Good article: Analyzing Bulldozer: Why AMD's chip is so disappointing

[Black96ws6]

Good read. Lots of reasons why BD is the way it is, and he covers them all (SMT design, scheduling, cache latencies, etc):

http://www.extremetech.com/computing/100583-analyzing-bulldozers-scaling-single-thread-performance

 

[triarii3]

meeh, poelpe built their expectations too high. it's still a great CPU...if it were cheaper say around $150 bucks. but At the $200 range i'd be getting the 2500k or the i7 920 anyday.

 

[lifeblood]

A lot of what they said we had also figured out when Bulldozer first came out. Cache latency and IPC.

Bulldozer isn't a bad CPU, it just isn't competitive compared to its competition. Hopefully piledriver will find some magic and come out swinging. I'm not holding out hope, though.

 

[StrangerGuy]

A lot of what they said we had also figured out when Bulldozer first came out. Cache latency and IPC.

Bulldozer isn't a bad CPU, it just isn't competitive compared to its competition. Hopefully piledriver will find some magic and come out swinging. I'm not holding out hope, though.

The last time AMD actually tweaked their existing uarch other than Llano was back in 2001 with the Athlon XP and they were winning back then. Other than that its just add more cache + smaller process + more cores. With all three options denied, I don't hold much hope.

 

[Ferzerp]

I keep seeing "it's not a bad cpu. It's just not competitive".

For you folks, what exactly would be a "bad" cpu if one that "is not competitive" is "not bad"?

 

[apoppin]

The last time AMD actually tweaked their existing uarch other than Llano was back in 2001 with the Athlon XP and they were winning back then. Other than that its just add more cache + smaller process + more cores. With all three options denied, I don't hold much hope.

Maybe you don't. The only things keeping FX-8150 from being competitive in gaming is low clockspeed and high TDP.

If it had launched at 4.2GHz we would not be discussing it as a failure
:whiste:

i just spent another week with it at 4.4GHz testing with HD 6970-X3 Tri-Fire and it scales with clockspeed.

 

[TemjinGold]

I keep seeing "it's not a bad cpu, it's just not competitive".

For you folks, what exactly would be a "bad" cpu if one that "is not competitive" is "not bad"?

If it's made by Intel, obviously.

 

[Ferzerp]

That is the vibe I am getting off people if they can actually say a product is "not competitive" and "not bad" in the same sentence.

 

[MrX8503]

If the CPU is not competitive then its a bad CPU.

 

[grkM3]

always waiting for the next cpu from amd,even after the new cpu comes out!

now we wait for piledriver mmmm

 

[nemesismk2]

I keep seeing "it's not a bad cpu. It's just not competitive".

For you folks, what exactly would be a "bad" cpu if one that "is not competitive" is "not bad"?

I agree bulldozer is not a bad cpu but it's not a good one either

AMD will have to try harder with Bulldozer II but by then I will be using Intel probably

 

[gevorg]

Bulldozer is still a modern CPU and not as bad many people/reviewers claim it is. At the end of the day, it can perform all CPU intensive tasks just fine, with little noticeable differences in real-world usage. AMD just needs to price it right, since at this time its quite overpriced relative to competition. The BD=crap nonsense needs to stop.

 

[veri745]

If the CPU is not competitive then its a bad CPU.

A) performance is dependent on workload
B) performace/$ is a factor

The i7-990X is not competitive with the i7-2600k on a performance/$ basis, particularly in gaming, but I would not say the 990X is a "bad" CPU. In fact, I would consider it to be technologically amazing.

 

[Ferzerp]

Bulldozer is still a modern CPU and not as bad many people/reviewers claim it is. At the end of the day, it can perform all CPU intensive tasks just fine, with little noticeable differences in real-world usage. AMD just needs to price it right, since at this time its quite overpriced relative to competition. The BD=crap nonsense needs to stop.

But see, that argument is that that's not really a valid metric because all the previous models did the things you claim just fine, and for cheaper.

They can't price it right because there is no way they can manufacture such a huge transistor count chip for less than they would have to charge for it to make sense. However, yes, I agree that there is a price point at which it would go from bad to making a little sense, but as I said, that would likely end up as selling at a loss. At that point there is no reason to make these things. There's not really any way to reconcile that as far as I can tell.

