AMD Zambezi news, info, fans !

[informal]

Because performance sucks?

[Alex-Ro]

You get a cookie

[xsecret]

AMD sales team is not dumb : if you want to have an idea of performance, just look at the price tag. AMD wants to be competitive with Intel. So you will find CPUs with same performance and same price tag than Intel. About stepping, things are quite easy : there was a performance issue in B1 and previous steps. Perfs were down 10-15% than expected on SOME benchmarks. This bug was solved in B2 stepping. So, if you use a B2 stepping WITH a BIOS newer than mid-Aug, you have "shipping" performances. If not, you have ~6-7% lower than expected overall. Protip : B2 step is CPUID F.1.2.

[informal]

AMD sales team is not dumb : if you want to have an idea of performance, just look at the price tag. AMD wants to be competitive with Intel. So you will find CPUs with same performance and same price tag than Intel. About stepping, things are quite easy : there was a performance issue in B1 and previous steps. Perfs were down 10-15% than expected on SOME benchmarks. This bug was solved in B2 stepping. So, if you use a B2 stepping WITH a BIOS newer than mid-Aug, you have "shipping" performances. If not, you have ~6-7% lower than expected overall. Protip : B2 step is CPUID F.1.2.

You mean this is shipping performance? B2 stepping,even if it's 6-7% slower or even 15% slower,it sucks badly since it is slower/or barely equal to 1100T. Rumored price from AMD themselves 300$. Rumored price from one dude having them listed on his own site : 260$. Todays 1100T price :190$ (will go down after Zambezi launches). If as you say price reflects performance then you will have 1100T performance (+-10/15%) with 30+% higher price. Is this logical?

In the link above (Vr-zone),just one glance at C10 64bit single core test tells you something is off. You have a single Bulldozer core using 256bit FPU for itself and running at 4Ghz.It gets 3769 pts with some of the features turned off in BIOS(best result they managed). Now ,take a look at single Thuban core, running at 3.7Ghz in same benchmark. It scores 4103pts. That is 17% faster than what Zambezi would get at 3.7Ghz and still faster (8%) than what Zambezi gets at 4Ghz. This is the brand new,double sized,improved,SMT capable FlexFP,with free reg-reg moves(no cost instruction according to AMD), and million other improvements versus K10? Yeah,call me crazy but I don't think so.

[xsecret]

You mean this is shipping performance? B2 stepping,even if it's 6-7% slower or even 15% slower,it sucks badly since it is slower/or barely equal to 1100T. Rumored price from AMD themselves 300$. Rumored price from one dude having them listed on his own site : 260$. Todays 1100T price :190$ (will go down after Zambezi launches). If as you say price reflects performance then you will have 1100T performance (+-10/15%) with 30+% higher price. Is this logical?

I will not comment those benchmarks, but the 1100T is not a competitor for the FX-8150. AMD don't want to be competitive with their old-gen EOL CPU (ie 1100T), they want to be competitive with Intel CPUs. So it makes sense to adjust the price of the new FX-8000 series to the price of an Intel CPU with similar performance. The price is the key : get the price, check on the Intel price-list for a similar price and you should have an idea of performances.

PS : Again, the $260-$300 range is just rumor...

[informal]

Ok,if the 260-300$ price range is just a rumor(from official AMD website BTW...) ,what is the price range then? Below 200$? Between 200 and 260? According to now removed vr-zone results,that poor fx8120 can't even touch 1100T in multithreaded applications,what chance it has against 2500K(if this is what you imply). Even 8150 has no chance against 2500K,maybe even 2400. Makes zero sense to me. 315mm^2 die,slower than 4 years old K10 in multithreaded workloads AND single threaded workloads,slower than 225mm^2 SB QC without SMT (with GPU counted in the die area!). Is this the brand new,area efficient high-performance x86 core that AMD has been working on for 6-7 years now. If it is,then they failed and need to rethink what they have been doing for 7 years.

[xsecret]

From a micro-architectural point of view, there is some nice ideas in Bulldozer. But it seems they failed to finally implement what they expected at first. The concept of "cores" for Bulldozer is just a marketing BS. A FX-8150 is a 4-cores CMT-based CPU with a dual Integer cluster. CMT architecture is not something new and noboby called a cluster a "core" before. AMD just renamed a core "a module" and a cluster "a core" in order to amaze ppls with "8-core CPU !!". Now what's next ? Calling an ALU "a core" ? After all, why not ? So the FX-8150 could be a 16-cores CPU as well.

Edit : As a proof, if you look at their own patent (http://www.freepatentsonline.com/20080209173.pdf), you see they know exactly what is a "Core" and what is a "Cluster". So why calling a cluster a core ? For marketing purpose of course, but that's still BS.

