TR Forums

http://www.xbitlabs.com/articles/cpu/di ... html#sect1

My gosh that is a sad story.

Edit: More blood:
http://www.anandtech.com/show/4955/the- ... 0-tested/5

Edit: Lost Circuits has some of the best concluding remarks of any of the reviews on the web. Awesome stuff. Not perfect writing, but excellent content:
http://www.lostcircuits.com/mambo//inde ... itstart=19

This really drives home the point that a module is not two cores and should not have been called two cores.

The comparison of Thuban (2 core) vs. Sandy Bridge (1 core + HT) just cements my belief that a die-shrunk Thuban with two more cores and a slightly higher clock would be able to compete on price... unlike poor Bulldozer.

Who knows, maybe AMD's investment in a new microarchitecture will pay off in a couple of years. But the sure thing would be paying off for them right now.

There's simply not a single redeeming aspect to this new CPU at this point, and for one to be suddenly unlocked by Windows 8 is extremely doubtful. This thing is a complete and utter disappointment from every angle. Shareholders have to be pissed. It's hard to describe how ridiculous this is. Substantially lower in IPC than their previous CPU. Wow. AMD truly is becoming irrelevant.

FuturePastNow wrote:

Who knows, maybe AMD's investment in a new microarchitecture will pay off in a couple of years.

Assuming they are still in the microprocessor business at all in a couple of years...

ALL benchmark presented on show that bulldozer is 50% slower core for core, clock for clock then a Thuban chip...

This actually seem to match the execution units difference.

BD 2core = 4 units
Thuban 2 core = 6 units

So raw compute per core can be 50% higher on thuban VS bulldozer, and that seem to translate 1:1 in real test.
So you need 50% more core and or clock rate to balance.

But look at those memory & cache write performance ! its absolutely shocking.

DDR3-1600 is rated at 25GB, and Intel get close to this theoretical peak... How much does AMD get out of their ddr3 memory controller? 9GB... ***9GB***
How can this be any other indication that bulldozer memory controller is broken ?

And the L1 cache copy performance. 40GB vs 115GB ? WHAT?! , this is where all x86 local structures and variable are written over and over in intensive loops.
Having 3 time less bandwidth to local data is catastrophic.

Why did AMD decided to use cache so slow on bulldozer? I'm puzzled...

flip-mode wrote:

There's simply not a single redeeming aspect to this new CPU at this point, and for one to be suddenly unlocked by Windows 8 is extremely doubtful. This thing is a complete and utter disappointment from every angle. Shareholders have to be pissed. It's hard to describe how ridiculous this is. Substantially lower in IPC than their previous CPU. Wow. AMD truly is becoming irrelevant.

Yeah... I used to be a AMD fan with rooting for the underdog and everything ya know... but BD is just unforgivable and I agree with everything you said. Compared to sandy bridge... more power hungry, slower, more expensive, runs hotter... Even compared to it's predecessor.... wow... the same

An 8-core Llano with L3 would have been both smaller and faster judging by this and the die shots.

Game_boy wrote:

An 8-core Llano with L3 would have been both smaller and faster judging by this and the die shots.

Except that Trinity (the successor to Llano) will be much better than Bulldozer in every possible way. In some ways Llano is better than Bulldozer now. Piledriver had better actually fix Bulldozer's issues STAT. What really worries me is that Ivy Bridge will do to AMD what Windows 95 did to Mac OS (and Apple) pre-Mac OSX. . . . . nuke it from orbit.

riviera74 wrote:

Except that Trinity (the successor to Llano) will be much better than Bulldozer in every possible way.

So Trinity will be almost as good as Thuban again?

Original post updated with a link to Anand's clock-for-clock tests. Anand did us the further courtesy of running one of the benches on only a single core for all the processors rather than X-bit's method of using two cores. Both methods are useful though since now the merit of AMD's "module" architecture needs to be tested.

But Anand's fully-threaded test shows the Phenom II x6 at 3.3 GHz beating the Bulldozer's 8 "cores" at 3.6 GHz. Embarrassing. Jezuz. It's gotta be hard to come to work in the morning when you can't even beat your own old CPU. Heck, forget catching up to the 2006 Conroe-core clock performance, does this thing even have the single core clock-for-clock performance of a Pentium 4?

flip-mode wrote:

... does this thing even have the single core clock-for-clock performance of a Pentium 4?

Hey, it's really bad, but not that bad

Look at it this way:

Multithreaded test:

Thuban 3.3ghz - 32465
BD 3.6ghx - 31768

Scale both numbers to match core clock & execution units:

Thuban normalized to 3.6ghz : 32465 / 3.3 * 3.6 = 35416
BD normalized to the same # of execution units as thuban = 35739

A 3.6ghz Thuban with 18 execution units match exactly a 3.6ghz BD with 18 execution units.

LEts just hope that AMD can lower voltage (increase clock) and increase IPC with better memory /cache performance ...
The Intel clock is ticking.. tick...tock..tick...tock... AMD better diffuse the IvyBridge bomb quickly!

If you consider performance per transistor, Bulldozer is way, way slower than Pentium 4 or maybe even any CPU ever. It's got 2 billion transistors. It gets it's azz spanked by Thuban's 0.904 billion transistors. Maybe it's even the lowest performance per transistor of all time? Dunno.

flip-mode wrote:

There's simply not a single redeeming aspect to this new CPU at this point,

I disagree- it's really good at unzipping compressed files.

FuturePastNow wrote:

flip-mode wrote:

There's simply not a single redeeming aspect to this new CPU at this point,

I disagree- it's really good at unzipping compressed files.

Thank god the new Bulldozer-based computer I'm building at home is going to be my "unzipping" machine. Otherwise I'd be pretty disappointed.

I updated the original post with a link to the Lost Circuits article conclusion. There are some really well though out comments there that I have to say make it the best conceived conclusion to a Bulldozer review that I have read so far, despite some grammar mistakes.