Bulldozers first screens

[XRL8]

Im stumped on the way some people are "sure" that BD will be a weakling.All information we as a community have points to it being a fresh and new "all bases covered" attempt at the cpu.
Of course thing CAN get bad, phenom I was a good cpu on paper, but it had been plagued by low clocks and TLB bug.However information we do have now points to it being somewhat comparable to sandy bridges available at this time (ON AVERAGE) .Probably a bit more powerfull multithreaded and a weaker single threaded but with more aggresive turbo to offset that.
So where this "its gonna fail" coming from ? Phenom I ?Intel had failed 1ghz PIII, it had netburst it had rambus, it still has itanium, chipset problems in i820 and lately in p67.Both Companies had their share of fail.
I really welcome good multithreading, because thats where wolrd is going for one ,and second im a power user, i do multitask heavily, and while i dont see a difference between game running 100 or 200fps ,i like to do all my threads fluidly.Besides, if theres no reason for 8 threads, so why 2600k ?
Nobody in their right mind expects BD to surpass sandybridge clock for clock with a core for core.Lets just get it out there.It doesnt mean thats going to be a weak cpu tho.

[h0bbes]

I assume that is a variation of the quote :

Never argue with stupid people, they will drag you down to their level and then beat you with experience.

And it applies to any fanboy (we have both types in this thread).

[SEA]

But we can break it down to these main groups.
1. Ordinary users, using Office, youtube, flash games and simple browsing and occasional games.
...

- Office does not seem to need any significant cpu at all, you can remove it from list of single for single cpu as well.
- Flash games are multithreaded now (from flash ver 10.1 i believe), particularly in Opera they use more than 2 cores.
- Same for simple browsing. Developers use worker threads in JS or regular web sites, and you even don't notice that. Chrome opens (usually) tabs in new process. You can always open Opera new process manually. IE9 is highly multithreaded. All that used in simple browsing.
- occasional games. It is like occasionally playing Crysis, right? OK, the more cores the happier it is.. Or flash games, again, more that 2 threads utilised in Opera.

Bottom line: forget about single core (and even dualcore) at all!

[Micutzu]

...

Your choice in words (basically making an advantage of turbo or multiple offerings in the market, even if the same applies to competition's products) proved, at least for me, that it would be very hard to change your mind about this, and the discussion will go beyond the boundaries of rationality.

[Postmodum]

First argument from a fanboys is :

Calling others fanboys and that he can't have a solid discussion with them... even if all he did was post an entry calling others fanboys & not having arguments what so ever ... Oh well... Another one bites the dust.

As for my 2 cents...

My thoughts are that BD will be a solid bet for a user that uses heavily the hyper threading ...
Even 990x can't be immensely superior to the SB ... SB was a work of art for the mainstream ... So... can't imagine the BD "trashing" the SB. But i'm sure it will be a viable competitor with a much stronger capacity for HT rendering ...

As for gaming... let's face it ... i believe it won't be the best performer but will deliver more than enough cpu raw performance for games... enough performance for you not being affected with cpu bottleneck's for sometime ...

[informal]

I would still look back at this leaked slide(from this perspective I'm fairly confident it's not a fake one):


Many people came to a conclusion,primarily from C11.5 numbers,that Zambezi will have very strong FP units.A total score of almost 11 pts tells us that per one single FMAC(in shared MT mode) we have roughly 30% more computational power than one whole Thuban core,roughly at the same clock.When Zambezi doesn't see load across all cores,FMACs can be "paired up" (per module) so that in single or poorly threaded fp workloads we have a lot higher fp resources,on top of those 30%. Now on top of all this comes new Turbo boost in poorly threaded workloads.All things summed up,at least in legacy (or FMA tuned) fp code,Zambezi will be a big step up from today's cores,yes even versus SB.

What is unknown is pure integer core performance.I'm hoping for at least 10-15% IPC uplift,in non-MT workloads.Turbo should be aggressive here,even with all cores loaded so we can expect another 10% from added Turbo headroom(4.1Ghz rumored Vs 3.7Ghz on Thuban today).So effectively 25% uplift versus Thuban 1100T is not out of realm of possibility.

[MrMojoZ]

Your Thuban is raped by the 2600K ...

