Performance/watt comparison

[Denithor]

Just a quick calculation from the x264 2nd pass charts of frames/sec and power consumption:

i7 2600k = 36/155.4 = 0.232
i5 2500k = 28.7/133.3 = 0.215
i5 2400 = 26.8/131.6 = 0.204
FX-8150 = 35.8/229 = 0.156
X6 1100T = 31.5/200 = 0.158
X4 975 = 23.5/183.8 = 0.128

Which shows, for x264 encoding, the 2600k does 49% more work for the same power and the 2500k does 38% more work for the same power as the new flagship Bulldozer chip.

Heck, even the PhII chip has a slight advantage in work/watt versus the new chip! And this is a benchmark that more or less favors the Bulldozer architecture...

 

[RussianSensation]

The most eye-opening to me was this:

This pretty much means if Intel sells us a $225 6-core Haswell in 2013 (we can hope), it's pretty much lights out for Bulldozer. AMD should start working on improving IPC at all costs.

 

[Arkaign]

The most eye-opening to me was this:

This pretty much means if Intel sells us a $225 6-core Haswell in 2013 (we can hope), it's pretty much lights out for Bulldozer. AMD should start working on improving IPC at all costs.

That chart just confirms that AMD needs a tweaked 32nm lineup of PhIII in 4/6/8 core options. BD goes in the garbage where it belongs.

 

[Dresdenboy]

The most eye-opening to me was this:

This pretty much means if Intel sells us a $225 6-core Haswell in 2013 (we can hope), it's pretty much lights out for Bulldozer. AMD should start working on improving IPC at all costs.

First those problems limiting BDs performance have to be sorted out. It's a matter of time.

 

[Dresdenboy]

That chart just confirms that AMD needs a tweaked 32nm lineup of PhIII in 4/6/8 core options. BD goes in the garbage where it belongs.

In my instruction execution analysis I've shown that BD isn't that far behind. But Husky looked really good in integer performance. Even compared to SB. But this result of synthetic AIDA tests doesn't remove other bottlenecks.

 

[Abwx]

First those problems limiting BDs performance have to be sorted out. It's a matter of time.

Can you please point the most probable of "those problems" ?..

Indeed , the perf/W doesnt seems to be up to expectations ,
not to talk about perfs/mm2..

 

[Ferzerp]

Can you please point the most probable of "those problems" ?..

Indeed , the perf/W doesnt seems to be up to expectations ,
not to talk about perfs/mm2..

I think he means the problem of it being AMD's NetBurst (MORE CORES! instead of MORE MHZ!, but still). Sort that out and it will work fine.

 

[NostaSeronx]

Can you please point the most probable of "those problems" ?..

Miss predictions

Stalls in the front-end

FMAC Bandwidth is getting choked by something

Some but not all legacy code is from the Mrom Engine and not the fastpath decoders

Faulty Transistor placement(crossing streams)

Over hyped

Latency in memory and caches

 

[ShadowVVL]

I guess we can expect weak cpu performance from trinity as well.
I was looking forward to trinity,If it turns out to be a slow power hog of a chip Im going IB and haswell for future builds

 

[veri745]

I guess we can expect weak cpu performance from trinity as well.
I was looking forward to trinity,If it turns out to be a slow power hog of a chip Im going IB and haswell for future builds

Trinity is supposed to include an updated (Piledriver) core, though. Hopefully there are some improvements in it.

This is a brand new architecture, which means there may be large gains to be had from relatively minor tweaks. Unlike Stars, which has pretty much been tweaked to death.

 

[Joseph F]

Trinity is supposed to include an updated (Piledriver) core, though. Hopefully there are some improvements in it.

This is a brand new architecture, which means there may be large gains to be had from relatively minor tweaks. Unlike Stars, which has pretty much been tweaked to death.

If it were that simple, AMD would have just tweaked it for the initial launch. I think that BD architecture's performance is going to be horrendous until AMD comes out with a whole new arch.

 

[AnandThenMan]

I think that BD architecture's performance is going to be horrendous until AMD comes out with a whole new arch.

Bulldozer is not that bad, and is definitely fixable.

 

[Despoiler]

Bulldozer is not that bad, and is definitely fixable.

Yup and already there is a rumor that AMD knows what the performance issue is, fixed it in Trinity engineering samples, but it didn't make it to BD for launch. If true, things might get interesting.