 

[lifeblood]

If it's made by Intel, obviously.

Is a 8150 "uncompetitive" if it costs $20? Hell no, its a steal at $20 bucks. It will kick the crap out of anything else I can get for $20. But at $279 it is "uncompetitive" in that I can get better for cheaper.

I guess a "bad" CPU is one that gives bad info, like the floating-point division bug in the original Pentium. If you did work that used those calculations, you got the wrong answer. That could be very bad.

Personally, as much as we slam on Bulldozer, or any other CPU for that matter, I'm still astonished how anybody designs something that complex and it actually works. I couldn't do it.

 

[Ferzerp]

So, by your definition the original pentium was a "bad" cpu? Interesting.

I guess we should all use atoms? They give the correct answer as far as I know. (hint, atoms are bad CPUs too)


There are many things I can't do, but I can still recognize when a thing is done poorly. Not having the ability to do something myself does not magically remove my comparative abilities. I surely can't play football, but I am quite capable of saying that the Colts are a horrible team this year. I don't need the ability to be able to play to see the result of that playing.

I'm not really going to award AMD with my money as some sort of participation award. They failed. They can go back and try again. Encouraging them to not excel by buying these things is a really bizarre notion.

 

[sm625]

We've known for 6 months that the cache was bad. This article does not explain wtf AMD was thinking by releasing a chip with 16MB of cache that really is not much faster than main memory. My god they could have put 16 cores 4M cache, and a 4 channel memory controller in the same amount of die space, and the thing would have flew.

 

[Vesku]

Their write up would have been more interesting if they had sourced the technical information releases on bulldozer prior to launch. They also took an odd approach to looking at CMT penalties by saying "Look at this performance hit from the CMT approach" and then saying "That actually matches up pretty closely to what AMD said the penalty would be", should have flipped that. Show what AMD was saying then show how it turned out in retail FX.

We've known for 6 months that the cache was bad. This article does not explain wtf AMD was thinking by releasing a chip with 16MB of cache that really is not much faster than main memory. My god they could have put 16 cores 4M cache, and a 4 channel memory controller in the same amount of die space, and the thing would have flew.

This is a server chip being flogged for performance desktops. The difference in L3 benefits between many HPC applications and regular desktop applications can be quite drastic.

 

[TakeNoPrisoners]

It is disappointing to people who play games and already have an overclocked Phenom II.

Even worse, after release Gigabyte pulled the AM3+ compatibility from my motherboard.

Man if a company keeps pulling this crap they might as well just go bankrupt.

 

[Ferzerp]

That's another line I keep reading.

All the other x86 server chips work wonderfully for desktop use. Intel and AMD offerings alike. The only time they would perform worse for desktop use is on some of the greater than dual socket and massive core count (e.g. the 10 core/20 thread E7 part from Intel that tops out at 2.4Ghz) sku's that have lower clock speed to support that, and single threaded performance will of course suffer. However, those processors are niche processors even in the server space. To need one, you have to need that massive core count on a single system and not have a workload that can run on a GPGPU. 4 core vs 4 core desktop vs server typically perform the same and are clocked similarly.

Bulldozer isn't a special server snowflake. We want similar things in desktop chips as we do in server chips for the most part.

 

[Vesku]

That's another line I keep reading.

All the other x86 server chips work wonderfully for desktop use. Intel and AMD offerings alike. The only time they would perform worse for desktop use is on some of the greater than dual socket and massive core count (8-10 on the intel side) sku's that have lower clock speed to support that. 4 core vs 4 core desktop vs server typically perform the same and are clocked similarly.

Bulldozer isn't a special server snowflake. We want similar things in desktop chips as we do in server chips for the most part.

Actually it is and you even cover why it is so. They've actively designed bulldozer from a core count density perspective and with server workloads in mind. We aren't talking small office servers, this is about high density computing.

Seems to me AMD has pretty much conceded the enthusiast desktop space to Intel. They are firmly focused on cpu+gpu integration as can be seen by their cries of "look we have Trinity working" during the buildup to the FX launch.

 

[Lonyo]

Maybe you don't. The only things keeping FX-8150 from being competitive in gaming is low clockspeed and high TDP.

