In defense of "Bulldozer"

[deadrats]

Perhaps the term "defense" isn't the most accurate, but after reading through numerous reviews it does seem apparent to me that Bulldozer isn't getting the reception AMD may have hoped for and most of it lies at the feet of AMD's marketing department.

What we, as enthusiasts, need to keep in mind is that AMD has long wanted to include some form of SMT (simultaneous multithreading) on it's chips for a while. Intel has long had HT and while Intel and AMD have a cross licensing agreement that allows any technology developed by one of them to be used by the other, AMD wanted to differentiate it's processor from Intel's by developing it's own SMT technology.

Bulldozer, is the culmination of that technology. While the AMD marketing machine wanted to establish a competitive marketing edge by counting each ALU as a separate core, they didn't do the processor nor the company as a whole any favors. AMD marketing beat it into our heads so thoroughly that these were 4 module 8 core processors that every review out their parroted the official company line without any critical thinking of their own.

If we look at the Phenom 2 architecture or the Core i architecture, we see that each core of either processor is composed of 3 ALU's and a FPU, in the Core i all 3 ALU's are capable of integer math and 1 of the ALU's is also used for Boolean comparisons (if/else, yes/no, that sort of thing).

Bulldozer, AMD analyzed their Phenom design and discovered that under most workloads only 2 ALU's were being used, with the 3rd barely being touched, so they dropped the 3rd ALU.

If we apply the same standard to Bulldozer as we do to both the Phenom 2 and Core i processors we see that a processor like the FX 8150 is really a quad core processor that has SMT capabilities. Following this logic the FX 6100 becomes a tri-core with SMT and the 4100 becomes a dual core with SMT.

When viewed in the correct way, without AMD marketing departments B.S. the Bulldozer doesn't look like such a failure, it stops looking like an 8 core that can barely beat AMD's older hexa core and Intel's quad core and instead becomes AMD's SMT capable quad core that is as fast or faster than their older 6 core and can keep up with Intel's quad cores.

In that regard, the Bulldozer becomes a winner.

Of course, once we factor in the price and the TDP, that skews our view again, but at least it's not the benchmarks that have us gagging.

I'm also of the firm belief that the true power of Bulldozer hasn't been shown in the benchmarks done so far, because benchmarks like Cinebench 11.5 default to spawning a number of threads equal to the number of cores and the x264 benchmark uses the threads = auto switch.

I'm fairly certain that of if Cinbench were rerun with the thread count manually set in the preferences to 12 or 16 threads, that the FX8150 might pull ahead of it's competitors; likewise I would be willing to bet if the benchmarking script of x264 benchmark were modified so that threads = auto read threads = 16 or 24, that the FX8150 may just surprise a few people.

Anyway, just some thoughts I wanted to share.

[Thermalzeal]

I too made a similar comparison, counting each model as a core. Unfortunately the issue is that even at 32nm, the die space taken up by the core is massive. On the G34 socket, AMD does have an Opteron that is 12 core (6-module) chip, but that core is literally a rectangle rather than your normal square due to the extra 2 modules.

Anyways, in defense of the offense on bulldozer's nonsense. Your right, AMD marketing is partially to blame for the negative dissent. And it is our duty to completely heckle and yell at AMD until the fire the n00bs that decided they were going to act like Intel and market "core counts" and "Guiness world record CPU Mghz". Both things which the company has ingrained in it's audience to be wary of (higher numbers = better that is).

I have to admit, I was more pissed off that I waited 6 months to replace a bluescreening Phenom board and CPU and ended up getting a Thuban core instead.

Not sure if this has been posted before...but best response EVER to Bulldozer. http://www.youtube.com/watch?v=SArxcnpXStE This video made my day after AMD ruined it.

[Accord99]

Originally Posted by deadrats

When viewed in the correct way, without AMD marketing departments B.S. the Bulldozer doesn't look like such a failure, it stops looking like an 8 core that can barely beat AMD's older hexa core and Intel's quad core and instead becomes AMD's SMT capable quad core that is as fast or faster than their older 6 core and can keep up with Intel's quad cores.

