AMD Thuban (6 core desktop) for Q2

[Voo]

I have some problems with understanding why larger cores, but fewer cores should translate to much higher frequences (you can get a bit higher because you have a bigger thermal headroom, but not that much)

I'm not that fit with the whole analogue electronic stuff, but shouldn't bigger cores make a lot of problems, like more capacitance, clock skew, etc.? Also larger, more complicated logic tends to be slower.

 

[cbn]

I have some problems with understanding why larger cores, but fewer cores should translate to much higher frequences (you can get a bit higher because you have a bigger thermal headroom, but not that much)

I was thinking larger cores would be able to do more work.

Double the core size (keeping frequency the same)= double the IPC

 

[Fox5]

I was thinking larger cores would be able to do more work.

Double the core size (keeping frequency the same)= double the IPC

That's very true if by double the core size you mean double the cores and then adopt an appropriate programming model.

Intel's current cores are already very wide, and have already reached the point of major diminishing returns, the only thing Intel could really double up on for a good performance boost would be SIMD units, and that would only work well in some apps.

 

[sandorski]

IPC and MHZ limits have been reached, pretty much. More Cores is the best way to improve Performance at this time.

 

[cbn]

Intel's current cores are already very wide, and have already reached the point of major diminishing returns, the only thing Intel could really double up on for a good performance boost would be SIMD units, and that would only work well in some apps.

IPC and MHZ limits have been reached, pretty much. More Cores is the best way to improve Performance at this time.

If this is true then I figure it won't be long before general purpose computers really do split off from fullsize CPUs.

We are already seeing this happen with LGA 1156 and LGA 1366.

I guess this would add another level of mainboard customer. (From most complex to least complex: Dual socket server board--->single socket server/high end consumer board---> lower power/lower core count consumer board (Mini-ITX or Mini-DTX)---> laptop--->ultra mobile)

Somewhere in the middle (Mini-DTX, etc) I am **hoping** Google or some other kind of OS competitor really shakes things up by working from the bottom up. That could really make competition interesting. Cheap enthusiast level hardware in the hands of more people might really help. I figure playing with hardware/software at a very young age is how a lot of people got their inspiration to go into IT.

 

[JFAMD]

I was thinking larger cores would be able to do more work.

Double the core size (keeping frequency the same)= double the IPC

That is like saying a smaller car will automatically have better gas mileage. It is not as simple as you believe.

Performance really falls into 2 categories: speed (how fast can I go) and throughput (how much work can I get done.)

Desktops live in the speed world, everyone is obsessed with "faster."

Servers live in the throughput world and everyone is obsessed with "more."

Servers will benefit more from more cores, more often. We are bumping up into the upper limits on "speed" but there are still some things to be done to improve that; unfortunately a lot of it will need to happen on the software side (i.e. GPGPU and OpenCL) vs. just cranking up the clock rate.

 

[Voo]

Servers will benefit more from more cores, more often. We are bumping up into the upper limits on "speed" but there are still some things to be done to improve that; unfortunately a lot of it will need to happen on the software side (i.e. GPGPU and OpenCL) vs. just cranking up the clock rate.

And not to forget that everything that works well with GPGPU is per definitionem easily parallelizable, so it all results in having to write multi-threaded software..

Ah we poor developers, I think I should look for a job that does some kind of data processing - you know, stuff that's easy to parallelize

 

[grimpr]

Mr.Fruehe wisely tells us that in the not so distant future, the "Throughput" paradigm is coming to the desktop. Personally i believe him since AMD has a proven record in grasping and foretelling the future trends.

In defending his words, i urge you to watch some videos from the Microsoft PDC 2009.

Microsoft Perspectives on the Future of Programming
http://microsoftpdc.com/Sessions/FT52

C++ Forever: Interactive Applications in the Age of Manycore
http://microsoftpdc.com/Sessions/FT19

F# for Parallel and Asynchronous Programming
http://microsoftpdc.com/Sessions/FT20

Manycore and the Microsoft .NET Framework 4: A Match Made in Microsoft Visual Studio 2010
http://microsoftpdc.com/Sessions/P09-09

The State of Parallel Programming
http://microsoftpdc.com/Sessions/P09-17

DirectX11 DirectCompute
http://microsoftpdc.com/Sessions/P09-16

Lighting up Windows Server 2008 R2 Using the ConcRT on UMS
http://microsoftpdc.com/Sessions/SVR10

 

[oatmealstorm]

Then when I would overclock it I could say my pc is turned up to 11~~~~!

You're gay!

You have PM. -AT Admin DrPizza

 

[soccerballtux]

Mr.Fruehe wisely tells us that in the not so distant future, the "Throughput" paradigm is coming to the desktop. Personally i believe him since AMD has a proven record in grasping and foretelling the future trends.

In defending his words, i urge you to watch some videos from the Microsoft PDC 2009.

Microsoft Perspectives on the Future of Programming
http://microsoftpdc.com/Sessions/FT52

C++ Forever: Interactive Applications in the Age of Manycore
http://microsoftpdc.com/Sessions/FT19

F# for Parallel and Asynchronous Programming
http://microsoftpdc.com/Sessions/FT20

Manycore and the Microsoft .NET Framework 4: A Match Made in Microsoft Visual Studio 2010
http://microsoftpdc.com/Sessions/P09-09

The State of Parallel Programming
http://microsoftpdc.com/Sessions/P09-17

DirectX11 DirectCompute
http://microsoftpdc.com/Sessions/P09-16

Lighting up Windows Server 2008 R2 Using the ConcRT on UMS
http://microsoftpdc.com/Sessions/SVR10

Awesome post I'm going to watch all of these thanks dude.

 

[JFAMD]

Well, I would never speak for the desktop world, I just don't know it well enough.

Throughput is a great measure for server because they spend so much time shoveling things from point A to point B.

I am not sure about throughput on the desktop. The reason I say this is that desktops generally rely on some type of user interaction (i.e. mouse, keyboard) which means that throughput might not be the issue. Things like video transcoding might be more of a throughput concern (think about apps that run unattended for long periods of time....)