AMD's Bobcat and Bulldozer

[madcho]

on the mcm part they are still bound by tdp ... now have 2 of those glued together and you need to lower your clocks considerably ... so you still stay in the desired tdp so i dont think you will see clocks higher then 2.3 for the 16 cores version ...

maybe 2.5 for the 12 cores mcm part on 32nm ... and higher the less cores they have

2.3ghz is available for 12cores mcm magny cour.

I don't think next magny cour is 3x2 modules only ... and it's already said 33% more core, so next magny cour ( interlago ) is 16 cores.

[Florinmocanu]

on the mcm part they are still bound by tdp ... now have 2 of those glued together and you need to lower your clocks considerably ... so you still stay in the desired tdp so i dont think you will see clocks higher then 2.3 for the 16 cores version ...

maybe 2.5 for the 12 cores mcm part on 32nm ... and higher the less cores they have

You already have 2.3 ghz for 12 full cores at 45nm. With smaller cores with BD and smaller process node, i can definitely say that we will see higher frequency than 2.3 for 16 and 12 core BD parts.

[-Boris-]

on the mcm part they are still bound by tdp ... now have 2 of those glued together and you need to lower your clocks considerably ... so you still stay in the desired tdp so i dont think you will see clocks higher then 2.3 for the 16 cores version ...

maybe 2.5 for the 12 cores mcm part on 32nm ... and higher the less cores they have

So you expect a 4 module BD to be as hot as a six core 10h at the same frequency?

[savantu]

<3GHz, Maybe 2,6GHz. Also, i expect IPC gain to be atleast 10% in integer code, more in float when running "normal" application, one that does not have only floating point calcs. That kind of code would be better to be ran on gpu's anyway.

Let's say 2.6GHz.

2.6GHz/2.2Ghz(MC)*16/12(cores)*1.1(IPC)= 1.733

Which means BD should be 73,3% faster than MC at least in INT code. I'd assume their performance estimates are based on Spec_Int_Rate and Spec_FP_Rate.

[Sn0wm@n]

So you expect a 4 module BD to be as hot as a six core 10h at the same frequency?

why not ??? more logic then 10H more cores etc... even with the shrink i doubt that the 8 cores desktop will first come at more then 3.2ghz ... maybe with later revision higher clocks ... but for release it wont go higher then that

You already have 2.3 ghz for 12 full cores at 45nm. With smaller cores with BD and smaller process node, i can definitely say that we will see higher frequency than 2.3 for 16 and 12 core BD parts.

but it wont be the same architecture as magny cours ... so only a 200mhz bump to 2.5 for the high end model wouldnt be hard to believe ... maybe 2.6 without turbo .... now how the turbo will kick in on the server is to be seen but i believe that stock with no turbo it wont go past 2.6 at the most for a 12 cores 32nm part ... 16 cores wont go much past 2.3ghz .. if my assumption is bad then its all good ...

[Chrono Detector]

AMD seems quite confident that Bulldozer can actually deliver, and I do hope they are right. They need to really compete in the high end against Intel.

I wonder, even if the top end Bulldozer does actually perform faster than Intel's top end or close, will they price it like $1000 or more like the Athlon FX.

[Sn0wm@n]

AMD seems quite confident that Bulldozer can actually deliver, and I do hope they are right. They need to really compete in the high end against Intel.

I wonder, even if the top end Bulldozer does actually perform faster than Intel's top end or close, will they price it like $1000 or more like the Athlon FX.

that's what i am afraid if it comes down like this :S ... at the time i got into computers more the athlon fx 62 was at around 900$ or so :S ... let's hope that they wont price it like that

[slaveondope]

I wont mind if the price isnt too high seeing as we should still have plenty of "wiggle" room with the fsb(htt) on the low end chips, assuming they dont pull a Sandy Bridge stunt.

If it delivers Id suspect both Intel and AMD to come down a bit.

[-Boris-]

why not ??? more logic then 10H more cores etc... even with the shrink i doubt that the 8 cores desktop will first come at more then 3.2ghz ... maybe with later revision higher clocks ... but for release it wont go higher then that

Because I think that an octocore 10h at 32nm would do 2.6GHz easily. And I think that a BD module uses less trannies than a dualcore 10h.

[Sn0wm@n]

[QUOTE=-Boris-;4533989]

why not ??? more logic then 10H more cores etc... even with the shrink i doubt that the 8 cores desktop will first come at more then 3.2ghz ... maybe with later revision higher clocks ... but for release it wont go higher then that/QUOTE]

Because I think that an octocore 10h at 32nm would do 2.6GHz easily. And I think that a BD module uses less trannies than a dualcore 10h.

where in hell have i said an 8 cores desktop cpu would only be at 2.6ghz ??? nowhere... i said 3ghz to 3.2 at first and maybe higher with later revision ....

[Florinmocanu]

He said that 8 10H if done on 32nm would be bigger than 8 BD modules. The rest you can figure out for yourself.

[superrugal]

Let's say 2.6GHz.

