Intel Details Nehalem uArch Improvements - 256KB L2, 8MB L3 Confirmed

[Movieman]

The stupidest claim of the month.
If native+ IMC suffers from low BW , how does a non-native non-IMC design do ? Crawls ?

Got bored with the rest of your post.Sorry , but it's obviously you had too much free time lately to write such rubbish.

You were doing fine till you decided to add the highlighted sections.
I keep saying this but I think no one's listening:
When you attack the person instead of the argument of that person it is YOUR credibility that is lost, not theirs."
Add in that we are supposed to be dealing as friends here and friends don't or shouldn't talk that way to one another.
Now, back to the discussion..

[Donnie27]

when did I say a 2.4ghz Nehalem will be $1000? I was implying that, assuming overclocking flexibility is gone in the mainstream version, a Nehalem with comparable performance to a oveclocked $300 Penryn (say 3.6ghz) will be much more expensive since you will have to buy the actual ghz you are shooting for.

I should have said $1000 for example, my bad. Either way, 2.4GHz Nehalem will not cost much more the 2.4GHz Penryn. Just because AMD couldn't get much speed out of Phenom, doesn't mean Intel will be stuck there (NOT saying you said that either). I expect Nehalem to start at higher speeds than Kentsfield or Penryn Quads. You'll not need to buy as much GHz though and that's my point. 2.4GHz Nehalem might hang with a 3 to 3.2GHz Penryn. So that's a free 600 to 800MHz overclock without lifting a finger.

I'd be disappointed if there wasn't a performance increase when moving to Nehalem. Then not being able to overclock becomes more of a negative. Again, like Phenom for example If Nehalem were like AMD's latest, then I'd agree with you.

For mixed up and confused folks, here's what I'm talking about.

http://xtreview.com/addcomment-id-45...hitecture.html

POSTER: computer news || AMD TALK ABOUT NEHALEM ARCHITECTURE
DATE:2008-03-19

Randy Allen , AMD Vice President , which heads the company subdivision for servers and work stations production : stated that Intel overtakes AMD in technical solutions sphere , proposing update in nehalem architecture . Barcelona processors already have the divided cache in the third level, and the integrated memory controller since in 2003. The high-speed system bus QPI is analogous to hyperTransport 3.0 witch AMD company uses since 2006. As asserts Randy Allen, in this sense intel do not proposes any update.

The processors shanghai will be released on 45 nm technology, and in 2009 they can find eight cores. The built-in memory controller with DDR- 3 support will be also added .

AMD reproach Intel (idea was first from AMD) the integration of graphic core in processor, in this case the corresponding products from Intel (Havendale) can appear earlier than AMD ( hybrid processors swift ) . In one world a war witch never end....

As we see from posts like KT's, many believe every lie this Spin Meister tells!
Let's see, QPI just like HTT comes from RAMBUS and EV6 from Alpha, not AMD. Ask AMD how much they pay RAMBUS?
Intel proposes faster processors=P
With QPI, AMD's Multi-socket Memory Bandwidth advantage disappears.
Intel has DDR3 now.
Intel is already at 45nm.
L3, they're kidding, right?
So has Randy forgotten Timna?

[Metroid]

Here is a roadmap I saw recently, http://pc.watch.impress.co.jp/docs/2...gai403_03l.gif.

You definitely have a super eye.

The PDF version is a much better text to visualize.

Edit: JPG.

Metroid.

[Donnie27]

when did I say a 2.4ghz Nehalem will be $1000? I was implying that, assuming overclocking flexibility is gone in the mainstream version, a Nehalem with comparable performance to a oveclocked $300 Penryn (say 3.6ghz) will be much more expensive since you will have to buy the actual ghz you are shooting for.

Here is a roadmap I saw recently, http://pc.watch.impress.co.jp/docs/2...gai403_03l.gif. The first Nehalem parts it shows are socket 1336 versions @ 3.2ghz and 3.0ghz, and if you look to the right at the price axis, they fall in the $800-1000 range. The mainstream parts don't show up until Q2 09, and they start with a 2.4ghz part in the $200-400 range. Assuming they don't OC and are 30% faster than Penyn clock for clock (generous assumption for the sake of argument), that still only makes the 2.4 Nehalem equivalent to a 3.2ghz Penryn which you can have right now for $300. Most of us here aren't memory bandwidth limited anyway (Nvidia wake the F up plz) so the IMC might not be enough of a reason to switch over if overclocking is bad.

Many people here OC to set records or to Fold, WCG, etc... They could probably use the memory bandwidth and SMT even if the chip doesn't overclock well. I do it to play games and save money. If the price/performance isn't there on Nehalem then it makes no sense for me to buy one. But if there is still room to OC despite the IMC, QPI, and PCIe on die then I'll be very happy .

I missed your edit! I disagree with it even more. I agree about seeing records but more will be done with less MHz, lower power draw and etc.... that's not bad, that's good and that will sell well. Some current selling processorss can't overclock, are slower even after they overclock, run hotter and still seem to be selling LOL!