But its <=4 thread performance is poor as compared to existing AMD quad-cores, let alone Sandy Bridge. The closest description that can be applied to Bulldozer is that it's a 8 weak core design.

I'm fairly certain that of if Cinbench were rerun with the thread count manually set in the preferences to 12 or 16 threads, that the FX8150 might pull ahead of it's competitors; likewise I would be willing to bet if the benchmarking script of x264 benchmark were modified so that threads = auto read threads = 16 or 24, that the FX8150 may just surprise a few people.

Since BD can't actually run more than two threads per module simultaneously, all you get is that with 4 threads is that 2 threads will run for a period of time, then the OS will swap it for the next set of threads and so forth. No extra work will be done per unit time and it'll do as much as good as trying to run 8 threads on a Phenom II X4.

[drizek]

I think we are all used to AMD CPUs delivering underwhelming performance.

What last straw for me, though, was looking at the power consumption numbers. There are some components that I wouldn't put put in my system even if they were free. Prescott, Fermi and FX 5800 are the first examples that come to mind, and Bulldozer has fully earned a place on that same list.

[LOL_Wut_Axel]

Yeah... no. Bulldozer sucks, especially since it needs 2 Billion transistors to do less than what Intel can do with just under a million PLUS a graphics processor on die.

There's really nothing to defend about Bulldozer in its current iteration. It's a chip that shouldn't have been released until they had a revision that was actually competitive.

[deadrats]

Originally Posted by Accord99

Since BD can't actually run more than two threads per module simultaneously, all you get is that with 4 threads is that 2 threads will run for a period of time, then the OS will swap it for the next set of threads and so forth. No extra work will be done per unit time and it'll do as much as good as trying to run 8 threads on a Phenom II X4.

this is actually not true, i have both an e7400 and an x4 620, if i use an app like avidemux that allow me to manually set the thread count, the encoding using x264 (from within avidemux) gets faster by about 2 or 3 fps if i manually set the number of threads to 4, at 5 it starts to slow down again, this is using the Penryn, the Phenom doesn't scale as well (no microfusion of instructions).

modern processors are out of order cpu's and the can fetch, execute and retire instructions as data becomes become available, the can easily keep more than 1 thread per core in flight simultaneously. SMT isn't about enabling the cores to handle more threads, it's a method of instructing the OS and apps that the cpu can handle more threads and get the software to comply.

furthermore, like Intel's processors have been able to do since Conroe, BD can fuse instructions after the have been decoded and since the front end is a 4 instruction wide architecture, each module, with properly written code and assuming the thread scheduler is properly tuned, should be able to handle at least 4 threads easily (the dead give away are those massive L2 and L3 caches).

[Silverforce11]

Originally Posted by LOL_Wut_Axel

Yeah... no. Bulldozer sucks, especially since it needs 2 Billion transistors to do less than what Intel can do with just under a million PLUS a graphics processor on die.

There's really nothing to defend about Bulldozer in its current iteration. It's a chip that shouldn't have been released until they had a revision that was actually competitive.

True, but consumers wouldn't care about the transistor count. However, that huge power usage for worse performance is beyond nuts.

[Ferzerp]

I'm a little confused? Is bulldozer a damsel in distress that needs defending? I thought it was just a processor.

[Accord99]

Originally Posted by deadrats

modern processors are out of order cpu's and the can fetch, execute and retire instructions as data becomes become available, the can easily keep more than 1 thread per core in flight simultaneously.

But in standard processors, they can only work on instructions from a single thread and nly one thread can run on a processor at a time. The OS controls which thread gets access and how long it can execute for. Every time a thread is changed, a context switch is made and registers and caches are flushed in order to load the context of the new thread.

A Hyperthreading processor or CMT BD can use instructions from two threads, but no more than that.

