Screw Quad Core! Eight Core anyone? FASN8ING!

[coldpower27]

Originally posted by: RaynorWolfcastle

Originally posted by: ericlp
Intel is not even a true quad core... Do some read up...

I'm no fan of either Intel NOR AMD.

But I am a fan of new technology, If Intel were to Dominate the chip market, then say good buy to competition and pushing the envelope. I mean... Last time Intel took the lead prices were high and tech advances become slow as Intel racked in the cash witch tended to make Intel Lazy.

I say lets liven it up a bit! I'm all for lower prices and more competition!

The only thing that matters is performance and power, the means by which those are achieved is irrelevant. My computer could be made of hamsters for all I care, as long as performance is up to snuff and power consumption is reasonable

QFTMFT

 

[dmens]

Originally posted by: Hard Ball

Originally posted by: dmens

Originally posted by: ericlp
Intel is not even a true quad core...

what's that

Well the most commonly accepted differentiator between MCM 2^n-core and native 2^n-core is that: the presence or absence of a difference in communication latency between the cores of a single socket versus cores from two different sockets.

sounds like amd's marketing definition of "native"

 

[Idontcare]

Originally posted by: dmens

Originally posted by: Hard Ball

Originally posted by: dmens

Originally posted by: ericlp
Intel is not even a true quad core...

what's that

Well the most commonly accepted differentiator between MCM 2^n-core and native 2^n-core is that: the presence or absence of a difference in communication latency between the cores of a single socket versus cores from two different sockets.

sounds like amd's marketing definition of "native"

Yep. I love how people attempt to add credibility to their opinions by prefacing them with clauses such as "the most commonly accepted..."

Accepted by who? Documented where and by whom?

I'm with the poster who commented about hamsters above, I don't care if it is quad, octo, giga, whatever...how fast will it run MY applications of interest and at what cost and at what heat and noise levels.

 

[DrMrLordX]

Originally posted by: bfdd
You do mean 8800GTS in SLI and not GTX, if you look a SINGLE GTX was holding it's own even betting it in some benchs...

hah, that's what I get for not reading Anand's graphs as closely as I should.

techreport's benchmarks do cast things in a different light, however.

 

[DrMrLordX]

Originally posted by: Borealis7

cons: twice the power consumption, twice the heat output of a quadcore (need two coolers...)

.

Not necessarily true. Look at Kentsfield/Clovertown for example . . . their power consumption per chip is twice that of Conroe/Woodcrest.

So, running two Woodcrests produces no more (or less) heat than running a single Clovertown. If anything, keeping the Woodcrests cool is probably easy due to the fact that heat is distributed between two separate CPUs.

 

[Hard Ball]

Originally posted by: dmens

Originally posted by: Hard Ball

Originally posted by: dmens

Originally posted by: ericlp
Intel is not even a true quad core...

what's that

Well the most commonly accepted differentiator between MCM 2^n-core and native 2^n-core is that: the presence or absence of a difference in communication latency between the cores of a single socket versus cores from two different sockets.

sounds like amd's marketing definition of "native"

Well, it doesn't really matter whose marketing initiative it sounds like; what's relevant is that this design decision actually makes a large difference in performance of multithread applications.

 

[Hard Ball]

Originally posted by: Idontcare

Originally posted by: dmens

Originally posted by: Hard Ball

Originally posted by: dmens

Originally posted by: ericlp
Intel is not even a true quad core...

what's that

Well the most commonly accepted differentiator between MCM 2^n-core and native 2^n-core is that: the presence or absence of a difference in communication latency between the cores of a single socket versus cores from two different sockets.

sounds like amd's marketing definition of "native"

Yep. I love how people attempt to add credibility to their opinions by prefacing them with clauses such as "the most commonly accepted..."

Accepted by who? Documented where and by whom?

I'm with the poster who commented about hamsters above, I don't care if it is quad, octo, giga, whatever...how fast will it run MY applications of interest and at what cost and at what heat and noise levels.

Accepted as being currently taught in computer architecture classes in ECE and CS departments of most major research universities around the country (as at mine).

If you have something useful to contribute; why don't you grace us by shedding light on the benefits and detriments of each microarchitectural element of CMA and K10.

 

[aigomorla]

Originally posted by: DrMrLordX

Originally posted by: JackPack

Except it offers true quad-core performance.

