AMD Zambezi news, info, fans !

[FlanK3r]

XLR:some SB chips are 5 GHz stable with good voltage (1.35V), of course, more of this can 5 GHz stable too, but with 1.45-1.55V. But dont forgot, SOI HK is diferent than Intels HK. I thinking, 1.45V at AMD will be still very safe for 24/7. More dangerous are temps than volts.

[XRL8]

XLR:some SB chips are 5 GHz stable with good voltage (1.35V), of course, more of this can 5 GHz stable too, but with 1.45-1.55V. But dont forgot, SOI HK is diferent than Intels HK. I thinking, 1.45V at AMD will be still very safe for 24/7. More dangerous are temps than volts.

I think youre simplyfing it.As it all depends, in general all extremes are not healthy for a cpu.
For instance i had a A64 with only 1.2v passively cooled which was pretty much always at 60-70C, and its kickin to this day.However if you put 1.6v into a phenom II, even good watercooling wont stop it from deteriorating (if you keep it at this voltage constans and put a strain on it).
On the intel side its even worse, i OC`ed my friends core2duo xeon (65nm), it had big ass cooler which kept it at 50C max, however increased voltage deteriorated it in just a year, it was running 3.3ghz, now it barely does 2.3.

The key relation for me is median, between voltage and temperature, if i need extreme measures to cool a cpu, that means voltage is too high for 24/7 operation.Of course benching is another matter, as its just shorts periods of time of higher voltage/temperature.Anyhow, not one of my personal cpus broke or deteriorated keeping it all in some balance.

However getting back to BD, of course whats healthy and whats not, relation between temperature frequency and voltage it all will be determined later.
However if lets say retail BD is able to hit 5ghz stable 24/7 and Sandy is the same at this moment.Its ok, but not as good as i hoped.Its pretty much certain IPC of sandy is higher, so BD needs more clock.
I was under impression SB`s arent really hitting 5ghz fully stable, all i saw was cpuz screens, or 3dmark runs (thats on air).But you prolly know better.I didnt had the opportunity to build SB system yet.

[Lightman]

....
However if lets say retail BD is able to hit 5ghz stable 24/7 and Sandy is the same at this moment.Its ok, but not as good as i hoped.Its pretty much certain IPC of sandy is higher, so BD needs more clock.
I was under impression SB`s arent really hitting 5ghz fully stable, all i saw was cpuz screens, or 3dmark runs (thats on air).But you prolly know better.I didnt had the opportunity to build SB system yet.

I might take it wrong but are you hoping to hit 5GHz on 8 cores stable for them to sit idle and run single threaded workloads to allow SB outperform it?
I agree SB ST performance most likely will be higher at the same clock than BD's, but BD strength should be MT. AMD focus for BD is on MT and scalability.

I see it like this:
if BD really can hit 4.8-5.x GHz on AIR stable for all 8 cores under load then it's great and will have amazing performance in MT. Scaling 8 cores to that speed is harder than 4. Just look at Intels SB vs Gulf. 6 core monster scales up to 4.4-4.6GHz on AIR.
When ST performance is your goal I bet you can squeeze extra few hundred MHz out of BD max MT clock.

[XRL8]

I might take it wrong but are you hoping to hit 5GHz on 8 cores stable for them to sit idle and run single threaded workloads to allow SB outperform it?
I agree SB ST performance most likely will be higher at the same clock than BD's, but BD strength should be MT. AMD focus for BD is on MT and scalability.

I see it like this:
if BD really can hit 4.8-5.x GHz on AIR stable for all 8 cores under load then it's great and will have amazing performance in MT. Scaling 8 cores to that speed is harder than 4. Just look at Intels SB vs Gulf. 6 core monster scales up to 4.4-4.6GHz on AIR.
When ST performance is your goal I bet you can squeeze extra few hundred MHz out of BD max MT clock.

Youre comparing it as if 8 BD cores were fully comparable to 4 intel cores with HT.They are not, its different approach.
Gulftown is based on older tech than SB, and less (probably much less) mature process, so again i dont see it as viable comparison.
AMD positions 8 core BD`s to 4 core SB`s , so thats where comparisons have to be made.Its more like 4 modules vs 4 cores.Or 8 cores vs 8 threads.
One AMD core is much smaller than one SB core.So its pretty much destined to have lower ipc.Thus it would be expected of them to reach higher clocks.
And its been known for some time that BD is architectured as high speed cpu.So its not unreasonable to expect it reach higher clocks.
I hope for fully customizable turbo.If i can set 8 cores to 5ghz, then maybe 5.5ghz in single/dual threaded work.That should help.But its yet to be known how turbo v2.0 works.

