Rumour: Bulldozer 50% Faster than Core i7 and Phenom II.

[Nemesis 1]

well, they'll have 16 core server cpus so it shouldn't be too hard to send some of those to the retail side, especially if yields are decent and/or they can can charge $1000 for them like intel does on their EE chips.

Ya that could work but you have to keep in mind Intel ring bus allows it to add cores at will. Intel can easily do 10 cores as there talking about over at extreme . It would aldo seem that the 6-8 core units are coming in the 3rd quarter . I believe there is a thread here on that . Also the nore ores AMD has to use unit to compete cuts margins . O/T anyone know yet whos hacking the fotum . I spent half the day yesterday on remote with MSN . But they won't tell me what they found they said they would handle it.

 

[Tuna-Fish]

So why do we have almost the same size with two (2)K10. Why did AMD spend money and time to create the BD module when they could install two (2) K10 cores and have almost the same core size and 100% CMP?

Is this a serious question? Just look at all the sweet, sweet sram in the branch predictor unit. One BD core is not equal to one Stars core.

 

[w0ng3r]

it is quite easy to hack sli on a non-sli motherboard. all of the hardware connections are there, you just need a small software tweak.

I've read about that, but does the SLI performance suffer as a result?

 

[GlacierFreeze]

I've read about that, but does the SLI performance suffer as a result?

Why would it? All you're basically doing is unlocking a lock that Nvidia wants on there.

 

[itsmydamnation]

Ya that could work but you have to keep in mind Intel ring bus allows it to add cores at will. Intel can easily do 10 cores as there talking about over at extreme . It would aldo seem that the 6-8 core units are coming in the 3rd quarter . I believe there is a thread here on that . Also the nore ores AMD has to use unit to compete cuts margins . O/T anyone know yet whos hacking the fotum . I spent half the day yesterday on remote with MSN . But they won't tell me what they found they said they would handle it.

ringbus does not allow you to add all the cores you want at will, infact once you hit a point (low teens) ring buses become very hard to maintain performace. you see it in all ring based topologies, its why there are complex things like bufferfly, torus , etc topologies. Ring buses have advantages and disadvanages, for something small like the 4 core SB it makes good sense. Start pushing up past 10 cores and it will be interesting to see.

 

[Phynaz]

ringbus does not allow you to add all the cores you want at will, infact once you hit a point (low teens) ring buses become very hard to maintain performace.

Got a link?

 

[dorion]

I liken AMD shrinking K10 for Llano on 32nm as Intel shrinking Pentium4 for Cedar Mill on 65nm even though they had Core2Duo in the works at the same time.

It makes for a good backup plan should BD bomb for any reason. Should BD take off, like Core2Duo did for Intel, then Llano gets relegated to the low-end ASP markets just as Cedar Mill was.

I question the wisdom of diluting such a small resource pool of R&D dollars which AMD's management did in order to develop BD and Llano at the same time as supporting Bobcat development...but they know their business far better than I'll ever understand it in my lifetime so it must have made sense on a whole number of levels of risk management and cost-benefits analyses.

(or maybe not, Dirk was asked to walk the plank for some good reason...and the timing...)

I think Llano is already the backup, it had to happen to get Fusion going else AMD would be 3+ years behind Intels on die graphics.

 

[Martimus]

Got a link?

Every stop in the ring bus adds one cycle of latency. So it really depends on how many cycles of latency the chip is designed to deal with.

 

[Nemesis 1]

Yep the more stop the more latency .The biggest problem is the more stops power efficiency goes down. On the other hand the more stops = more bandwidth its a trade off. But 10 cores from what I am seeing from current SB is very doable

 

[bryanW1995]

I've read about that, but does the SLI performance suffer as a result?

no, all it does is disable the drm that is required for multi pci-e 16 slot boards. if anything, the computers with that crap on them might run a miniscule amount slower until you disable the drm I suppose.

 

[Ajay]

Ya that could work but you have to keep in mind Intel ring bus allows it to add cores at will. Intel can easily do 10 cores as there talking about over at extreme . It would aldo seem that the 6-8 core units are coming in the 3rd quarter .

Intel is doing 10 cores for their next gen Xeons.

 

[videoclone]

if BD is a success and its faster then not only SB, but the LGA2011 SB...
does intel have anything else to counter it with? Does that leave a 2-3 year window for AMD to hold the worlds fastest CPU until intel can pump billions into developing ( Copying AMD's modul design ) something faster

 

[Accord99]

if BD is a success and its faster then not only SB, but the LGA2011 SB...
does intel have anything else to counter it with? Does that leave a 2-3 year window for AMD to hold the worlds fastest CPU until intel can pump billions into developing ( Copying AMD's modul design ) something faster

Possibly, but given how Sandy Bridge is both the most powerful x86 core ever made as well as the most power efficient, it seems unlikely that Intel can't counter with higher clock speeds and core counts.

 

[JFAMD]

When you consider that 95%+ of the market is NOT buying the top speed part, how do you define successful?

I think too many people get wrapped up in arguing benchmarks on parts they will never buy.

