upcoming AMD’s Bulldozer FX Launch Lineup revealed

[Arkadrel]

I quoted you, Idontcare, because I wanted to point out the ability to run multiple instances which hadn't been covered thus far.

To soccerballtux: Yeah, I'm afraid of that. But does that portend a good chance or a poor chance of getting the 8100 to 4.2GHz? Anyone have historical perspective on AMD's launch binning?

The 8100 does 3.7ghz with turbo (this makes it use 95watts).
Overclock it to 4ghz for turbo? sure, but you ll probably make it use like 110watts instead then 😛



Most AMD cpu's can easily get another 500mhz or so squeezed out of them with overclocking.

Even Anandtech managed (with not the best overclocking skills) to get:
the Phenom II x 4 955 from 3.2Ghz (stock) -> 3.8Ghz.


Im pretty sure, any of those bulldozers, you will be able to add another 500mhz, to their default clock's, and probaly also to their turbo modes.




I think we *could* be seeing the 8150 versions with a stock turbo of 4.2ghz,
overclock and maybe hit 4.7ghz when useing overclock+turbo (maybe even higher).

 

[Plimogz]

I think we *could* be seeing the 8150 versions with a stock turbo of 4.2ghz,
overclock and maybe hit 4.7ghz when useing overclock+turbo (maybe even higher).

4.7GHz with safe voltage would be good. Somehow I doubt that most top SKU chips will have 500MHz of headroom. Seems like AMD can ill afford to leave quite that much performance on the table -- assuming that BD isn't the improbable game-changer which I persist in hoping that it'll be.

I wonder how long it'll be before AMD launch a 140W SKU. Or maybe even take a page out of Intel's book and push their top SKU to 150W.

 

[soccerballtux]

I quoted you, Idontcare, because I wanted to point out the ability to run multiple instances which hadn't been covered thus far.

To soccerballtux: Yeah, I'm afraid of that. But does that portend a good chance or a poor chance of getting the 8100 to 4.2GHz? Anyone have historical perspective on AMD's launch binning?

poor. My 965 can't get an inch past 4.05ghz, it actually might not run at 4.05ghz either, it seemed stable but I run at 4g to call it a day.

 

[soccerballtux]

The 8100 does 3.7ghz with turbo (this makes it use 95watts).
Overclock it to 4ghz for turbo? sure, but you ll probably make it use like 110watts instead then 😛



Most AMD cpu's can easily get another 500mhz or so squeezed out of them with overclocking.

Even Anandtech managed (with not the best overclocking skills) to get:
the Phenom II x 4 955 from 3.2Ghz (stock) -> 3.8Ghz.


Im pretty sure, any of those bulldozers, you will be able to add another 500mhz, to their default clock's, and probaly also to their turbo modes.




I think we *could* be seeing the 8150 versions with a stock turbo of 4.2ghz,
overclock and maybe hit 4.7ghz when useing overclock+turbo (maybe even higher).

nah, the trick is that 500mhz IS the turbo. In my opinion, but I'm feeling pretty certain of this.

 

[eternalone]

Good looking Im getting one of the 95 watters. Probably the FX 4100 since it looks like a good bargain well that is till I see the prices. Anyone know if cores will be able to be unlocked????

 

[Arzachel]

Good looking Im getting one of the 95 watters. Probably the FX 4100 since it looks like a good bargain well that is till I see the prices. Anyone know if cores will be able to be unlocked????

It has been hinted at, but I wouldn't take it for granted. The quad is, imo, the most interesting chip in the lineup, due to the TDP and the (possible?) clocks. If the modules are fully gated, then it should always turbo unless faced with very FP heavy code. Might make decent overclockers.

 

[Voo]

Bitcoin mining via phoenix needs a dedicated core, half a core to a full core dedicated to browsing-music-internet voice. If I had more than a tri-core I'd keep at least one VM up 24/7.

And what exactly would you doing with your VM in the background? Bitcoin mining isn't especially useful anymore (but we'll let it there, can't harm), so this leaves.. browsing-music-VOIP which needs at most maybe a tenth of a core (and that's being quite lenient)

Yup seems like you really need those eight cores.