[TESKATLIPOKA]

Gulftown is based on older tech than SB, and less (probably much less) mature process, so again i don't see it as viable comparison.

Yet you don't have a problem comparing BD versus SB overclockability, not to mention BD 4 module will be bigger than Gulftown 6 core and way bigger than SB 2600K. If 4.8-5Ghz is possible then AMD did an excellent job and we should be happy.

[XRL8]

Yet you don't have a problem comparing BD versus SB overclockability, not to mention BD 4 module will be bigger than Gulftown 6 core and way bigger than SB 2600K. If 4.8-5Ghz is possible then AMD did an excellent job and we should be happy.

Why should i have a problem comparing mature 32nm of intel vs unknown 32nm of GF ? Its about whats available in the market for comparable price.
What i read 4 module BD is ~250mm2 ,Sb is 216mm2 , so thats not a huge difference.But getting back to the point.Its irrelevant how big it is for a consumer/enthusiast/oems.Price, and power consumption.Thats what matters.
Price/performance/power consumption.And im just hoping BD is comparable.And secretly i hope BD is going to be better than SB.

[Motiv]

Sorry, but this doesn't wash at all.

AMD positions 8 core BD`s to 4 core SB`s , so thats where comparisons have to be made.Its more like 4 modules vs 4 cores.Or 8 cores vs 8 threads.

AMD will position itself against a comparable Intel chip. The amount of cores/modules is irrelevant compared to the results that it gives.

One AMD core is much smaller than one SB core.So its pretty much destined to have lower ipc.

That statement is actually quite incorrect. SB cores are larger because of the overhead of HT, while BD can be relatively smaller due to the sharing of resources wihin a single module. IPC in this respect has nothing to do with core size.

Thus it would be expected of them to reach higher clocks.

This is down to the pipeline really, not the size of the core.

And its been known for some time that BD is architectured as high speed cpu.So its not unreasonable to expect it reach higher clocks.
I hope for fully customizable turbo.If i can set 8 cores to 5ghz, then maybe 5.5ghz in single/dual threaded work.That should help.But its yet to be known how turbo v2.0 works.

True, the BD cores are known to be higher in speed, due to the longer pipeline. However, this is a brand new chip and is an unknown with regards to speeds reached. I highly doubt you'll be seeing 5ghz in ST, let along MT.

I think you're taking the BD chip as some sort of messiah for AMD. What it is is a very capable MT chip. If its IPC is anywhere close to Intel, and the speed of the chip can exceed intels offering, then they'll have a good solid chip for the next 4/5 years to work with. I am not holding my breath on this, as I still see AMD as a low-med type of company (GPUs of course a separate issue). I do think it's MT performance will be quite spectacular due to the amount of cores it can have. Desktop wise, I still feel Intel will have the upper hand (something that a lot of fanboys will find quite distasteful to hear).

[XRL8]

AMD will position itself against a comparable Intel chip. The amount of cores/modules is irrelevant compared to the results that it gives.

Uhm, thats what i said.AMD isnt positioning this above SB ,its not aiming at SBE,AMD positions this against 2600K (the highest AMD FX).
And while amount of cores/modules is irrelevant/The results are.They can however differ very greatly (one app a win for amd ,another for intel etc.)

While Core size isnt some defining metric, its logical to assume amd wont pull any miracles, if core is smaller than SB (and it is) than its safe to assume it will be weaker.
HT and sharing resources is another thing, and more relevant to the MT tasks.Thats why we cant expect stunning superpi performance ;-).
Ergo, i dont agree that core size has nothing to do with performance that can be extracted from it :P.

As for the speed, yes pipeline is longer thus its a high frequency design, however the smaller the chip the easier it is (its smaller than thuban thats for sure)

True, the BD cores are known to be higher in speed, due to the longer pipeline. However, this is a brand new chip and is an unknown with regards to speeds reached. I highly doubt you'll be seeing 5ghz in ST, let along MT

Well, im already seeing 4.8ghz from three sources.So i dont think that 5ghz from retails isnt possible.Its not even a stretch,and thats MT.

