Bulldozers first screens

[Dimitriman]

Well tbh it would behoove AMD to market a Module as a core, with 2 threads inside each module...........

Why?

Because more often than not it will get compared to an equivalent intel part with 4 cores........which happens to have 8 threads.

Just like HWbot where wprime rankings are based on the amount of cores not threads due to it being a "feature".

Amd needs to term this as a "feature" or they are still in the same boat they have always been...........competing with 8 threads less in an 8 core benchmark........when socket 2011 hits.

4 threads less. Socket 2011 will be limited to hexcores for the time being.
Also AMD will have 5 module desktop variants in 2012, should arrived within 6months of s2011 intels.

[keithlm]

Well tbh it would behoove AMD to market a Module as a core, with 2 threads inside each module...........

It comes around to what I said about chip developer/designer INTENT.

Q: How many threads did the developers of the Intel chip with 4 hyperthreaded cores plan to optimally service?

Q: How many threads did the developers of the 8 core AMD chip, comprised of 4 cores, plan to optimally service?

In both cases the number of cores (or modules) is not as important as the number of threads the designer planned to support.

[chew*]

It comes around to what I said about chip developer/designer INTENT.

Q: How many threads did the developers of the Intel chip with 4 hyperthreaded cores plan to optimally service?

Q: How many threads did the developers of the 8 core AMD chip, comprised of 4 cores, plan to optimally service?

In both cases the number of cores (or modules) is not as important as the number of threads the designer planned to support.

The problem still lies in the fact that because of this most reviewers are going to head to head it not based on pricing but based on flagship multicored parts.

Bulldozer will go head to head with gulftown, 8 threads to 12 threads yet the review will get spun as a 6 core part beating a 8 core part in multithreaded performance if 8 cores is not enough to beat 12 threads which based on my personal calculations is going to be close.

The flip side of this is they will also compare to a cheaper sandy part which by my estimations will lose in multithreaded.......however IPC wise at same clocks lets face it, sandy bridge is going to be a tough nut to crack.....

[keithlm]

Bulldozer will go head to head with gulftown, 8 threads to 12 threads yet the review will get spun as a 6 core part beating a 8 core part in multithreaded performance if 8 cores is not enough to beat 12 threads which based on my personal calculations is going to be close.

AH... I predict that you are correct. That is exactly what might happen. Less knowledgeable people will be deceived.

In the past this has happened so often that I just kind of accept it and laugh at the people that will base their opinions on what the people screaming the loudest will be shouting.

But that is just the way things will be since there a MANY MANY more Intel advocates than there are AMD advocates. So we can expect people to award the fictional and pointless "crown" based on things that mean nothing. So in other words nothing has changed for a few years.

[w0mbat]

well, a FX8 is kinda close to a 8-core cpu and if the i7 2600k is faster that does indeed say something. it may be 8 threads vs. 8 threads, but its 4 int cores vs. 8 int cores.
we just have to find an point where we can say "yes, a module doesnt equal 2 cores but it equals xxx". i think the 1.8x thats floating around for CMT is quiet good.

[XRL8]

Nvidia and ati have vastly different approaches to their chips, the fact that ati has more shaders doesnt mean shader per shader it has to be faster.People seem to understand that.
So whats so hard to comprehend about this ?
Intel has bigger cores, more powerful ones, AMD has more of them but theyre less complex.Easy peasy.

Only metrics that matter are:
Chip size @ same process node
Performance per watt
Performance per $

[piquadrat]

But what all those 4/8/12/whatever cores/modules are for if you run single threaded application on them? They are wasted. You guys seem to silently assume that all of the algorithms your shiny new amd systems will be chewing up are easy scalable, multi threaded ones. There are only several classes of computational problems which show such a property. And in this regard amd is clearly suggesting now and then that they emphasize multi-core efficiency and put it above IPC performance in their BD architecture.
And this is very wrong. This is wrong especially because single threaded performance is exactly what their current offering is lacking in comparison to the closest competitor. This is a mistake today and will be tomorrow.

[BeepBeep2]

But what all those 4/8/12/whatever cores/modules are for if you run single threaded application on them? They are wasted. You guys seem to silently assume that all of the algorithms your shiny new amd systems will be chewing up are easy scalable, multi threaded ones. There are only several classes of computational problems which show such a property. And in this regard amd is clearly suggesting now and then that they emphasize multi-core efficiency and put it above IPC performance in their BD architecture.
And this is very wrong. This is wrong especially because single threaded performance is exactly what their current offering is lacking in comparison to the closest competitor. This is a mistake today and will be tomorrow.

