AMD sheds light on Bulldozer, Bobcat, desktop, laptop plans

[JFAMD]

That should be "1.8 times the performance of what two independent cores would have"...

The actual statement is two integer cores in a module would have 1.8 times the throughput of a single core in a module.

That is often repeated and rarely repeated correctely.

 

[JFAMD]

Are these the current TDPs of the six-core Istanbul opterons? If so, and those are the same TDP targets for their 16-core Interlagos, that's just amazing. Then again, that says nothing about performance, but I'm assuming they wouldn't be putting out too little performance - after all. this is still more than a year away, they probably wouldn't be clocking their 16-core monster at <1Ghz just to remain @ 55w.


Wouldn't that depend on how many cores they will offer and the clockspeed? Since we don't know both, we probably can't tell. That's almost two years away, so hard to tell. Interlagos will come out first, then speculations about desktop Bulldozer will proably abound or be leaked or previewed, since AMD releases for the server space first.

Those are the current TDPs. Magny Cours TDPs will be similar, but not exactly the same.

I think you'll be surprised by what we will be able to do with 12 cores in MC and keep in mind that MC will have the same platform (Maranello) as you will see with Bulldozer.

 

[JFAMD]

When AMD mentions the 2.1 GHz 8 core Bulldozer being 60-80% faster than Thuban are they are talking about the 2.8 Ghz hexcore?

Where was that ever said?

 

[Idontcare]

The actual statement is two integer cores in a module would have 1.8 times the throughput of a single core in a module.

That is often repeated and rarely repeated correctely.

Thats what was going thru my mind at the time, the way the register stated it they made it sound like single-threaded performance would be awesome in comparison to a single core (because they made the multi-threaded performance gain of 1.8x a comparison to a single core)...but my efforts to translate into multi-core numbers went astray.

But c'mon there's no such thing as bad publicity, right?

 

[jvroig]

Where was that ever said?

From Johan's article about AMD's roadmap:

We will discuss this core in more detail but here are some extra tidbits we managed to find out:

&#8226; Two integer clusters share fetch and decode logic but have their own dedicated Instruction and Data cache
&#8226; Integer clusters can not be shared between threads: integer cores act like a Chip Multi Processing (CMP) CPU.
&#8226; The extra integer core (schedulers, D-cache and pipelines) adds only 5&#37; die space
&#8226; L1-caches are similar to Barcelona/Shanghai (64 KB 2-way? Not confirmed)
&#8226; Up to 4 modules share a L3-cache and Northbridge
&#8226; Two times 4 Bulldozer modules (2 x 8 "cores" or 16 cores) are about 60 to 80% faster than the twelve core Opteron 6100 CPU in SPECInt_rate.
​

So not exactly "8 Bulldozer cores are 60-80% faster than 6 Thuban cores", but "16 Bulldozer cores are 60-80% faster than 12 Opteron 6100".

Computer Bottleneck must have mixed it up, but I am of the impression that the Opteron 6100 has similar performance profile to desktop Phenom II chips.

 

[Idontcare]

From Johan's article about AMD's roadmap:

So not exactly "8 Bulldozer cores are 60-80% faster than 6 Thuban cores", but "16 Bulldozer cores are 60-80% faster than 12 Opteron 6100".

Computer Bottleneck must have mixed it up, but I am of the impression that the Opteron 6100 has similar performance profile to desktop Phenom II chips.

I imagine John is more interested in where the specific clockspeed number for Bulldozer came from in CB's post.

 

[jvroig]

I imagine John is more interested in where the specific clockspeed number for Bulldozer came from in CB's post.

Ah, I suppose you are right. I simply discarded the clockspeed numbers in my head since they were similar (2.8 vs 2.8) and assumed that's how one would compare processors anyway to measure performance jump from old to new architecture (further assuming both clockspeeds exist for the old and new chips).

 

[cbn]

Where was that ever said?

http://forums.anandtech.com/showpost.php?p=28988386&postcount=284

How fast is a Opteron 6100?

I am also confused because during an internet search I was seeing MC in reference to Opteron 6100 (not Instanbul).

 

[Idontcare]

When AMD mentions the 2.1 GHz 8 core Bulldozer being 60-80% faster than Thuban are they are talking about the 2.8 Ghz hexcore?

http://forums.anandtech.com/showpost.php?p=28988386&postcount=284

How fast is a Opteron 6100?

I am also confused because during an internet search I was seeing MC in reference to Opteron 6100 (not Instanbul).

