The IDF Thread

[Hippiekiller]

Originally posted by: Stoneburner
I don't understand the hype for penryn, we already know it's not much better than conroe according to anand's preview. As for Nehalem, it could turn out to be unspectacular like barcelona, it could turn out to be a dud like pentium D's.

You don't understand the hype for penryn?

Think of G0 Quads, how many people have been going gaga over those?

Just because of a new production run that runs cooler and uses less voltage. This is far more significant then that with all the new manufacturing techniques.

 

[Pederv]

When each core can process twice as fast as a single core can currently process (AMD or Intel) wake me up.

 

[Arkaign]

Originally posted by: Pederv
When each core can process twice as fast as a single core can currently process (AMD or Intel) wake me up.

Moores law. Slightly skewed by the shift towards multi-core, but pretty accurate. Look at the time between Athlon 64 3000+, and what most would consider a proc that is twice as fast, the X2 6000+. I know you're talking core for core, but that is getting murkier as time goes on. I will say that you'll probably have to continue sleeping for a couple of years. The current best processor for IPC is the C2D, and to double the performance of one of those cores running at 3ghz/1333fsb, you'll need Nehalem, K11, or the like. Much more likely we'll be seeing 16-core 5ghz systems, of which one of the cores might APPROACH a 100% improvement comparing one core to one core.

Seems like you're bitter that AMD is getting pulverized, but you have to remember how incremental performance increases have classically been. When P3 was released, the 450mhz version barely edged out a P2-450. When P4 was released, it barely edged out (lost in many cases) the P3. The Athlon Slot A K7 tied with the P3 Slot 1 Katmai, the Athlon Thunderbird tied with the Coppermine P3, it's just been increment and stalemate for eons.

The only true divergences have been Athlon XP kicking the poo out of P4 Willamette, and Core 2 Duo kicking the crap out of AMD X2. I'm talking performance increases in excess of single digits, which seems to be the norm.

 

[Phynaz]

Originally posted by: Pederv
When each core can process twice as fast as a single core can currently process (AMD or Intel) wake me up.

Single core is dead.

Total machine throughput is what counts.

 

[firewolfsm]

Originally posted by: Phynaz

Originally posted by: Pederv
When each core can process twice as fast as a single core can currently process (AMD or Intel) wake me up.

Single core is dead.

Total machine throughput is what counts.

He means when performance per core is doubled.

 

[Arkaign]

Originally posted by: firewolfsm

Originally posted by: Phynaz

Originally posted by: Pederv
When each core can process twice as fast as a single core can currently process (AMD or Intel) wake me up.

Single core is dead.

Total machine throughput is what counts.

He means when performance per core is doubled.

I understood him, as my post detailed in full. It's a silly thing to say 'wake me up' when that happens, when doubling system performance is much more interesting.

 

[Stoneburner]

Originally posted by: SexyK

Originally posted by: Stoneburner
I don't understand the hype for penryn, we already know it's not much better than conroe according to anand's preview. As for Nehalem, it could turn out to be unspectacular like barcelona, it could turn out to be a dud like pentium D's.

I don't understand this post. Penryn is the "tick" in Intel's "tick/tock" so no one was expecting major changes to the architecture. The benefits will be SSE4, other minor arch tweaks, more cache, lower power consumption and higher clocks. Nehalem is the "tock" that will bring a complete redesign of the architecture as well as an IMC and CSI to the 45nm process that will be extremely mature by then. Based on Intel's execution since the inception of the "tick/tock" concept and the core and interface improvements that have already been discussed by Intel, there is no reason to think that Nehalem will be anything but a significant leap forward from the Core 2 architecture that's in Conroe and Penryn. Of course only time will tell, but it's hard to deny that Intel's execution has been nearly flawless since the Core 2 introduction.