[shiznit93]

lol i totally mistook a ? for a 3. Thx for the pdf. Scratch what I said about at 3.0ghz version.

Donnie, my point is not that a 2.4ghz nehalem will cost more than a 2.4ghz pernyn. I'm saying that if a 2.4ghz penryn overclocks to 3.6 but a nehalem doesn't, given equal price i'm not so sure the nehalem is better.

[Donnie27]

We all know Road Maps change are only rough estimates and can and do often change. But even there it shows at least one Bloomfield under $400.

[Donnie27]

lol i totally mistook a ? for a 3. Thx for the pdf. Scratch what I said about at 3.0ghz version.

Donnie, my point is not that a 2.4ghz nehalem will cost more than a 2.4ghz pernyn. I'm saying that if a 2.4ghz penryn overclocks to 3.6 but a nehalem doesn't, given equal price i'm not so sure the nehalem is better.

I got that but it is a pretty good reach to think Nehalem will not overclock at all. Plus with more than 4 threads, Nehalem will kill Penryn. I'm not so sure even at 3.6GHz Penryn run much faster than Nehalem with the same amount of threads. We'll see though, right?

[Metroid]

We all know Road Maps change are only rough estimates and can and do often change. But even there it shows at least one Bloomfield under $400.

Well in 2006 Intel launched only one Conroe based on that price(E6600) and I still have it, now Penryn Q9450 at same price, great performance processors. Hopefully will be the same with Nehalen, sure road maps may change. Probably these performance processors are the best selling chips for Intel.

Metroid.

[Shintai]

We all know Road Maps change are only rough estimates and can and do often change. But even there it shows at least one Bloomfield under $400.

I bet there wont be a 400$ bloomfield unless it is in a server version again Xeon. For desktop I think we only see 999$+ Bloomfields.

[shiznit93]

Yea it was stretch to say that Nehalem won't oc at all, but I can't help thinking that Intel would like to get rid of overclocking once and for all. With IMC, PCIe, and QPI on die now they might have found a way. I'll let smarter people speculate further.

[savantu]

Why do people bother about CPU prices and spread FUD about $1000 per CPU ? Intel's policy is crystal clear.
You have 5 bins , irrespective of the CPU type , features and so on.

So we have Nehalem Bin 1 to 5.

Bin 1 will be ~$180 , bin 2 $266 , bin 3 $316 , bin 4 $530 and bin 5 is the XE at $999.If there are more than 1 XE bin , up the lowest bin to the next step ( bin 1 $266 )

Rest assured that the Nehalem bin at $266 ( whatever frequency it will be ) is going to be faster than the current CPU bin at $266 ( whatever that is , Q6700 / Q9300 )

[Blauhung]

Shiz,

That is kinda how things are going. Basically, low end budget overclocking was an unintended side effect of the current platform architecture. (I feel this is another one of those statements I should include the "I'm not speaking on Intel's behalf here at all, please don't look at it that way")

At this point Intel has a chance to save massive production costs by consolidating much of the silicon that ends up in a platform onto the newer processes that are pulling in more money and filling out the factories running them to cover capitol costs of the capacity that was set in place. The side effect of all this most likely is that those out there who have the knowhow to push a current lower cost mainstream level platform beyond it's spec'ed performance have to make a choice:
1) either move up to higher end/higher cost platforms on the new architecture to get your fix, or...
2) stick with what you have now.
The second option might just require waiting for the mainstream level of performance to rise above the heavily tweaked mainstream systems we have out now.

To the average OEM buying computer user/business... This change results in a large improvement in performance and power usage. It's only this community it hurts, and only those that are looking for the cheap deal that they can run beyond factory specs.

[shiznit93]

Well put, couldnt have said it any better and that was basically what I was getting at. I guess ill have to wait for mainstream (ie low cost) nehalem performance to surpass low cost but overclocked core 2 duo to upgrade. But as you said, it will be a big plus for laptops and oem machines as the lowest common denominator will rise. I've been riding a wave of cheap performance since the Celeron 300a so I can't say I'm totally happy about it but whacha gonna do?

[Blauhung]

Why do people bother about CPU prices and spread FUD about $1000 per CPU ? Intel's policy is crystal clear.
You have 5 bins , irrespective of the CPU type , features and so on.

So we have Nehalem Bin 1 to 5.

Bin 1 will be ~$180 , bin 2 $266 , bin 3 $316 , bin 4 $530 and bin 5 is the XE at $999.If there are more than 1 XE bin , up the lowest bin to the next step ( bin 1 $266 )

Rest assured that the Nehalem bin at $266 ( whatever frequency it will be ) is going to be faster than the current CPU bin at $266 ( whatever that is , Q6700 / Q9300 )

Yes, but also keep in mind that there are 3 different platform architectures at play here, each based on unique silicon coming through the fabs.