[informal]

Well actually I do think they are cores. Each "core"/cluster can retire 4 cops(macro ops) so each module is ,like they call it,an optimized dual core. But all this doesn't matter if performance is not there. Why bother? They will end up behind intel even more and they will be 3 generations behind by the time IB launches. I couldn't care less how they call their cores if these cores performed at least better than K10 ones. By the look of things right now,these cores will be much slower (than K10) and frequency potential will not be nearly enough to catch up. I still have a hard time believing they knew this all along and still went with it. Note that we re not talking 5% slower than K10 here.We are talking A LOT slower in both integer and SIMD.

But let's just wait and see what happens. On paper it is indeed a novel idea and a promising one too. Current numbers do look very bad,but maybe things improve with the retail chips.

[xsecret]

Well actually I do think they are cores.

A core must be able to handle INT, FP and SIMD SSEx instructions.

Multithreaded applications generally use SIMD, so when all the "cores" are in use, that's equal to 4 computing units. Are you aware of many MT application that massively use Integer units ?

[informal]

A core must be able to handle INT, FP and SIMD SSEx instructions.

Multithreaded applications generally use SIMD, so when all the "cores" are in use, that's equal to 4 computing units. Are you aware of many MT application that massively use Integer units ?

Each "core" can do all of those you said it must do. It can do INT in the "core" part and FP/SIMD in the FlexFP part. FLexFP part can be dedicated or shared so it can be run in SMT mode (shared) or all 256bit (2x FMAC) can be "given" to a core that requests it if other core has no FP/SIMD instructions scheduled. So in MT case with SIMD you actually have 8 threads running on 8 FMACs,each of which is 128bits wide(and each of which can do add or mul or fma). The fact that each FMAC has (on paper) half the potential execution resources of one K10 core doesn't make it any "lesser core".It only *could* make it slower than K10,nothing more nothing less. It still runs 8 threads across 8 hardware execution units(8 128bit FMACs). Before,we were under impression these FMACs would end up being faster than each K10 core. Now,as recent leaks show,each FMAC will end up slower,sometimes a lot ,then one K10 core(actually FPU in this core).

[2good4you]

Ang once again SweClockers.com presenting "Bulldozer news"

http://translate.google.com/translat...mbezibulldozer

[repman]

But let's just wait and see what happens. On paper it is indeed a novel idea and a promising one too. Current numbers do look very bad,but maybe things improve with the retail chips.

This maybe be a dumb question but, is it possible that the OS (Windows) would need an update because of the BD design in order to properly support it?

[Mats]

Ang once again SweClockers.com presenting "Bulldozer news"

Honestly, who cares? It's just a reason for not posting or reading there.
I've never been interested in Swedish computer forums, I see no point in it because they're at least one step behind this forum 99 % of the time.
Don't you see how frustrated you get? Let it go.

There's always someone who is wrong on the internet, you can't cure them all.

[Mats]

This maybe be a dumb question but, is it possible that the OS (Windows) would need an update because of the BD design in order to properly support it?

Yeah, JF talks about "Final drivers"- what's that about?

[informal]

OS might need an update but this is not a reason why these samples perform like they do. Something else is wrong (or it's just they way the design works.. .we don't know).

[repman]

OS might need an update but this is not a reason why these samples perform like they do. Something else is wrong (or it's just they way the design works.. .we don't know).

I see, thank you for the reply. I think everyone should ask themselves why would AMD bother with something that is slower than what they currently have. They would drop the project long ago if it really was slower/same performance.

[informal]

In some aspects it does provide somewhat better performance than their previous six core design(yes,even these poor ES). But then again,if this is rather an exception then the rule,why would they bother?

[repman]

Exactly, even a die shrink of K10 would probably be better than wasting money for something you already have. It's best to wait for the final results.


EDIT:
Anand's twitter:

anandshimpi anandshimpi
Beware of any leaked Bulldozer benchmarks, unless you're running B2.G you're not looking at shipping performance
5 hours ago

anandshimpi anandshimpi
I'm not saying anything about absolute performance, just keep in mind that silicon that's older than ~2 weeks isn't production worthy
5 hours ago

anandshimpi anandshimpi
And I don't believe the final decision has been made to go to market (desktop) with B2.G either, will know for sure in the coming weeks
5 hours ago

anandshimpi anandshimpi
This is why we never did an early preview of Bulldozer on AT, no sense in putting out numbers that may not be representative
5 hours ago

[chew*]

From a micro-architectural point of view, there is some nice ideas in Bulldozer. But it seems they failed to finally implement what they expected at first. The concept of "cores" for Bulldozer is just a marketing BS. A FX-8150 is a 4-cores CMT-based CPU with a dual Integer cluster. CMT architecture is not something new and noboby called a cluster a "core" before. AMD just renamed a core "a module" and a cluster "a core" in order to amaze ppls with "8-core CPU !!". Now what's next ? Calling an ALU "a core" ? After all, why not ? So the FX-8150 could be a 16-cores CPU as well.