Have you really tried SB platform

He specifically mentioned rendering, have you never looked at Thuban rendering benchmarks? Did you really need to come into an AMD thread to tell someone that Intel "rapes" AMD? With the added bonus of being wrong at the same time?We still have people this pathetic on this forum? This thread really shines an ugly light on some people.

[[XC] gomeler]

trollolololol. g-bye.

[Micutzu]

Well, it's hard to have a solid discussion if an AMD official says 33% more cores brings up to 50% more performance => roughly up to 12.5% improvement in performance per core vs. current generation, and here we are talking about beating Sandy Bridge in desktop applications .

[Manicdan]

i think some people need to not make it so personal, otherwise bans could be coming...

for those arguing about BD in desktop single threaded uses, i feel sorry for the people who buy an octo-core cpu, when they really just needed a duel-core, and could save themselves a few hundred dollars. for gaming, its quite easy to expect the turbo from 2-4 threads will be strong enough to keep it up there with SB. just take thuban gaming scores, add 10-15% for IPC, add 10% for clocks if 3 cores or less (4.1ghz/3.7ghz), 24% if its 4 threads since thuban can only turbo 3 (4.1ghz/3.3ghz). were looking at 25 to 46% increase in gaming performance over the best thuban vs the first BD. (how well can SB turbo with 4 threads running?, i only know max turbo is 3.8ghz, but i have no clue what cases let it reach that) *these numbers are based on stock expectations and have no bearing on OCed results that are way to much guesswork to predict.

[informal]

Well, it's hard to have a solid discussion if an AMD official says 33% more cores brings up to 50% more performance => roughly up to 12.5% improvement in performance per core vs. current generation, and here we are talking about beating Sandy Bridge in desktop applications .

That's throughput server performance.You also fail to understand that when everything is loaded up in Orochi,due to shared design we have roughly 25% "penalty".This is average number since AMD stated BD shared design offers an average of 80% performance of CMP approach(hypothetical BD with no shared resources). To muddy the waters even more we have an aggressive new Turbo ability which will be more useful in desktop due to nature of workloads there. Then you have FMAC(floating point unit) vs integer IPC etc.

My point is you cannot use the "50% more throughput" statement to accurately extrapolate desktop performance.

[Micutzu]

He specifically mentioned rendering, have you never looked at Thuban rendering benchmarks? Did you really need to come into an AMD thread to tell someone that Intel "rapes" AMD? With the added bonus of being wrong at the same time?We still have people this pathetic on this forum? This thread really shines an ugly light on some people.

Ooops.

http://lab501.ro/procesoare-chipsetu...an-attitude/17
http://lab501.ro/procesoare-chipsetu...an-attitude/18
http://www.anandtech.com/show/4083/t...2100-tested/17
http://www.bit-tech.net/hardware/cpu...ridge-review/7
http://www.tomshardware.com/reviews/...k,2833-15.html
http://pcper.com/reviews/Processors/...w/Render-Tests
http://www.lostcircuits.com/mambo//i...&limitstart=14

[-Sweeper_]

That's throughput server performance.You also fail to understand that when everything is loaded up in Orochi,due to shared design we have roughly 25% "penalty".This is average number since AMD stated BD shared design offers an average of 80% performance of CMP approach(hypothetical BD with no shared resources). To muddy the waters even more we have an aggressive new Turbo ability which will be more useful in desktop due to nature of workloads there. Then you have FMAC(floating point unit) vs integer IPC etc.

My point is you cannot use the "50% more throughput" statement to accurately extrapolate desktop performance.

wasnt it 180% the performance of a regular BD "core"?

[Manicdan]

wasnt it 180% the performance of a regular BD "core"?

1 full core + .8 core = module performance
*ignores IPC differences, do no compare with old architectures*

[-Boris-]

- Office does not seem to need any significant cpu at all, you can remove it from list of single for single cpu as well.
- Flash games are multithreaded now (from flash ver 10.1 i believe), particularly in Opera they use more than 2 cores.
- Same for simple browsing. Developers use worker threads in JS or regular web sites, and you even don't notice that. Chrome opens (usually) tabs in new process. You can always open Opera new process manually. IE9 is highly multithreaded. All that used in simple browsing.
- occasional games. It is like occasionally playing Crysis, right? OK, the more cores the happier it is.. Or flash games, again, more that 2 threads utilised in Opera.

