In defense of "Bulldozer"

[zsdersw]

Sold out. It's not like a couple people bought the CPU. It completely sold out at one of the most well known and respected internet computer hardware shops. How can it do better than sell out?

That's not a relevant response to what I wrote. No one said what you claimed they said; that no one would buy it.

Actually, I think you have it backwards. The only people who really care about having the fastest and best CPU are enthusiasts. For the majority, whatever cheap CPU is included in Best Buy's $399 deal is what they are happy with. And sure, those $399 systems are not going to include FX-8150 CPUs, but they also aren't going to include i5 or i7 CPUs. .

I have nothing backwards. I said the only people who would buy it right now would be enthusiasts... and that's just about enough people to result in Bulldozer being sold out. That's not any different than what you said.

And they really don't care about CPU power usage, unless it is directly noticeable (louder fans, for example)

Yes, some enthusiasts do care about CPU power usage. Some because they're building HTPCs and want passive (or nearly passive) cooling, others because they like to see how high they can overclock.

 

[Ferzerp]

High power usage at stock means ridiculous power usage when overclocked.

 

[Nemesis 1]

There is no reason what so ever to defend BD. AMD choose gate first . They used SOI for along time and its getting less effective with each shrink .

AMD will fix their front end and other improvements to put smiles on the AMD base fanbios. The problem is intel is a moving target that has a big lead . So as AMD improves so shall Intel. This is AMDs real problem. Ya BD is an unpolished turd. But it has great potencial . Get over yourselves . The biggest problem I see with AMD right now and into the future 32nm products is leakage. That is going to be hard to get a handle on . As for everthing else the microarch is pretty dam good . Just need to polish the turd and it has great potencial.

 

[dac7nco]

No one was really saying, en masse, that nobody will ever buy it.

No, the reality is: Bulldozer sold out because the only people who would buy it right now, immediately at launch, are enthusiasts.

I'm saying it: If and when AMD can actually deliver an appreciable quantity, the FX-8150 will collect dust on the shelves, and no serious enthusiast will buy one. There are no AMD enthusiasts, and AMD has exactly zero high-performance CPUs to be enthusiastic about. The people who are likely to buy an FX are people with an overriding hatred of Intel, who will gladly suffer a loss of productivity for an inferiority complex... as long as they've done their part to "help the good guys".

In other words, the people who need counseling and medication.

Daimon

 

[Nemesis 1]

AMD recommends avoiding AVX when compiling with ICC. The 256b AVX ops (internally usually represented by 2 uops) just add overhead w/o adding throughput. GCC forces AVX128 to improve by ~3% on avg (SPECFP I assume).

I made no mention of AVX as you and others precieve I was adderessing The prefix of Vex only . A number of test show that AMD using FMA is no faster than AVX. Also alot of what you can find on the subject was is AMD supplied AVX recompiles. It will be awhile befor we see the true effects of AVX in the desktop . Servers will be the first programms. Multi media tho is looking rather good . almost double the performance with SB.

 

[Dresdenboy]

I made no mention of AVX as you and others precieve I was adderessing The prefix of Vex only . A number of test show that AMD using FMA is no faster than AVX. Also alot of what you can find on the subject was is AMD supplied AVX recompiles. It will be awhile befor we see the true effects of AVX in the desktop . Servers will be the first programms. Multi media tho is looking rather good . almost double the performance with SB.

I've downloaded the recompiled versions but didn't look at them any further. But I followed Dark_Shikari's developer chat log and saw, that he optimized some x264 subroutines during the last few weeks he had access to bulldozers (first remote, then on his desk).

