge R815 (AMD Opteron 6176 SE, 2.30 GHz) - 314.

The numbers say itselves.
0 0 [Posted by: Azazel  | Date: 08/04/10 05:49:49 AM]
Azazel give us some link for your proof ,your number look like fake numbers.What about price perfomance ratio,power consumption ,SQL benchmarketc...
0 2 [Posted by: Blackcode  | Date: 08/04/10 02:26:30 PM]
0 0 [Posted by: chipguy  | Date: 08/04/10 08:57:48 PM]
Link not work
0 1 [Posted by: Blackcode  | Date: 08/05/10 03:17:35 AM]
I reckoned everybody who's in touch with business knew what SPEC meant.
You can visit for the numbers which you call "fake".
0 0 [Posted by: Azazel  | Date: 08/05/10 12:23:42 AM]
Future of server market is CPU with many core ,you should know that. Anton underline this :"the new chip designs may allow to pack more cores into the next-generation processors without any running into thermal problems".
So if u know how many core have Ati and Nvidia with CUDA core ,GPU now is stronger than CPU,AMD go in right direction.Noone use single thread app on server market.Point is that number of core is more important than frequencies.
0 1 [Posted by: Blackcode  | Date: 08/05/10 03:25:15 AM]
Azazel, here are some more numbers that speak for themselves:
Test: SPEC_INT_rate
System - Peak / Base
1) IBM System x3690 X5 (Intel Xeon X7560) - 390 / 364
2) ASUS RS700-E6 server system (Intel Xeon X5680) - 382 / 356
3) Dell PowerEdge R815 (AMD Opteron 6176 SE, 2.30 GHz) - 401 / 314

So if you go strictly by the Base results, then yes the Intel processors come out on top. Once you add in Peak compiler optimizations then the AMD system comes out ahead.

Lets look at another test...
Test: SPEC_FP_rate
System - Peak / Base
1) IBM System x3690 X5 (Intel Xeon X7560) - 284 / 275
2) ASUS RS700-E6 server system (Intel Xeon X5680) - 255 / 247
3) Dell PowerEdge R815 (AMD Opteron 6176 SE, 2.30 GHz) - 323 / 295

Here the AMD processor beats the Intel parts in both base and peak by a nice margin.
0 1 [Posted by: mamisano  | Date: 08/05/10 06:17:11 AM]

Anton isn't a fanboy of any company. He's a mere blogger wanting some attention. Unfortunately he seems to be successful.

Fortunately the release is still far away so the mindless drivel along with the people that suck up the words can be ignored.
0 0 [Posted by: DavidC1  | Date: 08/04/10 09:14:03 AM]

This is worse than anyone could have expected, really.

Adding cores to gain performance is not a great technological feat, and not something Intel can't reproduce easily. So, sorry, going to 16 cores fails to impress.

A lot depends on the size of each core. I don't think anyone expects AMD to make as good a processor as Intel, they simply don't have the talent Intel does, but, if they can make it smaller and slower, unlike now, which is the same size and slower, it's fine.

I think a lot of people were hoping the Bulldozer would finally make AMD competitive, and get their design at least fairly closet to Intel's. This is a huge disappointment, and will continue to relegate AMD to their bottom-feeding habits. I just hope the processor is somehow smaller, although I don't know how it could be. If not, it's going to be hard to make a profit. Good thing they bought ATI.
0 0 [Posted by: TA152H  | Date: 08/04/10 05:01:32 PM]

Today's Magny Cours delivers better processor performance than westmere-base Xeons.

When Xeon went from Nehalem to Westmere, they added 50% more cores and had a 33% increase in performance (43% int, 24% FP, averages 33.5%).

We are adding 33% more cores and getting 50% more performance. AMD must have the talent if, generation on generation we have delivered more performance increase than Intel.
0 1 [Posted by: JF-AMD  | Date: 08/04/10 08:06:51 PM]
- collapse thread

I don't think so the reviews say another thing
0 0 [Posted by: chipguy  | Date: 08/04/10 08:33:40 PM]
Which reviews?
0 1 [Posted by: mamisano  | Date: 08/05/10 06:18:26 AM]
0 0 [Posted by: chipguy  | Date: 08/05/10 03:43:16 PM]
You're very conveniently overlooking some VERY important aspects of this, which scares me. It means you really have no understanding.

