Here comes the spin machine! (Our Take on AMD FX)

[BlueBlazer]

Gotta love this >> AMDUnprocessed: AMD FX Processors Go Head To Head | Competitive Performance For AMD's New Eight Core The comments got disabled (conspiracy theorists can apply). Take a look at the "Core i7 980X Cinebench score" (actually its from a Core i5 2500K).

 

[Jovec]

The fundamental flaw with the "more cores" strategy is that it generally requires developers to do more to see benefits. Programming is no different than most any other business and as such has to make due with limited budgets, compressed time tables, underskilled and/or inexperienced labor, a lack of priority and resources from management, etc.

This is roughly the same issue as the x86 uarch. It is cheaper and easier for Intel and AMD to throw billions at R&D, fabs, and production and for customers to throw billions purchasing x86 than it is to move away from it on the software end.

Now, as to BD itself... It works. It does everything a CPU is supposed to do. There is nothing you can do on Intel that you can't do on AMD. ATI/Radeon is doing fine. Bobcat and Llano seem to be doing fine in their respective markets. BD will likely do well in the server sector. It just seems to be a bad enthusiast level CPU.

BD obviously doesn't do enough to change that landscape, but AMD has been losing that battle for a while now. On 10/11, if you needed the best gaming CPU, best single-threaded IPC, and/or best multi-threaded performance you bought Intel. On 10/13, you still do, however I believe that the Bobcat (would anyone on this forum still buy Atom?), Llano (AMD is relevant in notebooks finally), and Radeon lines place AMD in a better position today than they have been in years past.

 

[drizek]

A really good read...spot on I think.

I wonder if that is why AMD's PR mouthpiece JF-AMD got all quiet after the launch? :whiste:

I would like for him to explain his last post:
http://forums.anandtech.com/showpost.php?p=32339565&postcount=1938

Seems like he was right about one thing though...AMD will now have to cherrypick benchmarks...

I don't see too many people arguing over Bulldozer's performance. I think everyone pretty much agrees that it sucks.

 

[sangyup81]

I don't see too many people arguing over Bulldozer's performance. I think everyone pretty much agrees that it sucks.

We've seem this before. Hopefully AMD can make it decent like how they fixed the Phenom and came out with a passable and well priced Phenom II.

 

[lOl_lol_lOl]

Didn't AMD have some form of periodical assessment of BD development? for example, to report to the CEO and shareholders. At what point did they bow down and say, lets just increase clocks and release the damn thing?

 

[drizek]

periodical assessment of BD development? ... report to the CEO and shareholders.

I think that was when the shareholders fired the CEO.

 

[drizek]

We've seem this before. Hopefully AMD can make it decent like how they fixed the Phenom and came out with a passable and well priced Phenom II.

AMD shrunk phenom after just one year. They need at least 2 years before they can do a die shrink of Bulldozer.

This is a complete disaster.

 

[lOl_lol_lOl]

AMD shrunk phenom after just one year. They need at least 2 years before they can do a die shrink of Bulldozer.

This is a complete disaster.

I may not be an expert in this field, but this is far worse than phenom.

BD = netburst + phenom I + fermi...

Even with the most optimistic of expectations, performance will not rise at a rate comparable to phenom I -> phenom II... with respect to the intel performance increases.

 

[Vesku]

I may not be an expert in this field, but this is far worse than phenom.

BD = netburst + phenom I + fermi...

Even with the most optimistic of expectations, performance will not rise at a rate comparable to phenom I -> phenom II... with respect to the intel performance increases.

It's not quite that bad, I posted this is in pre-launch discussion:

"Yeah, assuming the none clustered version of BD would have near Nehalem performance than the 8150 should be a bit slower in single threaded overall while being a bit faster in multithread overall.

8 (clusters) * .85 (BD to SB IPC ratio) * .8 (performance hit for clustering) = 5.44 (SB equiv MT)

4 (SB cores) * 1.3 (Gains from Hyperthreading) = 5.2 (SB equiv MT)

Allow for modest fudge factor in IPC ratio, performance hit from clustering and performance gain from Hyperthreading and you can see that with some clockspeed advantage the 8120 and 8150 should be a good alternative to the 4 core SB "K" series. Unless AMD pulled a real coup and a non clustered BD would have a theoretical IPC near SB level, seems quite unlikely though. "

That was assuming Nehalemish IPC for a non-module BD core. Unfortunately it seems with current software at least that IPC is roughly on par with Phenom II as a non-module BD core.