The prefix doesn't help or hinder FX. And it's sometimes difficult to code for FMA in tight assembler loops. Further there are many places to add FMA or AVX to before seeing bigger improvements. Some quotes:

2011-09-28 01:33:59 < Dark_Shikari> [b]SSE2: 342 cycles[/b]
2011-09-28 01:34:00 < Dark_Shikari> [b]AVX: 374[/b]
2011-09-28 01:34:05 < Dark_Shikari> [b]FMA4: 340[/b]
2011-09-28 01:34:18 < kemuri-_9> lol
2011-09-28 01:34:26 < Dark_Shikari> I guess this makes sense given that it only has 128-bit execution units
2011-09-28 01:34:34 < Dark_Shikari> and the INT16_TO_FLOAT code is obnoxiously slow because avx sucks
2011-09-28 01:34:41 < Dark_Shikari> i.e. avx has no way of doing int16_t -> float fast
2011-09-28 01:35:18 < Dark_Shikari> Hmm. I wonder if FMA4 supports sse registers?
2011-09-28 01:35:37 < Dark_Shikari> Oh. It *does*...
2011-09-28 01:35:38 < Dark_Shikari> Let me try that.
2011-09-28 01:37:45 * codestr0m ears perk up
2011-09-28 01:49:29 < Dark_Shikari> [b]FMA4: 314 cycles[/b]. Much better

2011-10-12 21:00:27 < Dark_Shikari> I know this, because AVX offers exactly zero benefit on Bulldozer
2011-10-12 21:00:36 < Dark_Shikari> because BDZ has move elimination

Sometimes FMA makes it worse when just used to quickly replace muls and adds due to different dependency chains:

2011-09-27 21:05:56 < Dark_Shikari> mul dst0, src0, src0
2011-09-27 21:05:59 < Dark_Shikari> mul dst1, src1, src1
2011-09-27 21:06:02 < Dark_Shikari> mul dst2, src2, src2
2011-09-27 21:06:02 < mru> i.e. throughput of 1/cycle
2011-09-27 21:06:06 < Dark_Shikari> mul dst3, src3, src3
2011-09-27 21:06:10 < Dark_Shikari> add accum, dst0
2011-09-27 21:06:12 < Dark_Shikari> add accum, dst1
2011-09-27 21:06:15 < Dark_Shikari> add accum, dst2
2011-09-27 21:06:16 < Dark_Shikari> add accum, dst3
2011-09-27 21:06:23 < Dark_Shikari> This is vastly faster than:
2011-09-27 21:06:31 < Dark_Shikari> mac accum, src0, src0, accum
2011-09-27 21:06:34 < Dark_Shikari> mac accum, src1, src1, accum
2011-09-27 21:06:35 < mru> wtf
2011-09-27 21:06:37 < Dark_Shikari> mac accum, src2, src2, accum
2011-09-27 21:06:40 < Dark_Shikari> mac accum, src3, src3, accum

http://akuvian.org/src/x264/freenode-x264dev.log.bz2

 

[exar333]

No one was really saying, en masse, that nobody will ever buy it.



No, the reality is: Bulldozer sold out because the only people who would buy it right now, immediately at launch, are enthusiasts... and that's a group of people of just the right size to result in it being sold out.

I remember the FX 5800 selling-out as well when that was launched. We all know how well that went...

 

[exar333]

Funny how people who wouldn't touch and AMD CPU with a ten-foot pole are all over the AMD threads. You don't like it, you don't buy it.

And AMD had the performance crown once, didn't help their sales, Intel has a stronghold on the retailers.

If Intel made the Bulldozer we would have seen much more friendly reviews titled "8-core Mega Beast". What happened to the Passmark* benches anyway, they vanished from the reviews just like 3DMark who was replaced by the Nvidia-friendly "Compute & Tessellation"


http://www.extremetech.com/computing/100173-when-cpu-brand-loyalty-goes-too-far

The thing is, it's not the AMD owners who are whining but the die-hard Intel fans who wouldn't buy an AMD chip whatever. The kind of dudes who recommend the Intel IGP over the Llano in a HTPC.


*If anyone has meaningful evidence on why the Passmark tests are irrelevant (while benches like "Monte Carlo simulation are) please feel free to post it.

I love the recent argument from 'fans' of BD.

-[Related to BD] Power consumption is not that important, as long as you get good performance.
-[Related to PhII] Performance of existing PhII chips are 'good enough' and you do not need a i5/i7 SB
-[Related to NV] Power consumption is absolutely horrid for only 10-15% more performance compared to AMD/ATI equivalents

Interesting how the arguments always seem to match the existing products? I agree with point #3 and think it's definiretely valid.

