12 Core AMD Processors

[matt100]

I can see more things being multi threaded as dual core becomes the norm and once they've multi threaded something it might not be that much of a jump to make the software use as many as are available.

So basically it'll become a speed thing again, it'll work ok on 2, faster on 4, faster still on 8 etc etc.

I could be totally wrong but I think the move to multi threaded software has already begun seeing as even laptops are now routinely sporting dual core cpus.

 

[Dark_Angel]

I can see more things being multi threaded as dual core becomes the norm and once they've multi threaded something it might not be that much of a jump to make the software use as many as are available.

So basically it'll become a speed thing again, it'll work ok on 2, faster on 4, faster still on 8 etc etc.

I could be totally wrong but I think the move to multi threaded software has already begun seeing as even laptops are now routinely sporting dual core cpus.

You are right. Windows Vista's support of multi cores has increases, even inside the explorer process is now multicored.

More efficient CPU designs and multicores are the way forward I think, hopefully with a reasonable boost to clock speed in future revisions.

 

[Chimerical]

I respect people's arguments and views on here.

But I find it annoying when people constantly go "what's the point of 12 cores, nothing will use it".

"640K (RAM) ought to be enough for anybody". - Bill Gates, 1981

That quote proves every point in the book. Progress is needed, no matter how small at first.

 

[jaykay]

Yes, that's true, but people who are actually going to use the 12 cores aren't the majority of the market.

Im pretty sure AMD make more money off their server market than any other. So this CPU could help them a lot.

 

[Corasik]

Intels Dunnington is still a Penryn based part. It will quickly be obsoleted by the 8 core Nehelem.

Remember too, that Nehelem brings back Hyperthreading, and as Core2 and Nehelem have a much greater parallelism than Pentium IV, hyperthreading should be better than the 25% gain you got with P4 on highly threaded apps.

A single Nehelem processor with 8 cores, processing 2 threads per core, should be a match for a 12 core chip without hyperthreading more or less. Not to mention the fact that the Nehelem adds an integrated memory controller, and CSI (intels version of hypertransport). It should be a good boost clock for clock compared to Penryn parts.

Take Suns Niagra2 processor for example. Its got 8 cores, but it can actually process 64threads concurrently, and it does if very efficiently. Throughput of server type applications (like Apache web server for example) can get massive benifits from the Niagra processor family.

Back to Intel, take a Skulltrail type system, add a pair of 8 core Nehelems, and you have a system which can process 32threads at any given time. Thats going to be a potent system for sure.

Good to see AMD are still in the game, but Intel knows it was a mistake to sit on P4 technology for so long. They are now developing new cores to keep performance on the up. They intend to keep the performance crown. Even if AMD to manage to take the crown at some point, I personally believe intel will actually make an effort to get it back. With P4 they just relied on their brand name, and it almost cost them bigtime.

 

[Dark_Angel]

Very well written post Corasik.

 

[FrankJH]

It does depend heavily on the interconnect between the cores (and to memory to a lesser extent)

If AMD get this to a degree better than Intel, it doesnt matter about Hyperthreading (in the sense that 12 true cores better connected will always be better than 8+8 that arent so well connected)

Intel could always get it done better, and then they would be sitting pretty

I moved from a dual core Intel to a Q6600 just after the Q's became available, and yes I think I can really tell just in Windows (XP and Vista) the fluidity improve drastically from having the two extra cores, Windows may not be written to optimally deal with multi-cores however background apps still use them, and the dog that is Windows is just more pleasant to use

 

[theforce]

Highly interesting stuff. Still nice to have AMD about with new ideas even just to motivate Intel not to rest on their current pefrormance crown. No amount of cores will keep me happy since most of the times at work i have a ridicoulous amount of of applications running. Photoshop, Coreldraw, Dreamweaver, media player which all require a serious processing power to run smoothly.
We have been near the 3GHz mark for cores for some years now and the road ahead was adopting multi-core systems which at first devided people who thought upping the frequency of a single was the way ahead. Taking into consideration that 32nm processors are going to see their way into consumer PC's in late 2009 (if AMD doesn't die, making Intel milk on their 45nm technology for ages) we could see a wild increase of cores per CPU if the yields are good or a good increase in Mhz which will also be welcomed.
To sum it up i think the number of cores at the moment is nice and it more than enough for most people even people who are heavy multitaskers. I think upping the frequency now should be the next logical step since the yields are not getting better as we near the barrier of silicon capabilities.

 

[Rroff]

Remember too, that Nehelem brings back Hyperthreading, and as Core2 and Nehelem have a much greater parallelism than Pentium IV, hyperthreading should be better than the 25% gain you got with P4 on highly threaded apps.

Hyperthreading on the old 800FSB P4s get you 25-50% gains in the likes of SETI, Folding, etc. and in some multithreaded games like COD2, ETQW more recently and probably COD4 tho I've not tested that with a P4 around 50% average gain - giving them a bit of a last laugh over the old Athlon XP line for gaming useage heh.

A single Nehelem processor with 8 cores, processing 2 threads per core, should be a match for a 12 core chip without hyperthreading more or less.

It should in theory be more than a match - as its likely to be running a higher clock speed and HT works quite well in general with most modern multi-threaded applications.

 

[Corasik]

It should in theory be more than a match - as its likely to be running a higher clock speed and HT works quite well in general with most modern multi-threaded applications.

True, but if you start off with a fairly realistic view, if it turns out to be better when it is released its a pleasant surprise, rather than a disappointment

25% gains with Hyperthreading were based on a blend of applications, some of which gained very little. Its certainly true that a few applications benifitted more, but others didnt do as well.