Edit : As a proof, if you look at their own patent (http://www.freepatentsonline.com/20080209173.pdf), you see they know exactly what is a "Core" and what is a "Cluster". So why calling a cluster a core ? For marketing purpose of course, but that's still BS.

Wow someone with common sense...........

Like i said before I discussed this face to face with engineers at AMD, BD is a native 4 core 8 thread part.

and then the lynch mob got mad..........

[Manicdan]

Wow someone with common sense...........

Like i said before I discussed this face to face with engineers at AMD, BD is a native 4 core 8 thread part.

and then the lynch mob got mad..........

new AMD tri-cores beats old hex-cores!!!! is it really possible!!! amd is f*ing awesome!!!!
lol

its all perspective, perf per dollar of 4 threads with a typical overclock is what i care about. that might mean a 4100 or 8100, not sure yet.

[informal]

I hope we get a price range soon for FX parts.This should tell us enough about performance.

[demonkevy666]

Wow someone with common sense...........

Like i said before I discussed this face to face with engineers at AMD, BD is a native 4 core 8 thread part.

and then the lynch mob got mad..........

the shared L1 instruction cache is what comes to mind for me, ever time i read your comments.

[xdan]

new AMD tri-cores beats old hex-cores!!!! is it really possible!!! amd is f*ing awesome!!!!
lol

its all perspective, perf per dollar of 4 threads with a typical overclock is what i care about. that might mean a 4100 or 8100, not sure yet.

At an estimate die size of 315-330mm^ it's f*ing FAIL....
And i don't think that BD X6 - tricore whatsoever will be better than Thuban. Equal i quess .
Thuban X8 on 32nm with 8MB L3 cache would probably came in same die size. With a better IMC and faster L2 cache, i might wonder if that 8 true core design could do...

And now about crippled ES, in every industrys there are so called "prototypes", but for an example a car initialy designed to run with 300KM/H as a prototype wouldn't go with 140KM/H. May be 240-260KM/H.
BD ES are running just pathetic. If that initial true performance than no new revision can boost the performance with 50%.
And why let false rumours spread out.
It's not AMD a company craving for PROFIT?
It's this a good strategy?

AMD at least should make an official anouncement saying: "Folks, retail will have much more performance, don't trust anything on net."

[Manicdan]

At an estimate of 315-330mm^ it's f*ing FAIL....
And i don't think that BD X6 - tricore whatsoever will be better than Thuban. Equal i quess .
Thuban X8 on 32nm with 8MB L3 cache would probably came in same die size. With a better IMC and faster L2 cache, i might wonder if that 8 true core design could do...

thuban x8 would not get the same single threaded perf of current competition, a 2100 would still be a better choice for gamers. BD however should have much stronger IPC when a second thread isnt running on the core/module, and its being designed for very aggressive turbos which the old stars cores were not built for.

why dont you think BD will be able to beat thuban thread/core vs core? the architecture is stronger and the frequency is higher. the only real issue is when it comes to having those extra threads running will it bring average perf down to be less than the old stuff, which i dont think would be a problem. an x8 BD should be much more than 33% faster than thuban with 33% more cores/threads. its also smaller than thuban by a pretty noticeable amount, around a whole cores worth. so a thuban x5 would have similar mm2 and give us a basic idea on perf/mm2 increase AMD was able to get.

[xdan]

And who says and who can prove that BD will have single thread performance equal to SB or better?
Or even better than Thuban?
I don't think that you understood what i meant .
An improved Phenom II arhitecture should have interesting performance: reduced latency L2& L3 cachem better IMC, larger L3 cache -8MB). Just some tweaks as Nehalem -> SB.
And probably would have better yelds, because is something well known and so on.

nd its being designed for very aggressive turbos which the old stars cores were not built for.

why dont you think BD will be able to beat thuban thread/core vs core? the architecture is stronger and the frequency is higher. the only real issue is when it comes to having those extra threads running will it bring average perf down to be less than the old stuff, which i dont think would be a problem. an x8 BD should be much more than 33% faster than thuban with 33% more cores/threads. its also smaller than thuban by a pretty noticeable amount, around a whole cores worth. so a thuban x5 would have similar mm2 and give us a basic idea on perf/mm2 increase AMD was able to get.

Aggresive turbos without a much better performance per clock means nothing.
X6 1100T has 400Mhz Turbo- enough, performance per clock it what lacks.
2 cores Phenom X2 against 1 module BD at same clock will be something interesting to compare.
You miss the point that Thuban is on 45nm and BD is on 32nm. As i sad probably Thuban X8 on 32nm shoud have same die size as BD.
All what you say was easy to believe in march- april, but not now, after 4-5 months of just patethic leaks, lies, delays, and so on.