I would completely support BD if the performance stayed similar to PhII with much better power usage, or if performance increased 10-20% with the same power usage, or if the performance gain was substantially higher than the power increase.

[Edit] If BD was across the board 10-15% better than the 2600k, people would be complaining about the power issues, but it wouldn't be as big of a deal. It would be like the NV GTX480 vs the AMD 5870. One faster but uses a lot more power, vs a little slower but much more efficient. When you are slow AND power-hungry, it is lose-lose.

 

[bryanW1995]

OS scheduler issue was known a long time ago, I actually had a conversation with John fruehe about it 6+ months ago. He told me they were working on it. Apparently they weren't working on it hard enough...

 

[Thermalzeal]

Oracle has their own CPU now after buying Sun. It's called the SPARC T3/T4. They have 16 and 8 cores respectively.

These chips were on hold for a while but now they're actually shipping.

Yes, I'm aware that Oracle has the Sun hardware unit still in operation. But Larry Ellison has been recently quoted about wanting to go after a chip company. In addition, Oracle is interested in beefing up the hardware side of their business.

 

[SickBeast]

You'll need to back this up with something.

I back it up with the fact that Bulldozer loses to pretty much anything intel makes in its price range, across the board, especially when you consider things like clock speed and power consumption.

If you want to look at their other products like the Phenom II and Llano, they generally perform the same or worse than Bulldozer.

AMD's entire CPU portfolio is obsolete upon arrival compared to what intel has on the market. It has been this way for years now.

Intel has such a dominant position in the CPU market that they have been able to disable overclocking and get away with it.

 

[Nemesis 1]

I've downloaded the recompiled versions but didn't look at them any further. But I followed Dark_Shikari's developer chat log and saw, that he optimized some x264 subroutines during the last few weeks he had access to bulldozers (first remote, then on his desk).

The prefix doesn't help or hinder FX. And it's sometimes difficult to code for FMA in tight assembler loops. Further there are many places to add FMA or AVX to before seeing bigger improvements. Some quotes:

2011-09-28 01:33:59 < Dark_Shikari> [b]SSE2: 342 cycles[/b]
2011-09-28 01:34:00 < Dark_Shikari> [b]AVX: 374[/b]
2011-09-28 01:34:05 < Dark_Shikari> [b]FMA4: 340[/b]
2011-09-28 01:34:18 < kemuri-_9> lol
2011-09-28 01:34:26 < Dark_Shikari> I guess this makes sense given that it only has 128-bit execution units
2011-09-28 01:34:34 < Dark_Shikari> and the INT16_TO_FLOAT code is obnoxiously slow because avx sucks
2011-09-28 01:34:41 < Dark_Shikari> i.e. avx has no way of doing int16_t -> float fast
2011-09-28 01:35:18 < Dark_Shikari> Hmm. I wonder if FMA4 supports sse registers?
2011-09-28 01:35:37 < Dark_Shikari> Oh. It *does*...
2011-09-28 01:35:38 < Dark_Shikari> Let me try that.
2011-09-28 01:37:45 * codestr0m ears perk up
2011-09-28 01:49:29 < Dark_Shikari> [b]FMA4: 314 cycles[/b]. Much better

2011-10-12 21:00:27 < Dark_Shikari> I know this, because AVX offers exactly zero benefit on Bulldozer
2011-10-12 21:00:36 < Dark_Shikari> because BDZ has move elimination

Sometimes FMA makes it worse when just used to quickly replace muls and adds due to different dependency chains:

