[OFFICIAL] Bulldozer Reviews Thread - AnandTech Review Posted

[Munky]

That depends how you measure IPC. If you count a multi-threaded scenario, I'm sure there are cases where BD does beat the Ph2.

Also, the whole IPC and clock for clock comparison is purely academic as opposed to practical. If the chip with a longer pipeline and lower IPC can reach much higher clocks, then why does it matter? I believe the real measure of cpu efficiency is performance/watt, and clearly BD loses that battle by a long shot.

 

[Magic Carpet]

http://vr-zone.com/articles/amd-fx-8150-cpu-overclocking-review-a-bulldozer-for-gamers-/13694-1.html

2600k @ 4.7 vs 8150 @ 4.7

Some food for thought.

 

[Munky]

Here's some more food for thought:
http://vr-zone.com/articles/amd-fx-8150-cpu-overclocking-review-a-bulldozer-for-gamers-/13694-6.html

In Cinebench, BD single-thread performance scales horrendously with OC'ing. It's almost like there's an insurmountable bottleneck.

 

[MentalIlness]

The Computer Base review in the OP, also tests the 6 core and 4 core cpu's.

 

[Magic Carpet]

Here's some more food for thought:
http://vr-zone.com/articles/amd-fx-8150-cpu-overclocking-review-a-bulldozer-for-gamers-/13694-6.html

In Cinebench, BD single-thread performance scales horrendously with OC'ing. It's almost like there's an insurmountable bottleneck.

Balance is key. Hopefully, a stepping away.

 

[Phynaz]

As usual , you re completely out of subject..

Since AMD said 80% of the theorical perfs of an usual
multi core their numbers are rather right providing
the softs are heavily multithreaded , wich is hardly
the case for many softs in this slide , specially some games..

Your humor bone broken?

 

[drizek]

Here's some more food for thought:
http://vr-zone.com/articles/amd-fx-8150-cpu-overclocking-review-a-bulldozer-for-gamers-/13694-6.html

In Cinebench, BD single-thread performance scales horrendously with OC'ing. It's almost like there's an insurmountable bottleneck.

I didn't want to read the whole article,

is Turbo enabled or not?

That could explain most of the poor scaling.

 

[drizek]

why does it matter? ... performance/watt

You answered your own question. When has a CPU that focused on max clockspeed been competitive in performance/watt?

Bulldozer seems like a great CPU for, if anything, a laptop. Run it at low clocks and low voltages and you get potentially lower idle power draw than Sandy Bridge with a lot of breathing room for performance when you plug it in and throw TDP out the window.

Most computers are idle most of the time.

Bulldozer seems like it would be particularly poor in servers due to its high power consumption, but I haven't seen anyone show how it scales with voltage/wattage at low frequencies.

 

[Abwx]

Your humor bone broken?

Not at all , just some reminder since you did compare
the wrong colones in the slide above but i guess that
it was to be expected...

 

[Abwx]

Bulldozer seems like it would be particularly poor in servers due to its high power consumption, but I haven't seen anyone show how it scales with voltage/wattage at low frequencies.

Yet , there s a 35W Interlagos part...

Frequencies will not exceed 2.5 ghz for the higher clocked
16C Interlagos , at least for the current stepping...

 

[frostedflakes]

You answered your own question. When has a CPU that focused on max clockspeed been competitive in performance/watt?

Bulldozer seems like a great CPU for, if anything, a laptop. Run it at low clocks and low voltages and you get potentially lower idle power draw than Sandy Bridge with a lot of breathing room for performance when you plug it in and throw TDP out the window.

Most computers are idle most of the time.

Bulldozer seems like it would be particularly poor in servers due to its high power consumption, but I haven't seen anyone show how it scales with voltage/wattage at low frequencies.

Power consumption only seems to get crazy at high clocks, 4GHz+. Seems to do fine at lower clocks, though. At stock the FX-8150 stays within the 125W TDP with all the cores loaded. 16-core Interlagos is clocked really low (2.3GHz base and 2.8GHz turbo for the highest model IIRC) and has a TDP of 115W. I don't see power consumption being an issue for it. As long as you keep clocks low it seems to be OK. But past a certain threshold power consumption seems to increase exponentially with clock speed, it's nuts.

 

[Phynaz]

Not at all , just some reminder since you did compare
the wrong colones in the slide above but i guess that
it was to be expected...

Since I don't know what a colone is, please excuse me for comparing the wrong one.

 

[drizek]

Power consumption only seems to get crazy at high clocks, 4GHz+. Seems to do fine at lower clocks, though. At stock the FX-8150 stays within the 125W TDP with all the cores loaded. 16-core Interlagos is clocked really low (2.3GHz base and 2.8GHz turbo for the highest model IIRC) and has a TDP of 115W. I don't see power consumption being an issue for it. As long as you keep clocks low it seems to be OK. But past a certain threshold power consumption seems to increase exponentially with clock speed, it's nuts.

Ya, I've been wanting to see some undervoltaged benchmarks.

I would also like to see Interlagos for desktops. I think at that point AMD might actually have an advantage. It should have the same single threaded performance as an 8150 but with the potential for double hte performance in multithreaded. If they can sell an Interlagos desktop part for $300, they might have a CPU that a lot of people would prefer over a 2600K.

 

[Abwx]

Since I don't know what a colone is, please excuse me for comparing the wrong one.

You are right about the spelling , read "column" , but still ,
with the correct wording , your comparison will be further
out of track , unfortunately.

