Intel/AMD Quad Core Head-to-Head and Platform Changes

[joset]

I've been running an AMD "4x4" since January this year 8) :lol:

Could you elaborate on that?


Cheers!

 

[gOJDO]

@pete4r
K8L is allready developed and it was taped out in august. AMD will clock their K8L processors as much as possible becouse they have strong competition on the road. So far, they can hardly break the 3GHz barrier, especialy becouse they will use the 65nm SOI3. This is the point on which I disagree with you:
"Im sure they can easily break the 3GHz barrier, but that would be Pentium 4 EE all over again."
Also
"Also they can push the clock speed of their cores up when needed, like the old FX series"
They pushed the clock speed of the old FX series with pushing the TDP to 125W which is also contradictory to your statement:
"4 Cores @ 2.9GHz is enough, they have to develop their processor with a strong Power Consumption in mind."
a 2.6GHz FX-60 has 125W TDP. While the 65nm SOI3 K8 dualcore chips will have almost same TDP as 90nm SOI2 K8 dualcore chips. How do you expect the 2.9GHz improved quadcore taking more than twice silicon space to have low power consumation?
That is another point at which I disagree.
And I also disagree on this part:
"So maybe they think 2.9GHz is enough... who knows? but we all just have to be patient to see what AMD can pull out of their bags. "
AMD don't think that 2.9GHz is enough, and we don't have to be patient to see what AMD can pull out of their a$$.

 

[Wombat2]

I've been running an AMD "4x4" since January this year 8) :lol:

Could you elaborate on that?


Cheers!

My current desktop machine is a dual socket 940 Opteron 275 2.2. Ghz server i.e. 4 cores. 4x4 is two socket F Opterons in a dual socket F mobo rebadged as "4x4" for the gamer desktop market.

Socket 939 and socket F Opterons perform identically.

Edit: Corrected 939 -> 940

 

[Heyyou27]

My current desktop machine is a dual socket 939 Opteron 275 2.2. Ghz server i.e. 4 cores. 4x4 is two socket F Opterons in a dual socket F mobo rebadged as "4x4" for the gamer desktop market.

Socket 939 and socket F Opterons perform identically.

There is no such thing as a dual socket 939 board, seeing as all 939 chips only have on Hyper transport link; you probably have dual socket 940, which uses registered memory and is not targeted at gamers, but servers and workstations.

Anyways, I'm not going to wait until 2008 for AMD to release K8L, just to have it defeated by Yorkfield.

 

[joset]

My current desktop machine is a dual socket 939 Opteron 275 2.2. Ghz server i.e. 4 cores. 4x4 is two socket F Opterons in a dual socket F mobo rebadged as "4x4" for the gamer desktop market.

Socket 939 and socket F Opterons perform identically.

Well, that's some rig, indeed!
But, a dual Socket F MB will also support QC... that's 8 cores.

I believe the plus in the 4x4 topology, might be the adaptation of third parties' co-processors, aside the scalability of the platform; the very downside, is cost. Interesting approach, versatility-wise, but hardly ubiquitous.


Cheers!

 

[GlacierFreeze]

Where does it say that DDR3 IMC won't be on AM2+ K8L quad cores?

You'd better read the links.


Cheers!

Okay, I just confused myself for a second that's all. AM2 CPUs won't be compatible with socket AM3. AM3 CPUs WILL be backwards compatible with socket AM2 (but use DDR2 only).

 

[Wombat2]

Well, that's some rig, indeed!
But, a dual Socket F MB will also support QC... that's 8 cores.

You can get an 8 core Intel machine (2x Clovertowns) ... next month ... for what an 8 core 4x4 system will cost in Q3 2007

I believe the plus in the 4x4 topology, might be the adaptation of third parties' co-processors, aside the scalability of the platform; the very downside, is cost. Interesting approach, versatility-wise, but hardly ubiquitous.

