Whilst Seronx is a bit chatty, stop giving him such a hard time guys.

If you disagree that's fine but your treating him like he's an idiot when we're in a thread that's entirely based on speculation.

Which is just as stupid.


It's kind of annoying coming in here and seeing people talk down to one person so much.

seronxAnd those issues weren't yield bent

Of course they were. When you do a respin, it's always related to yields in one form or another. When they say they cannot get high enough clocks, it means not enough chips yield at whatever the target clocks were.

I am mainly talking about Cinebench, wPrime, and those other benchies

Those are not desktop apps. Unless you spend all your time doing CAD rendering, in which case you are not an standard desktop user, and not even an standard enthusiast or gamer.

Also they are benchmarks with almost never represent actual performance, because they are highly optimized for MT and floating point math, which is not attainable on REAL code.

I happen to spend a lot of time doing CAD btw. For work. So I have always been highly interested in BD. That does not mean that I will believe in AMD's claims 100%, when they contradict everything I know about computer science, much less those claims coming from an AMD evangelist. I dont need AMD nor anyone to tell me what would be the performance and the trade offs, when I can look at their architecture and know what to expect. It's like I don't need an evangelist telling me how the universe was created or how the world was populated with different animals, when I actually know about the big bang and evolution.

he talks down back to us:D, im sorry seronx, but i do feel your sounding a bit like a tape recorder now and as i say stateing too much to be fact pre release:ohwell:

theoneandonlymrkhe talks down back to us:D, im sorry seronx, but i do feel your sounding a bit like a tape recorder now and as i say stateing too much to be fact pre release:ohwell:

I can imagine it being somewhat frustrating for him though having everyone post in ways where you can feel the >=( in the words :laugh:


To be honest I don't think there's anything else about clock speed to be discussed and everyone should go have a nice cup of coffee or tea. :laugh:

pantherx12To be honest I don't think there's anything else about clock speed to be discussed and everyone should go have a nice cup of coffee or tea

thats mine and a few others point i think:)

been work 6-12, il go av a lie down get out tother side o bed this time lol:D

I think this image sums up everything.

BenetanegiaOf course they were. When you do a respin, it's always related to yields in one form or another. When they say they cannot get high enough clocks, it means not enough chips yield at whatever the target clocks were.

The yields are okay, they were okay since AMD facepalmed and said this Good Chip deal wasn't a good idea Global foundries is like robbing us before the CPU even launches

^that was in April 2011

And there is a conference call on Thursday @ July 21st

I don't know what is about but the "yield" issue will be shot down in that conference call supposedly

BenetanegiaThose are not desktop apps. Unless you spend all your time doing CAD rendering, in which case you are not an standard desktop user, and not even an standard enthusiast or gamer.

I do gaming + recording and the recording application can address up to 12 cores

SSE3 -> SSE4.2 is going to mean I'll be able to record at 60fps
(I don't record to my hard drive)

www.twitch.tv/seronx/b/286984318
me playing a space game WITH nothing to do

BenetanegiaAlso they are benchmarks with almost never represent actual performance, because they are highly optimized for MT and floating point math, which is not attainable on REAL code.

Multithreaded performance is what I deal with everyday!

BenetanegiaI happen to spend a lot of time doing CAD btw. For work. So I have always been highly interested in BD. That does not mean that I will believe in AMD's claims 100%, when they contradict everything I know about computer science, much less those claims coming from an AMD evangelist. I dont need AMD nor anyone to tell me what would be the performance and the trade offs, when I can look at their architecture and know what to expect. It's like I don't need an evangelist telling me how the universe was created or how the world was populated with different animals, when I actually know about the big bang and evolution.

