Nehalem & Larrabee presentations from IDF

[Idontcare]

Updated the links, Intel changed the file links on me sometime today...links work again as of 8/21 7:15PM EST

http://intel.wingateweb.com/US.../SF08_TCHS001_100u.pdf

I did not know that Nehalem was designed to support 3 DIMMs per channel for a total of 9 DIMMs per Nehalem COU. That would be a sweet mobo.

edit: Here's another good presentation from IDF with more details on the power management architecture as well as how the turbo mode can be expected to behave from a low-technical detail consumer viewpoint (good news - all 4 cores can be turbo'ed if the power consumption is still below TDP when fully loaded...see page 126)

http://intel.wingateweb.com/US.../SF08_NGMS001_100t.pdf

edit2: here's the Larrabee presentation - http://intel.wingateweb.com/US.../SF08_VCTS001_100s.pdf

edit3: Intel projects DDR3 will reach price parity with DDR2 by Q4 2009, see slide 7 - DDR3: Moving To Mainstream

Also see slide 12 on MetaSDRAM...16GB DDR3 DIMMs possible. Mention of a 144GB Nehalem system demo.

 

[Tempered81]

Nice find

 

[Idontcare]

Originally posted by: Idontcare
edit2: here's the Larrabee presentation - http://intel.wingateweb.com/US.../SF08_VCTS001_100s.pdf

You guys should check out slide 29 of the larrabee presentation where they show scaling of "non-graphics applications" and then look at the extreme value of the abscissa...64 cores :shocked:

 

[VirtualLarry]

Originally posted by: Idontcare
Also see slide 12 on MetaSDRAM...16GB DDR3 DIMMs possible. Mention of a 144GB Nehalem system demo.

Now that would really be something to see.

I could probably use that much ram too, with a 64-bit newreader and some of these newservers that have insane yearly retention for the binary dvds groups.

 

[solofly]

Originally posted by: Idontcare
http://intel.wingateweb.com/US.../SF08_TCHS001_100t.pdf

I did not know that Nehalem was designed to support 3 DIMMs per channel for a total of 9 DIMMs per Nehalem COU. That would be a sweet mobo.

edit: Here's another good presentation from IDF with more details on the power management architecture as well as how the turbo mode can be expected to behave from a low-technical detail consumer viewpoint (good news - all 4 cores can be turbo'ed if the power consumption is still below TDP when fully loaded...see page 126)

http://intel.wingateweb.com/US.../SF08_NGMS001_100s.pdf

edit2: here's the Larrabee presentation - http://intel.wingateweb.com/US.../SF08_VCTS001_100s.pdf

edit3: Intel projects DDR3 will reach price parity with DDR2 by Q4 2009, see slide 7 - DDR3: Moving To Mainstream

Also see slide 12 on MetaSDRAM...16GB DDR3 DIMMs possible. Mention of a 144GB Nehalem system demo.

DDR3 isn't that much more than DDR2, today!

 

[StinkyPinky]

Originally posted by: solofly

Originally posted by: Idontcare
http://intel.wingateweb.com/US.../SF08_TCHS001_100t.pdf

I did not know that Nehalem was designed to support 3 DIMMs per channel for a total of 9 DIMMs per Nehalem COU. That would be a sweet mobo.

edit: Here's another good presentation from IDF with more details on the power management architecture as well as how the turbo mode can be expected to behave from a low-technical detail consumer viewpoint (good news - all 4 cores can be turbo'ed if the power consumption is still below TDP when fully loaded...see page 126)

http://intel.wingateweb.com/US.../SF08_NGMS001_100s.pdf

edit2: here's the Larrabee presentation - http://intel.wingateweb.com/US.../SF08_VCTS001_100s.pdf

edit3: Intel projects DDR3 will reach price parity with DDR2 by Q4 2009, see slide 7 - DDR3: Moving To Mainstream

Also see slide 12 on MetaSDRAM...16GB DDR3 DIMMs possible. Mention of a 144GB Nehalem system demo.

DDR3 isn't that much more than DDR2, today!

We're going to have some seriously powerful home pcs in a few years. SSD drives will become cheaper and affordable, multi-core cpus will be the norm, large amounts of fast ram, and strong performing graphics cards.

 

[aigomorla]

Originally posted by: StinkyPinky

We're going to have some seriously powerful home pcs in a few years.

wait 4.00ghz+ Quadcore Yorkie = not seriously powerful?
Cuz a lot of people have had these for a while now.

i wish i had a faster io sector. :\

 

[hans007]

i was at IDF today. I did see a nehalem encode a video about twice as fast as a kentsfield machine. It had to be at least twice at fast.

 

[Idontcare]

Originally posted by: hans007
i was at IDF today. I did see a nehalem encode a video about twice as fast as a kentsfield machine. It had to be at least twice at fast.

Pretty sweet.

What's the low-down on this metaram? Server stuff only or will we see it for Bloomfield systems on desktop?

