AMD's Bobcat and Bulldozer

[trinibwoy]

but thank you for pointing out the obvious that yes there was a loss and B it wasnt significant ... you just proved my point

There is no point to be made until we see some independent benchmarks, who knows what AMD considers a "significant loss".

[Florinmocanu]

We should also take into account that the 180% performance of a module compared to 2 normal cores is not a comparison of a BD module vs 2 Thuban cores.

The comparison is quite clear, if they wouldn't have chosen the module design and went for the classic core design, than 2 Bulldozer cores would mean 200% while a Bulldozer module means 180%.

But the saving in die space and power consumption is big so the choice of using modules is obvious.

So they allow a small performance loss, but they gain the ability to add a lot more modules/cores to a BD CPU. Which is really nice.

[Hornet331]

Could someone summarize what was learned from the info at HotChips?

Not that much new, ammount of caches has been revealed/confirmed, overall architecture layout of bobcat/bolldozer has been revealed and thats preaty much it. No performance numbers.

place your bets now!

1/ shintai
2/ terrance
3/ stargazer
4/ gallag
5/ savantu

heheheheee

Not a singel amdroid.. suprsie, suprise....

[Sn0wm@n]

There is no point to be made until we see some independent benchmarks, who knows what AMD considers a "significant loss".

yes the point was at first that terrace is getting more boring the more he points out that hes only pointing the bad sides of amd ... but you didnt see that one at first ....

[JF-AMD]

Hey John,any chance we can listen to the archived audio from the Hot Chips,like webcast or similar?

Not aware that it was recorded.

[JF-AMD]

Single-threaded performance will have significant improvement.

Ssssshhhhhh! Don't tell terrace. He'll say "Single-threaded performance will have significant improvement"

[blindbox]

Ssssshhhhhh! Don't tell terrace. He'll say "Single-threaded performance will have significant improvement"

Hahahahahahahaha

Whoa, 500 posts already

[Dresdenboy]

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.

[mongoled]

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.

[superrugal]

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.

Freq_ratio_of_half_#_of_Cores = 3.2/2.3 = 1.39

Does 3.2 means Interlagos freq is 3.2ghz? Is it a few too optimistic?

[madcho]

i hope you're right dresdenboy

And JF, is there a desktop guy ?

[Mats]

Freq_ratio_of_half_#_of_Cores = 3.2/2.3 = 1.39

Does 3.2 means Interlagos freq is 3.2ghz? Is it a few too optimistic?

READ! He said Zambezi, not interlagos!
It says "half the number of cores", he means a desktop BD would have half of that 16 core server statement, which is 8 cores.
3.2 GHz, 8 cores, 32 nm isn't unrealistic, given that we have 3.2 GHz, 6 cores, 45 nm today.

Thanks DB for the BD calculation!!

[Florinmocanu]

Freq_ratio_of_half_#_of_Cores = 3.2/2.3 = 1.39

Does 3.2 means Interlagos freq is 3.2ghz? Is it a few too optimistic?

Interlagos or Zambezi have the same cores. He meant Zambezi, if you read the whole post, which will probably go for 3+ Ghz. 3.2 or 3.4 is quite a good assumption i guess.

[-Boris-]

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.

Well 30% boost from turbo sounds a bit high. That is comparable with 4.4GHz turbo on Phenom II. Even if BD will have better turbo I doubt it will be that much better. I think it's safer to say 73% boost + turbo instead. That would put BD at the same speed as an Phenom II at 4-4.2GHz and then turbo on top of that.

[Manicdan]

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.

my assumption is that the first model of the 8 core desktop version will require a 5ghz quad PII to match it in gaming.

BD with 4 threads running around 3.8ghz = PII quad at 5ghz

just an assumption.

[superrugal]

Interlagos or Zambezi have the same cores. He meant Zambezi, if you read the whole post, which will probably go for 3+ Ghz. 3.2 or 3.4 is quite a good assumption i guess.

Ouch is my fault.
If Zambezi with deeper pipeline only reach 3.2ghz, I will think this is a little underestimate. For me I guess it will reach 3.5-4Ghz.

But it seems that amd emphasized it compared them with same power envelope, which means freq is not the same?

[Mats]

Yeah, 3.2 GHZ is a bit low, and 30 % turbo seems a bit high to me. But we're comparing it with todays K10 which wasn't designed for six cores and turbo from the beginning, yet it still got it today.
BD's 32 nm and power saving features may be enough for that high turbo. On the other hand, Intel only goes 400 MHz turbo with their upcoming i7 2600.

[-Boris-]

Ouch is my fault.
If Zambezi with deeper pipeline only reach 3.2ghz, I will think this is a little underestimate. For me I guess it will reach 3.5-4Ghz.

Just to make some things clear.

Deeper Pipeline != Higher Frequency
Shorter Pipeline != Higher IPC

I have a feeling that people have read some P4 articles and made some conclusions of their own. To achieve high frequency a deep pipe can help, but there are reasons to put more stages in a processor other than frequency. And the other way around, more stages don't automatically lower IPC. Depending on the nature of the stages, more stages can actually improve IPC.

[freeloader]

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.