That's not saying much considering it's been just over a year and they are still on the same architecture. LIke somebody mentioned, it's interesting how barcelona is considered a failure in the desktop market based on a preview on an incomplete platform but penryn is not despite, as of what we know now, barely performing 5 percent better than conroe.

I'm a bit troubled by intel as they seem to be going off the deep end in a different direction now. FIrst is was the mhz race, now it appears to be a core race. I'm more interested in learning about intel's version of torrenza as my guess would be that it's the future of computing.

 

[Arkaign]

Originally posted by: Stoneburner

Originally posted by: SexyK

Originally posted by: Stoneburner
I don't understand the hype for penryn, we already know it's not much better than conroe according to anand's preview. As for Nehalem, it could turn out to be unspectacular like barcelona, it could turn out to be a dud like pentium D's.

I don't understand this post. Penryn is the "tick" in Intel's "tick/tock" so no one was expecting major changes to the architecture. The benefits will be SSE4, other minor arch tweaks, more cache, lower power consumption and higher clocks. Nehalem is the "tock" that will bring a complete redesign of the architecture as well as an IMC and CSI to the 45nm process that will be extremely mature by then. Based on Intel's execution since the inception of the "tick/tock" concept and the core and interface improvements that have already been discussed by Intel, there is no reason to think that Nehalem will be anything but a significant leap forward from the Core 2 architecture that's in Conroe and Penryn. Of course only time will tell, but it's hard to deny that Intel's execution has been nearly flawless since the Core 2 introduction.

That's not saying much considering it's been just over a year and they are still on the same architecture. LIke somebody mentioned, it's interesting how barcelona is considered a failure in the desktop market based on a preview on an incomplete platform but penryn is not despite, as of what we know now, barely performing 5 percent better than conroe.

I'm a bit troubled by intel as they seem to be going off the deep end in a different direction now. FIrst is was the mhz race, now it appears to be a core race. I'm more interested in learning about intel's version of torrenza as my guess would be that it's the future of computing.

Just over a year and on the same architecture? Please. How long has AMD ridden K8? How long did they ride K7? Architecture stays for a while, a year is NOTHING. Core revisions, a la Conroe to Penryn are usually very incremental, and serve to increase yields, ramping, reduce costs, add minor features, etc. Barcelona is a failure unless a miracle happens because it does NOT equal what is required to be competitive in the desktop market. Even if Phenom is 15% faster than Barcelona clock-for-clock (how doubtful is that? .. look at previous Opteron vs. Athlon 64 IPC), we'd need to see a 3.3Ghz Phenom to be competitive at the time of release (November?). As it stands, K10 is just another utter disaster for AMD, primarily because Conroe was such a dynamo.

I loved seeing Athlon 64 pound P4 into the ground, but reality is reality. For the forseeable future AMD < Intel, until AMD brings K11 out or they file for bankruptcy and get sold off in piece or parcel.

 

[Roy2001]

Originally posted by: Stoneburner
I don't understand the hype for penryn, we already know it's not much better than conroe according to anand's preview. As for Nehalem, it could turn out to be unspectacular like barcelona, it could turn out to be a dud like pentium D's.

In same way, I don't understand all the hype for Barcelona for the past year, just like the sicked 2900XT and never born 2900XTX.

 

[firewolfsm]

Originally posted by: ArkaignBarcelona is a failure unless a miracle happens because it does NOT equal what is required to be competitive in the desktop market. Even if Phenom is 15% faster than Barcelona clock-for-clock (how doubtful is that? .. look at previous Opteron vs. Athlon 64 IPC), we'd need to see a 3.3Ghz Phenom to be competitive at the time of release (November?). As it stands, K10 is just another utter disaster for AMD, primarily because Conroe was such a dynamo.