8 (clusters) * .6 (BD to SB IPC ratio) * .8 (performance hit for clustering) = 3.84 (SB equiv performance)

So AMD actually didn't do to badly on implementing CMT from a performance hit standpoint. But they really missed the mark in terms of IPC for each core. Remains to be seen how much IPC will be eeked out of software customization.

 

[Patrick Wolf]

Gotta love this >> AMDUnprocessed: AMD FX Processors Go Head To Head | Competitive Performance For AMD's New Eight Core The comments got disabled (conspiracy theorists can apply). Take a look at the "Core i7 980X Cinebench score" (actually its from a Core i5 2500K).

HA! Brilliant. Their own marketing can't sugar coat this incredible failure.

 

[Axon]

Well, if you read some of the reviews closely, AMD told reviewers months ago that the 8150 was only going to compete with the i5 line. So they knew this was coming.

As to spin, they have to do the best they can. They do have a few competitive gaming benchmarks and they need to prop those up. Of course...those benches usually involve GPU intensive games.

I hope this can be like Fermi was for nvidia, where they took a laughingstock and made it not just respectable, but an excellent product.

 

[drizek]

Well, if you read some of the reviews closely, AMD told reviewers months ago that the 8150 was only going to compete with the i5 line. So they knew this was coming.

But... it doesn't compete with the i5 line.

I mean, it only beats the 2500K in multithreaded, and the "i5 line" are just i7 CPUs that are purposefully gimped to provide poor multithreaded performance.

If, like the 2500K, BD had excellent singlethreaded and poor multithreaded performance, it would be OK. The problem is that it can't compete in either. It just sucks, especially at this price.

 

[Blitzvogel]

Considering AMD is the underdog, they have to try and gain consumer attention from casual buyers by having these "AMD Unprocessed"-like events and advertisements. Intel is a pretty household name, AMD is not.

As for BD, I don't think of it as much of a failure as it is just another step for AMD. If anything, AMD will learn from this, apply it to the next version/revision of BD. Obviously AMD probably was aiming after servers and highly multithreaded workloads, which of course is the future, but not just yet. The release of a desktop chip was pretty much a requirement for them, since we have been with Phenom II and it's Athlon derived architecture for so long, and consumers were pretty much demanding it. The real worry for normal consumers is whether or not AMD can truly improve upon what they have now in order to catch Intel next year or year afterwards similar to the improvements going from Phenom I to Phenom II where AMD finally "caught up" (kinda) to Conroe and Core 2 IPC.

AMD needs to end these hollow victories. If you can't win in one area, you need to win in another. I almost want to say AMD should just focus on servers, graphics and Fusion. Fusion actually represents a real advantage for AMD when the price is right. Bleeding edge server and graphics tech can just trickle down to Fusion for mainstream consumer computing.

 

[Idontcare]

Gotta love this >> AMDUnprocessed: AMD FX Processors Go Head To Head | Competitive Performance For AMD's New Eight Core The comments got disabled (conspiracy theorists can apply). Take a look at the "Core i7 980X Cinebench score" (actually its from a Core i5 2500K).

Well, that's one way to be a predator and "attack the hill" I suppose.

 

[Lonbjerg]

Well, that's one way to be a predator and "attack the hill" I suppose.

I my mind it's more like pulling the pin out of a handgreande....and then throwing the pin, while clining on the the live grenade...a sure way to have something blow up in your face.

I wouldn't even cal that marketing...I would call it lying...and deliberatly faking data to appear better.

 

[Dadofamunky]

But... it doesn't compete with the i5 line.

I mean, it only beats the 2500K in multithreaded, and the "i5 line" are just i7 CPUs that are purposefully gimped to provide poor multithreaded performance.

If, like the 2500K, BD had excellent singlethreaded and poor multithreaded performance, it would be OK. The problem is that it can't compete in either. It just sucks, especially at this price.

Also, a 5.98 vs. 5.4 for an 8-core vs. a quad doesn't reflect well on single-core performance for the octo. But the worst thing is the thermals.

 

[Idontcare]

I my mind it's more like pulling the pin out of a handgreande....and then throwing the pin, while clining on the the live grenade...a sure way to have something blow up in your face.

I wouldn't even cal that marketing...I would call it lying...and deliberatly faking data to appear better.