4x4 is a transient stopgap for AMD until they can ship their own quad cores. Dual socket solutions are highly uneconomical for desktop use. Hence nobody is going to invest in developing specialised socket accelerators for a platform that is going to be around for 6-9 months as a tiny market niche. Ageia can barely get their physics accelerator going and that just needs a PCI slot.

 

[GlacierFreeze]

Would be cool though if they could implement a 2nd socket on all new boards (essentially making all AMD boards 4x4ish) on the cheap. It would open up the market for the accelerators. Price of the 2nd socket would be the main factor though.

 

[pete4r]

@pete4r
K8L is allready developed and it was taped out in august. AMD will clock their K8L processors as much as possible becouse they have strong competition on the road. So far, they can hardly break the 3GHz barrier, especialy becouse they will use the 65nm SOI3. This is the point on which I disagree with you:
"Im sure they can easily break the 3GHz barrier, but that would be Pentium 4 EE all over again."
Also
"Also they can push the clock speed of their cores up when needed, like the old FX series"
They pushed the clock speed of the old FX series with pushing the TDP to 125W which is also contradictory to your statement:
"4 Cores @ 2.9GHz is enough, they have to develop their processor with a strong Power Consumption in mind."
a 2.6GHz FX-60 has 125W TDP. While the 65nm SOI3 K8 dualcore chips will have almost same TDP as 90nm SOI2 K8 dualcore chips. How do you expect the 2.9GHz improved quadcore taking more than twice silicon space to have low power consumation?
That is another point at which I disagree.
And I also disagree on this part:
"So maybe they think 2.9GHz is enough... who knows? but we all just have to be patient to see what AMD can pull out of their bags. "
AMD don't think that 2.9GHz is enough, and we don't have to be patient to see what AMD can pull out of their a$$.

man over and over again u slapping your own face with what u say.

sure if we dont have to wait, where are the benchmarks of K8L vs Conroe? Where are the K8L processors in which shop? man stop ur wasting your time arguing with yourself here... I wont say more, waste of time trying to make a point for some1 who is locked in his own infinite little circle of understanding

 

[pete4r]

Would be cool though if they could implement a 2nd socket on all new boards (essentially making all AMD boards 4x4ish) on the cheap. It would open up the market for the accelerators. Price of the 2nd socket would be the main factor though.

yeah imagine a dual socket AM2 or 939 mobo, for €200, plus 2 Athlon 62 X2 3800 @ €150 each, thats only 200 + 150 + 150 = €500 u have a quad core system :lol:

 

[Wombat2]

yeah imagine a dual socket AM2 or 939 mobo, for €200, plus 2 Athlon 62 X2 3800 @ €150 each, thats only 200 + 150 + 150 = €500 u have a quad core system :lol:

4x4 is socket F only silly. It only takes extreme-gamer priced rebadged Opterons. You cant put cheap X2s into it ... AMD isnt stupid :lol:

http://www.dailytech.com/article.aspx?newsid=4422

 

[evilr00t]

Dream on, unless you want a dual-COMPUTER motherboard.

 

[joset]

4x4 is a transient stopgap for AMD until they can ship their own quad cores.

More in the line of this:
http://www.dailytech.com/article.aspx?newsid=1920

See, technologies come & go but there's a cumulative net effect, on what regards Technology; whatever AMD (or any other company) comes up with, it'll be good news for the big "T"; competition also has a net effect on selective filtering. Sure, 4x4 will vanish but so will LGA sockets/packaging & Next Generation [micro] Architectures; the beauty lies somewhere in between, from brainstorms to the drawing boards, from there to materialization & into our own laps. That's the way I see it, anyway.


Cheers!

 

[Wombat2]

Could you post a handful of benchmarks to give the readers who are following the 4x4 some idea of rough performance, preferably applications similar to the current obserations used with Kentsfield.

Jack

Just look up the past benchmarks for the 2.2 Ghz 1mb cache X2s 4400+ ... same performance as my 275s. I just get two extra cores.