WTF?

How the hell do you know that Bulldozer's IPC will not be better?

[richierich]

But what all those 4/8/12/whatever cores/modules are for if you run single threaded application on them? They are wasted. You guys seem to silently assume that all of the algorithms your shiny new amd systems will be chewing up are easy scalable, multi threaded ones. There are only several classes of computational problems which show such a property. And in this regard amd is clearly suggesting now and then that they emphasize multi-core efficiency and put it above IPC performance in their BD architecture.
And this is very wrong. This is wrong especially because single threaded performance is exactly what their current offering is lacking in comparison to the closest competitor. This is a mistake today and will be tomorrow.

How many applications are single-threaded today? lol. So many more programs are programmed for more than 1 thread now. Dual-cores have been around for so long. Same with quad-cores. Now where do you think the future is headed? And even then AMD has said there will be an IPC increase over K10.5.

[chew*]

I think what the poster was trying to say is AMD has no issues hitting high clocks, it has no issues getting more cores in a chip so what it really needs to focus on is IPC and in a big way.

If they got the IPC up to snuff even 2-3% less than intel........they wouldn't need to keep adding cores, the performance would work across the board including multithreaded, later dates they could add more cores as needed.

[BeepBeep2]

I think what the poster was trying to say is AMD has no issues hitting high clocks, it has no issues getting more cores in a chip so what it really needs to focus on is IPC and in a big way.

If they got the IPC up to snuff even 2-3% less than intel........they wouldn't need to keep adding cores, the performance would work across the board including multithreaded, later dates they could add more cores as needed.

I think there's gotta be a balance...it depends on the architecture design.

If we have a core that is 10% slower than Intel's but hits 5.5 Ghz on air then it is easily competitive...not only that but you then see higher Mhz and average user is tricked like in P4/A64 days. I think its more about single thread performance as a whole and not just IPC, even though more IPC looks nice on paper depending on how you look at it.

[chew*]

I think there's gotta be a balance...it depends on the architecture design.

If we have a core that is 10% slower than Intel's but hits 5.5 Ghz on air then it is easily competitive...not only that but you then see higher Mhz and average user is tricked like in P4/A64 days. I think its more about single thread performance as a whole and not just IPC, even though more IPC looks nice on paper depending on how you look at it.

The problem is currently what we have is a cpu that clocks less for normal temps.........and when it is cranked even 700 mhz higher cold than competitior. It's still losing.

Llano imo is a smart move for AMD, most users don't need all that cpu power, and don't want to buy a high priced vga.

Enter builds that can be done with vga excluded, and built into cpu, something AMD can beat intel at in laptop or low end desktop.

Bulldozer I can't comment on, and I doubt we will see to much even at our around the 7th pertaining to it.

I still think they need to fill that niche with a high IPC moderate cored cpu however.

[charged3800z24]

But what all those 4/8/12/whatever cores/modules are for if you run single threaded application on them? They are wasted. You guys seem to silently assume that all of the algorithms your shiny new amd systems will be chewing up are easy scalable, multi threaded ones. There are only several classes of computational problems which show such a property. And in this regard amd is clearly suggesting now and then that they emphasize multi-core efficiency and put it above IPC performance in their BD architecture.
And this is very wrong. This is wrong especially because single threaded performance is exactly what their current offering is lacking in comparison to the closest competitor. This is a mistake today and will be tomorrow.

Intel's strong single thread performance has a lot to do with their turbo setup. BD looks like it will have a very agreessive Turbo feature... and we won't know for sure how it does till we see real results.

[chew*]

Thats not what matters though..........

Maybe this will shed some light on what i'm trying to say.

Apples to apples cores/threads.

core i5 2500K 4 cores / 4 threads wprime at 5.9 = 5.109

Amd phenom II x4 955 4 cores / 4 threads wprime at 6.6 = 5.516

Bottom line IPC carries over to multithreaded performance............

[Nintendork]

Thats not what matters though..........

Maybe this will shed some light on what i'm trying to say.

Apples to apples cores/threads.

core i5 2500K 4 cores / 4 threads wprime at 5.9 = 5.109

Amd phenom II x4 955 4 cores / 4 threads wprime at 6.6 = 5.516

Bottom line IPC carries over to multithreaded performance............

At least compare same gen competitor chips.

i5 2500K
FX4000

[BeepBeep2]

At least compare same gen competitor chips.

i5 2500K
FX4000

Oh my god...

