AMD cuts to the core with 'Bulldozer' Opterons

[FlanK3r]

BORIS: my x6 1090T is 4300 MHz stable with air cooler XG 1283. Think, not all i5 750 can do this ,-) (yes, Denebs are whorse than Thubans for OC, with air about 3.9-4.1GHz).

[-Boris-]

superpi is your day to day task that you enjoy with the gf and the familly???


i bet you enjoy your collection of blu ray movies more right??? rip them and encode them ...this will require more cores ... wich amd is pretty close to intel in $/performance ..... but yeahh i get your point... single thread performance counts more then core count

I don't use blu-ray, optical media is soo 1995.
But gaming is probably the most performance demanding activity that people generally do with their home computers. So I think games is an important factor.

Personally, I think it's a good idea to buy a processor that has good gaming performance, and then only switch graphics cards the next couple of years, and then change processor again.

[savantu]

I didn't hear anybody complaining when your competitor achieved 33% more
performance with 50% more cores in Westmere....


Regards, Hans

That is perfectly inline with a simple shrink to a new process given the scaling limitations.

BD however is AMD's 1st new core since K7 and they had ample opportunity to start with a clean sheet design. 12.5% without knowing the frequency ( it could very well be higher than MC ) is nothing to write home about.

[-Boris-]

BORIS: my x6 1090T is 4300 MHz stable with air cooler XG 1283. Think, not all i5 750 can do this ,-) (yes, Denebs are whorse than Thubans for OC, with air about 3.9-4.1GHz).

No, not all of them, but not all 1090Ts either. Most i5 750s I worked with have done 4GHz with ease, since I built these systems for others I haven't really pushed them. And since an i5 at 2.66GHz can beat an 3.2GHz X6 in games, and many other apps. I bet a 4GHz+ i5 can beat a x6 4GHz+. And do you want us to take an i7 into the picture?

The thing is, if Zambezi operates at 3GHz-3.6GHz it would be nice if it could compete with i7 in the same range, current X6 Phenom II 3.2GHz is in the area around i5 2.66Ghz. An X4 needs 1GHz more to even get close to i5.

So, thats why people want more than 50% more performance with 33% more cores. 12% per core wont put Bulldozer in i7 range, let alone future intel products.

[-Boris-]

That is perfectly inline with a simple shrink to a new process given the scaling limitations.

BD however is AMD's 1st new core since K7 and they had ample opportunity to start with a clean sheet design. 12.5% without knowing the frequency ( it could very well be higher than MC ) is nothing to write home about.

12.5% per core is actually less than I would expect from just shrinking Phenom II to 32nm. And Phenom II should be capable of around 50% more cores on 32nm too.

[informal]

Barcelona to Shanghai,at the same clock, brought 10-15% increase in server workloads,even the desktop saw a good 6-8% jump(and in games up to 20%,which proves how these apps are cache sensitive).
I think we need to give AMD some credit and to wait for actual results instead of trying to extrapolate single thread performance from an average figure AMD gave this early in the game. This thing has so many improvements on so many levels.

The thing is, if Zambezi operates at 3GHz-3.6GHz it would be nice if it could compete with i7 in the same range, current X6 Phenom II 3.2GHz is in the area around i5 2.66Ghz. An X4 needs 1GHz more to even get close to i5.

Are you for real?1 Ghz? In which parallel world are you living... i5 (no SMT,with Turbo) @ 2.66Ghz can't compete with 3.2Ghz thuban in real world applications.

[-Boris-]

Are you for real?1 Ghz? In which parallel world are you living... i5 (no SMT,with Turbo) @ 2.66Ghz can't compete with 3.2Ghz thuban in real world applications.

Real world? Cinebench isn't something people work with everyday right?
That lostcircuits review is hardly representable for everyday use by ordinary people. But in the programs ordinary people use it was quite close. In videoencoding they won in one program each, in Photoshop they were close, and then there only two games. Personally I don't work with Cinebench or DIEP chess. Do you?

Check this review:
http://www.xbitlabs.com/articles/cpu...t_6.html#sect0

There you have much more user oriented benches. It's pretty tied in applications, i5 wins in games. I ignore the two synthetic categories here.

And check this one:
http://www.anandtech.com/show/3674/a...55t-reviewed/6

X6 is a notch better at videoencoding, it's tied in archiving, an i5 is a clear winner in games and is a tiny bit better at photoshop. I ignore 3D rendering, since I don't know anyone that has done that for years.



Read a bunch of reviews, and please post some here if you like. But as long as you don't play cinebench all day long or some synthetic bench, i5 750 an Phenom x6 is pretty equal, but i5 wins hands down in games, which might be most important category for most desktop users.