Nehalem to Westmere was a shrink, not a generational change like Bulldozer, which has been waited for, for years.

Nehalem is so much better to begin with, the point isn't how much better percentage wise this shrink is to its predecessor, compared to AMD's generational change compared to its predecessor, but how well the compare to each other.

With this disappointing improvement, it's clear Intel will have still have much better performance per core than AMD, and probably be smaller as well, since they are the same size now. I'm hoping this isn't true, but it seems very likely.

The important metrics are performance per core, size of core, and power use. Having to throw massive amounts of cores to compete with a competitor's fewer cores isn't very good. It's expensive, and less efficient, and can easily be matched when the competitor wants to throw more cores on it.

0 0 [Posted by: TA152H  | Date: 08/05/10 07:22:53 AM]
At least for now AMD doesn't worry about production expensiveness and efficiency due to AMD is forced to pay for the whole 45nm SOI production line to GloFo in any case. And i suspect GloFo'll compel AMD to make those payments for 32nm SOI too or GloFo'll not be interested in low volume production exclusively for AMD.
0 0 [Posted by: Azazel  | Date: 08/05/10 10:36:32 PM]

My prediction is that 4 Bulldozer core ,version for desktop will be faster than Intel Gulftown in Cinebench ,h264 encoding both pass,pov ray,but not in 7 zip...if we add price perfomance ratio,Bulldozer will cut price for Intel Gulftown,and Intel fanboy will be happy ,so go AMD goooo
0 1 [Posted by: Blackcode  | Date: 08/06/10 02:40:36 PM]
- collapse thread

My prediction says another thing
0 0 [Posted by: chipguy  | Date: 08/06/10 04:47:30 PM]

From what I've been reading, Bulldozer was being designed to compete with Intel's SandyBridge and behond.

Bulldozer should easily outperform Nehalem by at least 75% to 150%. Single Core performance, not sure, but Multi-Core, AMD really should have the upper hand Big Time.

Nehalem and Phenom II are yesteryear's technologies, it's time for another Hammer by AMD, but this time much much faster and efficient than.

AMD has no choice but to get serious and deliver something that is bloody fast.
0 0 [Posted by: nt300  | Date: 08/09/10 05:45:11 PM]
- collapse thread

Can you show me the numbers please?, or give me the link please?
0 0 [Posted by: chipguy  | Date: 08/10/10 09:29:36 PM]

Doesn't this violate Amdhal's law? Going from 12 to 16 cores should theoretically allow for a 33% odd increase with 95% parallelized code. You could assume the code is over 95% parallelized (which is a rather big assumption).
0 0 [Posted by: bjorno  | Date: 08/18/10 03:52:16 AM]


Add your Comment

Related news

Latest News

Thursday, August 28, 2014

12:22 pm | AMD Has No Plans to Reconsider Recommended Prices of Radeon R9 Graphics Cards. AMD Will Not Lower Recommended Prices of Radeon R9 Graphics Solutions

Wednesday, August 27, 2014

9:09 pm | Samsung Begins to Produce 2.13GHz 64GB DDR4 Memory Modules. Samsung Uses TSV DRAMs for 64GB DDR4 RDIMMs

Tuesday, August 26, 2014

6:41 pm | AMD Quietly Reveals Third Iteration of GCN Architecture with Tonga GPU. AMD Unleashes Radeon R9 285 Graphics Cards, Tonga GPU, GCN 1.2 Architecture

Monday, August 25, 2014

6:05 pm | Chinese Inspur to Sell Mission-Critical Servers with AMD Software, Power 8 Processors. IBM to Enter Chinese Big Data Market with the Help from Inspur

Sunday, August 24, 2014

6:12 pm | Former X-Bit Labs Editor Aims to Wed Tabletop Games with Mobile Platforms. Game Master Wants to Become a New World of Warcraft