Why Bulldozer's spotty performance is good news.

[Neuromancer]

Here's an interesting post from Overclock.net. It gives some perspective.

http://www.overclock.net/amd-cpus/1141188-asus-crosshair-v-formula-board-may-11.html#post15299874

Thanks I edifyied that post. Since Gigabyte counts parallel processing even for VRMs. ASUs' 8+2 = gigabyte 16+4. .

I recently reviewed the 990FX UD5 and the intersil chip it uses for VRM distribution is only a 6 phase parallel processing unit. (IE high end)

Funny the Chinese site calls it a 10 phase, but its not its 6 phases parallel. Providing 5*2 phases for cpu and 1*2 pahses for nb.

I get where they are wrong, the nomenclature is close, but intersil does not make a 10 phase VRM

 

[Inceptor]

Why? You quoted some random guy with no idea as a reliable source...

I'm scratching my head here.
I quoted someone who put up an interesting perspective on the situation. I made no claim about his veracity or expertise. I quoted a comment...
You on the other hand are looking for an argument with someone, preferably someone who will take the exact opposite viewpoint to you.
Unfortunately, I can't give that you, because I agree with you, in general, about BD.
I'm sorry, but who cares how Opterons have done in the past, AMD does not have a large share of the server and workstation markets... I point you back to your statement regarding AMD vs Intel resources. I don't even need to reference anything to know I'm right, and no one else here needs those references either.
AMD wants server market share, yes. Will they get much more of it? Maybe, if Interlagos turns out to be something special. A whole lot more of it? Doubtful.

Anything an executive says, to media, or even much of what they say to shareholders, is marketing bull@#$@. No point being bitter about that kind of stuff. That's the way things are.

I think you missed out on the fanboy rage exchanges. You should have been here a couple of weeks ago, you would have enjoyed yourself with them.

Let's see fanboy accusation list, what have people accused me of?
Intel fanboy? check.
AMD fanboy? check.
Next on the list NV and Radeon fanboy badges, then I think I'll try for mobo company fan badges...
This could be fun...

 

[John Doe]

All in all it seems to me that AMD remains the cheaper, less yet still quite fast performance option. And hell, you can still overclock any AMD chip... I expect to see lots of us here at TPU buying lower class BD chips and making them fly faster than the fastest stock BD chip... and the fastest stock Intel chip. That's still crazy performance.

They don't unlock now, do they? The Phenom X4 gave solid performance, especially in games considering it's price. Once AMD matures their chips, yeah. Until then, I doubt those on the higher-end side would pick BD over SB. Unless they're a fanboy of course...

 

[Neuromancer]

I agree that AMD rocks, but guys lets stop the animosity. (wow did I just say that?) I am having a hard time reading the facts from the love/hate relationship.

Here is my animosity:


You know what I think is stupid? SLI on AMD. F*bomb on nVidia and their crappy ass products that cant even game on one screen and play HD video on another. like my 6850 can. Its why my 580GTX sits idle until I need to bench or revie wsomething. Why? Cuz that is ALL nVidia is good for.

There you go, hate message for you that is gauranteed to drop some bombs. It happens to be fact too. But nVidia fans will argue that no gamer would have an HD video playing on a second monitor. (I RTS/RPG slow type games )

DX11 breaks on both cards BTW... not sure why yet. Still 6870 in my gamer 580 sits on shelf. (both 580s)

EDIT: only reason I upgrade from 6850 to 6870 was the heatpiped heatsink, 6850 is getting reused dont you fret

EDIT EDIT: Sorry, got so caught up in this stuff.

 

[John Doe]

You know what I think is stupid? SLI on AMD.

Uhm, yeah. That's why I had a pair of 480's on a Crosshair 2 yet it kicked S775 ass. Both the 780a and 980a (which is a rebadged 780) were good chips without the issues of S775 SLi systems. They were rare and made a good combo.

 

[hat]

I would, because it would be cheaper and still hellishly fast, even at stock, but we can overclock any BD chip. The current 6 core BD model is $190, a bit cheaper than the $220 Intel wants for the 2500k, with two more cores to spare... making it a bit less helpless in heavily multithreaded apps, although it would still lag behind in most things. Then you can overclock the 6 core BD beyond stock 2500k performance... of course then you could clock the 2500k to outclass the overclocked BD, but even though the 2500k is faster, the BD is still pretty damn fast, and cheaper, factoring overall system cost in.

The point I'm making is AMD is slower and cheaper, although slower doesn't really matter because even the slower performance is still pretty fast. In short, nothing really changed... I pretty much expected BD to turn out like this.

I wonder why AMD hasn't put a similar feature to Intel's turbo boost in BD, though. This makes me wonder about the overclockability of BD...

 

[Frick]

exactly!! even i dont know why but intels dont feel as fast in everyday application.