Check out the latest review on anandtech. The Barcelona is equal or better than Penryn in IPC AND it's more power efficient. Not to mention that Barcelona's power efficiency only improves with higher clock speeds (I'm sure 2.5GHz is more efficient than 2GHz.) Also, in the server market, AMD has a better 4 socket solution. Barcelona is not a failure in any way, especially in the server market. Also realize that Budget and mid-range processors sell much more than the high-end, it's not as important for AMD to ramp up clockspeeds as everyone thinks. Until they do, they have a fast, efficient architecture that can take on Penryn and Conroe.

 

[Stoneburner]

Arkaign, that's my point. As long as we are talking about the same basic architecture, there's no real worry (though prescott was not really new). Intel may have executed well with conroe (and penryn is still basically conroe with some bells attached) that doesn't mean nehalem will be successful. K10 v. k8 proves the point.

 

[Phynaz]

Word is Intel is showing 3.85Ghz Penryns in the demo area.

Normal cooling, normal voltage, not cherry picked.

 

[Roy2001]

Originally posted by: Phynaz
Word is Intel is showing 3.85Ghz Penryns in the demo area.

Normal cooling, normal voltage, not cherry picked.

Penryn A0 silicon runs at 3.7Ghz in lab last November.

 

[Pederv]

Originally posted by: Arkaign

Originally posted by: Pederv
When each core can process twice as fast as a single core can currently process (AMD or Intel) wake me up.

Moores law. Slightly skewed by the shift towards multi-core, but pretty accurate. Look at the time between Athlon 64 3000+, and what most would consider a proc that is twice as fast, the X2 6000+. I know you're talking core for core, but that is getting murkier as time goes on. I will say that you'll probably have to continue sleeping for a couple of years. The current best processor for IPC is the C2D, and to double the performance of one of those cores running at 3ghz/1333fsb, you'll need Nehalem, K11, or the like. Much more likely we'll be seeing 16-core 5ghz systems, of which one of the cores might APPROACH a 100% improvement comparing one core to one core.

Seems like you're bitter that AMD is getting pulverized, but you have to remember how incremental performance increases have classically been. When P3 was released, the 450mhz version barely edged out a P2-450. When P4 was released, it barely edged out (lost in many cases) the P3. The Athlon Slot A K7 tied with the P3 Slot 1 Katmai, the Athlon Thunderbird tied with the Coppermine P3, it's just been increment and stalemate for eons.

The only true divergences have been Athlon XP kicking the poo out of P4 Willamette, and Core 2 Duo kicking the crap out of AMD X2. I'm talking performance increases in excess of single digits, which seems to be the norm.

Remember way back before dual cores when we were looking for increased frequency and increased IPC? Just because we're getting 2, 3, 4, 8, 64 cores, does that mean we should settle for the same old IPC or worse?

A couple of years ago, when dual cores came out, everybody was talking about how in a couple of years software will take advantage of the dual cores, well I'm still hearing how in a couple of years we'll see software take advantage of dual or quad cores.

I used to upgrade my system when my performance doubled and it cost less than $300 for the hardware, it just isn't happening as fast as it used to.

Good night

 

[Nemesis 1]

Well its been a long day . Iread most the reviews comparing Harptown cloverton and K5x2

We found out that harptown at same clock avg. about 20% faster than cloverton . There are some good server reviews out there today.
We found out that sli is on Skulltrail .
As I said in other post here. Today would be an eye opener for you all .

But you haven't seen it all yet there is more to come,

ALL and ALL it was a good day for Intel people I would say.

 

[Keysplayr]

Originally posted by: Nemesis 1
Well its been a long day . Iread most the reviews comparing Harptown cloverton and K5x2

We found out that harptown at same clock avg. about 20% faster than cloverton . There are some good server reviews out there today.
We found out that sli is on Skulltrail .
As I said in other post here. Today would be an eye opener for you all .

But you haven't seen it all yet there is more to come,

ALL and ALL it was a good day for Intel people I would say.

Actually, it was just a good day of some answered questions for everyone. Not just "Intel people".

