AMD's Bobcat and Bulldozer

[cegras]

It's not fair to compare the 2 products Intel and AMD will be offering to the desktop enthusiast market segment because the Intel part supports twice the number of threads?

I'm sure the Intel parts *will* cost more, but they'll perform better, and in that, help justify the price differential.

We'll see about that when the benchmarks are out. In particular, AMD, although failures in terms of absolute performance, positioned their products nicely to offer competitive, alternative solutions to intel.

[JF-AMD]

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



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

Funny, we talk to a lot of cloud customers and the 6000 is just as interesting as the 4000 to them.

Here's the problem with ARM. It's really low power, but it is also really low performance.

So, while you can get really low power, you end up adding more and more servers. Suddenly, physical infrastructure becomes the problem.

Wouldn't you rather have a 35W (TDP) 6-core in a 2P than 12 individual 1P ARM servers?

The real issue is that ARM does not support 64-bit and it is limited to small micro servers. That becomes a management nightmare.

Some time, in the future, I can tell you a really funny story about this.

[JF-AMD]

There would be no point for anyone to choose Zambezi over octal Sandy, except price, possibly.

Also, JF admitted elsewhere that the 50% / 33% claim for Interlagos / MC has been substantially juiced by including, in the aggregate used to measure, serial workloads that benefit from the Turbo that MC lacked. While adding Turbo is a good thing, this means the "fully-parallel" throughput improvement is considerably less than 50%. In order to make sense of the claim, you're going to need to see exactly what they've chosen to average over, and at that point, you'll likely have access to simpler zambezi benchmarks of all sorts.

Actually, you are spreading FUD, so I will not reply to this.

[terrace215]

Actually, you are spreading FUD, so I will not reply to this.

Thankfully, your exchange remains:

Turbo (improved) in a server chip. Yeah!!!

That feature was totally unexpected and is a welcome one. I wonder if it accounts in anyway in the 50% improvement in performance. I hope not!!!!

F.
Last edited by franzius; 08-23-2010 at 11:02 PM. Reason: just clarifying my thoughts, man

Obviously it is part of the equation, but definitely not all of it.

http://www.semiaccurate.com/forums/s...75&postcount=4

If that is FUD, well, it came from you...

[qcmadness]

Funny, we talk to a lot of cloud customers and the 6000 is just as interesting as the 4000 to them.

Here's the problem with ARM. It's really low power, but it is also really low performance.

So, while you can get really low power, you end up adding more and more servers. Suddenly, physical infrastructure becomes the problem.

Wouldn't you rather have a 35W (TDP) 6-core in a 2P than 12 individual 1P ARM servers?

The real issue is that ARM does not support 64-bit and it is limited to small micro servers. That becomes a management nightmare.

Some time, in the future, I can tell you a really funny story about this.

When will AMD go for 12-core and 16-core monolithic chips?

[Chaserjzx100]

wow this thread has been derailed for some time now... apologist really brings hilarity in this thread haha

@JF
will it be possible or feasible if a company says that they would want a bobcat solution for there server/cloud needs?

[ajaidev]

Ok this is how i see the sandy bridge vs Bulldozer thing:

Sandybridge - Core + SMT "2T"
Bulldozer - Module "2T"

Technically i find bulldozer quite a refresh and very well made. But i had access to sandy bridge some time ago and well its no slouch either. The main feature in sandy bridge is the flexibility which i have been say's for months, relatively speaking bulldozer has less flexibility "Tough i have to see proper documentation before being 100% sure but it seems so"

Now if you look at how the module works with its power requirements is the reason why i say its equal to a single core +SMT of sandy bridge this is not in respect to performance but in design. The module implements Turbo, power gates, etc with respect to a single module and not a single core whereas the SNB implements it on core level.

