Intel's move to an on-chip northbridge.

[Rock Hydra]

I was reading the Anandtech article about moving the northbridge onto the chip itself. In another article I read that Intel's target TDP is something like 65 watts, but I'm thinking that's just for the core. I know Intel's northbridges get pretty warm. My 856PE NB does. So, I'm thinking that the chip package as a whole could put quite a bit of heat. Mabye as much as a Prescott or maybe something closer to a Northwood?? I'm hoping not. Anybody have thoughts about this?

 

[bunnyfubbles]

If AMD can do it, I think intel can...

 

[stevty2889]

It looks like it's only going to be used in notebook chips, to help save power. If they do it right it could also increase performance, if there is a direct link to the northbridge in the packaging, instead of it having to share the FSB, but still not as fast as AMD's on die. Actualy considering it would now be sharing a much larger heatsink than a chipset would normaly get, it shouldn't have much trouble keeping the northbridge cool.

 

[akugami]

Hrm...I had to re-read it cause I thought I read it wrong. It was on package for the northbridge and cmos voltage regulator and not actually on die. The package that holds the actual CPU is where they put the northbridge. They get some similar benefits to having it on die without as much complexity and probably a little more flexibility than having it on die. The package housing the cpu and northbridge shouldn't heat up too badly because supposedly Intel's new CPU's will output significantly less heat. We won't really know till it comes out.

 

[Lonyo]

Originally posted by: akugami
Hrm...I had to re-read it cause I thought I read it wrong. It was on package for the northbridge and cmos voltage regulator and not actually on die. The package that holds the actual CPU is where they put the northbridge. They get some similar benefits to having it on die without as much complexity and probably a little more flexibility than having it on die. The package housing the cpu and northbridge shouldn't heat up too badly because supposedly Intel's new CPU's will output significantly less heat. We won't really know till it comes out.

Plus with the CMOS regulator on the chip, from the graphs it looks like it will help produce a lot less heat due to being able to switch voltages quicker, so power consumption/heat should drop, in theory.

 

[Braveheart77]

I love to see Intel scientists turning head on again looking for sollutions instead of marketing (the Mhz run).

 

[Leper Messiah]

IMHO, heat isn't the real problem it the layer of extra complexity, and yet another thing that can go wrong. One of the PC's benefits, is that if a GPU fails, you replace it, if your NB fails, you get a new mobo, etc. If you have 3 things in one, all of which could fail, you have to replace all three if only one fails. That gets mighty expensive... :thumbsdown:

 

[Bona Fide]

Originally posted by: Braveheart77
I love to see Intel scientists turning head on again looking for sollutions instead of marketing (the Mhz run).

Amen.

Intel has always had strong chipsets, so why not push that area? I like how they're thinking.

 

[Wahsapa]

Originally posted by: Leper Messiah
IMHO, heat isn't the real problem it the layer of extra complexity, and yet another thing that can go wrong. One of the PC's benefits, is that if a GPU fails, you replace it, if your NB fails, you get a new mobo, etc. If you have 3 things in one, all of which could fail, you have to replace all three if only one fails. That gets mighty expensive... :thumbsdown:

these cpu/nb chips would probably cost the same as regular mobile cpus do now a days, maybe even cheaper.

the same thing can be sad about old cars vs new cars. something breaks on your old chevy rip if off and glue a new one on, but with cars now a days, you hit somebody going 5 mph in the right spot you might be looking at 1500 in damage to replace some plastic.

 

[Furen]

Originally posted by: akugami
Hrm...I had to re-read it cause I thought I read it wrong. It was on package for the northbridge and cmos voltage regulator and not actually on die. The package that holds the actual CPU is where they put the northbridge. They get some similar benefits to having it on die without as much complexity and probably a little more flexibility than having it on die. The package housing the cpu and northbridge shouldn't heat up too badly because supposedly Intel's new CPU's will output significantly less heat. We won't really know till it comes out.

