ED#57 : The AMD Phenom Goes Triple Core!

[Adrian Wong]

Originally Posted by peaz

Multicore optimizations aren't that simple as many developers have already mentioned. it's just odd that AMD decides to sell it as a tri-core processor when everyone else is multiples of 2. the usual dual-core, quad-core and octa-core (SUN).

Unless AMD's processor architecture is able to automatically split all the instructions load equally to all 3 cores, the performance of this tri-core processor might not be too far off a dual core.

Hmmm only time will tell when we finally get to see and test this tri-core procs.

And it's definitely made out of broken quad cores. There's no way AMD can design an odd tri-core proc and not announce it well before the release date. 3 cores... so unique. and knowing AMD, if this was in their roadmap 2 years ago, we would have heard how great and wonderful 3 cores is compared to 2 for years and years until we got sick out of hearing it. And not to mention... why no 'projected' benchmarks of this 3 core procs? LOL!

Update: from News.com - http://www.news.com/8301-13579_3-9780049-37.html

Exactly what I wrote (referring to the News.com report).

Well, I'm no programmer, but I believe the biggest problem with optimizing for multi-core is dependencies, where an instruction has to wait for data from another instruction. This prevents them from running simultaneously. I think it applies to all multi-cores, no matter whether it's 2, 3 or 4.

So, some apps should work better for multi-cores than others. Those that can issue simultaneous, independent threads should work well. Unless there is something else about multi-core optimization (that I don't know about, sorry!), apps optimized for quad-core should work well with 3-cores... and 3-cores should see less penalty for apps NOT optimized for more than 2 cores.

[Adrian Wong]

Originally Posted by The_YongGrand

Never mind if the processor is cheaper, and can fit AM2 sockets backward compatible. The main thing is if it reaches the consumers well and good, it'll be great.

But how can they split workloads into 3 parts? The parts couldn't be symmetrical. Four parts will be easy - the smaller parts are all the same in the end.

I don't believe the "load-balancing" is done by the processor. It's all based on processing independent threads simultaneously.

If you are running a single-threaded application, then only one core would be running, no matter whether you have 2 cores or 4 cores.

If there are 4 separate threads, then the tri-core processor should be able to run at full 3 cores while a quad-core processor should be able to use all 4 cores.

[dess]

The article says:

these triple-core processors will be the world's first and only triple-core processor.

Well, this is not true, just take a look on the Xbox360's triple-core Xenon by IBM.

[Adrian Wong]

Originally Posted by dess

The article says:

Well, this is not true, just take a look on the Xbox360's triple-core Xenon by IBM.

Ooops. Let me correct that. The only triple-core processor for desktop PCs.

[peaz]

it all depends on the way the application is coded for multi-core optimization. it's not that easy to support n-core. most applications are specifically designed to utilize dual core if I'm not mistake. 4 cores = 2x2. that makes it easy for the programmers to double the number of threads to process the instructions since it's all multiples of two. I could be wrong here though.

it just strikes me that 3 core would make things very odd. even in the SMP world. we hardly see 3 CPUs. It's usually multiples of 2.

[Adrian Wong]

Originally Posted by peaz

it all depends on the way the application is coded for multi-core optimization. it's not that easy to support n-core. most applications are specifically designed to utilize dual core if I'm not mistake. 4 cores = 2x2. that makes it easy for the programmers to double the number of threads to process the instructions since it's all multiples of two. I could be wrong here though.

it just strikes me that 3 core would make things very odd. even in the SMP world. we hardly see 3 CPUs. It's usually multiples of 2.

Yeah, right now many apps seem to be optimized for dual-cores. But if I'm not mistaken, if you optimize the app for quad-core (i.e. more independent threads), then it will not be as fast as a quad-core but there should not be any problem for all 3 cores to run those threads, since they are independent anyway.

[peaz]

Need someone who knows multi core optimizations to help us out here. What AMD is doing here is quite similar to SUN's Ultrasparc T1 architecture which have 4 or 8 independant cores connected via a onchip crossbar interconnect to the L2 cache. The difference being AMD's phenom/barcelona has dedicated L2 cache and a shared L3 cache. Sun's T1 does not have an L3 cache.

But similarly, Sun's T1 cores acts as independant processors like AMD's Barcelona. ANd we all know that the T1 ultrasparcs are superb short-instructions multithreading processors but lousy when it comes to long running instructions processing. That makes it a great web server but a lousy video processing/3d rendering server for example. Unless of course, if the application is designed to make use of all those 32 independant threads (4 hardware strands per core, similar to hyperthreading on Intel P4s).

Much is to be proven by AMD here. can't wait for the actual benchmark results.

[Adrian Wong]

I think that's the problem with going with more than 2 cores. Not many applications can support more than 2 independent threads at once. Many are even single-threaded. However, more cores will allow you to run several single-threaded applications simultaneously.

[dess]

Originally Posted by Adrian Wong

Ooops. Let me correct that. The only triple-core processor for desktop PCs.

Don't you intend to correct it in the article as well?

[Adrian Wong]

Originally Posted by dess

Don't you intend to correct it in the article as well?

Ooops! Corrected it to "world's first and only triple-core x86 processor", but forgot to upload it.

Thanks for the reminder!