[savantu]

Real world? Cinebench isn't something people work with everyday right?
That lostcircuits review is hardly representable for everyday use by ordinary people. But in the programs ordinary people use it was quite close. In videoencoding they won in one program each, in Photoshop they were close, and then there only two games. Personally I don't work with Cinebench or DIEP chess. Do you?

Check this review:
http://www.xbitlabs.com/articles/cpu...t_6.html#sect0

There you have much more user oriented benches. It's pretty tied in applications, i5 wins in games. I ignore the two synthetic categories here.

And check this one:
http://www.anandtech.com/show/3674/a...55t-reviewed/6

X6 is a notch better at videoencoding, it's tied in archiving, an i5 is a clear winner in games and is a tiny bit better at photoshop. I ignore 3D rendering, since I don't know anyone that has done that for years.



Read a bunch of reviews, and please post some here if you like. But as long as you don't play cinebench all day long or some synthetic bench, i5 750 an Phenom x6 is pretty equal, but i5 wins hands down in games, which might be most important category for most desktop users.

To reinforce your point :

http://www.anandtech.com/bench/Product/109?vs=146

[EvilOne]

People also need to take into account the intel compiler that alot of todays benchmarks used aren't kind on AMD chips. If they can run code on AMD chips with older slower instruction set's it will do by default. I cant see that changing any time soon, so AMD will have to put up with it for now.

Atleast FTC seems to be trying to fix this issue.

- Disclose to software developers that Intel computer compilers discriminate between Intel chips and non-Intel chips, and that they may not register all the features of non-Intel chips. Intel also will have to reimburse all software vendors who want to recompile their software using a non-Intel compiler.

[savantu]

People also need to take into account the intel compiler that alot of todays benchmarks used aren't kind on AMD chips. If they can run code on AMD chips with older slower instruction set's it will do by default. I cant see that changing any time soon, so AMD will have to put up with it for now.

Atleast FTC seems to be trying to fix this issue.

Back again to same old stuff : ICC 8.0 did a check for vendor ID, newer versions ( currently ICC 10 ) have the check removed and will check for feature flags ( basically whatever the CPU supports the compiler will throw at it ). However, Intel claim no responsibility for code quality and bugs.
They say the check in 8.0 was introduced simply because AMD did not give them the detailled errata list for their CPUs ( obviously that AMD refrains from sending samples to Intel for validation ).
It would be like AMD sampling now BD to Intel so future updates to Intel's compiler can support BD features.

[FlanK3r]

Real world? Cinebench isn't something people work with everyday right?
That lostcircuits review is hardly representable for everyday use by ordinary people. But in the programs ordinary people use it was quite close. In videoencoding they won in one program each, in Photoshop they were close, and then there only two games. Personally I don't work with Cinebench or DIEP chess. Do you?

Check this review:
http://www.xbitlabs.com/articles/cpu...t_6.html#sect0

There you have much more user oriented benches. It's pretty tied in applications, i5 wins in games. I ignore the two synthetic categories here.

And check this one:
http://www.anandtech.com/show/3674/a...55t-reviewed/6

X6 is a notch better at videoencoding, it's tied in archiving, an i5 is a clear winner in games and is a tiny bit better at photoshop. I ignore 3D rendering, since I don't know anyone that has done that for years.


Read a bunch of reviews, and please post some here if you like. But as long as you don't play cinebench all day long or some synthetic bench, i5 750 an Phenom x6 is pretty equal, but i5 wins hands down in games, which might be most important category for most desktop users.

U read maybe 2 reviews, i read 40-50 reviews about one product ,-) (little bit diference)
So, what is than this:


whats wrong now, eh..., Thuban is winner (and Intel fanboys crying)



comparsion OC 4200 MHz i7 870 (2000MHz RAM, 3600 uncore) vs x6 1090T 4200 MHz (1800MHz RAM, 2900MHz uncore)

[-Boris-]

U read maybe 2 reviews, i read 40-50 reviews about one product ,-) (little bit diference)
So, what is than this:


whats wrong now, eh..., Thuban is winner (and Intel fanboys crying)



comparsion OC 4200 MHz i7 870 (2000MHz RAM, 3600 uncore) vs x6 1090T 4200 MHz (1800MHz RAM, 2900MHz uncore)

I read lots of reviews, the thing is, I read more than one single bench!

Please check through the reviews again, ignore cinebench, sysmark and other synthetic tests, also don't mind looking at programs that most people don't really use. You have, games, archiving, Photoshop, videoencoding and programs like that. Now compare results.

EDIT:
Check savantus link a couple of post up.

[Hornet331]

lost circuit review is kinda a joke.. half the time the 980 or the 1090 don't run at there fullest due to the apps not utulizing them...

