AMD promises 16-core threading by 2012

[Zapata]

AN AMD panellist here in Vegas thinks that it will reach sixteen cores software programming within five years. This will be in time for the Playstation 4, the new Xbox and Geforce G2012 or Radeon whaetever they will want to call it.

http://theinquirer.net/default.aspx?article=36767

Are these guys for real? Why I ask that, is whilst I presume there is no great technical difficulty in producing 16 core chips, my understanding is that it is very difficult to produce software that utilizes dual cores effectively, logic would suggest to me that producing software to utilize a 16 core chip would be at least 8 times as difficult and very likely much more than that, since you tend to get a much steeper rise in difficulty as complexity increases. So with that in mind circa 2012, 16 core chips are going to look a bit pointless, if in 2012 we are using software that has only just managed to fully exploit the dual core chip and software engineers have only started working towards fully exploiting a quad core.

Best and Warm Regards
Adrian Wainer

 

[ERP]

http://theinquirer.net/default.aspx?article=36767

Are these guys for real? Why I ask that, is whilst I presume there is no great technical difficulty in producing 16 core chips, my understanding is that it is very difficult to produce software that utilizes dual cores effectively, logic would suggest to me that producing software to utilize a 16 core chip would be at least 8 times as difficult and very likely much more than that, since you tend to get a much steeper rise in difficulty as complexity increases. So with that in mind circa 2012, 16 core chips are going to look a bit pointless, if in 2012 we are using software that has only just managed to fully exploit the dual core chip and software engineers have only started working towards fully exploiting a quad core.

Best and Warm Regards
Adrian Wainer

I'm surprised they are projecting such a small number of cores.
Whether software will really utilise multi cpu systems in 5 years is another issue...

 

[Skrying]

Didn't AMD reject the idea of a massive amount of cores? They did not want to follow Intel into the 80 core territory and instead want to go with a gradual increase with more powerful cores inside.

 

[3dilettante]

If quad core comes out in 2007 on 65nm, then 16 would indicate AMD expects 2 further process transitions in the next 5 years, or they are counting on one transition and a 4x4 motherboard/package-level integration.

Despite AMD's claim that they want to do n-cores per chip "right" as opposed to Intel's use of package-level integration, it just seems to mean they're always doing it "second". By this pattern, it also means that the cores will be approximately the same number of transistors as the upcoming quad-cores at 65nm, which is disturbing from a single-thread standpoint.

If they mean native 16-core, then 2.5 years per transition seems tight, given AMD's history.
Even at this improved rate, it's not quite as good as Intel's record, and even Intel might be having issues at future nodes.

 

[Tahir2]

...if in 2012 we are using software that has only just managed to fully exploit the dual core chip and software engineers have only started working towards fully exploiting a quad core.

That is a big if and that if is very wrong.

 

[Sobek]

Well the only place I see AMD promising anything is in their article title. I sure hope people don't get all giddy and start expecting something...


*ps* And for the love of god, does everyone at the INQ try and type way too fast for their own skills? It's like reading the chatlogs of an underage WoW player.

 

[Blazkowicz]

I don't like how "core" is indiscriminately used to actually mean "thread", and that might be about an AMD CPU with eight SMT cores, or some new architecture with a pool of ressources and 16 visible threads, or who knows, that might be a real 16 core CPU but intended for servers and with FPU, SSE things cut down ; or two dies on a chip with really fast interconnect between them.

Off course all cores/threads will be used, there's no shortage of media encoding, server shit and simulation shit to be done. but they might not be used by you and me except when doing such shit at home or intentionnally maxing out the power bill. 16 cores seems stupid for the desktop and gaming workstation.

 

[suryad]

Isnt there a point of diminishing returns with the number of cores aspect? Last I read it was 8 or something.

 

[pjbliverpool]

I don't like how "core" is indiscriminately used to actually mean "thread", and that might be about an AMD CPU with eight SMT cores, or some new architecture with a pool of ressources and 16 visible threads, or who knows, that might be a real 16 core CPU but intended for servers and with FPU, SSE things cut down ; or two dies on a chip with really fast interconnect between them.

Off course all cores/threads will be used, there's no shortage of media encoding, server shit and simulation shit to be done. but they might not be used by you and me except when doing such shit at home or intentionnally maxing out the power bill. 16 cores seems stupid for the desktop and gaming workstation.

What I don't get is why would 16 (or 8) x86 cores be pretty useless for gaming while Cell and its SPE's are great? We hear about the many ways the SPE's can help accelerate gaming a lot but couldn't you off load tasks in the same way to x86 cores? Im not saying they would be as fast but surely they would still have some use? Afterall an 8 core Core2 CPU for example has as many seperate processing elements as Cell, couldn't you model your code in pretty much the same way?

 

[Killer-Kris]

Isnt there a point of diminishing returns with the number of cores aspect? Last I read it was 8 or something.

That's entirely dependent on the software, and ultimately the algorithm that software uses. Many applications can't scale beyond one core let alone two while others may have effectively infinite scaling. And of course there's the whole range in between those two ends.

 

[pascal]

If quad core comes out in 2007 on 65nm, then 16 would indicate AMD expects 2 further process transitions in the next 5 years, or they are counting on one transition and a 4x4 motherboard/package-level integration.

Probably, two transitions after 65nm. IIRC the 45nm is expected for 2008 and 35nm for 2011.

16 cores will definitelly not be mainstream by 2012.
My guess quad cores will be mainstream by 2012 and dual cores will be value.

 

[Skrying]

Probably, two transitions after 65nm. IIRC the 45nm is expected for 2008 and 35nm for 2011.

16 cores will definitelly not be mainstream by 2012.
My guess quad cores will be mainstream by 2012 and dual cores will be value.

Really? Dual core systems are already budget priced from OEMs. I would be surprised if we do not go further from there unless the progression of increasing number of cores stops.

 

[pascal]

Usually value doesnt use the latest technology/process.
Maybe a dual-core 45nm for sub $100

 

[Skrying]

Usually value doesnt use the latest technology/process.
Maybe a dual-core 45nm for sub $100

The vast majority of people get their systems from an OEM. A large number of OEMs are offering dual core systems in their budget platforms. Budget systems tend to have rather over powered CPUs and other areas are left off such as smaller hard drives (say only 250GB), integrated graphics, and lower amounts of RAM.

If you wish to build then you can still grab a $80 Pentium D 805, which is dual core and with some overclocking can actually give rather decent performance for its cost.

 

[pascal]

We are talking about AMD, arent we?
IIRC the value processors (<$100) from AMD are single core only right now.
See the newegg prices: http://www.newegg.com/ProductSort/B...rand=1028&Subcategory=343&name=Processors-AMD

The new core coming is the K8L. Initially with the 65nm process.
Hope to see dual K8L cores with 45nm by 2008/2009 in the 3GHz to 4.5GHz range.
What I am trying to say is that maybe by 2012 the 45nm dual K8L will be available at sub $100 price range.