[mosox]

Actually the 95W AMD FX-6100 looks good in here:

http://www.legionhardware.com/articl...fx_4170,1.html

And in gaming in full HD it's only a few fps behind Intel. Too bad the equally priced i-5 2400 is not in the review.

[voltronn]

My god the 4170 is slow!

[StrangerGuy]

Originally Posted by mosox

Actually the 95W AMD FX-6100 looks good in here:

http://www.legionhardware.com/articl...fx_4170,1.html

And in gaming in full HD it's only a few fps behind Intel. Too bad the equally priced i-5 2400 is not in the review.

Hey, lets us all cherry-pick benchmarks that are GPU limited to make AMD look good! More news at 11.

[Keysplayr]

Originally Posted by mosox

Actually the 95W AMD FX-6100 looks good in here:

http://www.legionhardware.com/articl...fx_4170,1.html

And in gaming in full HD it's only a few fps behind Intel. Too bad the equally priced i-5 2400 is not in the review.

Mosox. Look at all the comments below the article you linked to.

I'd like to see how you'd interpret them.

[-Slacker-]

What? AMD is free to use Hyper Threading for free? That and any other technology Intel has developed??

Then why didn't they do it by now?

[ShadowVVL]

amds performance per watt and price is bad compared to intel.
the 2500k is still a a much better buy then anything fx offers and there is no reason to bother with fx platform this year.

Even if the 8150 can out perform intel in a few tasks the power consumption is about 40%-60% higher.

Don't really understand why people are still talking about bulldozer since intel already destroys bulldozer in every way.


I would like to see how interlagos turns out. when does it launch?

[happy medium]

In defense of "Bulldozer" ?

Thats very hard to do at this point.
I would wait a week.

[mosox]

Originally Posted by Keysplayr

Mosox. Look at all the comments below the article you linked to.

I'd like to see how you'd interpret them.

I have a different style than 95% of the users in reading the reviews.

I'm far from being a HW guru but I know more or less what most of the benches mean in the real world.

For instance in that review I didn't even bother to read the performance in Excel 2010, that's for huge Excel spreadsheets. Not interested, couldn't care less.

Winrar compression - I use this every now and then. The FX-6100 being 25% slower that the more expensive 2500K is fine in my book.

Adobe Photoshop CS5 - Don't use the program but 26% slower than the 2500K is also fine.

Handbrake - I use that, 21% slower than the 2500K is nice. I have no problem with 50% slower, it's not like I can't do anything else on the PC while transcoding.

etc

The gaming performance in full HD is good, I would like to see the test with a less powerful video card (they used a Gigabyte GTX580 SOC).

But in the land "Intel wins in X benches while AMD wins only in Y, so Intel is much better" I guess the real performance once you start using it doesn't matter.

[Keysplayr]

Originally Posted by mosox

I have a different style than 95% of the users in reading the reviews.

I'm far from being a HW guru but I know more or less what most of the benches mean in the real world.

For instance in that review I didn't even bother to read the performance in Excel 2010, that's for huge Excel spreadsheets. Not interested, couldn't care less.

Winrar compression - I use this every now and then. The FX-6100 being 25% slower that the more expensive 2500K is fine in my book.

Adobe Photoshop CS5 - Don't use the program but 26% slower than the 2500K is also fine.

Handbrake - I use that, 21% slower than the 2500K is nice. I have no problem with 50% slower, it's not like I can't do anything else on the PC while transcoding.

etc

The gaming performance in full HD is good, I would like to see the test with a less powerful video card (they used a Gigabyte GTX580 SOC).

But in the land "Intel wins in X benches while AMD wins only in Y, so Intel is much better" I guess the real performance once you start using it doesn't matter.

Not going to touch my question with a ten foot pole, are ya...