In a single socket, yes. It's been demonstrated already that a dual-processor Clovertown system frequently loses performance contests against quad-processor Opteron (dual core) rigs. There seem to be some cache coherency issues among other problems.

dual-processor K10 systems with 8 cores and faster HT links (HT3.0) should rock the casbah, not that word of octal-core AMD desktops is really news. Word of AgenaFX-equipped QuadFX boards has been out for awhile (though the FASN8 marketroid-speak is new).

Also, why all the hate for the 2900XT? In single-card configurations, it makes no sense, but 2900XTs in Crossfire seemed very competitive with 8800GTX SLI, at least according to Anand anyway.

In theory, K10 QuadFX/FASN8 rigs should be quad-card Crossfire capable using HTX slots instead of PCIe slots for some ridiculous performance and power consumption.

Where are you getting these results? Im sorry but X2900XT on XFIRE is being benched along a SINGLE GTX solution.

The X2900XT in xfire in all reviews shown was pitted against a 8800GTS. Okey, it also shows a 8800 ULTRA killing it. Wow, thanks for the long delay ATI, your product sucks hardcore even with the delay.

It also draws more power then the GTX, produces way more heat. So HOW can you be glorifying this product? The only advantages i see in it is that XFire works on a Intel boards. But the performance gain from a intel board to a NVidia board in GAMING is very small.

Sorry, id rather take my 680i any day of a 975, or even a P35, becuase 8950GX2 in SLI = GG for any game to come.

Lets see ATI try to beat a 8950GX2 rig. It like asking a civic hybrid, to up against a Civic Type-R.


I have really lost faith in ATI, and AMD. There delays are horible. Even the X1900XT launch, was majorly delayed. However the product was "worth it" I dont see the X2900XT being worth it at all.

The introduction of the 8900GTX with smaller die, and heat comsumption will be something ATI needs to seriously worry about. But then again, its AMD's problem now i guess.

 

[dmens]

Originally posted by: Hard Ball
Accepted as being currently taught in computer architecture classes in ECE and CS departments of most major research universities around the country (as at mine).

that's weird, 6 years in school and 3 years in the field and ive never even heard the word "native" used about multicores until AMD picked it up.

Originally posted by: Hard Ball
If you have something useful to contribute; why don't you grace us by shedding light on the benefits and detriments of each microarchitectural element of CMA and K10.

what for, that's irrelevant. don't try to change the subject.

 

[Hard Ball]

Originally posted by: dmens

Originally posted by: Hard Ball
Accepted as being currently taught in computer architecture classes in ECE and CS departments of most major research universities around the country (as at mine).

that's weird, 6 years in school and 3 years in the field and ive never even heard the word "native" used about multicores until AMD picked it up.

Of course, no one would use such ambiguous term of marketeering as 'native', but much more like terms such as "single die integrated design", or "fully integrated MPU". I have not really disputed which design is a "true quad", whatever that means. I'm simply stating that there is a significant architectural difference between the two design choices, and will yield difference under a number of workloads.

Originally posted by: Hard Ball
If you have something useful to contribute; why don't you grace us by shedding light on the benefits and detriments of each microarchitectural element of CMA and K10.

what for, that's irrelevant. don't try to change the subject.

For this, I was responding to: "I don't care if it is quad, octo, giga, whatever...how fast will it run MY applications of interest and at what cost and at what heat and noise levels. "; which simply ignores the fact that this discussion about multicore design is based on technical premises, whether one microarchitecture and its design choices yield differences in performance. What's the use of interjecting into a discussion of technical matters with "screw it, I don't really care about any of this crap".

 

[The-Noid]

AMD has never had a problem with innovation, it is with getting products to market.

 

[DrMrLordX]

Originally posted by: aigomorla


Where are you getting these results? Im sorry but X2900XT on XFIRE is being benched along a SINGLE GTX solution.

Actually, I misread the benchmark. bfdd already pointed that out. Still, 2900XT Crossfire comes out looking a lot better in techreport's vista-based benchmarks. Or, should I say, 8800GTX SLI and 8800GTS SLI come out looking a lot worse. Prolly driver issues.

 

[aigomorla]

Well you do know the 8900GTX are around the corner.

Heres crossing my fingers there like there younger brothers the 7900GTX.


If thats the case, a simple volt mod here, EK FC8800 waterblock there.




Need i say more?

 

[DrMrLordX]

Anything to lower prices on existing cards.