2.6GHz/2.2Ghz(MC)*16/12(cores)*1.1(IPC)= 1.733

Which means BD should be 73,3% faster than MC at least in INT code. I'd assume their performance estimates are based on Spec_Int_Rate and Spec_FP_Rate.

bingo??

http://www.heise.de/ct/artikel/Proze...r-1064662.html

google translated：http://translate.googleusercontent.c...hbJHxdcLE54JUg

With its eight modules - so depending on the perspective, eight to 16 cores - should the bulldozers server chip Interlagos about 70 percent more integer performance (SPECint) than the 12-Kerner achieve Magny-Cours, which provides thus not really a "starving" front end out. Besides the thick Interlagos with up to 8 MB L3 cache for all modules on the chip AMD plans to release half as large chips for servers (Valencia) and high-end desktop PCs (Zambezi). For floating-point Interlagos has to offer in comparison to Magny-Cours, although a third less cores, but with him will be higher thanks AVX and FMA, and better memory connection SPECfp the processing power by a third. These are connected FPUs not even on the small L1 cache. An L1-bypass for FPUs, which had little success Intel Itanium also - hopefully this is not a bad omen

[Calmatory]

In this very thread people discuss frequencies of 3.3-4 GHz for BD which is significantly higher than MC ( max 2.3GHz ).

The 50% more performance, 33% more cores applies versus Magny Cours. You also need to factor in frequency since this was the unknown part in the AMD slide.

...except that the figure was supposed to be at the same frequency, no? Terrace threw few pages of entertainment in with "IPC ONLY 12.5% HIGHER THATS BAD!" stuff based on this. Now suddently the IPC gets worse and AMD is doomed?

[-Boris-]

where in hell have i said an 8 cores desktop cpu would only be at 2.6ghz ??? nowhere... i said 3ghz to 3.2 at first and maybe higher with later revision ....

I meant an octocore MCM = 16 cores. I think that an Octocore 10h at 32nm is cooler than a Hexacore 10h at 45nm. So, i believe that even without a new architecture AMD would be able to make an 16 core MCM processor at 2.3GHz and higher.

I also believe that a Hexacore 10h uses more trannies than an 4 module BD.

[AliG]

...except that the figure was supposed to be at the same frequency, no? Terrace threw few pages of entertainment in with "IPC ONLY 12.5% HIGHER THATS BAD!" stuff based on this. Now suddently the IPC gets worse and AMD is doomed?

haven't you heard?

The more they post the more Bulldozer's IPC decreases. They are secretly Intel spies trying to damage AMD by their fierce posting on an internet forum.

[Mats]

Here's another one:

A quick and raw estimation of single threaded performance for Zambezi based on the 50% number given for Interlagos (just to show, what has to be counted in at the least):

Relative_perf_1_thread_to_AMD_fam_10h = (Perf_Magny_Cours*1.5 * 12 / 16) * Freq_ratio_of_half_#_of_Cores * Perf_boost_single_core_in_Module * Perf_boost_single_module_on_chip

Freq_ratio_of_half_#_of_Cores = 3.2/2.3 = 1.39
Perf_Magny_Cours = 1
Perf_boost_single_core_in_Module = 1.11 (while going from 90% back to 100%)
Perf_boost_single_module_on_chip = 1.3 (some cheap turbo)

Relative_perf_1_thread_to_AMD_fam_10h = (1 * 1.5 * 12/16) * 1.39 * 1.11 * 1.3 = 2.26

So with some frequency scaling a Zambezi core will be about 126% faster than a core running in a 2.3GHz MC without turbo. This would equal a 5.2GHz PhII core.

This is just speculation. Anyone is invited to check this.

[Sn0wm@n]

He said that 8 10H if done on 32nm would be bigger than 8 BD modules. The rest you can figure out for yourself.

oki ... if we follow that route what kind of clocks are you expecting for a desktop 8 cores cpu ????

[AliG]

that's what i am afraid if it comes down like this :S ... at the time i got into computers more the athlon fx 62 was at around 900$ or so :S ... let's hope that they wont price it like that

AMD seems quite confident that Bulldozer can actually deliver, and I do hope they are right. They need to really compete in the high end against Intel.

I wonder, even if the top end Bulldozer does actually perform faster than Intel's top end or close, will they price it like $1000 or more like the Athlon FX.

Why not? I'm not saying I would ever buy their products if they priced it that high, as I go with the best low-mid grade product on the market regardless of brand, but once again, why not?

If Intel can remove fsb overclocking and it's understood around here why they are doing it (for profit reasons of course), then why shouldn't AMD be able to profit if they have a real winner?

Even still, if you can remember back to conroe pricing, Intel smashed AMD performance wise, but didn't really start to milk their products until now, unlike NVIDIA. They always had very powerful, very affordable products and also the ridiculously overpriced extreme editions. It's really up to us to decide the pricing anyways, as the market always determines a product's value, not the seller (at least in the US).

[Sn0wm@n]

I meant an octocore MCM = 16 cores. I think that an Octocore 10h at 32nm is cooler than a Hexacore 10h at 45nm. So, i believe that even without a new architecture AMD would be able to make an 16 core MCM processor at 2.3GHz and higher.