What the eternal heck are you guys on? This makes no sense, and it is not true.

 

[John Doe]

My point was nvidia sucked and you just named another reason. 780 to 980 chipsets.

thanks.

I was actually only referring to nVidia GFX and their inability to multimon well but that is a great point, they sucked at chipsets too for the most part (NF4 rocked) oh and NF200 snicker snicker

So that makes them "suck"? I had HD 4870 CrossFire and it was garbage. Such a driver nightmare. Before 2010, there was no application profiles so CrossFire was a PITA. It usually didn't work, even when it did, you had to force it by changing the application name, which kicked you out of Steam servers. Ever since X1950 cards, ATi (AMD) either lacked on GPU power or drivers.

 

[Neuromancer]

I would, because it would be cheaper and still hellishly fast, even at stock, but we can overclock any BD chip. The current 6 core BD model is $190, a bit cheaper than the $220 Intel wants for the 2500k, with two more cores to spare... making it a bit less helpless in heavily multithreaded apps, although it would still lag behind in most things. Then you can overclock the 6 core BD beyond stock 2500k performance... of course then you could clock the 2500k to outclass the overclocked BD, but even though the 2500k is faster, the BD is still pretty damn fast, and cheaper, factoring overall system cost in.

The point I'm making is AMD is slower and cheaper, although slower doesn't really matter because even the slower performance is still pretty fast. In short, nothing really changed... I pretty much expected BD to turn out like this.

I wonder why AMD hasn't put a similar feature to Intel's turbo boost in BD, though. This makes me wonder about the overclockability of BD...

Um AMD did put turbo boost in.

Also 6 core BD = 3 core phenom 2 with hyperthreading, but a better version of hyperthreading.

I do agree with the "slower" does not matter with more cores. Or slower overall if BD has hte same desktop feel phenthub and zosma had.


(have not actually used zosma just guesstimating)

BD requires to much power and has a lower IPC than thub/zos extensions are great, Intel lives on them, but generally not supported on AMD , lets face it., Where is hte AMD multicore optimizer download?

EDIT: Why talk about the 6 (lol) core Bulldozer when the 8 (lol) core only meets the 2500k intel processor?

 

[hat]

When I see on Newegg's spec list Multi-Core: Six-Core, I'm led to believe it actually has 6 cores... not that it's a 3 core with hyperthreading. Either someone's spewing bullshit about BD, or AMD's being really misleading with their spec sheets. Where can I read in detail about this supposed hyperthreading that's going on here? I don't see anything about any kind of turbo boost anything on Newegg's spec page for the BD chips either... you'd think whoever's in charge of that would be making damn sure that they're blowing their marketing horn about having that feature.

 

[Frick]

So that makes them "suck"? I had HD 4870 CrossFire and it was garbage. Such a driver nightmare. Before 2010, there was no application profiles so CrossFire was a PITA. It usually didn't work, even when it did, you had to force it by changing the application name, which kicked you out of Steam servers. Ever since X1950 cards, ATi (AMD) either lacked on GPU power or drivers.

They only time they lacked in power was probably the HD2xxx series. Drivers I can agree with though. I'm willing to put up with it though as they has always offered the best price/performance ratio and that's what I'm after.

 

[seronx]

Opinions?

I LOL'ed hard when I saw this okay

 

[laszlo]

amd make a mistake by launching this chip too soon as almost no soft is capable to use all cores and the new added instruction;this chip will become obsolete till we'll have majority of daily used programs/new games, will be able to use dozer at full capacity so i think is a good product but arrived in the wrong market and time;they should focus on quads and improve them or create something what the market demand;now they just show "we can do it we launched the 1st 8 core desktop CPU" and... so what? not average joe buyer will ask himself :can i use it fully? nope;when can i use it? ...after 3-5 years...;than why to invest in it? just to have it?

someone at amd has taken a v.bad marketing decision by forcing this chip and this will affect them

 

[scooper22]

When I see on Newegg's spec list Multi-Core: Six-Core, I'm led to believe it actually has 6 cores... not that it's a 3 core with hyperthreading. Either someone's spewing bullshit about BD, or AMD's being really misleading with their spec sheets. Where can I read in detail about this supposed hyperthreading that's going on here? I don't see anything about any kind of turbo boost anything on Newegg's spec page for the BD chips either... you'd think whoever's in charge of that would be making damn sure that they're blowing their marketing horn about having that feature.

BD FX-6xxx is:
- a 6-core integer/fp128
- a 3-core fp256
at the same time.

as long fp256-ops are somewhat sparse and integer/fp128 are the majority of all ops I think it is fully legitime to call it a six-core as six fully independent threads can be processed. So 6 threads run in parallel (6T) on six cores (6C).