I am not an AMD Evangelist sadly, I own 2 Pentium 4s ($2000 PCs)

theoneandonlymrkhe talks down back to us:D, im sorry seronx, but i do feel your sounding a bit like a tape recorder now and as i say stateing too much to be fact pre release:ohwell:

I know a couple things

pantherx12I can imagine it being somewhat frustrating for him though having everyone post in ways where you can feel the >=( in the words :laugh:


To be honest I don't think there's anything else about clock speed to be discussed and everyone should go have a nice cup of coffee or tea. :laugh:

>=(

Clock speeds are higher than 3.6GHz, how high I don't know but the sky is the limit(or until the heat melts the interconnects)

pantherx12I can imagine it being somewhat frustrating for him though having everyone post in ways where you can feel the >=( in the words :laugh:


To be honest I don't think there's anything else about clock speed to be discussed and everyone should go have a nice cup of coffee or tea. :laugh:

I am amazed these threads are even staying open. People are speculating about speculations. These aren't even official releases. Everybody has already decided how this processor will perform and we don't even know for 100% certain what the processor will be.

BenetanegiaIt's like I don't need an evangelist telling me how the universe was created or how the world was populated with different animals, when I actually know about the big bang and evolution.

nobody knows at this time...

their are but theories and scientists are fond of making new ones;)



topdocumentaryfilms.com/is-everything-we-know-about-the-universe-wrong/

much can and will change in the next month regardless of what any of us knows or realises and hence such surety is false

seronxThe yields are okay, they were okay since AMD facepalmed and said this Good Chip deal wasn't a good idea Global foundries is like robbing us before the CPU even launches

^that was in April 2011

And there is a conference call on Thursday @ July 21st

I don't know what is about but the "yield" issue will be shot down in that conference call supposedly

The yields were obviously bad, and was certainly backed up by the fact that AMD changed his deal with GF to a per worming chip deal.

I do gaming + recording and the recording application can address up to 12 cores

SSE3 -> SSE4.2 is going to mean I'll be able to record at 60fps
(I don't record to my hard drive)

For video encoding Intel's QuickSync, has proven to be better than any other MT apporach out there. If you really are into gaming + video, the best answer right now is going Intel.

Multithreaded performance is what I deal with everyday!

I do too, to a high extent, but because I do, I actually know how it works, so I know that AMD's claims (and yours by extention) about scaling and IPC etc, are complete bollocks!!

I am not an AMD Evangelist sadly, I own 2 Pentium 4s ($2000 PCs)

That's actually a proof supporting your AMD evengelism rather than the opposite. If instead of having bought AMD CPUs in the P4 era, I had bought P4s, I would be burnt too and I would most probably by an AMD fanboy lol.

The evidence seems to support the idea. You bought 2 P4s, but you have not used any other Intel CPU since. That's it you never actually used the good Intel chips, only the worst ones. That pretty much equals fanboyism. Read this:

techreport.com/discussions.x/21294

Only one purchase is needed to become a fanboy. Whatever you used before is irrelevant, it is the last purchase/election what matters.

theoneandonlymrknobody knows at this time...

their are but theories and scientists are fond of making new ones;)



topdocumentaryfilms.com/is-everything-we-know-about-the-universe-wrong/

much can and will change in the next month regardless of what any of us knows or realises and hence such surety is false

None of that actually contradicts the big bang, only the way it may have happened. It does not contradict, it does show some flaws in the model.

Anyway, that was not my point. i.e. the big bang theory being "wrong" does not make the Bible right.

BenetanegiaOnly one purchase is needed to become a fanboy. Whatever you used before is irrelevant, it is the last purchase/election what matters.

your sounding like youve a deffinate lean yourself:) I hope they bring out something better then intels but because intel will then bring out something better again and so on, I personally wont be goin intel till they settle on a socket for a while/at all as i value the option to upgrade in parts. roll on BD so i can see some reviews:)

jpierce55I am amazed these threads are even staying open. People are speculating about speculations. These aren't even official releases. Everybody has already decided how this processor will perform and we don't even know for 100% certain what the processor will be.