 

[trollolo]

so what is this "turbo", is it like auto overclocking?

 

[bryanW1995]

Wonder how tight the binning will be? Will I have good odds of getting an 8100 to or even beyond 8150 clocks?

That is key, if an 8100 is relatively cheap and oc's to a similar level to 8150 (after oc) then we could have budget overclocking back in force. Hopefully they'll at least be competitive with 2500k.

so what is this "turbo", is it like auto overclocking?

Yes. It bumps up to the turbo clocks under medium/heavy usage, then drops back down again.

 

[Arzachel]

so what is this "turbo", is it like auto overclocking?

Yeah, but without going over the TDP limits. If any modules are idle, they can be power gated and the TDP headroom can be then used to increase the clocks on those modules are under load. This might be server specific, but even with all integer units in use, turbo might still increase clocks if the fpu's see little work. Intel already has something similar.

 

[gevorg]

FX-6100 vs i5-2500K would be very interesting comparison. Wonder if the two extra cores can make up for the lack in single threaded performance.

 

[exar333]

Anyone else a little worried that all models top-out at the same frequency (3.6ghz) regardless of the number of cores?

The models are very odd to me as well...(8-core) max SKU 3.6 ghz; (6-core) max SKU 3.3ghz; and the (4-core) max SKU at 3.6 ghz.

Why is the 6-core freq so low? Why doesn't the quad come with standard freq > than the 8-core?

Why is the 6-core and 4-core turbos so low? These should be greater than 4 ghz, right? Something is odd here especially with the lower-SKU turbo levels.

 

[Idontcare]

Anyone else a little worried that all models top-out at the same frequency (3.6ghz) regardless of the number of cores?

I'm not surprised.

It probably means they are binning all the chips with the threshold that they validate as functioning just fine at 3.6GHz, and then they lock out cores that can't. Then they set the default multiplier as needed to fill up their inventory requests from resellers.

I see it as an artifact of a streamlined binning process more than anything else. Binning takes time too you know, and time is money. Its only natural that companies like Intel and AMD are seeking ways to reduce the expense of validation and test.

 

[Mopetar]

I'm not surprised.

It probably means they are binning all the chips with the threshold that they validate as functioning just fine at 3.6GHz, and then they lock out cores that can't. Then they set the default multiplier as needed to fill up their inventory requests from resellers.

I see it as an artifact of a streamlined binning process more than anything else. Binning takes time too you know, and time is money. Its only natural that companies like Intel and AMD are seeking ways to reduce the expense of validation and test.

The real question is if 3.6 GHz is an aggressive amount to get the most performance out of the chips or if 3.6 GHz is aggressive in the sense that it allows them to send out more working chips. If it's the latter, it should indicate that there might be some good overclocking potential.

 

[Idontcare]

The real question is if 3.6 GHz is an aggressive amount to get the most performance out of the chips or if 3.6 GHz is aggressive in the sense that it allows them to send out more working chips. If it's the latter, it should indicate that there might be some good overclocking potential.

I want to be hopeful.

But...from what we have seen of GloFo's 32nm SOI based on Llano, and from what we know to expect of a gate-first HKMG integration from device physics standpoint, I know my hopefullness is not really justifiable.

And yet...I really do want to be hopeful.

 

[Vesku]

And what exactly would you doing with your VM in the background? Bitcoin mining isn't especially useful anymore (but we'll let it there, can't harm), so this leaves.. browsing-music-VOIP which needs at most maybe a tenth of a core (and that's being quite lenient)

Yup seems like you really need those eight cores.

VM would be a deployment test bed, I bitcoin mine because electricity is cheap here and I have a couple of Radeons that mostly sit idle, browsing-music-VOIP uses anywhere from 20%-50% of a core combined usually. So yes I'd like at least 6 cores and why not go 8 if the price is right?

 

[sandorski]

Anyone else a little worried that all models top-out at the same frequency (3.6ghz) regardless of the number of cores?