I know the die size of SNB per core "+L2" but dont know about the die size of bulldozer's modules "+L2", i hope its close because SNB die size is not huge and it may be possible for Intel to make even higher count chips "This is the flexibility i was talking about". Now if we take the example of the 8c/16t sandy bridge vs 8m/16t bulldozer it is apparent to realize that all 16t of the bulldozer need to be fed for good gains but in situations where the thread count is <=8t the sandy bridge has a very good variable's for better performance.

Turbo wise i am very interested in knowing how much the bulldozer can do "its a module based service means the whole module will be oced not just a single core" the sandybridge is very good in turbo you people can expect some chips t come very near to the 4ghz mark "with turbo". Also you will be interested to know that intel's turbo implemented can be further increased with minor increase in TDP. If the turbo was core wise i do think AMD would win because most likely the AMD cores are smaller than the SNB core.

[Oliverda]

Now another question.

Why AMD did not go after 6-module Bulldozer CPUs?

AFAIK, the size of a Bulldozer module @ 32nm should be similar or smaller than a K10 core @ 45nm.
It would not be a difficult task to integrate 6 modules into a CPU die with ~300mm^2 size.

I'm fairly sure that we will see 6 modules versions still at 32nm.

[JF-AMD]

When will AMD go for 12-core and 16-core monolithic chips?

Probably not any time in the near future.

[JF-AMD]

wow this thread has been derailed for some time now... apologist really brings hilarity in this thread haha

@JF
will it be possible or feasible if a company says that they would want a bobcat solution for there server/cloud needs?

I have talked to all of the major OEMs about their cloud needs and Bobcat is "interesting" in the "I don't want to disregard any potential technology" way. But nobody is seriously making plans around it.

The problem is that with low power, those solutions bring along a lot more servers. So you lose power, but then you have to add proportionally more servers and managing the physical assets becomes more of a problem than power consumption.

[cegras]

Thankfully, your exchange remains:






http://www.semiaccurate.com/forums/s...75&postcount=4

If that is FUD, well, it came from you...

How is this different from Intel?

[informal]

AMD confirms the AM3+ as a necessity for a Zambezi product:
http://www.planet3dnow.de/cgi-bin/ne...&id=1282840508

"The existing G34 and C32 server infrastructure will support the new Bulldozer-based server products. In order for AMD’s desktop offering to fully leverage the capabilities of Bulldozer, an enhanced AM3+ socket will be introduced that supports Bulldozer and is backward-compatible with our existing AM3 CPU offerings."


"When we initially set out on the path to Bulldozer we were hoping for AM3 compatibility, but further along the process we realized that we had a choice to make based on some of the features that we wanted to bring with Bulldozer. We could either provide AM3 support and lose some of the capabilities of the new Bulldozer architecture or, we could choose the AM3+ socket which would allow the Bulldozer-base Zambezi to have greater performance and capability.

The majority of the computer buying public will not upgrade their processors, but enthusiasts do. When we did the analysis it was clear that the customers who were most likely to upgrade an AM3 motherboard to a Bulldozer would want the features and capability that would only be delivered in the new AM3+ sockets. A classic Catch-22.

Why not do both you ask? Just make a second model that only works in AM3? First, because that would greatly increase the cost and infrastructure of bringing the product to market, which would drive up the cost of the product (for both AMD and its partners). Secondly, adding an additional product would double the time involved in many of the development steps.

So in the end, delivering an AM3 capability would bring you a less featured product that was more expensive and later to market. Instead we chose the path of the AM3+ socket, which is a path that we hope will bring you a better priced product, with greater performance and more features - on time.

When we looked at the market for AM3 upgrades, it was clear that the folks most interested in an AM3-based product were the enthusiasts. This is one set of customers that we know are not willing to settle for second best when it comes to performance, so we definitely needed to ensure that our new architecture would meet their demanding needs, for both high performance and overclockability. We believe they will see that in AM3+."

It looks like they had a difficult choice to make and they went with more performance which is ,in the end,a right thing to do.Present AM3 system owners can use their AM3 CPUs in the new boards so not everything is broken compatibility wise. Use the old chip with the new board when AMD launches it and then just slide in the new 8 core Bulldozer when it comes next year.Not the perfect solution for present owners ,but if it is for performance reasons then it's understandable .