2011-09-27 21:05:56 < Dark_Shikari> mul dst0, src0, src0
2011-09-27 21:05:59 < Dark_Shikari> mul dst1, src1, src1
2011-09-27 21:06:02 < Dark_Shikari> mul dst2, src2, src2
2011-09-27 21:06:02 < mru> i.e. throughput of 1/cycle
2011-09-27 21:06:06 < Dark_Shikari> mul dst3, src3, src3
2011-09-27 21:06:10 < Dark_Shikari> add accum, dst0
2011-09-27 21:06:12 < Dark_Shikari> add accum, dst1
2011-09-27 21:06:15 < Dark_Shikari> add accum, dst2
2011-09-27 21:06:16 < Dark_Shikari> add accum, dst3
2011-09-27 21:06:23 < Dark_Shikari> This is vastly faster than:
2011-09-27 21:06:31 < Dark_Shikari> mac accum, src0, src0, accum
2011-09-27 21:06:34 < Dark_Shikari> mac accum, src1, src1, accum
2011-09-27 21:06:35 < mru> wtf
2011-09-27 21:06:37 < Dark_Shikari> mac accum, src2, src2, accum
2011-09-27 21:06:40 < Dark_Shikari> mac accum, src3, src3, accum

http://akuvian.org/src/x264/freenode-x264dev.log.bz2

The prefix is more about auto recompile than anything else. IB includes AVX improvements. I seen were AVX256 bit right now isn't doing alot or it would seem as thats the case . But AVX looks to be moving rapidly. Compared to other extensions when they were added.

 

[Idontcare]

OS scheduler issue was known a long time ago, I actually had a conversation with John fruehe about it 6+ months ago. He told me they were working on it. Apparently they weren't working on it hard enough...

Considering that it was "OS scheduler issues" that deep-sixed the original Cool'n'Quiet features on Phenom some 5 yrs ago...its a pretty sad excuse that they are still behind the ball all this time later.

You'd think they'd have a working relationship with Microsoft as well as leading *nix vendors to such an extent that this would have been robustly addressed well in advance of the product launch date.

There's just so many things about bulldozer this year that all point to it being "amateur hour" at club-AMD Which is sad, they really had their A-game together before C2D came along.

Forget barrier-to-entry concerns from a process technology angle, I'm beginning to wonder if we are seeing real evidence of a barrier-to-entry in just being able to competently design, test, debug and verify a modern competitive x86 design replete with all the requisite software/app eco-system management required for platform readiness.

The task, however colossal, appears to be beyond the means of AMD with this bulldozer launch. Via bowed out long ago. AMD might not be much longer in the running if they are hoping Microsoft is going to feel compelled to invest resources on their end just to get the OS "AMD friendly".

Microsoft wasn't compelled to do that for Itanium, if AMD's marketshare continues to tread water then just how much longer is Microsoft going to see an ROI in patching their OSes for the benefit of the few AMD customers out there? At some point the AMD ROI picture to Microsoft (and *nix) will look comparable to that of Itanium's.

Just a matter of time I suppose

 

[SickBeast]

Forget barrier-to-entry concerns from a process technology angle, I'm beginning to wonder if we are seeing real evidence of a barrier-to-entry in just being able to competently design, test, debug and verify a modern competitive x86 design replete with all the requisite software/app eco-system management required for platform readiness.

Yep, that is my concern as well. These CPUs have gotten so complicated that AMD can't even get a handle on creating a cohesive package that actually outperforms their previous designs.

It seems as though there are tools to mitigate this problem, but they cost at least 20% in efficiency, and even then the process is still astronomically complex.

They say that Bulldozer was a failure in its first incarnation, it was scrapped, and then they started over. It seems as though the second incarnation was a failed experiment as well. It will be interesting to see what AMD learns from it going forward.

 

[Obsoleet]

not true at all, AMD might do fine in some areas, but they do not own the laptop market at all.

I didn't say they own the market, I said they have the superior mobile solutions. More GPU is where it's at, especially considering power and usability of a mobile device.

The only part of laptop market which AMD has advantage is entry level gaming laptops. For corporate users Intel cpu reign superior. And even there GPU advantage might disappear as soon as 28nm mobile GPUs hit the market.

The one and only chip where they are clearly superior is Brazos (mostly because atom sux).

I have a high end i7 mobile chip. It's not that great, and I'd take an equivalently priced Llano or similar laptop over it anyday. The extra CPU power of my i7 is pointless.

That's what I've been saying, since C2D, this "war" is over. Intel won. Anything AMD or Intel puts out now is overkill for 98% of the users out there, they need to worry about ARM.