As for the interconnect between cores, C2D's use a shared L2 cache, Intels current quads use the old Frontside bus, which is getting on a bit now and it still is beating AMD's current processors. Nehelem gets rid of all the old baggage, bringing a native quad core, and the while the 8 core will still be 2 die's in one package, it will use CSI as the interconnect between the cores rather than the FSB.

Im not sure about the new "enthusiasts socket", it will certainly mean you cant simply overclock the entry level parts to get top performance. But thats not to say that an overclocked mainstream part wont still be a very fast system.

And yes, its GREAT new that AMD are still in the game, developing evolutionary improvements on their K8 architecture. If AMD hadnt had so much success with Athlon64, then Intel may well still be pushing its power hungry, inefficient P4 architecture. I hope for a long and competitive CPU market with both performance, price, and power efficiency wars between AMD and Intel for the forseable future. Long live AMD

 

[FrankJH]

As for the interconnect between cores, C2D's use a shared L2 cache, Intels current quads use the old Frontside bus, which is getting on a bit now and it still is beating AMD's current processors. Nehelem gets rid of all the old baggage, bringing a native quad core, and the while the 8 core will still be 2 die's in one package, it will use CSI as the interconnect between the cores rather than the FSB.

Very true - Im still hoping for an AMD win (its about time I would say) just to give Intel a kick up the whatsit

You never know, it would be nice to think of Nehelem (it self being a decent chip) coming off 2nd best

 

[tbz_ck]

I like the advances in number of cores (parallelism) but to be honest I find that hard disk access times are the bottlenecks in many systems now.

Don't think that because you haver 12 cores you can render stacks more at the same time if your disks can't handle it.

Personally I think we need a completely different method of accessing data to feed these multicore beasts and stop them from waiting around unnecessarily. Maybe solid state disks are the answer, but the price per Gb is prohibitive at the moment.

 

[Corasik]

Very true - Im still hoping for an AMD win (its about time I would say) just to give Intel a kick up the whatsit

You never know, it would be nice to think of Nehelem (it self being a decent chip) coming off 2nd best

Would be nice, history has proven in the past that if you give intel a hard enough kick in the .... they come out with astonishingly good products. But when times are easy for them they can get lazy.

 

[FrankJH]

Would be nice, history has proven in the past that if you give intel a hard enough kick in the .... they come out with astonishingly good products. But when times are easy for them they can get lazy.

Exactly my point - the original A64 was quite a while ago now, so hopefully AMD strikes back sometime soon

 

[HighlandeR]

Can you imagine the casual user or even hardcore gamer id imagine even 6 cores would be overkill

Cant wait frankly...

 

[NickK]

I havent found a game that REALLY makes use of dual core yet, 12 has to be an insane waste outside of servers.

Busses are a problem with CPUs such as these..

We're CPU bound in our products and that's with four quad cores.

 

[cozm01]

Hyperthreading is terrible, gives more fps but everything starts to stutter, much less smooth. Was many articles on this problem before.

 

[FatRakoon]

Who remembers those idiot adverts about Intel P4 HT CPUs?

"You can now browse AND EMa il and the same time.."

Give me strength!

Well, 12 Cores eh? - AMD certainly needed to have somethign up their sleeves and of course we will get the usual brigade of people who cannot see the use for that many cores, but the facts do indeed remain that this many cores might sound too much, but we will all be running them in 10 minutes down the line.... Thats exactly what happened with dual core, and now its the standard, in another years time we will all be saying 48 cores? Nah, I find my 12 core just fine thank you.

I have been looking for an excuse to go back to AMD for a while now, and with My No1 PC being hit by lightning the other day and its struggling to stay reliable now, this is certainly the boost I needed...

 

[NickK]

Agreed. No developer can guarantee what hardware your going to be on. They have to code for the lowest requirement.

That depends - if you're writing games this is true. If you're providing a scalable application then that observation isn't entirely true.

You can write an application that will scale with the number of cores - a lot of applications can and do split tasks into parallel tasks that can utilise more cores.

There's a general trend/realisation that the serial procedural languages are nolonger providing the right form of describing applications.

 

[Corasik]

Hyperthreading is terrible, gives more fps but everything starts to stutter, much less smooth. Was many articles on this problem before.

Never noticed any stutter with Northwood Core HT enabled. Skipped prescott altogether. But the wider the core the smoother and better scaling you'll get from hyperthreading.

Core 2 Duo for example has got a 4 issue core. P4 is a 3 issue core, but really functioned more like a fast 2 issue core. Many programs are unable to make full use of all 4 instructions that Core2 Duo can run in parallel, so hyperthreading would have made C2D even better at multi tasking, and as the core is wider, applications are less likely to get stuck in a deadlock with two applications fighting for the same cpu resources.

As you go multi core, the downsides of hyperthreading vanish almost entirely, as the OS can assign the thread to any available virtual cpu. The more virtual cores you have the less chance that you'll be stuck waiting for resources.

Sun's Nigra processors are able to process up to 4 threads per core, and the Niagra II can process 8 threads per core, and given the right application they scale insanely well.

Stuttering is more likely an issue with early graphics drivers making poor use of multithreaded programming, without realising that HT/Windows can assign threads to separate virtual cores. With more and more computers being built with true multicore, todays newest software is already being groomed to be tolerant to multicore hardware, and there should be no incompatiblites with Nehelems greatly improved hyperthreading systems.