I think you're taking the BD chip as some sort of messiah for AMD. What it is is a very capable MT chip. If its IPC is anywhere close to Intel, and the speed of the chip can exceed intels offering, then they'll have a good solid chip for the next 4/5 years to work with. I am not holding my breath on this, as I still see AMD as a low-med type of company (GPUs of course a separate issue). I do think it's MT performance will be quite spectacular due to the amount of cores it can have. Desktop wise, I still feel Intel will have the upper hand (something that a lot of fanboys will find quite distasteful to hear).

No im not.If that was the case, BD would have 8 comparable to SB cores with greater overclockability, and i dont think thats even possible.
However it can be "smart" enough chip to be comparable in ST, and have upper hand in most MT.
Desktop wise, theres really small amount of apps, that need ultra strong x87 FPU performance.From top of my head i can only think of superpi and pcsx2.
In the rest differences, while may be there, are just academic.In games, we are pretty much always GPU limited after certain threshold.
And now games are moving to MT.Everything is slowly moving to MT.Sysmarks and similar software are pointless benchmarks for 99% of population.
And so on.Only MT performance still matters much.Multitasking, compression, encryption, video encoding, seti and similar software.

@Informal
Last i read, there were estimates of ~250mm2.Maybe im wrong tho, could you point me to a source of this 300m2 estimate ?

[XRL8]

AMD will position itself against a comparable Intel chip. The amount of cores/modules is irrelevant compared to the results that it gives.

Uhm, thats what i said.AMD isnt positioning this above SB ,its not aiming at SBE,AMD positions this against 2600K (the highest AMD FX).
And while amount of cores/modules is irrelevant/The results are.They can however differ very greatly (one app a win for amd ,another for intel etc.)

While Core size isnt some defining metric, its logical to assume amd wont pull any miracles, if core is smaller than SB (and it is) than its safe to assume it will be weaker.
HT and sharing resources is another thing, and more relevant to the MT tasks.Thats why we cant expect stunning superpi performance ;-).
Ergo, i dont agree that core size has nothing to do with performance that can be extracted from it :P.

As for the speed, yes pipeline is longer thus its a high frequency design, however the smaller the chip the easier it is (its smaller than thuban thats for sure)

True, the BD cores are known to be higher in speed, due to the longer pipeline. However, this is a brand new chip and is an unknown with regards to speeds reached. I highly doubt you'll be seeing 5ghz in ST, let along MT

Well, im already seeing 4.8ghz from three sources.So i dont think that 5ghz from retails isnt possible.Its not even a stretch,and thats MT.

I think you're taking the BD chip as some sort of messiah for AMD. What it is is a very capable MT chip. If its IPC is anywhere close to Intel, and the speed of the chip can exceed intels offering, then they'll have a good solid chip for the next 4/5 years to work with. I am not holding my breath on this, as I still see AMD as a low-med type of company (GPUs of course a separate issue). I do think it's MT performance will be quite spectacular due to the amount of cores it can have. Desktop wise, I still feel Intel will have the upper hand (something that a lot of fanboys will find quite distasteful to hear).

No im not.If that was the case, BD would have 8 comparable to SB cores with greater overclockability, and i dont think thats even possible.
However it can be "smart" enough chip to be comparable in ST, and have upper hand in most MT.
Desktop wise, theres really small amount of apps, that need ultra strong x87 FPU performance.From top of my head i can only think of superpi and pcsx2.
In the rest differences, while may be there, are just academic.In games, we are pretty much always GPU limited after certain threshold.
And now games are moving to MT.Everything is slowly moving to MT.Sysmarks and similar software are pointless benchmarks for 99% of population.
And so on.Only MT performance still matters much.Multitasking, compression, encryption, video encoding, seti and similar software.

@Informal
Last i read, there were estimates of ~250mm2.Maybe im wrong tho, could you point me to a source of this 300m2 estimate ?

[JumpingJack]

@Informal
Last i read, there were estimates of ~250mm2.Maybe im wrong tho, could you point me to a source of this 300m2 estimate ?

Hans Devries arrived at a 300 mm^2 estimate from comparing common elements between the photoshopped die shot and that of Llano whose die size was known.

More recently, at the 2011 ISSCC conference AMD published 3 papers on BD. One was specifically about the module and it quoted 30.9 mm^2 for a single module. Another paper had a complete unadulterated and non-photoshopped die shot, from that one would estimate 285 mm^2 or 297 mm^2 depending on whether AMD considers the VSS gating block as a count in the total module area. Regardless, Hans was pretty close.