[Mats]

I know I've said this before, it's just that it makes me so mad...

It looks like they had a difficult choice to make and they went with more performance which is ,in the end,a right thing to do.

Yeah but this doesn't explain why they didn't launch AM3+ with this years 870/880G/890GX/890FX chipsets.
This would have been the only way to make some decent use of the remaining compatibility: You buy an AM3+ board for your AM3 CPU in 2010, and will be able to upgrade to BD in 2011.

I don't even own an AM3 board, I just feel sorry for all people who bought a new board this year and thought they would be able to use it with BD.

You need a new board for SB, but you also need a new board for BD which also happens to show up later. Tough choice.

[-Boris-]

I know I've said this before, it's just that it makes me so mad...

Yeah but this doesn't explain why they didn't launch AM3+ with this years 870/880G/890GX/890FX chipsets.
This would have been the only way to make some decent use of the remaining compatibility: You buy an AM3+ board for your AM3 CPU in 2010, and will be able to upgrade to BD in 2011.

I don't even own an AM3 board, I just feel sorry for all people who bought a new board this year and thought they would be able to use it with BD.

You need a new board for SB, but you also need a new board for BD which also happens to show up later. Tough choice.

This decision would have to be done a long time ago if that would be a possibility. I think it's a question of time.

[Mats]

This decision would have to be done a long time ago if that would be a possibility. I think it's a question of time.

Naturally, but the G34 was launched the very same month, which will work with BD..

[madcho]

Naturally, but the G34 was launched the very same month, which will work with BD..

check wiki ;

G34 was launched on March 29, 2010

AM3 was launched on February 9, 2009

G34 was thinked when BD was more advanced in work, so their work included the need for BD i think.

[ajaidev]

What pisses me is that the 8 series are not that great from the 7 series as it is so why launch a minor upgrade from the 7 series if you are not supporting the next chip that will come in months ?

What AMD could have done was to make a single cip compatible with both sockets "AM3 and AM3+" but disable the special abilities when use with the AM3 just like the case of AM2+ and AM2 ehh.

This move by AMD means that anyone who thinks of getting a bulldozer will also take a look at high end sandy bridge because he will have to buy a mobo and a cpu in any case.

EDIT: Forgot to add that the G34/c32 is suppose to get the added abilities also they were launched before AM3 and this means that AMD could have done something to retro fit the AM3 hell they could have delayed the AM3 and released a AM3+ type version i dont think most people would have cared.

[Mats]

G34 was thinked when BD was more advanced in work, so their work included the need for BD i think.

Come on, READ MY POST!
I said, AMD could've launched AM3+ with this years 870/880G/890GX/890FX boards, they showed up the same month as G34.
http://www.anandtech.com/show/2952

I'm not talking about the AM3 launch, you're making things up.

[superrugal]

AMD confirms the AM3+ as a necessity for a Zambezi product:
http://www.planet3dnow.de/cgi-bin/ne...&id=1282840508



It looks like they had a difficult choice to make and they went with more performance which is ,in the end,a right thing to do.Present AM3 system owners can use their AM3 CPUs in the new boards so not everything is broken compatibility wise. Use the old chip with the new board when AMD launches it and then just slide in the new 8 core Bulldozer when it comes next year.Not the perfect solution for present owners ,but if it is for performance reasons then it's understandable .

Is this an official confirmation or what? There is not AMD official link on that page.

Yeah but this doesn't explain why they didn't launch AM3+ with this years 870/880G/890GX/890FX chipsets.

Perhap they didn't want to unveiling any secret of bulldozer through the chipset & bios.

(I remember that there is an ASUS AM3 motherboard bios update several months before, inside the bios(source code or something else)there's a suspicious 8-core-cpu supply information.

[-Boris-]

Naturally, but the G34 was launched the very same month, which will work with BD..

Triple channel and PCIe added perhaps?