 

[-Slacker-]

The Computer Base review in the OP, also tests the 6 core and 4 core cpu's.

Mother of god the FX 4100 is slower than a Llano in everything! Probably slower than an athlon II x4 too.

This chip should be $80 ... what is AMD thinking ?

 

[Meghan54]

This chip should be $80 ... what is AMD thinking ?

That there are years of R&D to recoup?

 

[Idontcare]

That there are years of R&D to recoup?

Recouping past investments is one thing, but really what they are doing when selling above-cost is attempting to generate a cashflow that can sustain today's R&D on tomorrow's product.

Bulldozer was paid for on the back of Phenom I and II profits (as in gross margins, not net profits).

But Piledriver, steamroller, and excavator is entirely dependent on AMD"s cash reserves and cash flow. That is why they can't afford to sell Zambezi for $80. It would be a death sentence for their CPU R&D.

 

[Phynaz]

You are right about the spelling , read "column" , but still ,
with the correct wording , your comparison will be further
out of track , unfortunately.

What should I be comparing? Rather than just telling me I'm wrong, how about correcting me.

 

[jacktesterson]

delete

 

[Regs]

Recouping past investments is one thing, but really what they are doing when selling above-cost is attempting to generate a cashflow that can sustain today's R&D on tomorrow's product.

Bulldozer was paid for on the back of Phenom I and II profits (as in gross margins, not net profits).

But Piledriver, steamroller, and excavator is entirely dependent on AMD"s cash reserves and cash flow. That is why they can't afford to sell Zambezi for $80. It would be a death sentence for their CPU R&D.

You're dead wrong. R&D and other investment liabilities are proportionate the future sales forecasts of current products. Companies are in constant debt because they hope to use that debt to make more money. If they're not selling as well as they hoped, they're debt grows larger and interest payments are not forgiving on their cash flows.

 

[Idontcare]

You're dead wrong. R&D and other investment liabilities are proportionate the future sales forecasts of current products. Companies are in constant debt because they hope to use that debt to make more money. If they're not selling as well as they hoped, they're debt grows larger and interest payments are not forgiving on their cash flows.

Ah, I see. So in this world of yours then I suppose a company like AMD has never reduced its R&D budget, nor laid off R&D employees, in the wake of financial softness in its current quarterly revenue?

What about Nokia? DELL? HP?

I wish someone had clued TI into this, for decades our quarterly budget went up and down in tune with quarterly revenue because the shareholders wanted us to stick to a fixed target of R&D expense to revenue ratio. Including layoffs, hiring freezes, cancelled projects, and so on...even when we were profiting but not growing revenue.

Silly TI. They had no idea they were dead wrong for all those decades.

 

[Elixer]

But Piledriver, steamroller, and excavator is entirely dependent on AMD"s cash reserves and cash flow. That is why they can't afford to sell Zambezi for $80. It would be a death sentence for their CPU R&D.

Yeah, then need to recover costs, but it doesn't seem like the current price point will help them at all.

They would sell far more CPUs at a lower, more realistic price point than to gouge people with the current price point.

Only the clueless (or people who have a specific workload that works great with this CPU) would buy a lackluster CPU at the current price point, and that means no sale at all.
If you mean retail, then they are already offering all the big players huge discounts to get them to use the BD anyway.

 

[jacktesterson]

Yeah, then need to recover costs, but it doesn't seem like the current price point will help them at all.

They would sell far more CPUs at a lower, more realistic price point than to gouge people with the current price point.

Only the clueless (or people who have a specific workload that works great with this CPU) would buy a lackluster CPU at the current price point, and that means no sale at all.
If you mean retail, then they are already offering all the big players huge discounts to get them to use the BD anyway.

They have to take a Risk and price them where they are to start.

The 8150's are sold out everywhere.

NCIX.com has no bulldozers in stock.

 

[Blue Shift]

Yeah, then need to recover costs, but it doesn't seem like the current price point will help them at all.

They would sell far more CPUs at a lower, more realistic price point than to gouge people with the current price point.

Only the clueless (or people who have a specific workload that works great with this CPU) would buy a lackluster CPU at the current price point, and that means no sale at all.
If you mean retail, then they are already offering all the big players huge discounts to get them to use the BD anyway.

AMD doesn't need to manufacture and sell a massive number of their FX Bulldozer models, especially with GloFo's 32nm capacity being what it is.

Bulldozer has two purposes that are probably more important to AMD than the enthusiast market:
1) Servers. Interlagos should be quite effective at handling server workloads, as long as power consumption is reasonable.
2) Trinity. Since a modified form of Bulldozer will form the CPU portion of Trinity, AMD needed to see what the Bulldozer architecture's current strengths and weaknesses are in the realm of home computing. Now they have a better idea of what to address before Trinity tapes out.

As long as AMD can expect to sell every single Llano that they can possibly produce, it doesn't really matter all that much how Bulldozer sells; they can just produce more Llanos instead. Bulldozer may have more markup, but that doesn't mean Llano isn't profitable as well.

 

[Phynaz]

Since a modified form of Bulldozer will form the CPU portion of Trinity, AMD needed to see what the Bulldozer architecture's current strengths and weaknesses are in the realm of home computing.

So BD is a multi-billion dollar experiment?

Strengths and weaknesses are part of the design phase (simulation). You don't just design a product and throw it out there to see how it works. You should have a pretty clear understanding of how things work/perform before you ever have silicon in hand.