Intel Clarkdale previewed

[Idontcare]

Originally posted by: MODEL3

Originally posted by: Nemesis 1
I like this little info vid from intel its simple enough , Well done for what it intails.

http://www.youtube.com/watch?v...RROZmQ&feature=channel

Seems like a nice guy.


From Youtube comments:

Smeddog1: "ok,? is it me or is that a huge ass cpu???"

breadwithkorv : "lol thats not just one cpu thats a silcon waffer with more than 100 cpu chips on it? ".

LOL

Tha is funny, but the concept behind Smeddog1's thinking is not without grounds.

It's called wafer-scale integration and the approach is about as old as the DOE and DOD (government entities with the wealth to pay what it takes to acquire such products, as well as having the need for them). Top500 is for publicly disclosed supers, not what the NSA keeps tucked away.

Originally posted by: MODEL3
Let's be more clear about what i ask.
I ask if someone knows the GB/s range.

Lynnfield case:
The memory controller is within the CPU die.
With this kind of connection what is the communication speed?

Like this? (from above)

Originally posted by: IntelUser2000
Lynnfield: 16-18GB/s
Clarkdale: 10-11GB/s
Penryn: 6-7GB/s

 

[MODEL3]

Originally posted by: Idontcare
Tha is funny, but the concept behind Smeddog1's thinking is not without grounds.

It's called wafer-scale integration and the approach is about as old as the DOE and DOD (government entities with the wealth to pay what it takes to acquire such products, as well as having the need for them). Top500 is for publicly disclosed supers, not what the NSA keeps tucked away.

The yields must be great.
lol

Originally posted by: Idontcare
Like this? (from above)

Originally posted by: IntelUser2000
Lynnfield: 16-18GB/s
Clarkdale: 10-11GB/s
Penryn: 6-7GB/s

Yes i saw the numbers, that's why i asked IntelUser2000 to clarify.
For example:
Is the Lynnfield: 16-18GB/s number, on each direction?
Also how the numbers are calculated?
for example the 25,6GB/s QPI speed is calculated like this according to a nice Wikipedia link:

3.2 GHz
x 2 bits/Hz (double data rate)
x 20 (QPI link width)
x (64/80) (data bits/flit bits)
x 2 (two links to achieve bidirectionality)
/ 8 (bits/byte)
= 25.6 GB/s

 

[Idontcare]

Originally posted by: MODEL3

Originally posted by: Idontcare
Tha is funny, but the concept behind Smeddog1's thinking is not without grounds.

It's called wafer-scale integration and the approach is about as old as the DOE and DOD (government entities with the wealth to pay what it takes to acquire such products, as well as having the need for them). Top500 is for publicly disclosed supers, not what the NSA keeps tucked away.

The yields must be great.
lol

They are if you DFM. If you don't, well then you are probably already out of business anyways regardless the size of your IC.

 

[IntelUser2000]

Originally posted by: MODEL3

Originally posted by: Idontcare
Like this? (from above)

Originally posted by: IntelUser2000
Lynnfield: 16-18GB/s
Clarkdale: 10-11GB/s
Penryn: 6-7GB/s

Yes i saw the numbers, that's why i asked IntelUser2000 to clarify.
For example:
Is the Lynnfield: 16-18GB/s number, on each direction?
Also how the numbers are calculated?
for example the 25,6GB/s QPI speed is calculated like this according to a nice Wikipedia link:

3.2 GHz
x 2 bits/Hz (double data rate)
x 20 (QPI link width)
x (64/80) (data bits/flit bits)
x 2 (two links to achieve bidirectionality)
/ 8 (bits/byte)
= 25.6 GB/s

I'm sorry. No, those are the measured memory bandwidth numbers from synthetic benchmarks. Again, sorry for confusion.

The speed of the QPI between the CPU die and the GMCH die in Clarkdale is exactly same as the current highest QPI link speed or 6.4GT/s. Yea, you are right, 12.8GB/s in each direction or 25.6GB/s bi-directional. At least according to early diagrams anyway.

 

[Nemesis 1]

Did ya read the small mini review AT did . Notice how the reviewer is A little unsure of himself about just how good Clarkdale is . Its better than he thinks . I know . This chip is going to rock everyone. Why no 32nm 4 cores announced? Because Clarkdale is that good and Intel knows it . This thread stays on front page I will find a way to be sure of that.

 

[IntelUser2000]

Yea, imagine if Clarkdale reaches QX9550/9650 performance. Maybe the reason to name "i5" for both Lynnfield without HT and Clarkdale has some merit.

 

[Idontcare]

Originally posted by: Nemesis 1
Because Clarkdale is that good and Intel knows it .

Is it "that good" or is it simply that it is "good enough"?

IMO 2C/4T is simply "good enough" for the mainstream market that clarkdale is targeting. The clockspeeds are there to drive single-threaded performance where it needs to be, and the multi-threading is there for those few occasions where the person who buys a mainstream computer might just so happen to use an app or two that really challenges a multi-threaded rig.