I think I'm done posting in this thread.
Technically you can compare i7 920 to AMD Phenom II 940BE, or if you feel that those are not "same gen" then 955BE, the Phenom II chip is gonna fail hard to the intel chip stock OR overclocked even on LN2 to the 920 despite the fact that its BCLK limited.

Now what are we supposed to do, compare a chip released several months ago to a chip that doesn't even exist in public yet?

...and I'm the furthest thing from an intel fanboy.

[chew*]

At least compare same gen competitor chips.

i5 2500K
FX4000

FX 4000 is not out nor do I have accurate numbers nor do I intend to guess.....

My compare was of what I could apple to apple core per core thread for thread.......

Do you think if i chopped 2 cores off thuban it would be any diff?

Yes i'm very interested how 4 core bulldozer will stack up in the same compare.

Tbh tired of benching intel............

Llano I can do the compare on soon......but I already know the outcome, it's deneb with an IGP tossed in......

[Solus Corvus]

But what all those 4/8/12/whatever cores/modules are for if you run single threaded application on them? They are wasted. You guys seem to silently assume that all of the algorithms your shiny new amd systems will be chewing up are easy scalable, multi threaded ones. There are only several classes of computational problems which show such a property. And in this regard amd is clearly suggesting now and then that they emphasize multi-core efficiency and put it above IPC performance in their BD architecture.
And this is very wrong. This is wrong especially because single threaded performance is exactly what their current offering is lacking in comparison to the closest competitor. This is a mistake today and will be tomorrow.

I would argue that the case of a single threaded program with no multitasking OS is just as rare as the case of the MT program that scales perfectly with extra cores.

Every modern OS has many threads running in the background. And when you run a single threaded program a multicore processor lets you run other programs without taking cpu resources from the first program. Most people people won't fall in the extreme single core or extreme multicore case but rather somewhere in between. Where they fit in that spectrum depends on what applications they use and how they use them.

I'm glad that Intel and AMD are taking different approaches because designing a processor that is optimal for that entire spectrum of computer usage would be highly impractical. If you are running a single threaded program and can't do other stuff at the same time, say benching with superpi, then Intel's processors would probably be the better choice. Or if you are running a MT program that scales with cores or running a large number of simultaneous processes, such as encoding or many VMs, then AMD's processor might be the better choice.

I think it is a huge flaw of logic to expect one processor to be the best at everything. Not all programs are the same, not all algorithms are the same, and not all users are the same. IMO, it is good that there will be some diversity in the marketplace so we can pick the processors that are best for what we need them to do.

[jjj]

But what all those 4/8/12/whatever cores/modules are for if you run single threaded application on them? They are wasted. You guys seem to silently assume that all of the algorithms your shiny new amd systems will be chewing up are easy scalable, multi threaded ones. There are only several classes of computational problems which show such a property. And in this regard amd is clearly suggesting now and then that they emphasize multi-core efficiency and put it above IPC performance in their BD architecture.
And this is very wrong. This is wrong especially because single threaded performance is exactly what their current offering is lacking in comparison to the closest competitor. This is a mistake today and will be tomorrow.

Besides the fact that plenty of programs are multithread and that 95% of the users could do with a dual core, you need to understand that BD is a server chip.AMD share in server is very low (around 7% can't really remember the precise number) and server chips have great margins.AMD has around 29% share in desktop and they are ok with that but in notebooks they are at ~13.5% so they are bringing in Llano and in server BD.

[FlanK3r]

ok, first in desktop

[Apokalipse]

Bulldozer is the first 8 core die (as opposed to putting two 4-6 core dies on one chip)

[muziqaz]

Can someone tell me, why Daimler's post was deleted? Were those shots fake?

[Daimler]

Sandy is being Bulldozered. At last ^_^
enjoy:



PS Post restored. Had to clear some things up.
PSS Can't confirm that this is 100% true, but it's from blog of a real Bulldozer owner here: http://obrovsky.blogspot.com/2011/05/wow-enjoy.html

[chew*]

Sandy is being Bulldozered. At last ^_^
enjoy:



PS Post restored. Had to clear some things up.
PSS Can't confirm that this is 100% true, but it's from blog of a real Bulldozer owner here: http://obrovsky.blogspot.com/2011/05/wow-enjoy.html

Without a cpu-z and a rev version even if thats not a fake its pointless.

Your not going to be able to extrapolate any correct data from silicon that any endusers has at this point in time.

[muziqaz]

oh, ok.