The models are very odd to me as well...(8-core) max SKU 3.6 ghz; (6-core) max SKU 3.3ghz; and the (4-core) max SKU at 3.6 ghz.

Why is the 6-core freq so low? Why doesn't the quad come with standard freq > than the 8-core?

Why is the 6-core and 4-core turbos so low? These should be greater than 4 ghz, right? Something is odd here especially with the lower-SKU turbo levels.

This also struck me as Odd. I suspect in time there may be more 6 and 4 core processors released, but it seems to me that AMD is primarily wanting 8 core processors and only releasing 6 and 4 cores in order to Sell what doesn't handle 8 cores.

 

[386DX]

Anyone else a little worried that all models top-out at the same frequency (3.6ghz) regardless of the number of cores?

The models are very odd to me as well...(8-core) max SKU 3.6 ghz; (6-core) max SKU 3.3ghz; and the (4-core) max SKU at 3.6 ghz.

Why is the 6-core freq so low? Why doesn't the quad come with standard freq > than the 8-core?

You kinda have to ignored the 125W FX-8150 when looking at the SKU. That chip is obviously cherry picked to make BD look good in Benchmarks. AMD knows review sites are going to test the highest end chip to judge the CPU design and that's what this chip is for.

If we ignore the 125W chip from the list then nothing is odd. The 8 core runs at a max speed that's less then a 6 core, that's less then a 4 core. It makes sense because they all have to run within the 95W thermal envelope. An 8 core CPU at the same frequency as a 4 core CPU is obviously gonna produce more heat when all cores are running. Probably exceed the 95W TDP limit hence why they have to run the 6/8 cores at lower frequency. It doesn't mean the 8 core or 6 core will run slower then the 4 core when only 4 cores are being used (remember that's what turbo is for), the 8 core may run at 2.8/3.1 GHz when all cores are used, it may run at 3.3 GHz when 6 cores are in use, or 3.6GHz when only 4 cores are needed.

Why is the 6-core and 4-core turbos so low? These should be greater than 4 ghz, right? Something is odd here especially with the lower-SKU turbo levels.

The turbo level is obviously done to make sure the 4 core chips don't beat the 6/8 cores. Imagine if they made the 4 core turbo faster then the 8 core, you''ll get in a situation where the 4 core chips beats your 8 core chips in all the benchmarks that's not heavily multi-threaded. It'll be a marketing disaster. Your cheaper 4 core CPU beats your higher end 8 core CPU in half the benchmarks.

 

[Vesku]

Other question is will the desktop BDs let you run higher than TDP limit? It seems that Llano has a combined CPU+GPU TDP cap that has made overclocking both CPU and GPU at the same time difficult.

 

[VirtualLarry]

The turbo level is obviously done to make sure the 4 core chips don't beat the 6/8 cores. Imagine if they made the 4 core turbo faster then the 8 core, you''ll get in a situation where the 4 core chips beats your 8 core chips in all the benchmarks that's not heavily multi-threaded. It'll be a marketing disaster. Your cheaper 4 core CPU beats your higher end 8 core CPU in half the benchmarks.

This makes sense, but if all of the SKUs have unlocked multipliers, then does it matter? If you can overclock the quad-core to the same speeds as the 8-core, then what?

 

[exar333]

This makes sense, but if all of the SKUs have unlocked multipliers, then does it matter? If you can overclock the quad-core to the same speeds as the 8-core, then what?

They have this same issue today. A higher-clocked PhII x4 is faster at a lot of tasks than the slower-clocked x6.

It makes me think the 'quad' core isn't very good, and they are really pushing the 8-core CPUs. Or, they want to keep selling their existing 4 and 6-core PhIIs, so they don't want too much overlap on quads and hexes?

 

[exar333]

I'm not surprised.

It probably means they are binning all the chips with the threshold that they validate as functioning just fine at 3.6GHz, and then they lock out cores that can't. Then they set the default multiplier as needed to fill up their inventory requests from resellers.

I see it as an artifact of a streamlined binning process more than anything else. Binning takes time too you know, and time is money. Its only natural that companies like Intel and AMD are seeking ways to reduce the expense of validation and test.

