Why did AMD release bulldozer if they can still make good CPU's?

[JimKiler]

Althon's were good (or decent) and Ryzen appears to be solid so why did AMD release the underperforming Bulldozer CPU's? Was it a question of too much money sunk into the project before they realized it was a failure or was it one person at the top not willing to change the path or perhaps at the time they did not have a better solution?

Inquiring minds want to know.

 

[scannall]

The theory and the reality were ummm... divergent. Intel has had a few duds as well after all. Netburst and Itanium for example.

 

[JimKiler]

I read one theory that Itanium was the cause of AMD's initial success ~17 years ago and on the surface it makes sense.

 

[jpiniero]

Yeah pretty much sunk cost. They had already delayed it once before it's release so they knew it was a dud. I suppose they could have scrapped things and continued with a K10-based design but it's not like K10 was that competitive either once Sandy Bridge arrived.

 

[podspi]

The root cause was a failure of management. AMD has had a history of terrible management - and while Intel's anticompetitive actions against them didn't help, they've made a number of mis-steps since then that made things much worse.

 

[Topweasel]

Althon's were good (or decent) and Ryzen appears to be solid so why did AMD release the underperforming Bulldozer CPU's? Was it a question of too much money sunk into the project before they realized it was a failure or was it one person at the top not willing to change the path or perhaps at the time they did not have a better solution?

Inquiring minds want to know.

Well not having Keller for the K9 and K10 were a start. AMD had planned to replace the K8 (Athlon64 and Athlon 64 x2) long before Bulldozer. But the found they were making it too wide and wasn't going to work well with the future of multi-processing. So they worked the Althon64 into what we know as the Phenom and tried to create a new chip aimed at flexible and high clocking multi-processes with Bulldozer by using CMT. They bet the farm on the idea that core usage would continue to rise and that they may even be able to keep the arch going by changing up the modules later (4,18, 16 and so on) giving them the most flexibility and the perfect VM chip. They ran into two problems they could even catch quite up the the Phenom in per core IPC, clock speeds were only ok. But that still didn't seem like an issue for them as core usage was going to go up and it would leave Conroe and Phenom based chips in the dust. Then the i7 hit as they were preparing to launch . It devastated the performance and was so bad in comparison on just about every level that the amount of cores didn't matter. But they had to push through and hope that core competition on workstations and servers would keep the sales going and the rest of the market would catch up. But then biggest issue hit. Between the 3 and 2 core solutions with the Xbox and the Wii/WiiU and the complex Cell CPU (which didn't translate well to anything that BD did) and Intel refusing to budge on consumer core counts. Development simply stopped at 2-4 core usage by most games and apps. Programs started using less overhead further driving games as the only reason to get a powerful CPU outside workstation tasks and Intel offering several high core and multisocket solutions for them.

Basically AMD did what they had to do. They had nothing in the bin for further advancement of the K8 core. It was already like 8 years old. Even if they wanted to hold off (and they did for like almost a year to continue to tweak BD) the fact was they would have been even more screwed had they canceled BD and trudged on till they could get a new arch out. They did the smart thing. They released it they worked on getting the refreshes that were already baked pretty hard into their roadmap out and pretty much stopped development after that to focus on alternate solutions while rehiring Keller to come in and start fresh on a new CPU. Which is why he left before Ryzen launched. That's all he wants to do is fresh arch changes, he loves that adventure and challenge. When his new CPU from Apple comes out expect it to blow up the industry.

 

[Phynaz]

If they didn't release Bulldozer what would they sell? 5 year old Athlons?

 

[Shivansps]

If they didn't release Bulldozer what would they sell? 5 year old Athlons?

Well they could sell failed tablet APUs, oh wait, they did that as well.

AMD should have stopped selling FX about 3 years ago and AM1 should have never happen, those 2 things are a scam, at best. They should have concentrated on real desktop APU, and launch Carrizo on desktop.

 

[VirtualLarry]

The cores that made up AM1, actually aren't bad at all, close to Core2 I think, when clocked up past 2.0GHz.

 

[formulav8]

Hector Ruinz and Intel anti-competitive junk is what happened to help AMD initially for their future downfall. Intel is still trying to appeal the UE fine iirc. The fab costs, expecting fusion to just come together and other unforeseen's as well. They probably also expected multi-threading everything under the sun to happen in the immediate future.

But more Hector IMO.

 

[whm1974]

Hector Ruinz and Intel anti-competitive junk is what happened to help AMD initially for their future downfall. Intel is still trying to appeal the UE fine iirc. The fab costs, expecting fusion to just come together and other unforeseen's as well. They probably also expected multi-threading everything under the sun to happen in the immediate future.

But more Hector IMO.

Well I hope the woman now running AMD can turn it around.