I don't think anyone is questioning your consignment of 2.8GHz to Istanbul or Magny-Cours (whichever it is that you ultimately decide you had meant) but rather the we are curious when and where AMD said anything about Bulldozer being 2.1GHz for any SKU.

 

[cbn]

I don't think anyone is questioning your consignment of 2.8GHz to Istanbul or Magny-Cours (whichever it is that you ultimately decide you had meant) but rather the we are curious when and where AMD said anything about Bulldozer being 2.1GHz for any SKU.

Magny-Cours is double a Thuban/San Paolo right? It's clockspeed is 2.1 Ghz.

AMD is saying 16 core Interlagos (eight dual module bulldozers) will be 60-80&#37; faster than MC.

So unless we know Interlagos clockspeed @ x TDP it is impossible to know the improvement in IPC.

(Sorry for the confusion. I am clueless on AMD server nomenclature)

 

[Idontcare]

So unless we know Interlagos clockspeed @ x TDP it is impossible to know the improvement in IPC.

Yes, what you are noticing is the fact they (AMD marketing/PR) have been intentionally selective on the info they have provided thus far so that we can't make such conclusions or expectations when it comes to clockspeeds, TDP, or IPC.

It is not by accident that the needed info is seemingly absent from the public domain.

edit: besides, with SKU's ranging from 1.7GHz to 2.3GHz, who is to say the performance comparison of Interlagos isn't being made to the 1.7GHz version of Magny-Cours? Lots of intentionally built-in wiggle room for these guys with their performance estimates.

http://www.xbitlabs.com/news/cpu/di..._Server_Microprocessors_in_March_Sources.html

 

[cbn]

Yes, what you are noticing is the fact they (AMD marketing/PR) have been intentionally selective on the info they have provided thus far so that we can't make such conclusions or expectations when it comes to clockspeeds, TDP, or IPC.

It is not by accident that the needed info is seemingly absent from the public domain.

The problem was I thought Magny Cours was based on Bulldozer (rather than Istabul). That is why I thought the Bulldozer clockspeed would be 2.1 Ghz.

 

[cbn]

c domain.

edit: besides, with SKU's ranging from 1.7GHz to 2.3GHz, who is to say the performance comparison of Interlagos isn't being made to the 1.7GHz version of Magny-Cours? Lots of intentionally built-in wiggle room for these guys with their performance estimates.

http://www.xbitlabs.com/news/cpu/di..._Server_Microprocessors_in_March_Sources.html

http://www.channelregister.co.uk/2009/12/14/amd_bulldozer_preview/page2.html

It could be very interesting to see how the desktop CPUs come out when according this article ( I linked earlier in the thread) Interlagos will come in 55 watt to 105 watt TDPs. This sounds like Interlagos will be a much more efficient design than Istanbul.

With Desktop Bulldozers sporting fewer modules (than server chips) AMD will have more TDP headroom for higher clocks on the mainstream consumer sockets. The rumors about an AMD Turbo mode gives me hope these modules will scale well with additional voltage.

 

[jvroig]

It could be very interesting to see how the desktop CPUs come out when according this article ( I linked earlier in the thread) Interlagos will come in 55 watt to 105 watt TDPs. This sounds like Interlagos will be a much more efficient design than Istanbul.

With a two-year gap between their releases? Of course the new one would be more efficient. Limiting it to just 105W (instead of 125 - 140; I am assuming the Opterons sport the same max TDP as the Athlons/Phenoms for some models) is interesting though. Certainly a move to make data centers love them, and if this efficiency and (assumed great) performance carries over well to the desktop space, a clear win for us, too.

The rumors about an AMD Turbo mode gives me hope these modules will scale well with additional voltage.

Didn't hear/read those rumors, but that would be a logical "necessity" for AMD to remain competitive with Intel's processors. It goes a long way to mitigate the "single-threaded vs multi-threaded" concerns people have. More cores are available for multi-threaded work, but a more powerful single core is also made available at times when a single core is all that is needed.

 

[JFAMD]

The problem was I thought Magny Cours was based on Bulldozer (rather than Istabul). That is why I thought the Bulldozer clockspeed would be 2.1 Ghz.

Not sure where you got the 2.1 though. We have never commented on clock speed.

Also, I would be VERY careful in trying to assign desktop performance attributes based on server performance.

Server and workstation use different workloads, so comparing performance attributes across the two may not be accurate. We focus on throughput, they will probaly look more at peak.