 

[exar333]

Bulldozer is not that bad, and is definitely fixable.

Only took one day to kick-in damage control...

 

[exar333]

Yup and already there is a rumor that AMD knows what the performance issue is, fixed it in Trinity engineering samples, but it didn't make it to BD for launch. If true, things might get interesting.

Not likely. They have had years to work out issues and all of a sudden they 'know what the issues are' right at launch? If this was the case, they would have delayed and shipped with better performance. Initial product reception is HUGE.

PhII ended-up being a decent architecture, but the whole Phenom debacle killed the brand off the get-go, and dampened reception to the whole Phenom line. BD is doing that again.

That's exactly why Intel switched to 'Core' as their performance line, following the netburst debacle.

 

[AnandThenMan]

Only took one day to kick-in damage control...

What are you talking about? Damage control by who?

 

[kdubbs]

Bulldozer, you are a strange bird. Performance/$$$ isn't abysmal relative to intel's current midlevel offerings, and will likely improve with tweaks and software/OS optimization. If it was a tiny, efficient chip, I would be quite pleased.

But DAMN, sooo many xtors and it is huge, even with at 32 nm. And you can't sugarcoat the energy demands--inexcusable for a new process. There just seems to be so much that needs to addressed to make this a competitive product--and that is assuming the competition is standing still. I'm not optimistic for future BD-based products.

IDC has stated it repeatedly in the past: you can't expect miracles from AMD engineers when their resources are 1/10 of Intel's and the two companies are essentially playing the same game. Either change the game or gain access to greater resources through acquisition by a deep-pocketed corporation. We all would have liked to see AMD pull themselves up by the bootstraps and go toe to toe with Goliath, but that may have been a fairytale all along.

 

[Despoiler]

Not likely. They have had years to work out issues and all of a sudden they 'know what the issues are' right at launch? If this was the case, they would have delayed and shipped with better performance. Initial product reception is HUGE.

PhII ended-up being a decent architecture, but the whole Phenom debacle killed the brand off the get-go, and dampened reception to the whole Phenom line. BD is doing that again.

That's exactly why Intel switched to 'Core' as their performance line, following the netburst debacle.

The TLB bug shipped and the BIOS fix crippled the performance. Major issues can sneak past all of your validation testing. Not only sneak past, but also be ignored if you are working on things like getting power usage under control. If the CEO says ship it you ship it.

 

[Ferzerp]

I don't think they worked very hard on getting power usage under control.

 

[exar333]

I don't think they worked very hard on getting power usage under control.

Beat me to it...

 

[Nemesis 1]

Yup and already there is a rumor that AMD knows what the performance issue is, fixed it in Trinity engineering samples, but it didn't make it to BD for launch. If true, things might get interesting.

To believe anything AMD says at this time is lacking in common sense. The First BD samples showed up in early dec of 2010. AMD and Its people knew its performance at that time. As did others . We thought AMD might get another 10% by launch . But they only went for clock speed . I thought BD would be a bit better not alot .

 

[Nemesis 1]

Bulldozer, you are a strange bird. Performance/$$$ isn't abysmal relative to intel's current midlevel offerings, and will likely improve with tweaks and software/OS optimization. If it was a tiny, efficient chip, I would be quite pleased.

But DAMN, sooo many xtors and it is huge, even with at 32 nm. And you can't sugarcoat the energy demands--inexcusable for a new process. There just seems to be so much that needs to addressed to make this a competitive product--and that is assuming the competition is standing still. I'm not optimistic for future BD-based products.

IDC has stated it repeatedly in the past: you can't expect miracles from AMD engineers when their resources are 1/10 of Intel's and the two companies are essentially playing the same game. Either change the game or gain access to greater resources through acquisition by a deep-pocketed corporation. We all would have liked to see AMD pull themselves up by the bootstraps and go toe to toe with Goliath, but that may have been a fairytale all along.

It was AMD that decided to use gate first in their 32nm fabs befor they sold them . SOI as you scale down in size becomes less effective. AMD needs to get off IBMs tit.

 

[Vesku]

We'll have to see how GF 32nm improvements and their 28nm turn out, they have even more incentive as a separate company to make their process attractive. Especially since they have been trying to get low power chip maker business for their 28nm