4C/8T Hyper-Threading (Intel only) is a FOUR-core where every core can hold two threads. These are however not processed at the same time. Therefore it's a four-core (4C) with eight threads (8T) where half of them is stalled at all times. This is a true 4-core.


Schematics of one BD-module (FX-6xxx has three of these, FX-8xxxx four):

 

[John Doe]

Opinions?

I LOL'ed hard when I saw this okay

lol. That's PassMark, right? IIRC, they use some sort of browser tests and stuff to come up with that score... it's nonsense, not a real benchmark at all.

 

[nt300]

amd make a mistake by launching this chip too soon as almost no soft is capable to use all cores and the new added instruction;this chip will become obsolete till we'll have majority of daily used programs/new games, will be able to use dozer at full capacity so i think is a good product but arrived in the wrong market and time;they should focus on quads and improve them or create something what the market demand;now they just show "we can do it we launched the 1st 8 core desktop CPU" and... so what? not average joe buyer will ask himself :can i use it fully? nope;when can i use it? ...after 3-5 years...;than why to invest in it? just to have it?

someone at amd has taken a v.bad marketing decision by forcing this chip and this will affect them

AMD had no choice but to launch. It was the massive rumours that made Bulldozer seem like a killer CPU.

I don't see why people don't put aside its performance right now and look at the design as Pure Innovation on AMD part. They did something unique and completely different, and hopefully in time this will pay dividends.

With enough tweaking, cache latency timing fixes and the like, the upcoming Piledriver should look and feel like a monster in performance. Mark my words, Piledriver won't be no slouch. Hopefully by then MS will patch Win 7.

 

[Super XP]

Interesting, though it does feel like beating a dead horse over and over again.

Hopefully Piledriver will be tweaked and fixed in time for an aggressive Q1 2012 release on the Socket AM3+ platform. I expect at least a 35% to 50% performance boost in Multi-Threading performance and a 10% to 15% performance boost in Single Threaded Performance.

Bulldozer currently suffers from various issues including but not limited to Branch Prediction, Pipeline Flushing, Cache Trashing, Decode unit not wide enough and so on. If and when AMD fixes this, Bulldozer should have the ability to Bulldoze aas it was meant to do.

That said, most of those review sites which rated Bulldozer low did not perform proper testing. What I mean is they did not perform proper Format of the Harddrive, did not Install a fresh copy of Win 7 x64, did not have the right motherboard bios update etc.

In order to determine the proper speed and performace of a new CPU especially Bulldozer being completely build from the ground up, all above factors should have been meet and they were not.

Check out Bulldozer's Benchmarks via Rage3D, Hardware Heaven, Hardocp and Overclockers.com where they followed the proper practice for testing hardware. Bulldozer performed quite well, though it could have been better, it stood its ground.

That said, I give AMD 5*****'s for Pure Innovation, now FIX the Bulldozer issues and release Piledriver Thank YOu.......

 

[AphexDreamer]

http://www.tweaktown.com/news/21168...x.html?utm_source=dlvr.it&utm_medium=facebook

BD to be Fixed with a B3 Revision.

 

[erocker]

http://www.tweaktown.com/news/21168...x.html?utm_source=dlvr.it&utm_medium=facebook

BD to be Fixed with a B3 Revision.

Nice, I'll be sure to get on AMD about giving me a B3 as a replacement to my "broken" B2.

 

[Steevo]

http://www.tweaktown.com/news/21168...x.html?utm_source=dlvr.it&utm_medium=facebook

BD to be Fixed with a B3 Revision.

Not at all true.

The rumor mill, based on a BIOS paper for motherboard makers shows a B3 revision, and the site assumes they "(if the do)" fix the scheduling issues. The B3 could be lower power parts due to fixed leakage, they could be lower speed actual die cut 3 module X6 parts, they could be a birthday cake in a box.


Somehow you magically read "it will be fixed"?

What do you smoke so I can get some?

The flaw is in the design, AMD apparently never spent any good time on branch prediction, so they made up for it with a huge cache, that huge cache slowed down the whole chip in speed and sucked up power, that drove the clocks even further down and forced AMD to put a thermal limit on the chips.

It was and has been a horrible event of one failure leading to another failure in design, implementation, and finish. If they started with a known good CPU and asked the simple "what works, and why, what doesn't and why" and came up with the best 10 things they would have a monster, but they didn't. They have a netburst clone.

K8, with higher IPC and die shrinks, microcode updates and even having to run large in place calculations as 2 or 4 clock procedures would be faster for most users.

Look at what Intel did for core. Back to the P III design with the improvements they learned on P4 and branch prediction improvement. If AMD had done this instead of "real men use real cores" we would be looking at this from a whole different angle. Unfortunately AMD got lost somewhere before they bought ATI and haven't been the same since.

 

[AphexDreamer]

Not at all true.