 

[hans007]

Originally posted by: Idontcare

Originally posted by: hans007
i was at IDF today. I did see a nehalem encode a video about twice as fast as a kentsfield machine. It had to be at least twice at fast.

Pretty sweet.

What's the low-down on this metaram? Server stuff only or will we see it for Bloomfield systems on desktop?

i think the metaram was really for just servers. I only walked around the floor with all the hardware stuff for like an hour.

there was this one pc gaming thing going on (with what looked like quake 4 ) but i think for the most part the show was more about servers and atom and embedded.

ididnt get a chance to go to like every single booth since i had to sit in all these presentations most of the day.

 

[BRDiger]

hmm, "Extreme SKU has overspeed protection removed for overclocking"... http://www.computerbase.de/bild/news/20525/9/
Seems like overclocking nehalem is going to be expensiv, if it´s really limited to the EEs

 

[PliotronX]

Originally posted by: BRDiger
hmm, "Extreme SKU has overspeed protection removed for overclocking"... http://www.computerbase.de/bild/news/20525/9/
Seems like overclocking nehalem is going to be expensiv, if it´s really limited to the EEs

Let's just hope "overspeed protection" is the same as it has been with C2; merely limiting the multiplier. If the only other means of overclocking the CPU is limited in Core i7 since the memory controller is now local, it'd be a pretty disappointing turn of affairs.

 

[Nemesis 1]

Ya . I have heard alot about O/C. Will it won't it. The 3 cpus being released for X58 will all O/C . The 2.66 is $284+- . Can't say about the others. But this price for enthusiast is not bad.

Intel FSB today can do 400 easily DDR3 1600. Intel is using 133. I see lots of head room . This is Intel . Clearly in the lead . Give good performance and stability first. But have overhead for the future. For 3 years this is what we have seen from Intel.

Until Intel proves to all otherwise. The smart money would say these are O/Cing beast.


Its interesting how turbo boost works. From a enthisiast point of view. I lot has been said the multi. be locked up. Because of turbo boost we can't think like that . Because turbo boost goes up to the next multi. To increase MGz . It hasn't yet been revealed how man multi.up it goes . That I know about. So can't say. But this does mean . a 2.66 has multi unlock up working. How this relates to O/C has yet been seen.

 

[MarcVenice]

No, overclocking will NOT be limited to EEs only. The overspeed protection is some kind of turbo mode. I've read it in Dutch, and have no link to it in English. But this is roughly what happens with the turbomode. If there is a single threaded code running, 1 or 2 cores can get disabled, a special chip will monitor the heat output of the chipset, and if it's within limits, it will SPEED UP the single threaded code being run on the remaining active core. This will happen in steps of 133mhz, to a max of 266 on normal nehalem cpu's. The EE's though have no such limit, and could virtually go up to let's say 5ghz, on a single core. In fact, the EE's are pretty sweet, because you can 'tell' the CPU that it can keep speeding itself up, as long as it doesn't exceed X amount of heat output, so let's say you watercooling setup can handle 200w of heat, you can put the treshold at 200w.

Now, as for 'normal' overclocking, like we do now with the FSB, we will be able to mess with the speed of the Quickpath. I don't know exactly what it is, but it has something to do with the clockgenerator, and we will be able to manually increase it. It's said that there was a 3.9ghz nehalem at the IDF, and intel expects nehalem to overclock roughly the same as conroe/penryn.

 

[nevbie]

Originally posted by: Nemesis 1
Intel FSB today can do 400 easily DDR3 1600. Intel is using 133. I see lots of head room .

But you are comparing 400FSB to 133not-FSB...

If you think about the limits, it's about the hardware, as the numbers could represent anything.

 

[Nemesis 1]

Na what I am thinking is bumping the nehalem cpu up to 200 is a no brainer.

 

[Idontcare]

Originally posted by: Nemesis 1
Na what I am thinking is bumping the nehalem cpu up to 200 is a no brainer.

That would require a substantial amount of stability margin to have been built into every component in the process involved in the bus.

200/133 = 50% overclock.

Imagine if someone told you that it should be a no-brainer to take a long-existing well understood and implemented FSB system like the current 400MHz FSB and make it run at 600MHz...you'd be laughed out of town for being a crazy.

A 50% overclock is at the very edge of the limits of pushing a well understood, well matured technology like the quad-pumped FSB.

It will hardly be a no brainer to make such a miracle occur on first generation implementation of Nehalem's reference bus...not just because clocking the bus 50% higher will be a challenge but because of ALL the other components that you are expecting to function with the 50% higher clocked reference clock.

(note I am using "bus" and "reference clock" interchangably in very bad form here but until I purge FSB vernacular from my brain it will always be happening, clearly with Nehalem we are always referring to the reference clock when we speak of any bus other than QPI)

 

[seanp789]

An interesting note on Nehalem overclocking in Cinebench R10.