A 3 Ghz C2D is about 40% faster than an equivalently clocked X2. So I expect a 4x4 with 3 Ghz chips to be roughly the same speed as one QE6700 at 2.66 Ghz. However, the QE6700 will be VASTLY more overclockable than the 4x4. The k8 arch is near its limit so offers very little OC room. OCing on dual socket mobos is also much more difficult. The kentsfield on the other hand has been OCed to 3.8 Ghz on air cooling. At that clock it will totally anihillate the 4x4.

As gamers are OC addicts I expect them to prefer the Kentsfield route. It also uses less power.

 

[pete4r]

yeah imagine a dual socket AM2 or 939 mobo, for €200, plus 2 Athlon 62 X2 3800 @ €150 each, thats only 200 + 150 + 150 = €500 u have a quad core system :lol:

4x4 is socket F only silly. It only takes extreme-gamer priced rebadged Opterons. You cant put cheap X2s into it ... AMD isnt stupid :lol:

http://www.dailytech.com/article.aspx?newsid=4422

did u not notice the IMAGINE before everything else was said?

 

[Wombat2]

Cinebench is an very poor multicore benchmark at it scales very poorly.

On the scientific software I use which is designed to scale to thousands of cores, the scale up is linear over 4 cores. Opterons have huge memory bandwidth so scaling is excellent.

So on properly multithreaded sw you can expect 4x4 to scale linearly.

Kentsfield scales linearly over 4 cores on properly multithreaded sw.

 

[joset]

Could you post a handful of benchmarks to give the readers who are following the 4x4 some idea of rough performance, preferably applications similar to the current obserations used with Kentsfield.

Jack

Just look up the past benchmarks for the 2.2 Ghz 1mb cache X2s 4400+ ... same performance as my 275s. I just get two extra cores.

And, most of the times, reality pops up & bumps into you.

There's no AMD 4x4 (yet...), you know that & there's no way you can argue back.


Cheers!

 

[Wombat2]

The last Cinebench benchmark I saw showed a 3.2x speedup on 4 cores which is crappy. 3d studio max showed excellent 4 core scaling but that isnt exactly cheap.

The problem is that there simply isnt a good consumer-orientated freely available multicore benchmark that scales properly to 8+ cores.

 

[ricardo]

I am looking forward to 4x4 personally. I think that it will bring a new level to high-end desktops and to workstations. Also, it is going to give users enormous potential to upgrade in 12 months by adding 2 quad cores - changing the chips gets you 8 cores and a new (upgraded) chip architecture!

Also, if the 4x4 platform is basically a tweaked server platform, so much the better! I have an old fx51 system - now 3 years old - which is basically an opteron server tweaked and rebadged (with a server class Mobo, 2.2GHz clawhammer cpu under the hood, and came complete with registered memory). Its still a great machine. Rock solid, reliable, and still pretty fast (for a single core system). A lot of that is down to the fact that it's basically a server. It's built for stability & reliability, and it's also pretty fast! Yes, you sacrifice a little speed because of the server class mobo and RAM, but you're never really going to notice that.

 

[joset]

I am looking forward to 4x4 personally. I think that it will bring a new level to high-end desktops and to workstations. Also, it is going to give users enormous potential to upgrade in 12 months by adding 2 quad cores - changing the chips gets you 8 cores and a new (upgraded) chip architecture!

That would be AMD's 4x4 'platform' (mpjesse's [forum member] words... & I concur) main advantage, towards any other platform: its upgradeability. However, both price & AMD's own concurrence (Altair) in the upcoming year, might seriously limit the scope of such a platform; server/workstation-wise, it's understandable & desirable (RAS, improved FP on the upcoming FX-7x, scaleability); DT-wise, it would seem like committing suicide in two steps: High price premium at start up, w/1 QC & decide between the rope & the knife within 12 months, or rather, between a second 4x4 QC or Altair. I'm, purposefully, leaving Intel outside of the ballpark.


Cheers!