An 8-core 32nm HKMG CPU that uses the FX+Black Moniker
An 6-core 32nm HKMG CPU that uses the FX+Ultimate Moniker
An 4-core 32nm HKMG that will likely get replaced next year by an APU that uses the FX+Ultimate Moniker

The Processor is from AMD

The CPU is set to release in 2011, no real date on launch since it's a surprise!!!!

and there is a Flex FPU and what it does is magic, you know the unicorn kind

:pimp: Unicorn Magic.... #$%^ Yeah

BenetanegiaThe yields were obviously bad, and was certainly backed up by the fact that AMD changed his deal with GF to a per worming chip deal.

AMD added the new pricing agreement doesn't mean it is expecting lower yields from Global Foundries' 32nm process but that it will give "better protection" if yields are lower than expected.

AMD said that 32nm production is ramping up better than it forecasted last year and it tried to play down the significance of its new deal with Global Foundries.

If is the question, and we won't find out will we since Zambezi will be cheap because of the "Better Protection"

BenetanegiaFor video encoding Intel's QuickSync, has proven to be better than any other MT apporach out there. If you really are into gaming + video, the best answer right now is going Intel.

False, SSE is where it is at

You are talking about Conversion

I am talking about capturing

BenetanegiaI do too, to a high extent, but because I do, I actually know how it works, so I know that AMD's claims (and yours by extention) about scaling and IPC etc, are complete bollocks!!

3 IPC to 4 IPC is not complete bollocks

You aren't looking at the right areas

Scaling factors are the easiest to calculate with AMD CPUs

theoneandonlymrkyour sounding like youve a deffinate lean yourself:) I hope they bring out something better then intels but because intel will then bring out something better again and so on, I personally wont be goin intel till they settle on a socket for a while/at all as i value the option to upgrade in parts. roll on BD so i can see some reviews:)

Hmm according to the TechReport blogpost, yes I am a Sandy Bridge fanboy, since in my last purchase, I was forced to buy a SB when my previous build died. I was actually waiting for BD to make my purchase, not that I was going to buy BD, but I was waiting for it. When the BD module idea was first presented I was totally sold out, but then we got actual architecture info, and it's not exactly what they first claimed.

The difference is that I don't post every single AMD (Intel in my case I supose) PR stuff I come across and pretend it's a fact.

seronxFalse, SSE is where it is at

You are talking about Conversion

I am talking about capturing

To my knowledge and actual experience capturing (without any encoding going on) does not require ANY CPU time. That's what DMA is for.

3 IPC to 4 IPC is not complete bollocks

You aren't looking at the right areas

Scaling factors are the easiest to calculate with AMD CPUs

3 IPC to 4 IPC IS completely bollocks. Since AMD went from 3 IPC per core to 4 IPC per module. Each module can only issue 4 instructions, so when all 8 cores are used only 2 IPC per core. It will be able to do 4 IPC when only 4 are used though.

BD can do 8 threads and 4 IPC, but not at the same time. The BS comes from the fact that they are pretending to sell the idea that BD has 8 cores AND 4 IPC at the same time which is completely false.

BenetanegiaTo my knowledge and actual experience capturing (without any encoding going on) does not require ANY CPU time. That's what DMA is for.

Capture is SSE/Encoding is SSE for realtime

Benetanegia3 IPC to 4 IPC IS completely bollocks. Since AMD went from 3 IPC per core to 4 IPC per module. Each module can only issue 4 instructions, so when all 8 cores are used only 2 IPC per core. It will be able to do 4 IPC when only 4 are used though.

BD can do 8 threads and 4 IPC, but not at the same time. The BS comes from the fact that they are pretending to sell the idea that BD has 8 cores AND 4 IPC at the same time which is completely false.

