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

[itsmydamnation]

They come from the same die.

So, workloads are very different, you can have far more variation in power and heat output in the consumer space, which means more aggressive clocking + turbos. You also have things like registered ECC ram tends to be of lower speed then whats available in the consumer space. There are blogs and post by JF-AMD saying that they originally didn't intend to release the 8 core part for servers only 12 and 16. The only reason an 8 core server part is if clocks and power usage scale well.

 

[Mopetar]

They come from the same die.

The likely do and there are a lot of similarities, but there's still a distinction. The high-end desktop chips are just server chips that didn't pass the quality control tests. The level of expected quality for high-end desktop chips is lower than what's expected for server-grade chips, so Intel can recoup some losses by disabling a few features and selling them as enthusiest consumer parts.

The Gulftown chips are almost identical to their server counterparts. The biggest difference is that the Xeons have 2x QPI rather than only 1x. The 4 core Xeons have additional L3 cache compared to the high end desktop i7's that use socket 1366. There's probably a few other things disabled as well, but the high end Intel desktop processes are essentially Xeons.

There're going to be both consumer and enterprise grade BD CPUs. What we don't know is where the line is blurred between the two. It's fairly obvious that any 2 module BD chips were probably not intended for server use. 4 module BD chips may have originally been intended for server use. The consumer chips are going to come out first so that AMD has some additional time to work on yield issues or iron out any remaining problems. The best chips from the first runs will likely be saved up for the server part release, but the ones the fail quality control tests will have features disabled and be sold as high-end consumer parts.

 

[videoclone]

Anyone know when this thing will be coming out? I'm hoping it turns out to be the king of CPU's

 

[JFAMD]

Q2 for client, Q3 for server.

 

[itsmydamnation]

Q2 for client, Q3 for server.

can you do an NV for us and give us a release date for the release date of the products release?

 

[ShadowVVL]

I hope we get some good looking boards and some in red or green or maybe even a white and black board, blues not so fun anymore.

 

[drizek]

I would love an off-white board with dark grey lettering.

 

[videoclone]

i would like a blond, blue eyes, big rack, small backside,, wait... what are we talking about? give me a black board please i dont like bling... stealth for the win!

 

[HW2050Plus]

Some additional official information of AMD including some performance statements:

http://www.eetimes.com/electronics-...na-eyes-petaflops-IBM-hits-5-GHz?pageNumber=2

Interesting part of this is the 90% performance of previous cores. If you add the doubling of cores you come to 180% and this 80% gain from CMT.

Clocks are stated to be e.g. 3.5 GHz for same TDP envelope.

Still very little information and AMD is very tight lipped.

A 80% performance increase over current CPUs is now (again?) confirmed by AMD. However this is regarding integer only.

What makes this ISSCC statement really interesting is the following:
With Bulldozer we have a 80% increase of performance by the doubling of cores. But this assumes 10% slower cores. As JFAMD stated that single core performance increases this could mean that we could see an even larger performance increase in actual parts.

Again the 30.9 mm² Module size incl. 2 MB L2 Cache I stated earlier in this thread has been confirmed. That is roughly the same of a Sandy Bridge Core with HT, however the larger L2 increases the overall size.

 

[JFAMD]

The 80% number is not a comparison to current cores.

 

[AtenRa]

Any links from the actual isscc presentations ??

 

[Idontcare]

The 80% number is not a comparison to current cores.

Is it safe to say you have rued the day you ever uttered that number?

I've prolly seen you explain it, deftly and consistently, close to a hundred times now across various forums...and it seems like no matter how clearly you spell it out, nor the frequency with which you repeat the message, there is an endless pool of individuals out there in the world who are seemingly determined to totally misunderstand what the heck you were originally trying to communicate when you started the conversation about "80%" all those many months ago.

If you haven't done it already, you could probably devote an entire blog entry to "Mythbusting Bulldozer".

Hey I heard that you said the IPC is going to be lower AND it gets its low TDP rating because it is powered by the blood of slaughtered baby lambs instead of electricity from the wall. Its true! Confirmed by AMD!

 

[tincart]

Hey I heard that you said the IPC is going to be lower AND it gets its low TDP rating because it is powered by the blood of slaughtered baby lambs instead of electricity from the wall. Its true! Confirmed by AMD!

Really? That's horrible! I am totally not buying this product when it comes out.

 

[wuliheron]

I heard they were breeding microscopic gerbils in squirrel cages and feeding them methamphetamine to power the thing and the whole blood of innocent lambs thing is disinformation from Intel.

 

[podspi]

Pretty sure that article mixed up Bulldozer and Bobcat (there is another presentation on Bobcat). Bobcat has 90% of the performance of previous cores in a smaller area...

 

[JFAMD]

We get higher IPC via the flux capacitor.

 

[bryanW1995]

Is it safe to say you have rued the day you ever uttered that number?

I've prolly seen you explain it, deftly and consistently, close to a hundred times now across various forums...and it seems like no matter how clearly you spell it out, nor the frequency with which you repeat the message, there is an endless pool of individuals out there in the world who are seemingly determined to totally misunderstand what the heck you were originally trying to communicate when you started the conversation about "80%" all those many months ago.