Bottom line: forget about single core (and even dualcore) at all!

These simple tasks don't require a 8 Core BD. Probably works fine on any CPU capable of 4 threads out there.

Your choice in words (basically making an advantage of turbo or multiple offerings in the market, even if the same applies to competition's products) proved, at least for me, that it would be very hard to change your mind about this, and the discussion will go beyond the boundaries of rationality.

The first step out from rationality is when you stopped using relevant arguments and started with this Ad Hominem.
I don't say that turbo or multiple offerings is an advantage over the competition at all. I just say that these are aspects you have to consider when you talk about BD. And that an aggressive turbo might matter more to AMD than to Intel. I also say that there is no reason to believe that BD will be significantly worse than competing products in the same segment in poorly threaded applications. This seems to be a misunderstanding based on how many cores it has. No one uses this argument against 4C/8T i7, so why use it against BD?

Why won't BD fit the same market as quad core i7s?

[-Boris-]

Well, it's hard to have a solid discussion if an AMD official says 33% more cores brings up to 50% more performance => roughly up to 12.5% improvement in performance per core vs. current generation, and here we are talking about beating Sandy Bridge in desktop applications .

Has anyone actually made these claims or is this just a straw man?

[informal]

wasnt it 180% the performance of a regular BD "core"?

AMD slide stated 80% of CMP approach.
.

This is CMP approach(of hypothetical BD):


80% of 100 is 80,or 25%(1.25x) less. This is an average number though,so the penalty can be as low as 5% or as high as 40%,we simply don't know. My guess is that usually we will see a 10-20% penalty which will be offset by Turbo mode.This is for pure integer workloads. For fp ones we have over-provisioned FPU which will offer tangible increase in performance and which won't suffer that much in shared mode.

edit: the "throughput advantages ..." line talks exactly about this "penalty". Workload components won't suffer much when module is fully loaded.

[-Sweeper_]

1 full core + .8 core = module performance
*ignores IPC differences, do no compare with old architectures*

yeah, the fact is, I dont see a 25% penalty there..

even 2 regular cores dont scale linearly, more like 1.95x..

edit: got it

[h0bbes]

80% of 100 is 80,or 25%(1.25x) less.

20%? (sorry)

[informal]

20%? (sorry)

100/80=1.25 or 25% less. . If BD's approach achieves 80% of hypothetical non shared approach then we would have : ie. 200pts in some benchmark for hypothetical design; 80% of 200 is 160pts. Now hypothetical design is 200/160=1.25x or 25% faster.

[Sparky]

Aren't percentages sometimes so wonderfully confusing

[XRL8]

Ooops.

http://lab501.ro/procesoare-chipsetu...an-attitude/17
http://lab501.ro/procesoare-chipsetu...an-attitude/18
http://www.anandtech.com/show/4083/t...2100-tested/17
http://www.bit-tech.net/hardware/cpu...ridge-review/7
http://www.tomshardware.com/reviews/...k,2833-15.html
http://pcper.com/reviews/Processors/...w/Render-Tests
http://www.lostcircuits.com/mambo//i...&limitstart=14

I looked at results you posted.Theres no "raping" there of 1100T.2600K is faster, 2500K depends, sometimes faster sometimes slower.
Only thing raping there everyone is gulftown.

[h0bbes]

Aren't percentages sometimes so wonderfully confusing

Apparently yes :P It's all about the context (whose number's percentage you are talking about:P)

[Aerwidh]

100/80=1.25 or 25% less. . If BD's approach achieves 80% of hypothetical non shared approach then we would have : ie. 200pts in some benchmark for hypothetical design; 80% of 200 is 160pts. Now hypothetical design is 200/160=1.25x or 25% faster.

No offense, but... what?

200 is 25% faster than 160, sure. 160 is only 20% slower than 200, though. The numbers change depending on which number (the bigger or the smaller) you base the comparison on

Edit: Simply put, the "faster than" or "slower than" establishes that the second number is always the one used as a base amount, so you can't make the calculation the other way

[Manicdan]

x/y=z
y/x= reciprocal of z

2/1 = 2
1/2 = reciprocal of 2 which is .5

.8 and 1.25 are reciprocals of each other.