 

[podspi]

When you consider that 95%+ of the market is NOT buying the top speed part, how do you define successful?

I think too many people get wrapped up in arguing benchmarks on parts they will never buy.

Personally, if BD can cover the $200~$400 dollar range competitively (I think the X6's were ok a year ago, but now they are just not competitive in their price range, and that's including the price of the platform) then it'll be a success.


If the BD 2 or 3 module can compete with i5's in performance and price, OEMs should jump on board, and AMD will (should) have a success.

Don't get me wrong, I'd love for the 4 module BD to clean house and go for over $1000, but if it doesn't that wouldn't make it a failure in any way.

 

[Mopetar]

I'd love for the 4 module BD to clean house and go for over $1000

I like the first part what you have to say, but not so much the second part

If they have the performance to command $1000 per chip, it will hopefully drive Intel's high end chips down in cost and make AMD come down a little bit as well.

 

[Mopetar]

When you consider that 95%+ of the market is NOT buying the top speed part, how do you define successful?

I think too many people get wrapped up in arguing benchmarks on parts they will never buy.

That is completely true, but this forum probably has a large number of members who fall into that 5% that want to buy the top speed part or at least upper shelf.

Of course it really depends on the price/performance difference as well. For a lot of people the 2500K has such incredible value that it makes something like that 990X seem silly. The extra performance doesn't justify the extra $800+ to get that chip.

 

[996GT2]

That is completely true, but this forum probably has a large number of members who fall into that 5% that want to buy the top speed part or at least upper shelf.

Of course it really depends on the price/performance difference as well. For a lot of people the 2500K has such incredible value that it makes something like that 990X seem silly. The extra performance doesn't justify the extra $800+ to get that chip.

BD doesn't have to be the absolute fastest CPU on the market for me to consider it successful. I'll consider BD a success if it can be competitive with a similarly priced SB processor; e.g. if a $225 Bulldozer-based CPU can meet the following criteria:

1) Perform on par with (or better than) a Core i5-2500k (in all types of apps, not just in heavily threaded ones)
2) Overclock to the same level as a 2500k (that is to say, about 40% OC over stock, since most 2500Ks have absolutely no trouble doing 4.6 on air cooling).
3) Have similar (or lower) power consumption at idle and at load.

 

[podspi]

I like the first part what you have to say, but not so much the second part

If they have the performance to command $1000 per chip, it will hopefully drive Intel's high end chips down in cost and make AMD come down a little bit as well.

If AMD actually releases a (client) chip that can command a $1000 price-point, that puts them in a fantastic position wrt Intel's response. There is no way Intel is going to push the current $1000 performance level down to a price-point where AMD has trouble making money.

Now, I realize Intel really likes margins, and so this may not be an issue anyway (I expect Ivy Bridge is going to be high-end for a while) but if AMD released a solid $200 dollar part, but that's it (4 module BD, as fast as yields allowed) -- they would be in serious trouble once Intel brought in Ivy Bridge and made good on their plans to sell mainstream quad-cores (current i3-level prices). I'd just like to see AMD in a competitive position where Intel's next response doesn't send them into the bargain-bin again...


Personally, I think the 4-module BD will go for ~ $400 to $500 (perhaps much more at release, but after the 6-core SBs launch I expect them to stabilize at around this price). 3-module BD will compete with the i7's we know and love, and the 2-module will compete with the i5s.

ATM, Llano is AMD's response to the i3 and probably the real money-maker (or not). We'll see how OEMs react to it.

 

[Ajay]

if BD is a success and its faster then not only SB, but the LGA2011 SB...
does intel have anything else to counter it with? Does that leave a 2-3 year window for AMD to hold the worlds fastest CPU until intel can pump billions into developing ( Copying AMD's modul design ) something faster

With LGA2011 CPUs being 22nm, I doubt BD will retain it's lead if in fact it is faster than SB.

 

[PreferLinux]

The first LGA2011 CPUs will be 32 nm.

 

[drizek]

I thought 2011 were still 32nm, and the IB chips are going to stay 1155, right?

 

[drizek]

Personally, I think the 4-module BD will go for ~ $400 to $500 (perhaps much more at release, but after the 6-core SBs launch I expect them to stabilize at around this price).

That's insane. The 2500K is $180.

There is no way that BD will be fast enough to justify that price.

 

[Anexate]

BD doesn't have to be the absolute fastest CPU on the market for me to consider it successful. I'll consider BD a success :[/U]

1) Perform on par with (or better than) a Core i5-2500k (in all types of apps, not just in heavily threaded ones)
2) Overclock to the same level as a 2500k (that is to say, about 40% OC over stock, since most 2500Ks have absolutely no trouble doing 4.6 on air cooling).

I would add:
-motherboards having both SLI as Crossfire. I'm not going to buy a CPU that ties the customer to only one GPU brand; besides being this an unfair market practice (immagine intel supporting only Nvidia GPUs)

 

[Castiel]

I thought 2011 were still 32nm, and the IB chips are going to stay 1155, right?

That is correct.