[Lokinhow]

U read maybe 2 reviews, i read 40-50 reviews about one product ,-) (little bit diference)
So, what is than this:


whats wrong now, eh..., Thuban is winner (and Intel fanboys crying)

So, this benchmark says that an Athlon II X4 620 is faster than a Core i7 980X at video encoding

OK...OK..........

the one from techreport is pass 1. Pass 2 is much more real world, isn`t?

the last one is ok, show how much a Thuban is powerfull at video encoding, and it is pass 2.

[-Boris-]

Modified Savantus link. Removed programs I never heard anyone use, then removed duplicates, (20 benches in one single intel or AMD favorable program can skew the result a bit.) Then I removed synthetic like sysmark. And all without regard to if the bench I removed helped prove or disprove my point, for example, removed a heap of sysmark which favored i5.

http://www.anandtech.com/bench/Produ...58.59.60.61.62

[informal]

@Boris
And what happens if you've put 940 instead of Thuban there?Yeah difference is minuscule ,proving i5 and i7 give very similar performance(9% difference on average-wow and the difference in clock speed is in that range,who would have though? ).What I am trying to say,it makes no difference how the product is called(i"5") when it performs closely to much more expensive product(940) while costing less.SMT ,in the test selection you made,brings around 10-15% improvement when you count in the clock difference between the 940 and 750.And if Thuban is there with 750(it's actually faster than 750 but whatever) it's there with much more expensive products like the higher end i7s,too.

[-Boris-]

@Boris
And what happens if you've put 940 instead of Thuban there?Yeah difference is minuscule ,proving i5 and i7 give very similar performance(9% difference on average-wow and the difference in clock speed is in that range,who would have though? ).What I am trying to say,it makes no difference how the product is called(i"5") when it performs closely to much more expensive product(940) while costing less.SMT ,in the test selection you made,brings around 10-15% improvement when you count in the clock difference between the 940 and 750.And if Thuban is there with 750(it's actually faster than 750 but whatever) it's there with much more expensive products like the higher end i7s,too.

I suggest that you compare i7 to X6 directly, as i5 often is a bit of a middle ground it makes it hard to see the differences clearly. Phenom II X6 can't compare with i7.

[FlanK3r]

Why u craying guys....Once again, with every new CPU i rading most review from world (ussually its from 15 to 50 reviews at one product). Need not some teching about performance modern CPUs. But to time i have not all time for right graphs, thats all. It was only example.
If i will normal user, dont need expensive CPUs, but only dualcore (or dual with HTT) and triplecores. Maybe some low priced Quad. We are not mostly classic users!
Lostcircuits is not joke, problem with 12 threads (6+6) is jumping form threads to next threads. Is not much aplications using more than 4 cores at "100%" stressing. And second, Games are not to ideal for more than quadcores, look at some comparsion i7 HT on and HTT off.
Lol, Quadcores and more are ideal job for videoencoding, 3D graphics, CAD etc etc. And Intel with Gulftown launch was big brother with Cinema Studio and new Cinebench R11.5 (yes, this simulating is very good, better than in R10)

[informal]

I suggest that you compare i7 to X6 directly, as i5 often is a bit of a middle ground it makes it hard to see the differences clearly. Phenom II X6 can't compare with i7.

Oh so now X6 can't compare(you mean compete?) with i7s?Right.
Lost Circuits is not a joke,it's probably the best HW website out there.The guy who runs it is the VP of OCZ tech. And 3D rendering is one major part of performance when it comes to modern day desktop chips,just like A/V encoding.Gaming comes last since you can game just as well with lower end QC like Q6600 or Phenom 9850/810.

[FlanK3r]

I suggest that you compare i7 to X6 directly, as i5 often is a bit of a middle ground it makes it hard to see the differences clearly. Phenom II X6 can't compare with i7.

¨

Boris: little kid...informal has more better experience and arguments than u. Why are u so "pro-Intel" ? Look at it more realistic. Im in love with AMD, but know, Intel have great CPUs as example i5 750 (performance, price, OC) and i7 800 series (some i3, i5 6xx, i7 950 and higher are too much expensive)

[Lokinhow]

Why u craying guys....Once again, with every new CPU i rading most review from world (ussually its from 15 to 50 reviews at one product). Need not some teching about performance modern CPUs. But to time i have not all time for right graphs, thats all. It was only example.
If i will normal user, dont need expensive CPUs, but only dualcore (or dual with HTT) and triplecores. Maybe some low priced Quad. We are not mostly classic users!
Lostcircuits is not joke, problem with 12 threads (6+6) is jumping form threads to next threads. Is not much aplications using more than 4 cores at "100%" stressing. And second, Games are not to ideal for more than quadcores, look at some comparsion i7 HT on and HTT off.
Lol, Quadcores and more are ideal job for videoencoding, 3D graphics, CAD etc etc. And Intel with Gulftown launch was big brother with Cinema Studio and new Cinebench R11.5 (yes, this simulating is very good, better than in R10)

yeah, a problem with 12 threads
is there a problem with 8 threads too? because X4 620 is faster than a i7 920 on this benchmark...

i7 870 is faster than i7 965 in this bencharmk, problems with triple channel?
don't know about the rest, but THIS benchmark from lostcircuits is a joke. or do you really think that a X4 620 is faster than i7 920 at video encoding?

thuban IS really good for video encoding, and even better for 3D rendering. But this benchmark makes no sense.