Basically anyone who knows 2C/4T isn't enough firepower for them is going to already know they want gulftown or thuban anyways...so why try and carry a cross-over 4C sku that is way too many horses for the mainstream crowd but too few for the power-user?

It makes sense to me that they would want to delineate their market segments with a 2C vs. 6C chasm.

Question is - if sandy is 4C/8T then presumably such a part would be viewed as a step-up for the mainstream 2C/4T crowd...versus a step-down for the 6C power-crowd...so question is what does Sandy hold for the power-user crowd? 8C/16T? Even more?

Just as Gulftown doesn't have the MCM'ed IGP that clarkdale does, it is hard for me to convince myself that SandyBridge for enthusiast platform as well as enterprise will retain the on-die IGP that the mainstream part will have. Complete waste of silicon and money.

So they remove the IGP and put more cores/cache in its place, seems logical, but how many and how much?

 

[ilkhan]

Im thinking 4c/IGP for s1156 and 8c/no-IGP for s1366. Or equivalent sockets, anyway. I still doubt they are going to merge the sockets, Aside from a few enthusiasts bitching there's no reason to.

 

[Idontcare]

Originally posted by: ilkhan
Im thinking 4c/IGP for s1156 and 8c/no-IGP for s1366. Or equivalent sockets, anyway. I still doubt they are going to merge the sockets, Aside from a few enthusiasts bitching there's no reason to.

So long as the enthusiast socket is basically the XEON 1S/2S socket the answer there is obvious as Intel will always have a XEON socket that is not the same as the mainstream socket.

The two-socket deal (1156 vs 1366) in the consumer realm has little to do with splitting up the demographics or artificially bifurcating the consumer segments...Intel just transitioned to using the XEON socket and the XEON chips themselves as their enthusiast chips in a top-down approach rather than continue the prior path of building enthusiast parts from the mainstream parts.

 

[dmens]

memory <-> memory controller <-> QPI <-> CPU

i think the memory controller talks directly to the CPU via queues without going though QPI.

==================================

for those who are interested in using this platform as a HTPC (and i think there's a lot of people who are), i was assured by an internal source that the graphics portion of clarksdale will be able to handle 1080p + high def audio via HDMI output to a receiver. pair that with a cablecard on windows 7 and it should be a kickass HTPC. im going to start building one soon with a freebie CPU, ill try to post a mini-review without breaking NDA.

preliminary details:
- top-bin 2C+graphics westmere, downvolt/downclock so I can use a passive heatsink
- micro-ATX board (dunno which brand, ill go with something nice) and 4GB RAM
- 80GB SSD for boot, quiet 1TB for recording, no internal optical drive
- 250W power, maybe even lower. anything to keep it quiet.

given the power usage ive seen from westmere, i think this entire setup can be cooled by a single rear 120mm fan.

 

[Idontcare]

Originally posted by: dmens

memory <-> memory controller <-> QPI <-> CPU

i think the memory controller talks directly to the CPU via queues without going though QPI.

On clarkdale? The mem controller is off-die, its over with the GPU on Ironlake.

How else would Ironlake shunt IO to Clarkdale cpu if not thru the IO provided by QPI? Is there another IO bus at play there?

Or are you thinking Lynnfield maybe?

 

[alyarb]

gulftown is such a non-issue if it's going to be EE only. for the cost of one gulftown CPU months from now you could build a dual socket istanbul machine today.

 

[MODEL3]

Originally posted by: dmens
- micro-ATX board (dunno which brand, ill go with something nice) and 4GB RAM

You mean H57 motherboard? (with P55 you can't connect the CPU (with the on-package IGP), to a HDMI port (or VGA, or DVI).

if you mean H57, then you can find many sample H57 motherboards? (man, you have a nice job)

I saw some H57 motherboards photos from the Computex show, look a strange one from Shuttle:

http://www.tcmagazine.com/imag...uttle_H57_board_01.jpg

 

[IntelUser2000]

Originally posted by: Idontcare
Question is - if sandy is 4C/8T then presumably such a part would be viewed as a step-up for the mainstream 2C/4T crowd...versus a step-down for the 6C power-crowd...so question is what does Sandy hold for the power-user crowd? 8C/16T? Even more?

Just as Gulftown doesn't have the MCM'ed IGP that clarkdale does, it is hard for me to convince myself that SandyBridge for enthusiast platform as well as enterprise will retain the on-die IGP that the mainstream part will have. Complete waste of silicon and money.

So they remove the IGP and put more cores/cache in its place, seems logical, but how many and how much?

I don't think the core counts are increasing in Sandy Bridge. Probably in the extreme high end like the Westmere-EX successor, but I doubt for EP and consumer CPUs. Meaning it would stay 6 cores for desktop high end.

The reason is they still see dual cores profitable and will be a big part of the revenue. May seem a bit disappointing, but take a look at Gulftown pre-release benchmark. How many programs gain from 50% more cores? Even the ones that were thought to be "well-threaded" don't gain from it. The numbers are getting very small here.