 

[Doom2pro]

Althon's were good (or decent) and Ryzen appears to be solid so why did AMD release the underperforming Bulldozer CPU's? Was it a question of too much money sunk into the project before they realized it was a failure or was it one person at the top not willing to change the path or perhaps at the time they did not have a better solution?

Inquiring minds want to know.

AMD was trying to create a high clocking uArch but got burned with a combination of bad nodes from Global Foundries, and relying on automated CAD/Routing tools for Development which produced slower and less dense designs instead of "hand crafting" transistors and routing.

Now they had a lot of good ideas and some of those are in Zen and are responsible for getting Zen to clock as high as it does, so Bulldozer isn't a complete failure.

 

[Rayniac]

If they had just continued making Phenom II CPUs with improved manufacturing process they could have had a better product than Bulldozer but then they wouldn't be where they are today with Zen.

 

[formulav8]

Well I hope the woman now running AMD can turn it around.

She's a smart female. She's also a high end cpu/process/uarch nerd who knows her stuff.

 

[BeepBeep2]

AMD was trying to create a high clocking uArch but got burned with a combination of bad nodes from Global Foundries, and relying on automated CAD/Routing tools for Development which produced slower and less dense designs instead of "hand crafting" transistors and routing.

Now they had a lot of good ideas and some of those are in Zen and are responsible for getting Zen to clock as high as it does, so Bulldozer isn't a complete failure.

It seems they are still relying quite a bit on those tools - look at the amount of synthesized logic that look like blooms vs. Intel's cluster-of-chunks.
I think they've gotten very clever with their power saving advances / technologies and layout wherever possible, but even Ryzen from the surface doesn't seem to be much more "hand crafted" than Bulldozer as far as the use of higher-level design.

Deneb was already a high clocking design on the mature 45nm process, I have a chip that does 4.4 GHz stable on water. We all think GloFo's 32nm was a dud but I do wonder what a K10.5 shrink would have done (I don't really count Llano, since they made some changes that seemed to increase IPC but hurt clockspeed)

 

[IntelUser2000]

The crux of the matter is that they, the CPU designers(includes engineers and management) don't fully know the direction they want to go themselves.

Intel guys were SO sure of Netburst quickly scaling to 10GHz. After the failure of clock scaling it was very interesting to visit earlier rumors of Prescott at 5GHz speeds increasing 400-500MHz every few quarters.

With Bulldozer remember AMD claiming wild things about CMT.

 

[unseenmorbidity]

If they had just continued making Phenom II CPUs with improved manufacturing process they could have had a better product than Bulldozer but then they wouldn't be where they are today with Zen.

I am not sure that logically follows. Faildozer didn't do them any favors.

 

[Haserath]

I think they already had a contract for the Titan supercomputer so they had to get something for it. They were led to believe Bulldozer was better than it ended up too.

 

[sirmo]

Yeah Bulldozer really damaged AMD's reputation. Up until that point they had a string of pretty decent CPUs and some awesome ones like the Athlon Thunderbird, and the Athlon64 Hammers. Barcelona and the TLB bug (Phenom I) was the only blemish on their reputation before Bulldozer really. I mean even though Phenom II wasn't really breaking any records.. it was a very solid CPU.

Bulldozer was the only AMD generation I didn't buy. And I really thought AMD were done at that point.

 

[Fanatical Meat]

She's a smart female. She's also a high end cpu/process/uarch nerd who knows her stuff.

They replaced the youngish guy with the punchable face?

 

[Rayniac]

@unseenmorbidity but it did. Zen is using at least one if not more of the design ideas that came out with Bulldozer. Mainly the cores that can execute 2 threads at the same time. Bulldozer was the trial and error and now the new design shows what they learned of that.

 

[scannall]

@unseenmorbidity but it did. Zen is using at least one if not more of the design ideas that came out with Bulldozer. Mainly the cores that can execute 2 threads at the same time. Bulldozer was the trial and error and now the new design shows what they learned of that.

Same with Intel and Netburst. While Netburst was a bust overall, they did learn some things they took over to Core, along with the good parts of the P3.

 

[AnandThenMan]

Why did Intel make Netburst when they can make a good CPU? Why did Intel try and come out with Larrabee? Because companies are not perfect, people are not perfect.

 

[Topweasel]

Why did Intel make Netburst when they can make a good CPU? Why did Intel try and come out with Larrabee? Because companies are not perfect, people are not perfect.

Yeah let's not forget the nearly 10 billion dollar bomb that IA64 was.

 

[AnandThenMan]

Yeah let's not forget the nearly 10 billion dollar bomb that IA64 was.

Also a very pivotal time overall in computing history. AMD64 probably saved us all from being forced into an Intel only instruction set.

BTW quote is broken for me so have to do it manually.