Nehalem @ unknown overclock scored 45864 (IDF 64-bit)
Core 2 Quad X9770 @ 5.367 Ghz scored 24067 (64-bit)
Nehalem @ 2.93Ghz scored 16445 (Hexus.Net 64-bit)

This implies an overclock of the IDF system close to 8 Ghz. This is a crazy number that I'm having a hard time believing in. What am I missing? Was the IDF machine using 8 cores or 4?
Hexus.net claims a 26% advantage to Nehalem clock for clock. LINK

*EDIT* Some talk about the IDF demo running on the CineBench 11 engine which is apparently 2x as fast as the R10. This would put the Nehalem at a more reasonable 4Ghz+ OC. LINK

 

[Nemesis 1]

You need to reread the material. XT has good thread going on this . All sorts of trickery going on say some.

 

[Nemesis 1]

Originally posted by: Idontcare

Originally posted by: Nemesis 1
Na what I am thinking is bumping the nehalem cpu up to 200 is a no brainer.

That would require a substantial amount of stability margin to have been built into every component in the process involved in the bus.

200/133 = 50% overclock.

Imagine if someone told you that it should be a no-brainer to take a long-existing well understood and implemented FSB system like the current 400MHz FSB and make it run at 600MHz...you'd be laughed out of town for being a crazy.

A 50% overclock is at the very edge of the limits of pushing a well understood, well matured technology like the quad-pumped FSB.

It will hardly be a no brainer to make such a miracle occur on first generation implementation of Nehalem's reference bus...not just because clocking the bus 50% higher will be a challenge but because of ALL the other components that you are expecting to function with the 50% higher clocked reference clock.

(note I am using "bus" and "reference clock" interchangably in very bad form here but until I purge FSB vernacular from my brain it will always be happening, clearly with Nehalem we are always referring to the reference clock when we speak of any bus other than QPI)

A qoute from who. Believe him or not I don't care.



Today, 06:48 AM #137
Drwho?
Registered User




Join Date: Dec 2007
Posts: 52 Quote:
Originally Posted by bowman
So now both of you guys are here - are the mainstream chips (non-XE Bloomfield) worth bothering with for overclocking? Will they clock well at all? You're not going to tell, though, are you..


The main clock (it goes by increment of 133Mhz) can be changed ... so,
Asus and co will make it a feature, iam sure.

Thinking that it would have been impossible to OC is an insult to the skills of the OC cimmunity.

The multipliers are all indepentant from each other, no Phenom like issues!

who?

The thread at XT just got a little hot. Who is pissed. But From what he is saying . turbo mode is going to be great. If it works as he says . It sure looks to be.

 

[Idontcare]

Originally posted by: Nemesis 1

Originally posted by: Idontcare

Originally posted by: Nemesis 1
Na what I am thinking is bumping the nehalem cpu up to 200 is a no brainer.

That would require a substantial amount of stability margin to have been built into every component in the process involved in the bus.

200/133 = 50% overclock.

Imagine if someone told you that it should be a no-brainer to take a long-existing well understood and implemented FSB system like the current 400MHz FSB and make it run at 600MHz...you'd be laughed out of town for being a crazy.

A 50% overclock is at the very edge of the limits of pushing a well understood, well matured technology like the quad-pumped FSB.

It will hardly be a no brainer to make such a miracle occur on first generation implementation of Nehalem's reference bus...not just because clocking the bus 50% higher will be a challenge but because of ALL the other components that you are expecting to function with the 50% higher clocked reference clock.

(note I am using "bus" and "reference clock" interchangably in very bad form here but until I purge FSB vernacular from my brain it will always be happening, clearly with Nehalem we are always referring to the reference clock when we speak of any bus other than QPI)

A qoute from who. Believe him or not I don't care.



Today, 06:48 AM #137
Drwho?
Registered User




Join Date: Dec 2007
Posts: 52 Quote:
Originally Posted by bowman
So now both of you guys are here - are the mainstream chips (non-XE Bloomfield) worth bothering with for overclocking? Will they clock well at all? You're not going to tell, though, are you..


The main clock (it goes by increment of 133Mhz) can be changed ... so,
Asus and co will make it a feature, iam sure.

Thinking that it would have been impossible to OC is an insult to the skills of the OC cimmunity.

The multipliers are all indepentant from each other, no Phenom like issues!

who?

The thread at XT just got a little hot. Who is pissed. But From what he is saying . turbo mode is going to be great. If it works as he says . It sure looks to be.

Why wouldn't I believe Francois?

What I don't understand is how what Francois stated in his post (that you pasted above) is meant by you to support your post stating that a 50% overclock of the reference clock (133->200) was a "no brainer".

If you can explain that it would be appreciated.

 

[Nemesis 1]

It is not unrational of me to believe that Intel can run the Cpu at 200mgz. Wheather or not the cpu will run at that speed is another thing all together. But I believe the headroom is there. depending on multi. I am sure lock down is open.

As for the other systems they have there own clocks.