You got to cut them some slack though Lonbjerg. If you were employed at AMD and tasked with polishing bulldozer in the marketing slides you know darn well you'd be showing up to work already blitzed out of your mind, empty beer cans rolling around inside your briefcase just to numb the pain of the task that was ahead of you that day. Its easy to make honest mistakes in that condition.

Drunk posting is one thing, drunk marketing is not any prettier

 

[Lonbjerg]

You got to cut them some slack though Lonbjerg. If you were employed at AMD and tasked with polishing bulldozer in the marketing slides you know darn well you'd be showing up to work already blitzed out of your mind, empty beer cans rolling around inside your briefcase just to numb the pain of the task that was ahead of you that day. Its easy to make honest mistakes in that condition.

Drunk posting is one thing, drunk marketing is not any prettier

You got that right...but...then again.
Isn't this AMD marketings own fault?
For hyping Dulldozer to much before launch...and then "private" posters like JFAMD posting stuff also hyping Dulldozer a bit too much

 

[Idontcare]

I can't rationalize the hype, I can neither defend it nor can I condemn it, I simply don't understand it. I believed John when he said IPC would not decrease, I took it as a hard and fast incontrovertible boundary condition for months and months.

I'm disappointed, but its not clear what I should be disappointed with here. I really did not expect a Randy Allen repeat, but its hard to argue that's not what we got treated to, and yet it is unfair to hold any one person's feet to the fire.

I just don't understand the magnitude of that swing-and-a-miss. Gonna have to let time pass and the smoke to clear on this before I can come to any sensible understanding with much confidence.

 

[frostedflakes]

Once leaked benches started showing up I think it started to become pretty clear that IPC was lower. I think I even suggested this a couple times and invariably someone would reply that the benches were wrong because John said IPC would not go down. As if it was completely inconceivable that John could be lying or distorting the truth, could have been fed bad or incorrect information, etc.

 

[NostaSeronx]

 

[Lonbjerg]

I can't rationalize the hype, I can neither defend it nor can I condemn it, I simply don't understand it. I believed John when he said IPC would not decrease, I took it as a hard and fast incontrovertible boundary condition for months and months.

I'm disappointed, but its not clear what I should be disappointed with here. I really did not expect a Randy Allen repeat, but its hard to argue that's not what we got treated to, and yet it is unfair to hold any one person's feet to the fire.

I just don't understand the magnitude of that swing-and-a-miss. Gonna have to let time pass and the smoke to clear on this before I can come to any sensible understanding with much confidence.

I come to think of this:

 

[mrbrout]

You got to cut them some slack though Lonbjerg. If you were employed at AMD and tasked with polishing bulldozer in the marketing slides you know darn well you'd be showing up to work already blitzed out of your mind, empty beer cans rolling around inside your briefcase just to numb the pain of the task that was ahead of you that day. Its easy to make honest mistakes in that condition.

Drunk posting is one thing, drunk marketing is not any prettier

Long time AnandTech Reader and fan here (hi)

I have no horse in this race, I buy what is the fastest stable platform. But I really don't understand how you can even in Jest Defend fraudulent Marketing AMD marketing Videos claiming I980 when it's a I5-2500. And then to go on to say you can't defend nor condone nor hold "feet to fire" an AMD employee person that came on Your Site advertising grossly false IPC claims who is now conveniently silent. Are you serious? As you come from a respected website that prides itself on factual reviews I respectfully am taken back by your dismissive attitude. I appreciate your diplomacy and by all means treat them civil but these claims and people should be taken to task.

 

[zlejedi]

AMD has lost me as a customer on this one.

The straw the broke the camel's back was their alteration of the AM3 motherboard bioses to disallow Bulldozer as a drop in upgrade.

The part that really gets me, though, is that even if I were to drop a Bulldozer into my AM3 board, I would only see a benefit in applications that I don't even use.

Reminds me how pissed I were when i read on xbit am3 mobos won't support BD at all just a few months after spending cash on fresh 890GX.

But I guess it makes sense - Thuban already burned some mobos, 148Amps BD needs in Turbo mode would be a holocaust of any <8 phase VRM mobos.

 

[krumme]

I have been very hard on AMD marketing for years before most of you guys even got your diapers off. But man relax, this is just a minor stupid error. Nobody in the sane mind would publish results together with documentation that is wrong, when the effect will surely be the opposite. Blueblazer and L&#248;nbjerg knows that for sure, they just continue to frag away. So think again who is spinning now?