Zambezi can do 8 thread and 4 IPC per core

Phenom II can only do 1 of these scenarios
3 ALU ops
3 AGU ops
Not both
6 IPC in theory but the architecture didn't allow it so half/half was done 3 ALUs ops or 3 AGUs ops

Zambezi can do 2 ALU ops and 2 AGU ops per cycle thus it has a max of 4 IPC

but utilization of both increased(The IPC doubled realistically)

And I am pretty sure the last ALU and AGU weren't used often because they were called an AMU whatever the f that means
(Phenom II)
and most programs didn't even support it it was there just for the third floating point pipeline

Benetanegia3 IPC to 4 IPC IS completely bollocks. Since AMD went from 3 IPC per core to 4 IPC per module. Each module can only issue 4 instructions, so when all 8 cores are used only 2 IPC per core. It will be able to do 4 IPC when only 4 are used though.

BD can do 8 threads and 4 IPC, but not at the same time. The BS comes from the fact that they are pretending to sell the idea that BD has 8 cores AND 4 IPC at the same time which is completely false.

bit too irate though dude,

1 Each module can only issue 4 instructions 2 integer and 2 fpu= 4x4 =16 ipc max so how is it being limited to 8??

each module is doing its own scheduling? so it shouldnt need to drop below 3-4 ipc per module(2 integer 2 fpu or 1 integer 2fpu etc etc) still making anything between 4 - 16 ipc total not 8?

seronxAnd I am pretty sure the last ALU and AGU weren't used often because they were called an AMU whatever the f that means
(Phenom II)

thats why im saying your wrong though dude as performance benches etc might not best make use of this new archtecture for a while if at all.

To both of you. Each module can do 4 issues, so 2 per core when both cores in a module are used.

2 IPC x 8 "cores" == 16 IPC for the entire chip <- this is also true for SB with HT

or

4 IPC x 4 cores == 16 IPC <- this is also true for SB w/o HT

What is not posible and is being claimed is:

4 IPC x 8 cores == 32 IPC

Benetanegia4 IPC x 8 cores == 32 IPC

:laugh:

so is AMD sayin that?:laugh:

i hadnt heard that.

theoneandonlymrk:laugh:

so is AMD sayin that?:laugh:

i hadnt heard that.

Well if I'm 100% honest, I don't know if AMD trully said that or if it's being said by seronx and/or one of his links/quotes. But you gotta understand my confusion, I think that I've seen 20x more Bulldozer related claims made by seronx, than on the rest of the internet. In any case I'm responding to seronx and not AMD themselves so if at all, I'm being lenient to seronx by assuming that AMD did said that and that he is not making that up (too). :laugh:

as i read it, it could be poss to do 4x32b AVX calcs of the same type on one fpu and 4x32b AVX calcs of a different type on the other fpu plus 2 integer instructions at a max per clock? per module which does make for some interesting ipc figures single precision obv, and coded for obv,.

imho thats why its got a scheduler on the flex fpu as it will likely be used to allow double precision 64b x 2 per 128b fpux2 + 2 integer in use and that is then impressive no?.

BenetanegiaTo both of you. Each module can do 4 issues, so 2 per core when both cores in a module are used.

2 IPC x 8 "cores" == 16 IPC for the entire chip <- this is also true for SB with HT

or

4 IPC x 4 cores == 16 IPC <- this is also true for SB w/o HT

What is not posible and is being claimed is:

4 IPC x 8 cores == 32 IPC

theoneandonlymrk:laugh:

so is AMD sayin that?:laugh:

i hadnt heard that.

BenetanegiaWell if I'm 100% honest, I don't know if AMD trully said that or if it's being said by seronx and/or one of his links/quotes. But you gotta understand my confusion, I think that I've seen 20x more Bulldozer related claims made by seronx, than on the rest of the internet. In any case I'm responding to seronx and not AMD themselves so if at all, I'm being lenient to seronx by assuming that AMD did said that and that he is not making that up (too). :laugh:

IPC has increased per core

How it increases is unknown till it releases

It is 4 Complex Instructions per clock per module

The decoders are a fusion of Complex and Simple which is like



What is the difference between

3 Simple Instructions(Phenom II)
to
4 Complex Instructions(FX)

seronxIPC has increased per core

How it increases is unknown till it releases

Yes sir. :laugh:
The more you tell me that, the more I believe. Faith is starting to grow on me!