You only get two of the benefits by having the northbridge in the same package: the traces will be shorter from the cpu to the northbridge, though they would be shorter if the CPU die was even closer to the Northbridge die. This will lead to two things: slightly decreased latency (not that much, I'd guess, but anything is better than nothing) and you'll probably be able to clock the FSB higher (since there's less chance of signal noise). AMD's on-die northbridge also clocks it at the CPU' clock-speed, which reduces latency tremendously (much more than just shortening the trace distance), so it wont be even close in this regard. The voltage regulator thing is interesting but it will only work with very-low-power CPU/Northbrige sets, since these things get can get pretty hot at high loads. I'm pretty sure these repackaged cpus will be VERY expensive at first, since they'll probably be targetted at insanely "mobile" platforms, much more so than current ULV pentium Ms. They will never be at a price parity with normal Mobile CPUs (even if they become more mainstream) because packaging will be significantly more expensive and Intel probably doesnt like the idea of giving away free northbridges.

 

[Viditor]

To me, the voltage regulator was the best news for Intel that I saw come out of IDF!
It's a very realistic goal, and a brilliant idea...should help with the heat quite a bit.

 

[Acanthus]

It was inevitable that both Intel and AMD will make these moves eventually. EMI is going to become a great concern as bus speeds ramp up, so you need at least this move, and eventually everything on chip.

 

[Furen]

Originally posted by: Viditor
To me, the voltage regulator was the best news for Intel that I saw come out of IDF!
It's a very realistic goal, and a brilliant idea...should help with the heat quite a bit.

Though in theory it would lower heat and power consumption, I think that the lowered powere consumption is the more useful part of it, since it leads to increased battery life. I will help with managing heat since three of the most heat-producing components would be under the same heatsink but it probably wont help too much in the heat-generation department, since that level of fine voltage control would only be beneficial when facing periods of sporadic loads (which would be easy for the heat-dissipation system to handle, anyway).

Edit: replaced intermittent with sporadic, since intermittence is inherently cyclical.

 

[Acanthus]

This would also decrease the cost of motherboards by a noticable amount.

 

[DAPUNISHER]

Originally posted by: bunnyfubbles
If AMD can do it, I think intel can...

That has certainly been the pattern for some time now; AMD does it, then Intel follows :clock:

 

[Viditor]

Originally posted by: Furen

Originally posted by: Viditor
To me, the voltage regulator was the best news for Intel that I saw come out of IDF!
It's a very realistic goal, and a brilliant idea...should help with the heat quite a bit.

Though in theory it would lower heat and power consumption, I think that the lowered powere consumption is the more useful part of it, since it leads to increased battery life. I will help with managing heat since three of the most heat-producing components would be under the same heatsink but it probably wont help too much in the heat-generation department, since that level of fine voltage control would only be beneficial when facing periods of sporadic loads (which would be easy for the heat-dissipation system to handle, anyway).

Edit: replaced intermittent with sporadic, since intermittence is inherently cyclical.

Good point Furen...perhaps I should have called it "aid for the effects of leakage" to be more accurate.
I must say that I find Intel's power management to be second to none...unfortunately for them, with Netburst this was like putting a bandaid on a shattered femur. However, it should help quite a bit as we get near the end of the decade on their new architecture.

AMD hasn't had to deal with these problems nearly as much yet, and we can only hope that they are able to utilize Intel's advancements in a couple of years when they start to hit their OWN wall...

 

[Furen]

Originally posted by: Acanthus
This would also decrease the cost of motherboards by a noticable amount.

Maybe... It would certainly decrease the cost of the parts but getting all those traces off the cpu would probably make up for it. AMD motherboard should be significantly cheaper as well and they are reaching cheaper prices just now. Also, even if there IS some significant cost reduction--if this is used in ultra-mobile systems--the consumers will not see any of it,it will all go to the manufacturers' or the OEMs' margins, the last of which do need some improvement (IMO).