Once Nvidia has a product refresh on the market, the only saving grace of the 2900XT will be its apparent overclockability:

http://www.nordichardware.com/news,6331.html

Phase change or bust (LN2? feh).

AMD better roll out the 65nm R600 refresh soon . . .

 

[lopri]

Hey, Hard Ball: I agree with your points and I also hate to see a sound argument going down the hill towards pissing match.. Then again, you could have expected what this thread was going to be about, just by the title alone.

 

[aigomorla]

even if amd could pull off a uber barcelona / agenda ... and phenom was great.

they cant fight neph. Neph will most likely be out when native quadcore AMD's come out.

So what is amd going to do again?

EDIT: im sorry the op didnt even mention a native 8 core: WHERE ARE YOU GUYS GETTING NATIVE 8 CORES??

"platform includes two true quad-core AMD Phenom processors, "

Bah, screw that, id rather have dual yorkies. how does 16 cores sound to you now?

And to my understand, a dual cloverdale machine is already ready however if you want to nick and pick at the native factor, Welcome penryn.

I'll take a dual Cloverdale @ 3.4ghz over an AMD phem for now. Unless the phenom overclocks spectacularly.

And even then, i'll take dual yorkfields any day over 8 cores. You cant really fight against the power known as intel.

 

[DrMrLordX]

Originally posted by: aigomorla
even if amd could pull off a uber barcelona / agenda ... and phenom was great.

they cant fight neph. Neph will most likely be out when native quadcore AMD's come out.

If you mean Nehalem, it won't be out until late 2008. Quad-core K10 chips will be out this year, at least in the server segment anyway. On the desktop, maybe not until Q1 2008. Still before Nehalem, though.

EDIT: im sorry the op didnt even mention a native 8 core: WHERE ARE YOU GUYS GETTING NATIVE 8 CORES??

I think that's the first time anyone's mentioned native octal-core CPUs in this thread. AMD will be first to market with a desktop octal-core system this year utilizing two sockets with AgenaFX/Phenom processors, unless you count the dual-Clovertown Macs.

Bah, screw that, id rather have dual yorkies. how does 16 cores sound to you now?

Yorkfield will be a quad-core processor, just like Kentsfield. Yorkfield will also probably not be available in dual sockets unless Intel decides to release dual-socket desktop boards.

http://en.wikipedia.org/wiki/Yorkfield#Wolfdale_and_Yorkfield

And to my understand, a dual cloverdale machine is already ready however if you want to nick and pick at the native factor, Welcome penryn.

Clovertowns are only available on server boards that generally require FB-DIMMs. There might be boards out there utilizing registered DDR2 for Clovertown instead, but if there are, I haven't seen them. You can't get Clovertown on the desktop.

I'll take a dual Cloverdale @ 3.4ghz over an AMD phem for now. Unless the phenom overclocks spectacularly.

And even then, i'll take dual yorkfields any day over 8 cores. You cant really fight against the power known as intel.

I think the dual-cpu version of Yorkfield for servers and workstations will be . . . Harpertown? Anyway, some food for thought:

http://www.anandtech.com/cpuchipsets/intel/showdoc.aspx?i=2957

Sadly, Anandtech didn't have 3 ghz Clovertowns available for testing (bah), but given the fact that dual-core Conroes are generally held to be about 20% faster per clock than dual-core K8s, a pair of 2.33 ghz Clovertowns should be about the equivalent of a quad Opteron 2.8 ghz machine just given the superiority of the Clovertown's individual cores.

For some reason, the Opteron 880 system was top performer in most of the article's benchmarks. Clovertown is not scaling well beyond four cores, and Harpertown will likely suffer a similar fate. Intel's CSI bus should correct that problem.

Anyway, given the scaling problem current Intel systems seem to experience beyond four cores, were I sufficiently lacking in sanity to want an octal-core system, I'd look more seriously at K10 solutions. In a single socket situation, a 3.4 ghz Yorkfield would offer much stiffer competition versus a single quad-core K10 chip. Another odd example of Clovertown scaling badly:

http://www.anandtech.com/cpuchipsets/intel/showdoc.aspx?i=2897&p=9

At least Clovertown comes out looking better in Specjbb 2005 until the number of instances is increased:

http://www.anandtech.com/cpuchipsets/intel/showdoc.aspx?i=2897&p=5