I also believe that a Hexacore 10h uses more trannies than an 4 module BD.

let's forget server part for a minute shall we .... and lets focus on desktop ... 8 cores @ 3.2 on release and later on maybe even higher without turbo boost ....


and if you so love the server part then you could see how wrong my statement was when JF-AMD makes an official statement about the clocks of the bulldozer cpu lineup ....

Why not? I'm not saying I would ever buy their products if they priced it that high, as I go with the best low-mid grade product on the market regardless of brand, but once again, why not?

If Intel can remove fsb overclocking and it's understood around here why they are doing it (for profit reasons of course), then why shouldn't AMD be able to profit if they have a real winner?

Even still, if you can remember back to conroe pricing, Intel smashed AMD performance wise, but didn't really start to milk their products until now, unlike NVIDIA. They always had very powerful, very affordable products and also the ridiculously overpriced extreme editions. It's really up to us to decide the pricing anyways, as the market always determines a product's value, not the seller (at least in the US).

well because i wouldnt buy a 1000$ gpu ... let alone a 1000$ cpu .... and most people wouldnt ... the market for those type of cpu is quite low and unrealistic ... so if amd goes that route ill stick with what i have now ....

[AliG]

let's forget server part for a minute shall we .... and lets focus on desktop ... 8 cores @ 3.2 on release and later on maybe even higher without turbo boost ....


and if you so love the server part then you could see how wrong my statement was when JF-AMD makes an official statement about the clocks of the bulldozer cpu lineup ....

If you're looking purely at desktop, do you really need 4+ modules? Honestly I'm more interested in seeing the 2 module version, I really could care less for anything beyond 4 threads as I just can't think of anything I do that needs the processing power.

It's the samething I brought up in the Sandy Bridge thread, I'm most interested in seeing both how many additional multipliers you get beyond turbo and also how the 2500k is priced. Because quite frankly I don't need an i7, especially not the additional motherboard costs, so if I can get a quadcore (non ht) sandy bridge that overclocks to 4.5 ghz on air, then I'll be a really happy camper.

Same goes for AMD, if their 2 module processor is only 5% behind SB (unlikely, but we can hope), and they clock to 5ghz on air, then that's what I'm getting. As I said before, in the end I could really care less how well the top end stuff costs and performs as I can't afford it anyways. If one company offers a significantly better valued product around the ~$200 level, then I'll be very content.

[Sn0wm@n]

well 2 programs that i use scale well with more cores ... so yes i wouldnt mind having 8 cores ... and i could also re-install my vmware workstation and set up a virtual machine to try out linux and do some other bits ....

[Motiv]

If you're looking purely at desktop, do you really need 4+ modules? Honestly I'm more interested in seeing the 2 module version, I really could care less for anything beyond 4 threads as I just can't think of anything I do that needs the processing power.

It's the samething I brought up in the Sandy Bridge thread, I'm most interested in seeing both how many additional multipliers you get beyond turbo and also how the 2500k is priced. Because quite frankly I don't need an i7, especially not the additional motherboard costs, so if I can get a quadcore (non ht) sandy bridge that overclocks to 4.5 ghz on air, then I'll be a really happy camper.

Same goes for AMD, if their 2 module processor is only 5% behind SB (unlikely, but we can hope), and they clock to 5ghz on air, then that's what I'm getting. As I said before, in the end I could really care less how well the top end stuff costs and performs as I can't afford it anyways. If one company offers a significantly better valued product around the ~$200 level, then I'll be very content.

It's a strange one isn't it. If AMD can't get close to SB in terms of out and out performance (per core), then they will need more GHZ to get to that point.

you then hit the problem of a two module chip, being 80% the power if all cores on those modules are being run; so you need even more ghz to match Intel (all estimates of course).

So would the ideal solution, for a powerful desktop, be a 3 module system?

[-Boris-]

you then hit the problem of a two module chip, being 80% the power if all cores on those modules are being run; so you need even more ghz to match Intel (all estimates of course).

It's 90%!

[Florinmocanu]

me personally i want 12-16-32 , as many the better since my workflow requires a lot of CGI rendering. But that's me.

[Florinmocanu]

It's a strange one isn't it. If AMD can't get close to SB in terms of out and out performance (per core), then they will need more GHZ to get to that point.

you then hit the problem of a two module chip, being 80% the power if all cores on those modules are being run; so you need even more ghz to match Intel (all estimates of course).

So would the ideal solution, for a powerful desktop, be a 3 module system?

not really. Ipc may be lower than SB, or not, but, BD will be a higher count CPU that SB. for the 8-10 SB core versions you will have 12-16 cores BD. Probably in future BD CPUs we will get 12-16 cores for desktop as well, if they need it.


So, in multi-threaded scenarios, BD will be really good. In single thread scenarios, that penalty you say does not apply anymore, since 1 core in a module is used, not both. The penalty applies only when both of them work. And they would work at 90%, not 80%. A module will be 180% of 2 theoretical BD cores, which would be 200%.