Bins 4 and 5 will be Bloomfield with the 3 channel memory and best chance of overclocking support
Bins 2-4 will be Lynnfield with PCI on die and a good chance for limited overclcokability (we hope not, but might happen)
Bins 1 and 2 will be dual core Havendale with the IGP and probably even less of a chance for pushing.

Each bin will be a boost in stock performance over it's counterpart in the last generation.

[Donnie27]

I bet there wont be a 400$ bloomfield unless it is in a server version again Xeon. For desktop I think we only see 999$+ Bloomfields.

I'm saying Intel knows full and well that folks have gotten too use to Conroe type prices, #1! As I said, I think the Top Desktop model will cost 1200 US dollars or higher. Sad part is, the first batch will quickly seel out LOL!

#2. I'm NOT sure, got no proof and could easily be wrong as hell but I don't see Intel helping AMD by jacking the prices up (on all Nehalem models). They know AMD needs them (Intel) to increase prices. Makes no sense for Intel to help them (AMD). All of the other price cutting measures would have been a waste of time, effort and money. That's how Robber Baron Capitalists work. It makes more sense for Intel to have a very WIDE array of products, not a few expensive models.

[Donnie27]

Yea it was stretch to say that Nehalem won't oc at all, but I can't help thinking that Intel would like to get rid of overclocking once and for all. With IMC, PCIe, and QPI on die now they might have found a way. I'll let smarter people speculate further.

OK cool man! I don't think Intel will bother with folks in THIS market. The NO overclocking people at Intel always complained about VAR and shady OEM overclocking and lying about it. Intel didn't get all buddy up with Fugger to squander that good-will gained in this market.

[Donnie27]

Yes, but also keep in mind that there are 3 different platform architectures at play here, each based on unique silicon coming through the fabs.

Bins 4 and 5 will be Bloomfield with the 3 channel memory and best chance of overclocking support
Bins 2-4 will be Lynnfield with PCI on die and a good chance for limited overclcokability (we hope not, but might happen)
Bins 1 and 2 will be dual core Havendale with the IGP and probably even less of a chance for pushing.

Each bin will be a boost in stock performance over it's counterpart in the last generation.

Makes sense!

[Donnie27]

Why do people bother about CPU prices and spread FUD about $1000 per CPU ? Intel's policy is crystal clear.
You have 5 bins , irrespective of the CPU type , features and so on.

So we have Nehalem Bin 1 to 5.

Bin 1 will be ~$180 , bin 2 $266 , bin 3 $316 , bin 4 $530 and bin 5 is the XE at $999.If there are more than 1 XE bin , up the lowest bin to the next step ( bin 1 $266 )

Rest assured that the Nehalem bin at $266 ( whatever frequency it will be ) is going to be faster than the current CPU bin at $266 ( whatever that is , Q6700 / Q9300 )

Where we disagreed though is; Will that low-end $266 Nehalem that's faster than that $266 Penryn still be faster after overclocking both (if you can), what do you think?

[Blauhung]

Where disagreed though is; Will that low-end $266 Nehalem that's faster than that $266 Penryn still be faster after overclocking both (if you can), what do you think?

We won't really have a chance to know the answer to that question till the motherboard makers get samples.

[Donnie27]

We won't really have a chance to know the answer to that question till the motherboard makers get samples.

Yea, I know you're right. Some leaks would be nice uh?

[onewingedangel]

http://arstechnica.com/news.ars/post...e-part-ii.html

"The presentation also includes some details about Intel's 32nm "Gesher" CPU, due out in 2009. In brief, it's 4-8 cores, 4GHz, 7 double-precision FLOPs/cycle (scalar + SSE), 32KB L1 (3 clocks), 512KB L2 (9 clocks), and 2-3MB L3 (33 clocks). The cores are arranged on a ring bus, just like Larrabee's, that transmits 256 bytes/cycle. Gesher is due out sometime in 2009."

Considering Gesher (sandy bridge) is the nehelam derivitive the cache latencies are likely to be similar for nehelam.

[Blauhung]

Yea, I know you're right. Some leaks would be nice uh?

Hehe, can't help ya there

[Blauhung]

http://arstechnica.com/news.ars/post...e-part-ii.html

"The presentation also includes some details about Intel's 32nm "Gesher" CPU, due out in 2009. In brief, it's 4-8 cores, 4GHz, 7 double-precision FLOPs/cycle (scalar + SSE), 32KB L1 (3 clocks), 512KB L2 (9 clocks), and 2-3MB L3 (33 clocks). The cores are arranged on a ring bus, just like Larrabee's, that transmits 256 bytes/cycle. Gesher is due out sometime in 2009."

Considering Gesher (sandy bridge) is the nehelam derivitive the cache latencies are likely to be similar for nehelam.

Westmere is the nehalem derivative, Sandy Bridge (formerly Gesher), is the whole new architecture.

edit: but i guess for the purposes of your aurgument, the cache latencies are probably similar.

[Donnie27]

Hehe, can't help ya there

As I told someone else here, a guy can wish can't he, hehehehe!

[xlink]

well doing PCI-e 130 is possible
so not all is lost. Let's see.