[SEA]

12.5% per core is actually less than I would expect from just shrinking Phenom II to 32nm. And Phenom II should be capable of around 50% more cores on 32nm too.

There is no such a number like 12.5%. This was a result of incorrect direct calculation... You forgot scaling and frequencies. So you don't know about single core IPC. Also desktop BD will have less cores, and it will result in different scaling numbers.
Also you speak of 16-core BD in line with Phenom. This is wrong again.
Interlagos is not suppose to be clocked high rather it has to have more cores and lower frequency to fit server market performance needs most.
Example: a top phenom is 3.4 GHz. Top server MC is 2.3 GHz. Compare Intel's top desktop and server chips yourself...

[-Boris-]

Oh so now X6 can't compare(you mean compete?) with i7s?Right.
Lost Circuits is not a joke,it's probably the best HW website out there.The guy who runs it is the VP of OCZ tech. And 3D rendering is one major part of performance when it comes to modern day desktop chips,just like A/V encoding.Gaming comes last since you can game just as well with lower end QC like Q6600 or Phenom 9850/810.

When I upgraded from Phenom 9850 to Phenom II 940 a few games suddenly became alot more playable. For instance GTA IV.
Even if modern games are multithreaded, it doesn't change the fact that performance/thread is the most important thing.

And IF bulldozer is 50% faster from 33% more cores, I feel it won't beat i5 at games. Since performance/thread isn't that much over Phenom II.

Watch this, i5 vs Phenom II, same amount of cores, around the same clockspeed. i5 wins it all by a huge margin.
http://www.anandtech.com/bench/Product/109?vs=85

So, why do I compare at the same clockspeed? Because Intel and AMD currently makes processors around the same clockspeed, but intel seems to have even more room for high frequencies than AMD.
Thus, AMD needs major improvements over current generation in singlethreaded performance to match intel here.

So, 12% more performance per thread won't be enough to even beat current generation CPUs from intel. And when it comes to multithreaded performance, AMD will possibly be able to add more cores than Intel, around 2 more per die perhaps. But that wont compete with HT combined with much stronger cores.

8 core bulldozer vs 6 core i7 with HT? I can't see the bulldozer win this one if it already looses singlethreaded performance by such a margin.

[Manicdan]

comparing a server chip built for super perfect thread scaling, with gaming where a duel core still offers the best fps/dollar, is fail i think

also if the 50% faster with 33% is with same clockspeeds, please someone say it, cause so far i dont think it has been mentioned, and will greatly affect the IPC calculations people like so much

[-Boris-]

There is no such a number like 12.5%. This was a result of incorrect direct calculation... You forgot scaling and frequencies. So you don't know about single core IPC. Also desktop BD will have less cores, and it will result in different scaling numbers.
Also you speak of 16-core BD in line with Phenom. This is wrong again.
Interlagos is not suppose to be clocked high rather it has to have more cores and lower frequency to fit server market performance needs most.
Example: a top phenom is 3.4 GHz. Top server MC is 2.3 GHz. Compare Intel's top desktop and server chips yourself...

Server Bulldozer is to desktop bulldozer what server MC is to Phenom II.

And the reason to why I say 12% in singlecore boost is that server loads is very parallel. AMD gave us numbers in server loads. So, when utilizing every core a Bulldozer is 12.5% quicker than a MC per core. BUT, that is still very optimistic, since the thing with bulldozer is that it's meant to scale well. So if Bulldozer scales with amount of cores better than MC. It will have even less performance in singlethreaded applications.

So i'm hoping for the opposite to prove me wrong, that Bulldozer scales bad due to modules instead of cores, and due to shared FPU, It would be fantastic if these 50% total output was due to extreme singlethreaded performance that just happens to scale bad. That would mean over 50% performanceincrease in singlethreaded apps, but "only" 50% in multithreaded. However, a bit unlikely.


I want to make some things very clear here.
Everything is based on 50% performance increase over MC. I hope that number is extremely conservative. And that is for initial bulldozers and that they increase in clockspeed fast.

Personally, i believe Bulldozer will be much more capable than these numbers suggest. 4 ALUs, loads/stores per clock, shared FPU, 32nm, and totally new core probably capable of much higher performance/watt suggest that bulldozer will be very very good.


So I'm hoping I'm wrong here. Since I am already planning to throw my AM2+ system away and buy an AM3+ system next year.