Let's face it, in this market, 2011, Bulldozer is fine.. if not preferable to having a cheat of sorts in counting cores. I'd gladly market it. And I'd gladly trade my C2Q for one straight up. It's not that bad of a CPU, and has all the power (and more) than almost everyone here needs.

I use mine for H.264 conversion, heavy gaming, simultaneous streaming, and I have far more than enough CPU power. Many people (even here) just sit on Anandtech and post on an i7.. it's pretty absurd. What the market needs, and wants, are more integrated GPU solutions to reduce heat and power, along with price.

I understand the excitement and circle jerking, especially for the Intel fans who had to listen to Bulldozer hype for years.. but the reality is x86 CPUs are no longer a hot market for advancement.
Most of Intel's advancements in the performance department for most people are completely wasted at this point. Guys like me are no longer rushing out to buy an FX or i7 because we don't need one. I have plenty of resources too, it's not a lack of money.. it's just that it is wasted money. Especially if you have anything post-C2D era like a quadcore PhenomII or C2Q/i5/i7 already. Sandy Bridge and Bulldozer are utterly pointless.

But lets stop kidding ourselves, Bulldozer isn't slow.

I agree the power requirements are a bit high, and for that, the price on this gen should be a little lower. But this is from my standpoint only, prices should NOT be dropped on it due to marketing picking up slack in that area. Most people don't know about power requirements in Fry's.

Spend your money on a larger SSD or a better video card if you're gaming, even pickup a Brazos powered ultrabook with great battery life. If you can truly maximize the value of a high end i7 or FX, more power to you.. the product's (high end CPU advancements) appeal to the mass market is dwindling.
Say hello to the ARM era.

 

[Keysplayr]

"But lets stop kidding ourselves"

Do lets.

 

[Obsoleet]

"But lets stop kidding ourselves"

Do lets.

So the 8150 would be slower, than say, the CPU you'd have today if you weren't handed the one you use now?

I very much doubt you'd have a CPU as "slow" as a FX-8150. -.-

You said you make your wife and kids use lower/midrange Phenoms, I don't think you're an especially spendy guy. You might be frugal to use AMD, but you're not entirely stupid either.
Rather, you seem to agree with me that almost anything post the C2D era is quite alright for most people.

I think the 8150 would rape the doors off what you'd be using if you walked into Fry's to build yourself a rig on your own dime.

 

[dac7nco]

Forget barrier-to-entry concerns from a process technology angle, I'm beginning to wonder if we are seeing real evidence of a barrier-to-entry in just being able to competently design, test, debug and verify a modern competitive x86 design replete with all the requisite software/app eco-system management required for platform readiness.

This is a daunting thing to consider. I'm also pretty sure that all of this nonsense of "Intel keeping AMD just barely alive to keep from being declared a monopoly" is ridiculous. The US considers Intel "critical infrastructure"... I'm pretty sure AMD cashed those chips in a few years back (AMD to UAE: What you doin tonight?! (Hugs).:wub:

Daimon

 

[Obsoleet]

Another thing people forget, that this isn't 1997 anymore. As if actually matters in this market if AMD comes back or not. Or if Intel gets complacent.

These are dinosaurs. The age old fanboy circlejerk times are over. This is Apples time, this is ARMs time. The PC market is dying. No one outside of us age-old x86 dogs give a damn about Intel and AMD CPU designs which are both grossly overpowered.
A Bulldozer for all intents and purposes, is equal to an i7 for almost all of us.

The consumer wants a Macbook Air or an iPad. Both great products, and the former to be guaranteed to be running on an A6 sometime in the near future. The time is up for Intel, and the bickering back and forth between AMD/Intel is a debate that's dead. AMD has a less than optimal, but sufficient chip here to market and sell. Money is better spent on a large SSD or a decent GPU, but whatever. I hope they do well with it. I wish Intel could stop ARM, but they aren't going to be able to.

Times have changed, like they eventually always do.

 

[drizek]

The consumer wants a Macbook Air or an iPad. Both great products, and the former to be guaranteed to be running on an A6 sometime in the near future.