The rumor mill, based on a BIOS paper for motherboard makers shows a B3 revision, and the site assumes they "(if the do)" fix the scheduling issues. The B3 could be lower power parts due to fixed leakage, they could be lower speed actual die cut 3 module X6 parts, they could be a birthday cake in a box.


Somehow you magically read "it will be fixed"?

What do you smoke so I can get some?

The flaw is in the design, AMD apparently never spent any good time on branch prediction, so they made up for it with a huge cache, that huge cache slowed down the whole chip in speed and sucked up power, that drove the clocks even further down and forced AMD to put a thermal limit on the chips.

It was and has been a horrible event of one failure leading to another failure in design, implementation, and finish. If they started with a known good CPU and asked the simple "what works, and why, what doesn't and why" and came up with the best 10 things they would have a monster, but they didn't. They have a netburst clone.

K8, with higher IPC and die shrinks, microcode updates and even having to run large in place calculations as 2 or 4 clock procedures would be faster for most users.

Look at what Intel did for core. Back to the P III design with the improvements they learned on P4 and branch prediction improvement. If AMD had done this instead of "real men use real cores" we would be looking at this from a whole different angle. Unfortunately AMD got lost somewhere before they bought ATI and haven't been the same since.

Don't get your panties in a bunch. I was merely posting the fact that a B3 revision will be coming out and the article states that they hope it will provide better performance hence the "fix", fixed being a lose term in where I personally don't define what gets fixed but surly the revision fixes something else why the revision?. Both I and the article remain optimistic in the B3 revision as it is still too soon to same with certainty anything regarding it.

And no I can not provide you with illegal narcotics, sorry.

 

[Dent1]

^
This B3 stepping stuff is starting to seem more credible.

After many were left disappointed by the performance of AMD's FX-Series processors based on the Bulldozer architecture, the chip maker is now working on a new revision of its CPUs that will carry the B3 stepping.

This information was uncovered by Hardcore-Hardware in a public AMD document entitled “BIOS and Kernel’s Developers Guide (BKDG) for AMD Family 15h Models 00h-0Fh Processors.”

The document included a table that mentioned the current stepping of AMD's Bulldozer processors, but also a new revision of the architecture that was referred to as B3.

Sadly, the report doesn't detail the changes that AMD wants to bring to its chips with the introduction of this new stepping, but users should expect the usual tweaks that deliver slightly improved performance and better energy-efficiency.

This is not the first time that AMD is having troubles with a new processors architecture.

Shortly after the company launched it first quad-core CPUs from the Phenom line, it has been discovered that the chips suffered from a TLB (translation lookaside buffer) flaw which incurred a performance penalty of at least 10%.

These design problems required AMD to build a new chip stepping, known as B3, which arrived about four months after the initial Phenom release.


AMD launched the first FX-Series processors based on the Bulldozer architecture the last week, on October 12, but despite being eagerly awaited by many PC enthusiasts the performance of the chips has disappointed.

In most tests that were run by reviewers, AMD's flagship CPU from the FX-Series lineup, the FX-8150, lagged behind the Intel's Core i5-2500K, a few tests even showing it coming behind a last-gen Phenom II X6 1100T processor.

Right now, no information regarding the release date of the B3 stepping is available, but considering AMD's current position in the CPU market it makes sense for this revision to be placed on top of the company's priority list.

http://news.softpedia.com/news/AMD-Bulldozer-B3-Revision-Is-in-the-Works-228347.shtml

 

[Steevo]

You guys want to buy a AMD Extreme Edition FX? Yeah it still performs slower than the Intel chip running hundreds of Mhz per core slower, but it EXTREME!!!!!

The flaw is the design, the chip, the whole thing.

Our only hope is the learn something from it and perhaps move the best parts to a design that more reflects their K8 designs, with a good fat fast interconnect between cores.

 

[Dent1]

You guys want to buy a AMD Extreme Edition FX? Yeah it still performs slower than the Intel chip running hundreds of Mhz per core slower, but it EXTREME!!!!!

When you say "you guys" whom is it directed at specifically?

The flaw is the design, the chip, the whole thing.

Our only hope is the learn something from it and perhaps move the best parts to a design that more reflects their K8 designs, with a good fat fast interconnect between cores.

Or software developers can stop being lazy and start writing code to utilise multi core processors in general.

 

[Super XP]

When you say "you guys" whom is it directed at specifically?

Or software developers can stop being lazy and start writing code to utilise multi core processors in general.

The issue here is some people can't credit where credit is due. Bulldozer may have not been what people expected, but it's in no way a bad design. I've said this once, and I will say it again, AMD has Balls of Iron for creating something completely new and innovative. Now they need to learn from it and make Piledriver better.

It would also help if Developers would start writing code or Multi-Cores that would benefit both AMD and Intel.