 

[Phynaz]

Originally posted by: keysplayr2003

Originally posted by: Nemesis 1
Well its been a long day . Iread most the reviews comparing Harptown cloverton and K5x2

We found out that harptown at same clock avg. about 20% faster than cloverton . There are some good server reviews out there today.
We found out that sli is on Skulltrail .
As I said in other post here. Today would be an eye opener for you all .

But you haven't seen it all yet there is more to come,

ALL and ALL it was a good day for Intel people I would say.

Actually, it was just a good day of some answered questions for everyone. Not just "Intel people".

QFT

 

[Nemesis 1]

True enough Keys. I was disappointed over the info on Larrabee. But more will come I suppose . But I am not expecting a lot until spring IDF .

Nehalem info was pretty good. 32nm process info was Good.

I was hoping for more desktop stuff on penryn reviews wise but it was still good.

Yes it was a good information day for all. Thanks for the reminder .

 

[pm]

Originally posted by: jones377
And Otellini claims it taped out 3 weeks ago. Just how fast can you rush prototypes through production?

It didn't tape out three weeks ago - the first prototype parts came out of fab three weeks ago.

IDF transcript:
http://download.intel.com/pres...iptOtelliniKeynote.pdf

The section on Nehalem is at the bottom of page 8.

 

[IntelUser2000]

Check out the latest review on anandtech. The Barcelona is equal or better than Penryn in IPC AND it's more power efficient. Not to mention that Barcelona's power efficiency only improves with higher clock speeds (I'm sure 2.5GHz is more efficient than 2GHz.) Also, in the server market, AMD has a better 4 socket solution. Barcelona is not a failure in any way, especially in the server market. Also realize that Budget and mid-range processors sell much more than the high-end, it's not as important for AMD to ramp up clockspeeds as everyone thinks. Until they do, they have a fast, efficient architecture that can take on Penryn and Conroe.

The latest review is not too relevant in desktops. AMD has tremendous advantage in multi-processors because of their Hypertransport interconnect and the integrated memory controller. Significant improvements that Barcelona has over Opteron is essentially nullified in desktops.

Now the Nehalem tape-out. Interesting. It means we can expect Nehalem to be in production stage by 1 year from now. Remember Barcelona taped out at August 2006. Add a couple of months delay in addition to how long it takes from tape out to production, and October/November launch seems right. Nehalem might be pretty close to AMD's 45nm launch. I don't know about how Nehalem will turn out. The team that created Banias/Dothan/Yonah/Conroe executed flawlessly. Nehalem will be done the the same team that designed Pentium 4. Hopefully they don't screw up.

Prescott seemed just an extension to Netburst but internally it revealed some flaws. Instruction latency tests have shown that Prescott doesn't have the double pumped ALU aka Rapid Execution Engine.

 

[geoffry]

One thing I'm interested in is the breakdown of the 20 penryn chips being released on Nov 12th.....the only confirmed desktop chip is the Q9650 which will prob be $999.

Has anyone seen anything on confirmation of slower yorkfields being released alongside the extreme unit?

 

[Nemesis 1]

Of the reviews I read today this one here was the best .

http://www.hardware.info/en-UK...a_vs_Intel_Harpertown/

 

[jones377]

Originally posted by: pm

Originally posted by: jones377
And Otellini claims it taped out 3 weeks ago. Just how fast can you rush prototypes through production?

It didn't tape out three weeks ago - the first prototype parts came out of fab three weeks ago.

IDF transcript:
http://download.intel.com/pres...iptOtelliniKeynote.pdf

The section on Nehalem is at the bottom of page 8.

Thanks, I figured there was something fishy there.

 

[TuxDave]

Originally posted by: IntelUser2000
The team that created Banias/Dothan/Yonah/Conroe executed flawlessly. Nehalem will be done the the same team that designed Pentium 4. Hopefully they don't screw up.

:|

 

[Acanthus]

Originally posted by: Phynaz
When you have 4 dedicated fabs for this kind of thing, pretty quickly I guess!

fixed.