That would be worth the wait then.

[-Boris-]

Yeah, 3.2 GHZ is a bit low, and 30 % turbo seems a bit high to me. But we're comparing it with todays K10 which wasn't designed for six cores and turbo from the beginning, yet it still got it today.
BD's 32 nm and power saving features may be enough for that high turbo. On the other hand, Intel only goes 400 MHz turbo with their upcoming i7 2600.

And power usage isn't the only limitation to high frequency. Even if you have the headroom power wise you can't just clock higher. If that were the case it would be possible to clock an Intel Atom to insane performance.

So even if you give one module four times as high power headroom it will not necessary improve turbo capabilities much more than just 50% more headroom would.

[=SOC= Admiral]

I will never use Turbo as I am a folder and it only limits the OC that a cpu can hit. But as for me give me the best Hydra capable AM3+ mobo and the best Bulldozer. Thats all Im after and thanks to bulldozer AMD can say hello to BigAdvanced F@H because of the 8 cores.

[Manicdan]

why use hydra if its for folding?
you might have to wait a LONG time before hydra hits AM3+, or its gonna be a super expensive board with bells and whistles you may not care about

[Mechromancer]

@JF-AMD - Have had a chance to read the Arstechnica writeup on Bulldozer/Bobcat?

Fighting the last war

AMD always succeeds when it attacks Intel not where the latter is strong, but where it is weak. Historically, AMD's biggest wins have come when the company moved into an obvious hole in Intel's product line. For example, when Intel announced that EPIC and Itanium would be its 64-bit upgrade path, AMD countered with x86-64 and scored a huge victory in the server market. Or, when delays with the QuickPath Interconnect forced Intel to stick with its aging frontside bus architecture for way too long, AMD exploited its superior HyperTransport interconnect to pursue the multisocket server market. When Intel was pushing RAMBUS and, later, the power-hungry FB-DIMM, AMD stuck with cheaper DDR and gained a platform-level performance/watt advantage.

Right now, there are no obvious weak spots in Intel's conventional server platform; indeed, Intel's Xeon line is as strong as it has ever been. (Mobile is a different story, but that's a topic for later.) Insofar as Bulldozer is aimed at the server market, AMD is attacking Intel when and where the larger chipmaker is at its absolute strongest.

But notice that I said "conventional server platform" above. There is one obvious gap in Intel's current suite of datacenter offerings: Intel isn't directly pursuing low-power, high-density cloud servers, and this is a gap that both ARM and startups like SeaMicro are looking to fill with very dense server offerings based on mobile technologies (e.g., physicalization solutions).

If I ran AMD, I would redirect the company's effort toward building a low-cost, low-power, high-density, flash-based cloud server platform around Bobcat. Intel's Justin Rattner has admitted that for certain cloud workloads, these types of high-density solutions are superior to a monolithic server chip like Xeon. So AMD should stop obsessing over netbooks and monolithic server parts—both of these amount to fighting the last war—and just jump straight into the cloud server market that ARM is set to tackle with its upcoming Eagle part.

To do this would be to attack Intel where it is weak, because Intel's current answer to this is still in the labs. Intel will probably keep puttering away at its experimental Single Chip Cloud Computer, while pushing Xeon at cloud vendors and losing rack space to ARM-based systems. AMD could jump right in with something like Bobcat and be well-established as the go-to maker of high-density x86 servers before the SCCC makes it to market.

Will AMD take this advice? Probably not, and if it doesn't, Bulldozer better be very good.

Does AMD have plans to attack this new segment in the server market?

[Particle]

Just to make some things clear.

Deeper Pipeline != Higher Frequency
Shorter Pipeline != Higher IPC

I have a feeling that people have read some P4 articles and made some conclusions of their own. To achieve high frequency a deep pipe can help, but there are reasons to put more stages in a processor other than frequency. And the other way around, more stages don't automatically lower IPC. Depending on the nature of the stages, more stages can actually improve IPC.

Indeed. I think people are hung up on that era due to the latency incurred by a misprediction on a P4 with its very deep pipeline. This is a new era though, and there are two things going for us: modernized branch predictors which hit better and a shorter pipe than the P4. The P4s had a 20 or 31 stage pipeline depending on what era of P4 we're talking about. Bulldozer looks like what, 15? It's shorter right off and with better prediction. What each stage does is also going to be different, so let's hope that the stages were designed for the best case performance scenario and not just clock speeds.

Since it's hard to tell on forums, I just want to note to you: I'm not disagreeing with anything you said, so please don't feel compelled to defend yourself. I'm just expanding upon it.

[=SOC= Admiral]

why use hydra if its for folding?
you might have to wait a LONG time before hydra hits AM3+, or its gonna be a super expensive board with bells and whistles you may not care about

Im a gamer too and currently I can only use one gpu for gaming as I have amd but another limitation is that none of my cards are of the same generation. I have 3 cards. I will most likely be selling my 8800 and will be getting the CIVE when it launches in September. Hopefully I can get my 1055T to 4.0GHz on that board. Just a question where is the FSB located for AMD? Is it on the mobo or the cpu?

GTX470
GTX275
8800GTS 512