God, I've been so wrong. How could I have assumed it had anything to do with tech, when it's all magic! It's something mystical that nobody knows until the truth is revelaed to us by the Lords. The answer is not on the architecture, it's on our faith. Through faith we will be stronger and beat competition!!

No, now seriously. Pretty much everything has been revealed about the architecture and there's no magic formula. Actual/runtime IPC (as oposed to theoretical IPC) in BD might end up higher than on previous AMD architectures because of 2 main reasons:

1- use of multi-threading.
2- better branch prediction.

Point n. 1 is what we are mostly discussing. How much better is CMT over SMT? AMD will obviously want make you believe it's much much better and so much better in fact that it equals 2 complete cores. I call BS.

Point n. 2 is from where most serious IPC improvement claims come from. Most of the improvements for Sandy came exactly from increased IPC due to better scheduling and branch prediction. Theorerical IPC remains the same as Nehalem after all, and the 2500k can often times match 6 core i7's on threaded apps. Westmere already had much improved front end, and SB supposed an even bigger jump.

Bulldozer does introduce a much stronger branch predictor, but it's still to be seen if it can match or even come close to Intel's, which is its strongest point since Core 2.

seronxIt is 4 Complex Instructions per clock per module

The decoders are a fusion of Complex and Simple which is like

What is the difference between

3 Simple Instructions(Phenom II)
to
4 Complex Instructions(FX)

Read the link I posted, please.

At some point you will have to stop posting BS. Istanbul could do 3 simple and 3 complex too and Bd does 4 simple/4complex per module. So does Intel (mixing simple and complex), although differently.

BenetanegiaThe difference is that I don't post every single AMD (Intel in my case I supose) PR stuff I come across and pretend it's a fact.

Thats been my point since day one. Marketing is marketing is marketing.

What worries me about BD is the fact they said an over all increase of 30%......Thats AMD cherry picking stats to get that 30% increase for marketing. So in the real world its more like 15% or 20%......maybe.

This has been my number one concern. Not the IMC. Not the Mhz. Not the architecture. That little marketing number. You can tell far more about a pre-release product on the marketing you see IF you know what to look for in marketing. Both AMD and Intel are famous for inflating numbers. More then car manufactures! So everything you see. Everything you read. Its all FUD until we see real world numbers. No math in the world can explain how a CPU will work based of marketing slides.

BenetanegiaBulldozer does introduce a much stronger branch predictor, but it's still to be seen if it can match or even come close to Intel's, which is its strongest point since Core 2.

Read the link I posted, please.

At some point you will have to stop posting BS. Istanbul could do 3 simple and 3 complex too and Bd does 4 simple/4complex per module. So does Intel (mixing simple and complex), although differently.

I read it already get out of town

The increase is like this

1000 for Phenom II per core

1000 for Zambezi per core

It is the same, the performance didn't drop like you think it would it actually increased

1150 for Zambezi per core

There is another increase but I would wait till benchmarks to see this increase
(there is a huge difference between the Bulldozer/Zambezi to the Bulldozer/Zambezi we would have gotten in 2009, and it doesn't help that it only says this in the software guide)

seronxI read it already get out of town

The increase is like this

1000 for Phenom II per core

1000 for Zambezi per core

It is the same, the performance didn't drop like you think it would it actually increased

1150 for Zambezi per core

There is another increase but I would till benchmarks to see this increase

No I actually think the increase is more like this:

:laugh::laugh::laugh: for Phenom 2 per :nutkick:
:laugh::laugh::laugh::laugh::laugh: for Zambezi per :nutkick: plus :roll::roll: when the :banghead: is activated and a good deal of Turbo:slap: too.

Ok, ok I'm done with feeding the troll. Sorry guys.