[Ares1214]

I have a feeling AMD lost a lot of fans...and rightfully so. Intels design released 10 months earlier uses less power, costs less, and is faster, and now with IB and SB-E coming out in the next 4-6 months, AMD will be so far behind they wont even be a viable option. Not to mention by the time AMD release any form of BD that fixes their mistakes way down the line, Intel will probably have tri-gate or something. I mean didnt AMD have BD based CPU's down the line until like 2015? By the looks of it, AMD will get very little enthusiast money until then, but they could always fix things.

[Keysplayr]

Originally Posted by Ares1214

I have a feeling AMD lost a lot of fans...and rightfully so. Intels design released 10 months earlier uses less power, costs less, and is faster, and now with IB and SB-E coming out in the next 4-6 months, AMD will be so far behind they wont even be a viable option. Not to mention by the time AMD release any form of BD that fixes their mistakes way down the line, Intel will probably have tri-gate or something. I mean didnt AMD have BD based CPU's down the line until like 2015? By the looks of it, AMD will get very little enthusiast money until then, but they could always fix things.

AMD will be dismantled or bought out by another company by that time. (2015).
For the AMD enthusiasts here, I am NOT wishing this to happen. For all intensive purposes, it may even be a good thing to happen (the buyout part that is) if it means improving or infusing a boatload of talent and money to what is currently AMD. It's one thing for Intel to keep AMD afloat by keeping their prices higher so AMD becomes the cheap alternative and guaranteeing them some sales, but it's quite another for AMD to stumble upon stumble upon stumble. Eventually (and by eventually I mean right about now) they won't recover from the stumble. I would almost bet, that there is a team at AMD right now working on some sort of exit strategy in the event nobody buys them up.
This is not a doom and gloom observation on my part. It is based on several lackluster and plain underwhelming performance of this company over the last half of a decade. Time is indeed running out. Something NEEDS to happen one way or another. If AMD stock gets to a dollar (not far off) and is downgraded to a junk stock, it's over.

[LOL_Wut_Axel]

Originally Posted by mosox

I have a different style than 95% of the users in reading the reviews.

I'm far from being a HW guru but I know more or less what most of the benches mean in the real world.

For instance in that review I didn't even bother to read the performance in Excel 2010, that's for huge Excel spreadsheets. Not interested, couldn't care less.

Winrar compression - I use this every now and then. The FX-6100 being 25% slower that the more expensive 2500K is fine in my book.

Adobe Photoshop CS5 - Don't use the program but 26% slower than the 2500K is also fine.

Handbrake - I use that, 21% slower than the 2500K is nice. I have no problem with 50% slower, it's not like I can't do anything else on the PC while transcoding.

etc

The gaming performance in full HD is good, I would like to see the test with a less powerful video card (they used a Gigabyte GTX580 SOC).

But in the land "Intel wins in X benches while AMD wins only in Y, so Intel is much better" I guess the real performance once you start using it doesn't matter.

You may as well just buy a Sempron 145 since you're "fine with waiting". And the Core i5-2500K is only 5% faster stock than the i5-2400, so the 2400 is still a much better choice than the FX-6100.

[Thermalzeal]

I'm ok with Oracle buying AMD. Larry Ellison and Mike Hurd are the right guys to do the job.

[Keysplayr]

Originally Posted by Thermalzeal

I'm ok with Oracle buying AMD. Larry Ellison and Mike Hurd are the right guys to do the job.

Or maybe even Electronic Arts.

Seriously though, software companies should stick with software. Unless they've already proven themselves with both hardware and software.

[Nemesis 1]

Originally Posted by -Slacker-

What? AMD is free to use Hyper Threading for free? That and any other technology Intel has developed??

Then why didn't they do it by now?

When AMD is allowed to use the vexprefix you be sure and let me know about it . As Its not all about hardware Its a trinity. and in that trinity is a holy place were AMD can't enter.

[DaveSimmons]

Nice to see all the new members joining and old accounts being reactivated to defend the damsel in distress, but do we need a bunch of extra "it's really not as bad as all the benchmarks show" threads?

For 99% of home and business use, the price, performance and power consumption make Bulldozer inferior to Sandy Bridge. That's reality. Accept it and move on.