If you haven't done it already, you could probably devote an entire blog entry to "Mythbusting Bulldozer".

Hey I heard that you said the IPC is going to be lower AND it gets its low TDP rating because it is powered by the blood of slaughtered baby lambs instead of electricity from the wall. Its true! Confirmed by AMD!

naw, they use baby seals. they're more convenient, they just reopened the old ati offices in canada and took a short road trip to get the poor animals. I actually personally saw verifiably with my own eyes a picture of dirk meyer clubbing one of the seals, that's the REAL reason he's gone.

edit: FLUX CAPACITOR!! of course! that explains the 80% number, what you really said was "These new chips are great, but they really explode once you get them to 88 mph". I have the quote lying around here somewhere...

 

[Scotteq]

We get higher IPC via the flux capacitor.

<smile> at one point twenty one gigawatts? <smile>

 

[Scotteq]

naw, they use baby seals. they're more convenient, they just reopened the old ati offices in canada and took a short road trip to get the poor animals. I actually personally saw verifiably with my own eyes a picture of dirk meyer clubbing one of the seals, that's the REAL reason he's gone.

I thought Intel were the ones clubbing baby seals?

AMD's IPC cures cancer.

 

[HW2050Plus]

The 80% number is not a comparison to current cores.

Yes it is what you get from CMT. So we do not have 80% more from previous cores but only 62% (10% slower cores with 80% gain from CMT). Thanks to point this out. Actually it could be even lower because of scaling issues.

Therefore from ISSCC in combination with your CMT statement: AMD Bulldozer should be (design target) 62% faster than current generation of AMD CPUs using twice the core count.

Have a look at the ISSCC statement from AMD's engineer Michael Golden:

The unit helps the core meet its target of delivering 90 percent of performance of past AMD cores with a significant reduction in area and power, said Michael Golden, another AMD engineer.

Really quite interesting:
Michael Golden: Bulldozer core 90% of performance of previous cores
John Fruehe: Bulldozer core more performance than previous cores

So it is possible, that both are correct as Michael Golden states it was a design goal to reach 90% performance of previous cores while you say that the Bulldozer is actually faster than previous cores and has a higher IPC than previous cores.

But another explanation could be that Michael Golden is correct because he refers to integer and your statement maybe needs to include FPU, or needs to include the fact that a single core has an advantage if the other core in the module does not run in a thread or it just applies only to some artifical code segment where the Bulldozer can play out is 4-wide execution core, though for more practical code it will only be 2-wide.

Therefore while the ISSCC statement of Michael Golden is quite clear your's is not because we do not know the conditions you meant - but did not disclose - when the Bulldozer core will be faster and/or has higher IPC (e.g. Integer?/Other core idle?/ special code sequence?/ etc.).

Very much open questions.

Anyway the ISSCC disclosures brought some information and they are from engineers no private talk from marketing guys.

From ISSCC we know the following:
Core performance drops by 10%. Overall performance increases by ~60%.
All in regard to current AMD cores.

 

[hamunaptra]

Well AMD is in for a world of hurt or at least their customers are if the BD core is only 90&#37; as fast as current stars core is at the same clock. Because to mean anything on the desktop front AMD will effectively need to double the clock speed compared to current phenom II's to give any sort of performance close to what sandy bridge / ivy bridge, does / will.
And I just dont see 5-6ghz happening with BD LOL!
Or, if AMD is able to scale up to turbo an additional 1ghz on a single core to achieve high performance with single threaded or low thread count apps, this may be able to balance the performance offset as well.
We know 500mhz turbo is concrete for when more than one core being used as long as its under the TDP and we know that it may be able to scale further with even less cores used as long as its under TDP, the highest turbo clock...we dont know officially.

One would think with all the R&D put into optimizing the BD's front end and all the changes theyve made to it, that they would at least be able to boost peformance enough to make up for the lack of the quite useless 3rd INT pipe the stars core had. But, it would seem thats not the case =(


SAD AMD, VERY SAD you couldnt get better performance per core per clock than stars...ugh, why keep dissapointing? You are going backwards.

 

[busydude]

Really quite interesting: Michael Golden: Bulldozer core 90&#37; of performance of previous cores John Fruehe: Bulldozer core more performance than previous cores

I think you got it mixed up.. that quote was in reference to the Bobcat cores.

EDIT: I think the author got mixed up!!

 

[Anexate]

I can see we are reaching the moment before eruption of the volcano here; if AMD does not forward at least some data at Cebit....:twisted:

 

[Edrick]

AMD are marketing gods!

They were able to create a vaporware product that has been talked about, discussed, disected, argued, etc, all over the internet for just about 3 years now. Now that is the most impressive thing about BD.

 

[podspi]

It's pretty clear the author got Bobcat and Bulldozer mixed up, especially because they discussed BOTH CPUs. AMD has said many times that Bobcat was supposed to reach 90&#37; of current cores performance with a much smaller die and less power. It makes no sense for Bulldozer to be slower...