And they just went to 6 cores. 6 core Sandy Bridge for high end desktop and EP version should end up similar in die size to the 4 core with IGP, at 220-230mm2.

Regarding Gulftown:

The current rumors are there will be at least two versions-

EE: 2.4GHz, 2.53GHz Turbo
non-EE: 2.4GHz, 2.66GHz Turbo

I do believe that the base clocks will be higher than 2.4GHz though. Turbo won't be much better than Bloomfield due to the TDP.

 

[Idontcare]

Originally posted by: alyarb
gulftown is such a non-issue if it's going to be EE only. for the cost of one gulftown CPU months from now you could build a dual socket istanbul machine today.

Gultown - as in the consumer/enthusiast labeled cpu - might have just the usual 2 or 3 flavors but since Gulftown is really a derivative of the much broader XEON product line I would expect there to be quite a lot of XEON SKU's with more clockspeeds and features enabled/disabled...and the "W" versions should be 1366 compatible.

 

[alyarb]

that doesn't really help. for a given clock freq, xeons are way, way more expensive than their desktop counterparts, and considering the outrageous cost of dunnington, i would expect a 12-thread CPU to intro at well over $2k.

maybe there will be a 1.86 GHz 12T xeon that will be closer to $1500, but for real performance and more reasonable frequencies, it will be one hell of an expensive chip.

 

[IntelUser2000]

I DOUBT that they'll make one Gulftown model(a EE) for high-end consumer and call it a day. Since the Xeons and the regular models go through different validation process, I would be surprised if they don't introduce 3 like they did with Bloomfield. But even if it isn't 3, there's supposed to be at least 2.

That's one way to sort of bridge the gap for not having 4 core Westmere. A cheaper 6 core! This kinda reminds of the times when people were getting upset that lower end Bloomfields won't be overclockable.

 

[Idontcare]

Originally posted by: IntelUser2000
That's one way to sort of bridge the gap for not having 4 core Westmere. A cheaper 6 core! This kinda reminds of the times when people were getting upset that lower end Bloomfields won't be overclockable.

And Lynnfield too, remember Intel was going to "lock out" the OC'ing community by moving the NB on-die.

Oh the things that we have wasted our time debating and being concerned over...

 

[Nemesis 1]

Man why you guys comparring against Intel 4 core . Intel Vs. Intel not likely.

Ya want to compare Clarksdale to PHII. But for the sake of debate . The 2.33. GHZ Nehalem . Is = by a 3.2 ghz clarkdale. So that would mean a Clarkdale running at. 3.2 beats PHII running at 2.8ghz.

Doesn't matter what you guys think . OEMS will buy the clarkdale in droves . No amount of talk will chamge that . Intel not trying to sell these to us but to OEMs and they have a winner here. Power/performance is an overmatch for PHII.

 

[dmens]

Originally posted by: MODEL3
I saw some H57 motherboards photos from the Computex show, look a strange one from Shuttle:

http://www.tcmagazine.com/imag...uttle_H57_board_01.jpg

whats up with the socket angle? wack.

ill try to grab an intel board for my build.

 

[alyarb]

Originally posted by: Nemesis 1
Man why you guys comparring against Intel 4 core . Intel Vs. Intel not likely.

Ya want to compare Clarksdale to PHII. But for the sake of debate . The 2.33. GHZ Nehalem . Is = by a 3.2 ghz clarkdale. So that would mean a Clarkdale running at. 3.2 beats PHII running at 2.8ghz.

Doesn't matter what you guys think . OEMS will buy the clarkdale in droves . No amount of talk will chamge that . Intel not trying to sell these to us but to OEMs and they have a winner here. Power/performance is an overmatch for PHII.

he's trying to say something in English, but who can translate it into English?

 

[Idontcare]

Originally posted by: dmens

Originally posted by: MODEL3
I saw some H57 motherboards photos from the Computex show, look a strange one from Shuttle:

http://www.tcmagazine.com/imag...uttle_H57_board_01.jpg

whats up with the socket angle? wack.

ill try to grab an intel board for my build.

Somebody's going all brooklyn ghetto on them Manhatten design rules :laugh:

Yeah it does look totally kludged, one can only presume they know what they are doing and the rotation was done to minimize something bad (EMI?) or maximize something good (passive cooling on the components around the socket?).

I recally people having similar "is that for real?" type responses when that R600 rotated chip came out.

 

[alyarb]

heh PCIe and RAM are colinear. don't tell me they're that cheap.

 

[IntelUser2000]

Regarding the rotated sockets, couple of boards did that back in the Nforce 2 and the 865/875 days. The reason they do that, the manufacturers say, is to lower the distance between memory controller and the memory(however so slightly) to improve performance.

But before they did that with the MCH, which kinda looked cool. With the CPU socket it looks downright silly. :laugh:

 

[drizek]

Originally posted by: IntelUser2000
Yea, imagine if Clarkdale reaches QX9550/9650 performance. Maybe the reason to name "i5" for both Lynnfield without HT and Clarkdale has some merit.

I thought Clarkdale was i3?