True. The BD quad sure doesn't seem very interesting though. You can already get a PhII at the same or higher clock speeds. This will be a good measure of how the BD architecture works. Will a similarly-clocked BD quad beat a PhII in common tasks?

 

[386DX]

This makes sense, but if all of the SKUs have unlocked multipliers, then does it matter? If you can overclock the quad-core to the same speeds as the 8-core, then what?

It matters on the reviews, and to OEM's. Since benchmarks will be run at the default speed. Remember when the Phenom II 965 BE was released? There was no reason for it to be release since the 955 BE and 945 BE that were released previously were unlocked anyways. But since those CPU's got beat by Intel's offering they released the 965 BE (at 140W instead of 125W) so the benchmarks would look more favourable. The 8-core BD lineup looks eerily similar. Why release 3 different 8-core CPU if they are all unlocked anyways? Most likely reason the FX-8100 can't run faster then 2.8 GHz under the 95W TDP. At that speed BD probably won't be very competitive with Intel's offering, so put out two more CPU's with the 125W TDP. The FX-8120 will probably be released at the 125W TDP then eventually get lowered to the 95W TDP (then FX-8100 will be discontinued) when steppings improved (much like the 965BE went from 140W to 125W) but the FX-8150 frequency is to high to ever fit inside the 95W TDP with there current manufacturing process so it'll remain at 125W for its lifetime.

 

[Arkadrel]

True. The BD quad sure doesn't seem very interesting though. You can already get a PhII at the same or higher clock speeds. This will be a good measure of how the BD architecture works. Will a similarly-clocked BD quad beat a PhII in common tasks?

AMD Phenom II X4 975 Black Edition (stock 4x 3.6ghz, 125watts TPD) ~ 150$
AMD Bulldozer FX-4100 (stock 4x 3.6ghz (4.2ghz turbo), 95watts TPD) ~ ??$


People are "guessing" around 15% more IPC increase,
From the "50%" more "performance" figours, they claim with the 16 core server versions over the current optertrons top end (when it was announced).


I think its likely we ll see these chips more than 15% faster, than the Phenom II's mhz pr mhz. Because of all the good new stuff, theyre getting.

So the AMD Bulldozer FX-4100, esp in programs that are "only" useing 2 threads or so, when its running 4.2ghz (with turbo), will kick the liveing sh*t out of the Phenom II x4 975BE in performance.

in "single" thread opterations, the FX-4100 will probably be like ~30% faster thane the Phenom II, because of the mhz differnce (800mhz (17% differnce with turbo on), and the ~15% or so IPC differnce, giveing a total probably in the 30%+ region).

And it ll do it useing 30% less energy (95w TPD vs 125w TPD).


So my belief : FX-4100 vs Phenom II x975BE

1-2 threads = ~30% faster
3-4 threads = ~15% faster




If you compair for gameing (Anandtech's bench):
Phenom II x4 980BE vs i5 2500k

Fallout 3; Amd=87,5fps vs Intel=90,3fps (~3% slower)
Left4Dead; Amd=130.8fps vs Intel=142.7fps (~9% slower)
Crysiswarhead; Amd=83,8fps vs Intel=91.6 (~9% slower)

Dragonage; Amd=121.4fps vs Intel=159.4fps (~31% slower)
DawnofWar; Amd=60.1fps vs Intel=82.3fps (~35% slower)
WoW; Amd=80.6fps vs Intel=119.4fps (~47% slower)

ect.

I dont expect the FX-4100 to beat the i5-2500k, at most Im hopeing it ll come close to it.

Maybe in stuff where only 2 threads are used, the FX-4100 will beat the 2500k, simply because of its turbo boost to 4.2ghz (still only rumors though, that its turbo will hit 4.2ghz).

 

[Cookie Monster]

Wait so by turbo mode up to 4.2GHz, is that for all cores? I know people harp over 4.5~5GHz "overclocks" for their sandy bridge chips but infact these overclocks aren't for all the cores (I think its just a single core?). Does the same apply to BD?