If it had launched at 4.2GHz we would not be discussing it as a failure
:whiste:

i just spent another week with it at 4.4GHz testing with HD 6970-X3 Tri-Fire and it scales with clockspeed.

If Prescott had launched with high clockspeed and low TDP, it would have also been a success.
It didn't. People called it as it was, a CPU which failed to do what it needed to.

Bulldozer is very similar. It's slower clock for clock than its predecessor, runs hot, and is clocked lower than it needs to be. Doesn't matter if it would be a success if it had clocked properly, fact is it didn't clock where it needed to, so it's bad.

Maybe AMD can make it work, or maybe it'll just be a screwup that needs throwing out, and AMD might not be able to afford that.

 

[Ferzerp]

Not really. The 8 core count system, at best performs like a quad core on the intel side. In the server space this will still be the case. You can't just throw a lot of weak cores at a problem unless you're throwing a whole lot of weak cores at a problem (see the failure of larrabee for an example, compare that to the "core" count of GPGPU offerings from AMD and Nvidia). Unless they're going to offer something like a 16 module offering (IIRC, they're talking 8 module?), they aren't going to be competitive.

I've also seen a lot of misinformed "but, but virtualization!" arguments as well. If you think that in all but a few cases we really need more dense processing power for general purpose virtualization, you're either working in one of the few fields that that is the case, or you are just assuming VMs need processing power. They don't. Most of our VM systems are sitting with loads similar to the following example.

2, quad-core HT capable i7 Xeon's @2.93Ghz (X5570), 72Gb ram, 17 VMs currently sitting on that server.

Utilization:
Proc - 2.742Ghz processor time utilized (out of 8 physical cores, we are using load that could be provided by 2 or 3 cores with no noticeable change in performance), memory 46GB utilized.


We run out of memory volume (less of an issue), memory I/O (hard to identify, but it becomes a bottleneck), or disk I/O (more common), far, far earlier than we run out of processing power.

edit: and actually in the low utilization many thread environment of virtualization, I'd rather have stronger individual cores with HT than almost twice the weaker cores with no HT. We get around much of the penaltiy for context switching with HT, while not giving up peak single-threaded performance for those rare times that we have a spike in demand for a busy process.

 

[lifeblood]

So, by your definition the original pentium was a "bad" cpu? Interesting.

If you were doing work that required floating point than the Pentium was a "bad" CPU, at least until Intel fixed it. I remember at least one group of researchers suing Intel over it because they had to throw out 6 months of research because of the bug. If you were using it for games, office productivity, or anything else like I was than its was fine.

The Colts are horrible only relative to every other NFL team. I'm sure they would be awesome playing High School football teams (or maybe not :hmm. The point is that "bad" means its broke, while "uncompetitive" is relative to the alternative or competition. Again, at $20 BD would kick butt.

There are many things I can't do, but I can still recognize when a thing is done poorly. Not having the ability to do something myself does not magically remove my comparative abilities. I surely can't play football, but I am quite capable of saying that the Colts are a horrible team this year. I don't need the ability to be able to play to see the result of that playing.

I respect the individual who goes out on the field (or stage, or whatever) and puts forth the effort and takes the hits that come with it far more than the armchair quarterback who sits on his butt and just passes sentence. Yes, the Colts are doing badly this year, but I temper my criticism of them (like I do of AMD) until I have walked a mile in their shoes.

I'm not really going to award AMD with my money as some sort of participation award. They failed. They can go back and try again. Encouraging them to not excel by buying these things is a really bizarre notion.

Few people say to reward AMD with your money for a subpar CPU. I certainly don't and I certainly won't. My argument is that AMD should drop the price of a BD to where it equals in price an Intel CPU that it equals in performance.

 

[SlowSpyder]

I have hope for AMD's update, Piledriver. I think the architecture makes sense, it just didn't come together right for Bulldozer. I think if they tweak things and can get the clockspeed up, along with a better scheduler in Win8, AMD will have a solid part.

PhI wasn't good, PhII corrected a lot of the problems (cache, clockspeed) and was a solid part. Piledriver won't have the benefit of a new manufacturing process, though. But I still believe there is a lot of performance to be had by tweaking Bulldozer. But, as it is, as it launched, it's just not a tempting part compared to Intel's offerings.