[madcho]

Triple channel and PCIe added perhaps?

Just a stupid question, HOW old AM3+ PGA-ZIF will be able to plug old dual channel CPUs ?

Extra pins ? i don't believe so.

[Mats]

Perhap they didn't want to unveiling any secret of bulldozer through the chipset & bios.

(I remember that there is an ASUS AM3 motherboard bios update several months before, inside the bios(source code or something else)there's a suspicious 8-core-cpu supply information.

It's possible, but in this situation they will lose customers to Intel instead, since now you need a new board no matter which CPU you chose. People expected that it would work with AM3, and now they're pissed off.
Besides, compatible BIOS wouldn't need to show up until next year, and they would still have this situation if the original AM3 would have been used with BD.

It's the worst of both worlds:
- BD doesn't work in AM3 boards, because they made a new socket, but
- It's still not a new socket, no LGA, no PCIe, no added memory channels, nothing. I'm not saying that AMD would add all these features in BD from the start, but they could've added the pins for future use.
AM3+ is mechanically the same as the seven year old original K8/Sledgehammer/Opteron/Athlon FX-51 socket 940, you can't fit any of those features into it electrically, you need more pins.

Now we're stuck with yet another AM iteration, which obviously will need a replacement quite soon anyway, because it will be outdated before it gets launched.

[-Boris-]

Just a stupid question, HOW old AM3+ PGA-ZIF will be able to plug old dual channel CPUs ?

Extra pins ? i don't believe so.

An AM3 won't need to use all available pinholes in AM3+.
Hope you don't take my comment to seriously.

[Manicdan]

as a platform AMD has done an incredible job. simplistic upgrade paths, and low cost boards with good quality.
when i bought my 790FX board 3 years ago, i had NO idea i was going to put a 4ghz hex into it (i was expecting just a 3.6ghz quad)

and i honestly forget what the 800series boards brought anyway, just some USB3 and SATA6gbs in a few higher end models?

also there are almost always combo sales for buying a cpu+mobo on newegg. im expecting you can have one you like for 30-40$ less, if you wait for the cpus to come out before upgrading to AM3+

if they want to make us really happy, tell us what AM3+ will offer that we couldnt get otherwise

and is it possible to build a socket that has more pins, but the same pin layout to be compatible with older?

[blindbox]

It's possible, but in this situation they will lose customers to Intel instead, since now you need a new board no matter which CPU you chose. People expected that it would work with AM3, and now they're pissed off.
Besides, compatible BIOS wouldn't need to show up until next year, and they would still have this situation if the original AM3 would have been used with BD.

It's the worst of both worlds:
- BD doesn't work in AM3 boards, because they made a new socket, but
- It's still not a new socket, no LGA, no PCIe, no added memory channels, nothing. I'm not saying that AMD would add all these features in BD from the start, but they could've added the pins for future use.
AM3+ is mechanically the same as the seven year old original K8/Sledgehammer/Opteron/Athlon FX-51 socket 940, you can't fit any of those features into it electrically, you need more pins.

Now we're stuck with yet another AM iteration, which obviously will need a replacement quite soon anyway, because it will be outdated before it gets launched.

Still not a new socket? Well they've already said they had to do AM3+ and make bulldozer incompatible with AM3 to ensure a performance boost(EDIT: More performance boost compared to sticking to AM3, I mean). No LGA, no PCI-E, is that something to complain? Do tell me if there's a performance difference beyond the margin of errors in putting the PCI-E lanes on the chipset or on the processor. Memory channels, yes, that's only one thing though.

Replacement probably comes when the next generation comes. Heck, do we know if AM3+ is the same socket as AM3? We know that AM2+ and AM2 are the same sockets. Nothing that's stopping them from adding a wee bit more pins around the corners here and there and still maintain compatibility like how AM3 processors can fit into AM2+.

At least AM3 can fit into AM3+. Kinda like AM2+ and AM3, only backwards lol, though people prefer to change processors rather than mobo.