That's insane. I'm a consumer, and I'm going to get an Ivy Bridge Macbook Air, not an A6-powered one.

In all likelihood, I will get an A6 iPad though.

There is no benefit to ARM on a 13" form factor anyway. The display is using almost all the power, so power savings on the CPU side are kind of irrelevant, especially if you have to make extreme sacrifices in performance and application compatibility, as would be the case with an A6.

 

[StrangerGuy]

So the 8150 would be slower, than say, the CPU you'd have today if you weren't handed the one you use now?

I very much doubt you'd have a CPU as "slow" as a FX-8150. -.-

You said you make your wife and kids use lower/midrange Phenoms, I don't think you're an especially spendy guy. You might be frugal to use AMD, but you're not entirely stupid either.
Rather, you seem to agree with me that almost anything post the C2D era is quite alright for most people.

I think the 8150 would rape the doors off what you'd be using if you walked into Fry's to build yourself a rig on your own dime.

OK, so BD is good because it beats those old C2Ds instead of SB? You are making me laugh.

 

[zlejedi]

Most of Intel's advancements in the performance department for most people are completely wasted at this point. Guys like me are no longer rushing out to buy an FX or i7 because we don't need one. I have plenty of resources too, it's not a lack of money.. it's just that it is wasted money. Especially if you have anything post-C2D era like a quadcore PhenomII or C2Q/i5/i7 already. Sandy Bridge and Bulldozer are utterly pointless.

But lets stop kidding ourselves, Bulldozer isn't slow.

While BD or Phenom II might be fast enough (well imho they aren't considering gameplay feels much smoother on 2500K in World of Tanks) it still leaves the question why would you buy it when 2500K cost the same as 8120 and less than 8150 and simply doesn't have any weakpoints.

I like getting best value my limited money can get me.

 

[dma0991]

Times have changed, like they eventually always do.

ARM may have gained considerable traction in the current situation but PC will never disappear. If it did then where would content be produced? Would you prefer to transcode a video on an iPad or a PC which would be miles faster than a ARM could ever be. Any ultra mobile that comes out from the factory will be no more than a content consuming device, it will never be a full fledged content producing device like a PC could. All that Apple revolutionized was touch interface, performance was never Apple or any ARM based devices forte.

 

[3DVagabond]

ARM may have gained considerable traction in the current situation but PC will never disappear. If it did then where would content be produced? Would you prefer to transcode a video on an iPad or a PC which would be miles faster than a ARM could ever be. Any ultra mobile that comes out from the factory will be no more than a content consuming device, it will never be a full fledged content producing device like a PC could. All that Apple revolutionized was touch interface, performance was never Apple or any ARM based devices forte.

I agree. The PC is not dead, or even dieing. Other sectors are growing faster because of overall market maturity with the PC. As long as programmers are willing to create apps that will use more and more resources, more performance will always be a selling point.

 

[mosox]

I'm not an AMD fan, I'm a fan of the low prices and competition (that keeps them low).

Any firm constantly evaluates the market's response to their products and not only, to those of their competitors. The more hype, the more expensive the product (demand and offer).

Repeating everywhere that Bulldozer is worse than it actually is (and it's not that good but far from being a dud or a fail) will lead to Intel raising the prices. That's Capitalism 101.

Apple has a history of selling good but overpriced products based on the hype perpetrated by their fanboys everywhere and those fans were hurting themselves. The more they praised the Apple products and the more they trashed the competition the more money they had to pay for their favorite brand.

You can bet that somewhere in the Intel HQ some marketing guys are arguing for price increases for the Sandy Bridge line and this Bulldozer thinghy will affect the pricing of the Intel future products as well and not in a good way.

Imagine AMD announced a year ago that Bulldozer will be mainly for heavy multithreaded apps, how much would you had to pay for your 2500K? $350? $400?


So, if you like Intel and think that Bulldozer stinks you may want to keep that to yourself. Of course if somebody asks for your help in deciding what platform, recommend Intel but don't unnecessarily put down AMD or your next Intel CPU might be $600 and be named "i-Vybridge" or i-Ntel.