The QuadFather is pissed off...

[nenforcer]

He wants his cores back.

 

[nenforcer]

Instead of this.

 

[nenforcer]

We got this.

That's not appropriate for the tech forums, particularly as an embed
-ViRGE

 

[Ancalagon44]

Ewwww

 

[nenforcer]

Oh the irony.

 

[Ferzerp]

For some reason that last picture makes me think of Kerbal Space Program

 

[soccerballtux]

haha

 

[soccerballtux]

Oh the irony.

if only we had that instead...

 

[Idontcare]

if only we had that instead...

Near as I can gather, from Llano core/cache analyses versus bulldozer module/cache analyses...2 Llano cores w/2MB L2$ is nearly identical in footprint to 1 bulldozer module w/2MB L2$.

In other words, AMD really has a choice to produce 8-core Zambezi w/16MB$ or 8-core Llano (no GPU) w/16MB$ for about the same die size (an octo-llano would be slightly smaller).

And from everything we've seen so far, an octo-llano would thump an 8-core zambezi.

Everyone makes mistakes. Look where Larrabee went. Bulldozer is not looking like a step forward. Firing Dirk was.

 

[Vesku]

Yep, seems likely Dirk got canned for plowing on with Bulldozer. Which is odd considering his background was in the predecessors to Llano (wasn't it?). Perhaps he had the vision to see that the time is ripe for Pentium 4x2. We'll know next year, if IPC actually drops more with Piledriver than it will be pretty eerie mirroring.

 

[soccerballtux]

Near as I can gather, from Llano core/cache analyses versus bulldozer module/cache analyses...2 Llano cores w/2MB L2$ is nearly identical in footprint to 1 bulldozer module w/2MB L2$.

In other words, AMD really has a choice to produce 8-core Zambezi w/16MB$ or 8-core Llano (no GPU) w/16MB$ for about the same die size (an octo-llano would be slightly smaller).

And from everything we've seen so far, an octo-llano would thump an 8-core zambezi.

Everyone makes mistakes. Look where Larrabee went. Bulldozer is not looking like a step forward. Firing Dirk was.

From an architecture standpoint I don't understand why it was a step backwards. On paper it seems like it would be better.

 

[drizek]

Bulldozer cannot be considered a step backwards. STARS doesn't support modern CPU extensions. You can get by with one for a little bit, but the addition of features like AES-NI on competing Intel products gives intel a huge leg up.

I don't think a Phenom III is an option really. AMD only has one path forward: to make Bulldozer not suck.

 

[IntelUser2000]

From an architecture standpoint I don't understand why it was a step backwards. On paper it seems like it would be better.

I doubt it was better on paper either. Things are only as it seems to the eyes of the beholder. Here's MO:

1. Remember when on AMD analyst day they revealed the architecture of Bulldozer? Few might remember them saying both Bulldozer and Bobcat is designed for "Knee of the Curve" performance.

Knee of the Curve: In a graph showing exponentials that shows trends, there is a point where the graph has a steep gain, then it gains a bit more, but is mostly leveled out.

In traditional CPU designs, the engineers pulled heroic feats to gain 1% extra performance. What AMD is saying is with Bulldozer and Bobcat, they aimed for the easy hanging fruit more or less. From there I figured they are not aiming for Intel's performance per clock. To be honest, I thought it would have been doing better though.

2. Related to #1, they focused on clocks. They simplified the design to save die size. Then suddenly they said the clock time is reduced something like 20%. The most likely way is they increased pipeline stages. The problem with high clock speed designs is when you can't clock them high.

In other words, AMD really has a choice to produce 8-core Zambezi w/16MB$ or 8-core Llano (no GPU) w/16MB$ for about the same die size (an octo-llano would be slightly smaller).

I doubt they had better choice. Remember they seem to be having problem clocking Llano high at desktop frequencies. Now imagine that with 2x the cores.

 

[nenforcer]

Yep, seems likely Dirk got canned for plowing on with Bulldozer. Which is odd considering his background was in the predecessors to Llano (wasn't it?). Perhaps he had the vision to see that the time is ripe for Pentium 4x2. We'll know next year, if IPC actually drops more with Piledriver than it will be pretty eerie mirroring.

Dirk was pretty well entrenched at AMD and had meant a lot to the company since the first Athlon's. I knew it had to be something more serious than just a "lack of a mobile / tablet" strategy.

I definitely don't think IPC can possibly drop worse with Piledriver. In fact, I'm hoping with GF manufacturing improvements / yields we will see the foreward looking architecture reach its potential.

I never buy new chips / motherboards / architectures when the first come out anyways. I like to see how they pan out after a little time in the market to spot any other bugs. Glad I avoided the entire P67 chipset 3GBps SATA glitches fiasco earlier this year.

 

[Vdubchaos]

I never buy new chips / motherboards / architectures when the first come out anyways. I like to see how they pan out after a little time in the market to spot any other bugs. Glad I avoided the entire P67 chipset 3GBps SATA glitches fiasco earlier this year.

Agreed, I usually give AT LEAST few months. That's exactly why I went with 2500k, I didn't want to wait few months for "BD crap to pan out.

And now we have Z68 as well.

And IVy Bridge being 1155.

No brainer, it was a perfect time to go for SB